Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects...

23
Birth Order, Educational Attainment, and Earnings An Investigation Using the PSID Jasmin Kantarevic Stéphane Mechoulan abstract We examine the implications of being early in the birth order, and whether a pattern exists within large families of falling then rising attainment with respect to birth order. Unlike other studies using U.S. data, we go beyond grade for age and look at racial differences. Drawing from OLS and fixed effects estimations, we find that being first-born confers a significant educa- tional advantage that persists when considering earnings; being last-born confers none. These effects are significant for large Black families at the high school level, and for White families of any size at both high school and college levels. I. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine whether being early in the birth order implies a distinct educational and professional advantage, and whether within large families a pattern exists of falling then rising attainment with respect to birth order. The empirical results presented here, drawn from the Panel Study of Income Dynamics (PSID), show that being first-born does confer an advantage, while being last-born confers none. In particular, we stress the importance of controlling for the age of the mother at childbirth. The age of the mother at childbirth is positively cor- related with a child’s education. At the same time, it is mechanically, positively Jasmin Kantarevic is a senior economist at the Ontario Medical Association and a research affilate at the Institute for Labor Studies (IZA). Stéphane Mechoulan is an assistant professor of economics at the University of Toronto. The authors thank Gadi Barlevy, Chris Jepsen, Nancy Qian, Imran Rasul, seminar participants at the University of Toronto and UQAM, and two anonymous referees for valuable comments and suggestions. The data used in this article can be obtained beginning March 2007 through February 2010 from Jasmin Kantarevic, Ontario Medical Association, 525 University Ave, Suite 300, Toronto, Ontario, M5G 2K7 Canada, [email protected]. [Submitted February 2005; accepted December 2005] ISSN 022-166X E-ISSN 1548-8004 © 2006 by the Board of Regents of the University of Wisconsin System THE JOURNAL OF HUMAN RESOURCES XLI 4

Transcript of Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects...

Page 1: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

Birth Order Educational Attainmentand EarningsAn Investigation Using the PSID

Jasmin KantarevicSteacutephane Mechoulan

a b s t r a c t

We examine the implications of being early in the birth order and whethera pattern exists within large families of falling then rising attainment withrespect to birth order Unlike other studies using US data we go beyondgrade for age and look at racial differences Drawing from OLS and fixedeffects estimations we find that being first-born confers a significant educa-tional advantage that persists when considering earnings being last-bornconfers none These effects are significant for large Black families at thehigh school level and for White families of any size at both high school andcollege levels

I Introduction

Whether birth order affects performance has been an open empiricalquestion for decades In this study we examine whether being early in the birth orderimplies a distinct educational and professional advantage and whether within largefamilies a pattern exists of falling then rising attainment with respect to birth order

The empirical results presented here drawn from the Panel Study of IncomeDynamics (PSID) show that being first-born does confer an advantage while beinglast-born confers none In particular we stress the importance of controlling for theage of the mother at childbirth The age of the mother at childbirth is positively cor-related with a childrsquos education At the same time it is mechanically positively

Jasmin Kantarevic is a senior economist at the Ontario Medical Association and a research affilate at theInstitute for Labor Studies (IZA) Steacutephane Mechoulan is an assistant professor of economics at theUniversity of Toronto The authors thank Gadi Barlevy Chris Jepsen Nancy Qian Imran Rasul seminarparticipants at the University of Toronto and UQAM and two anonymous referees for valuable commentsand suggestions The data used in this article can be obtained beginning March 2007 through February2010 from Jasmin Kantarevic Ontario Medical Association 525 University Ave Suite 300 TorontoOntario M5G 2K7 Canada Jasmin_KantarevicOMAORG[Submitted February 2005 accepted December 2005]ISSN 022-166X E-ISSN 1548-8004 copy 2006 by the Board of Regents of the University of Wisconsin System

THE JOURNAL OF HUMAN RESOURCES XLI 4

correlated with a childrsquos birth order The omitted variable bias results in a clear offsetof the birth-order effect and represents a simple yet unrecognized source of modelmisspecification

A causal interpretation of the previous analysis would be premature Total numberof siblings the age of the mother at childbirth and other covariates such as parentaleducation are likely correlated with unobservable socioeconomic characteristics Inparticular the precise causal determination of early motherhood on childrenrsquos aca-demic outcomes has received considerable attention (for example GeronimusKorenman and Hillemeier 1994 Hofferth and Reid 2002 Lopez-Turley 2003) fol-lowing an even larger debate on the consequences of early pregnancy on mothers them-selves1 Yet even if early motherhood does not cause lower educational attainment fora child it is still possible that first-borns perform relatively better conditional on earlymotherhood

It would be very difficult to find compelling instrumental variables for all ourpotentially endogenous regressors Therefore to provide additional credibility to ourresults we use a fixed effects (FE) model which by construction removes variablesthat are constant within a family As such we take care of unobserved family-levelheterogeneity The results on birth order are broadly consistent with our initial ones

The PSID enables us to check whether those patterns vary by ethnicity and whetherthe effect we find is in the higher educational realm where financing matters In partic-ular we investigate whether birth order influences secondary or postsecondary educationWe find that birth-order effects are relatively stronger for White families Furthermoreboth ordinary least squares (OLS) and FE estimations show that the first-born lead isalready revealed at the high school stage2 Yet the exact mechanism through which first-borns appear to be advantaged is not fully identifiable from our data

Lastly the PSID gives us an opportunity to track outcomes over a longer periodthan just school years Therefore as a final check of the robustness of the results weestimate the impact of birth order on hourly earnings The same patterns emerge sothat when we omit the age of the mother at birth we find no effect whereas when weinclude it we find a strong positive influence of birth order on hourly earnings We donot find compelling evidence of differential birth-order effects on earnings betweenWhite and Black families

Our work relates to an active literature in the economics of the family that is fun-damental to our understanding of the intra-household allocation of resources3 Ourresults are consistent with those found by Black Devereux and Salvanes (2005) inNorway Booth and Kee (2005) in the United Kingdom and Conley and Glauber(2004) in the United States Yet unlike Conley and Glauber (2004) we are able to gobeyond grade for age

The Journal of Human Resources756

1 Obviously the age of the mother at childbirth is linked to a number of variables that should affect a childrsquoseducational attainment Younger mothers are more likely to be single have less human capital etc Alsoadverse effects of unplanned motherhood may dissipate over time (Bronars and Grogger 1993)2 Specifically birth-order effects are significant for large Black families at the high school level only andfor White families of any size these effects are significant at both the high school and college level We alsolook at the probability of repeating a grade conditional on high school completion which does not seem sig-nificantly influenced by birth order3 Birdsall (1979) Behrman (1986) Behrman and Taubman (1986) Kessler (1991) We elaborate on someother studies in the paper

Our contribution is perhaps most closely related to the work of Hanushek (1992)who used a sample of school children from low-income Black families in early 1970sIndiana Hanushekrsquos paper advances that while being early in the birth order impliesa distinct advantage it is entirely due to the higher probability of coming from a smallfamily Following Lindert (1977) the paper also highlights within large families adistinct and sizeable pattern of falling and then rising attainment with respect to birthorder to the point that it becomes best to be last-born In contrast to Hanushek oursample is more representative We are thus able to examine longer-run outcomes andwe also look at racial differences Our empirical results first present a close versionof Hanushekrsquos findings before challenging their robustness by introducing age ofmother at birth in the model and running fixed effects estimations

The rest of the paper is organized as follows Section II presents our data SectionIII shows how birth order affects various educational outcomes through different esti-mation strategies Section IV then extends the analysis to earnings Finally Section Vconcludes

II Description of the Data

Our data come from the Childbirth and Adoption History File(CAHF) a special supplemental file of the PSID The CAHF covers eligible people4

living in a PSID family at the time of the interview in any wave from 1985 through2001

The population examined here (henceforth the ldquoindex personsrdquo) consists of allthose for whom the CAHF sample contains records of the childbirth histories of atleast one of their parents The CAHF allows us to compile information on their birthorder and the total number of children that their parent(s) report(s)

The index persons with missing information on their birth order or for whom thenumber of siblings is not ascertained are necessarily excluded from the sample Toensure that all mothers have completed their fertility so that we correctly identify thetotal number of siblings we further limit the sample to those index persons whosemother was older than 44 in the last year she reported

Siblings are defined based on the childbirth histories of mothers5 In addition tothe birth order and the number of siblings of the index persons we have obtainedadditional demographic information on them and their parents from other PSID filesusing the unique individual identifiers that are present in the PSID main and supple-mental files

Notably the PSID suffers from an important attrition bias More educated peopletend to stick with the questionnaire over longer periods of time thus it appears that

4 Eligible persons are defined as heads or wives of any age and other members of the family unit aged12ndash44 at the time of the interview These individuals are asked retrospective questions about their birth andadoption histories at the time of their first interview In each succeeding wave these histories are updated5 The sample shrinks by 30 percent if siblings are defined based on the childbirth histories of fathers Incase both parents report we were able to identify between siblings and half siblings however this distinc-tion did not change any of our results We include a variable expressing whether both parents report in ourregressions

Kantarevic and Mechoulan 757

Tabl

e 1

Des

crip

tive

Sta

tist

ics

Var

iabl

eSt

anda

rd

Inde

x in

divi

dual

Obs

erva

tions

Mea

nD

evia

tion

Min

imum

Max

imum

Yea

rs o

f co

mpl

eted

edu

catio

n8

147

126

22

131

17Pe

rcen

t com

plet

ed h

igh

scho

ol8

147

082

mdash0

1L

og h

ourl

y ea

rnin

gs in

200

13

028

267

071

minus12

599

Age

(in

200

1)8

318

385

78

7625

89Pe

rcen

t mal

e8

318

05

mdash0

1Pe

rcen

t Whi

te8

318

047

mdash0

1N

umbe

r of

sib

lings

831

84

862

722

16Pe

rcen

t firs

t-bo

rn8

318

027

mdash0

1Pe

rcen

t sec

ond-

born

831

80

27mdash

01

Perc

ent t

hird

-bor

n8

318

017

mdash0

1Pe

rcen

t fou

rth-

born

831

80

11mdash

01

Perc

ent fi

fth-

born

831

80

07mdash

01

Info

rmat

ion

on a

ll si

blin

gs8

318

056

mdash0

1B

oth

pare

nts

repo

rt th

e ch

ildbi

rth

831

80

60mdash

01

Fam

ily in

com

eA

ge 1

ndash62

695

136

399

283

508

976

60A

ge 7

ndash14

457

619

993

173

601

173

255

393

Age

1ndash1

43

474

188

5014

026

109

217

848

0M

othe

r co

ntin

uous

ly m

arri

edA

ge 1

ndash68

318

025

mdash0

1A

ge 7

ndash14

831

80

36mdash

01

Age

1ndash1

48

318

021

mdash0

1M

othe

rA

ge a

t bir

th8

292

261

85

8815

48Y

ears

of

com

plet

ed e

duca

tion

810

211

03

302

120

Fath

erA

ge a

t bir

th5

000

293

26

5417

60Y

ears

of

com

plet

ed e

duca

tion

489

211

13

367

117

Incl

udes

ind

ex p

erso

ns w

ho h

ave

at l

east

one

sib

ling

who

are

25

year

s or

old

er i

n 20

01 a

nd w

hose

mot

her

is a

t le

ast

44 y

ears

old

in

the

last

yea

r sh

e re

port

ed T

he n

umbe

r of

dis

tinct

fam

ilies

is

311

2

The Journal of Human Resources758

Kantarevic and Mechoulan 759

education is decreasing over cohorts which is of course untrue according to the USCensus Because a first-born is older than other siblings by definition this alonecould in theory produce a spurious positive impact of being first-born on educationWe have checked that this problem is of no consequence for our results6

The summary statistics of our sample are presented in Table 1 The detaileddescription of variables is relegated to Appendix 1 We found more than 8000 indexpersons (from more than 3100 distinct families) older than 24 in 2001 with at leastone other sibling and whose mother has completed her fertility This is to be con-trasted with Conley and Glauber (2004) who use larger sample from the US Censusbut focus only on children under the age of 20 living at home

The main dependent variablemdashyears of completed educationmdashhas an average of1262 years7 About 82 percent of those selected index persons have at least 12 years ofeducation that is to say have completed high school8 Our measure of earningsmdashloghourly wage in 2001mdashshows an average corresponding to $145hour However theinformation on hourly wages or salary income is not available for most index persons

The average age in our sample is 39 About half of respondents are male and 47percent are White The average number of siblings is 486 This high number is con-sistent with the PSID oversampling minorities and low-income populations Fifty-sixpercent of the index persons have all their siblings reporting and 60 percent have boththeir parents reporting their childbirth history

We improved our analysis by including important observed family level-effects thatvary across parity family income and whether the mother is married Wheneveravailable we constructed the corresponding information for each of the first 14 yearsof life of each index person9 The main limitation is that information on familyincome cannot be recovered for many index individuals

Lastly we include two variables describing characteristics of index individualsrsquo par-ents namely education (11 years for both mother and father on average) and the age ofthe parents at birth of index persons (26 for mothers and 29 for fathers on average)

III Methods and Results

A The First-Born Effect

We first use an OLS estimation with robust standard errors clustered by family unit(identified by the mother) which relaxes the independence assumption between the

6 The regressions presented in this article contain age controls that separate cohort from birth order effects 7 A more appropriate variable would be education at age 25 However the number of observations avail-able would drop considerably and we would not be able to run estimations by sibling size All of our otherresults when running estimations that control for siblings size hold when replacing education with educationat age 25 for those where such information can be traced8 A negligible fraction of those who declare a certain education level at some point in their life declare lesseducation later Removing such observations did not alter our results We therefore choose the latest educa-tion level reported as our variable of interest9 We then ran our regressions using the average of these variables for three age groups 1ndash6 7ndash14 and1ndash14 For each age group the averages are respectively calculated only if the information is present in atleast half of the years of the age group

The Journal of Human Resources760

error terms and requires only that the observations be independent across clusters10

In Table 2 we first test the hypothesis that being early in the birth order implies a dis-tinct advantage that is entirely due to the higher probability of coming from a smallfamily

Columns 1 and 2 of Table 2 reject that claim but help us understand why it mayhave been made In Column 1 we omit the number of siblings therefore the signif-icant coefficient on first-born reflects not only the birth order effect but also the prob-ability of coming from a small family In Column 2 the inclusion of the number ofsiblings leaves the coefficient on first-born insignificant

In Column 3 we include age of the mother at childbirth and find a positive andhighly significant effect of being first-born on years of education This effect is con-firmed when including the fatherrsquos characteristics in Column 4 where both a father anda mother report Often not all siblings report this is especially the case for large fam-ilies To check if we are biasing our results by including such families in Column 5 werestrict our attention to families with complete information on all siblings The findingsare similar there too further showing that they could not be driven by selective attri-tion within families by birth order In all specifications we find a stronger effect amongWhite families

The results presented here are for the impact of being first-born in families of morethan one child This particular procedure takes advantage of the full size of our sam-ple and can be useful when there are not enough observations to run separate estima-tions for different siblingsrsquo sizes11 We see that it allows us to reveal a significant androbust birth order effect

However we still obtain similar results when looking at individual birth ordereffects by siblingsrsquo size in Table 3 where all specifications include age of the motherat childbirth Although large families show higher birth order point estimates theeffect is present in two-sibling families as well Note that the coefficient on first-bornis only weakly significant in Column 4mdashfamilies of five siblingsmdashlikely because ofthe small sample size

The reason why the inclusion of the age of the mother at childbirth makes the coef-ficient on first-born larger in magnitude and more significant is clear The age of themother at childbirth is mechanically positively correlated with birth order and evenmore strongly across large families Conversely we see that it is positively correlatedwith a childrsquos education Then if having a high birth order carries a negative impacton education the two effects of birth order and age of the mother at childbirth com-pete against one another Therefore the coefficient on first-born in Table 2 Column 2reflects an omitted variable bias The results hold for both males and females formothers with or without more than a high school education We also found that spac-ing between births be it that of the first-born child with respect to the second-born orto the last-born child does not alter the conclusions either Additional OLS regres-sions confirmed that the effect is significantly present among White families of any

10 We also used random effects procedures but because they yield almost identical results as those withthe family clustered standard errors those are not reported11 For example we found a significant birth-order effect when considering separately mothers who firstgave birth early and mothers who first gave birth ldquolaterdquo (using various definitions of ldquoearlyrdquo and ldquolaterdquo) yetwe do not have enough observations to split the sample by maternal age and individual siblings size

Kantarevic and Mechoulan 761Ta

ble

2O

LS

Reg

ress

ion

wit

h D

epen

dent

Var

iabl

e E

duca

tion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

016

20

012

028

80

239

026

8(0

068

)

(00

69)

(00

72)

(0

098

)

(00

84)

To

tal n

umbe

r of

sib

lings

minus01

04minus0

121

minus01

08minus0

125

(00

15)

(0

015

)

(00

19)

(0

029

)

Age

of

mot

her

at c

hild

birt

h0

058

006

10

065

(00

05)

(0

011

)

(00

08)

d[

Mal

e]minus0

205

minus02

1minus0

226

minus01

86minus0

159

(00

47)

(0

046

)

(00

46)

(0

058

)

(00

57)

A

ge0

121

015

90

162

014

30

087

(00

27)

(0

027

)

(00

27)

(0

039

)

(00

40)

A

ge2

minus00

01minus0

001

minus00

01minus0

001

minus00

03(0

000

3)

(0

000

3)

(0

000

3)

(0

001

)

(00

001)

d[W

hite

]0

119

minus00

03minus0

099

minus02

95minus0

107

(00

72)

(00

73)

(00

71)

(00

87)

(0

098

)d[

first

-bor

n] times

d[W

hite

]0

126

019

30

184

028

60

172

(00

92)

(00

92)

(0

092

)

(01

15)

(0

104

)M

othe

rrsquos

educ

atio

n0

213

019

80

199

011

40

236

(00

12)

(0

012

)

(00

12)

(0

017

)

(00

18)

d[

all s

iblin

gs r

epor

t]0

261

009

60

181

014

4(0

068

)

(00

69)

(00

69)

(0

089

)d[

both

par

ents

rep

ort]

052

30

553

052

00

533

(00

63)

(0

062

)

(00

61)

(0

081

)

Fath

errsquos

edu

catio

n0

125

(00

12)

A

ge o

f fa

ther

at c

hild

birt

h0

006

(00

09)

Con

stan

t6

713

671

75

073

516

25

848

(05

28)

(05

31)

(0

541

)

(07

75)

(0

776

)

R2

015

780

1691

018

890

2009

019

82N

umbe

r of

obs

erva

tions

792

87

928

792

87

766

454

1N

umbe

r of

fam

ily c

lust

ers

311

23

112

311

21

869

173

2

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources762

Tabl

e 3

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Edu

cati

on

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

027

50

698

071

50

534

091

5(0

104

)

(01

36)

(0

185

)

(03

16)

(02

36)

d[

seco

nd-b

orn]

043

40

463

033

40

578

(01

17)

(0

16)

(02

82)

(01

95)

d[

thir

d-bo

rn]

017

10

212

036

8(0

14)

(02

47)

(01

75)

d[

four

th-b

orn]

minus00

120

466

(02

14)

(01

52)

d[

fifth

-bor

n]0

221

(01

33)

Age

of

the

mot

her

at c

hild

birt

h0

048

007

90

067

006

80

07(0

012

)

(00

13)

(0

013

)

(00

23)

(0

013

)

d[M

ale]

minus02

87minus0

115

minus00

35minus0

34

minus00

33(0

102

)

(00

91)

(01

07)

(01

31)

(0

087

)

Age

021

70

190

156

007

50

167

(00

56)

(0

057

)

(00

49)

(0

044

)(0

057

)

Kantarevic and Mechoulan 763

Age

20

002

000

20

002

3times

10minus4

000

2(7

times10

minus4)

(7

times10

minus4)

(6

times10

minus4)

(5

times10

minus4)

(8times

10minus4

)

d[W

hite

]0

119

minus01

244

times10

minus5minus0

03

minus01

14(0

12)

(01

3)(0

13)

(01

81)

(01

37)

Mot

herrsquo

s ed

ucat

ion

025

70

250

179

012

40

182

(00

25)

(0

026

)

(00

23)

(0

035

)

(00

22)

d[

all s

iblin

gs r

epor

t]0

147

021

2minus0

066

035

10

21(0

154

)(0

136

)(0

128

)(0

178

)

(01

49)

d[bo

th p

aren

ts r

epor

t]0

241

061

20

755

054

20

371

(01

18)

(0

118

)

(01

26)

(0

189

)

(01

27)

C

onst

ant

333

92

694

425

86

607

412

(11

48)

(1

184

)

(10

11)

(1

293

)

(14

15)

R

20

185

021

019

011

40

127

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources764

size greater than one whereas it is only present within large families among BlacksTherefore the ethnic differential disappears when considering large families only

As noted earlier a causal interpretation of the age of the mother at childbirth wouldhinge on the assumption of its exogeneity Without instrumental variables or a treat-ment versus control quasi experiment it is difficult to draw conclusions12 The age ofthe mother at childbirth itself positively correlated with birth order could easilyproxy for other unobserved variables such as level of human capital and parentalresources13

To address this problem our fixed-effect estimation (Table 4a and b) removes fam-ily characteristics and unobserved family-level heterogeneity14 Family fixed effectsaddress family unobservables to the extent that they are constant over time While wetry to incorporate observables that vary across birth order to affirm the robustness ofour results we are constrained by the availability of such variables in the data set

Unfortunately the coefficients on age or on age of the mother at childbirth are unin-formative in those fixed effects regressions Deviations from family means for ageconvey the same information as deviations from family means for age of mother atchildbirth We thus do not provide separate estimations with age and age of motherat childbirth The age of the mother at first childbirth is certainly relevant but here itis differenced out Still to assess this issue better we ran separate regressions split-ting the sample by maternal age15 To summarize the following results are robust toexcluding mothers who first gave birth as teens but we do not have enough observa-tions to meaningfully run the fixed effects regressions by sibling size on that lattergroup

In Tables 4a and 4b the fixed effects estimations that only control for age and gen-der confirm the previous results for the most part16 We also provide estimations includ-ing the marital status of the mother17 during the childrsquos first 14 years18 The results donot change by much19 Clearly the suggestion that first-borns are most likely to live

12 Also while it is possible in the context of larger samples to instrument for siblings sizemdashusing twinbirths (Black Devereux and Salvanes 2005) or using the fact that parents of two same sex children are morelikely to have a third child (Conley and Glauber 2004)mdashone cannot instrument for birth order per se13 This is further evidenced by the fact that including a dummy variable indicating whether the mother wasmarried at the time of childbirth instead of (and obviously also along with) age of mother at childbirth alsoresults in highly significant birth order coefficients Results are available upon request14 However it is worth noting that those fixed effects do not solve all endogeneity issues For example itmay be the case that first-born ldquoqualityrdquo is affecting subsequent fertility We thank an anonymous referee forpointing out this caveat15 We thank an anonymous referee for this suggestion16 In Column 1mdashfamilies of two siblingsmdashthe coefficient on first-born becomes significant at the 10 per-cent level when including index persons of 23 and 24 years of age suggesting the nonsignificance whenrestricting at age 25 stems from a small sample size Separate FE regressions for White and Black familiesconfirmed the ethnic differentials found earlier17 We tried two definitions number of years marriednumber of years considered in the age group and = 1if continuously married over the years considered in the age group 0 otherwise Since the results on birthorder do not change qualitatively with either of those we only report the results with the latter definition18 We also ran similar estimations with different age ranges and obtained similar results19 Because the information on marital status is retrospective we do not lose any observation by includingthis covariate Unfortunately this is not the case when adding average family income or average employmentstatus of the mother during the childrsquos first years

Kantarevic and Mechoulan 765

Table 4aFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Not Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0216 0437 0683 0594 0477(0181) (0206) (0265) (0379) (0224)

d[second- 0282 0412 0421 0199born] (0141) (0214) (0323) (0195)

d[third- 0138 0361 0055born] (0159) (0261) (0172)

d[fourth- minus0031 0181born] (0207) (0149)

d[fifth- minus0012born] (0125)

Observations 1422 1743 1477 922 2487groups 934 835 562 388 573

Table 4bFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0224 0436 0696 0585 0492(0181) (0206) (0267) (0379) (0226)

d[second-born] 028 0423 0413 0212(0141) (0216) (0323) (0197)

d[third-born] 0146 0363 0063(0159) (0261) (0173)

d[fourth-born] minus0025 0188(0207) (0149)

d[fifth-born] minus0016(0125)

Mother minus0196 minus0167 0112 0381 minus0105Continuously (0315) (0221) (0221) (0274) (0178)married aged 1ndash14

Number of 1422 1743 1477 999 2487observations

Number of 934 835 562 328 573groups

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources766

their critical development years in a stable household as opposed to later-borns whomay experience the divorce of their parents cannot entirely explain the first-bornadvantage At the same time the persistence of birth-order effects naturally poses theproblem of their origin

The literature is not able to distinguish between different theories on the topic ofbirth order For example schooling circumstances play a large role in educational out-comes and may be related to birth order20 Our analysis is limited in the sense that itcannot discriminate between many competing hypotheses on why birth order appearsto be important21

Nevertheless we checked whether what appears as a first-born advantage predom-inantly comes from financial constraints for example parents sending their first-bornto college and running out of money for the following siblings Conley offers the fol-lowing argument ldquo[I]n terms of parental investment the cup starts to run dry as wego down the line Parental resources it appears are allotted on a first come first-served basisrdquo22 Yet if it turns out that first-borns perform better beforehand then atheory based on budget constraints cannot fully account for our results In Tables 5ndash7we estimate the probability of completing high school following the same methodol-ogy as in Tables 2ndash4 We find that first-borns have a higher probability of completinghigh school than later born siblings

Specifically Table 5 shows again that the first-born effect is not an artifact of fam-ily size that it increases in magnitude and significance when including age of motherat birth that it is robust to including characteristics of father and to restricting thesample to families in which all siblings report Table 6 shows that these results alsohold when regressions are estimated separately by family size (except for families offive presumably because of the small sample size) Tables 7a and 7b are the counter-parts of Tables 4a and 4b Fixed effects regressions controlling for age and gendersupport the results found in Tables 5 and 623

Also we estimated education at age 18 conditional on high school completion (at18 or older) to see if later-borns are more likely to repeat grades which would sup-port a theory of birth-order effects based on cognitive development differences Wedid not find any evidence for this However the small sample size resulting fromselecting index persons with available information at age 18 warrants some caution

Finally we found evidence among White families only that conditional on com-pleting high school first-borns are more likely to receive postsecondary educationYet for all races conditional on postsecondary education we found no clear advan-tage to being first-born24 In summary financial constraints do seem to play a role butsome factors early in life contribute to the first-born premium puzzle

20 We thank an anonymous referee for this point21 We refer the reader to the survey of those theories presented in Black Devereux and Salvanes (2005)22 Conley (2004) p 6923 The results held when including motherrsquos marital status We ran into the same small size problems whenadding family income and motherrsquos employment status for different age ranges of each child The ethnic dif-ferential disappears when considering all families (but persists when focusing on smaller families) suggest-ing that some of the birth order effect among White families comes from postsecondary education24 Results are available from the authors

Kantarevic and Mechoulan 767

Tabl

e 5

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

004

40

024

005

50

039

004

4(0

014

)

(00

14)

(00

14)

(0

018

)

(00

17)

To

tal n

umbe

r of

0

014

001

60

011

001

2si

blin

gs(0

003

)

(00

03)

(0

004

)

(00

05)

A

ge o

f th

e m

othe

r 0

006

000

70

005

at c

hild

birt

h(0

000

1)

(0

002

)

(00

01)

d[

Mal

e]0

047

004

80

049

004

70

037

(00

09)

(0

009

)

(00

09)

(0

009

)

(00

10)

A

ge0

024

002

90

029

003

20

023

(00

04)

(0

004

)

(00

04)

(0

007

)

(00

07)

A

ge2

000

020

0003

000

020

0003

000

02(5

times10

minus5)

(5

times10

minus5)

(5

times10

minus5)

(8

times10

minus5)

(8

times10

minus5)

d[

Whi

te]

000

50

012

002

30

042

001

9(0

014

)(0

014

)(0

014

)(0

017

)

(00

19)

d[fir

st-b

orn]

timesd

001

5minus0

005

minus00

060

011

000

6[W

hite

](0

017

5)(0

017

)(0

017

)(0

021

)(0

021

)M

othe

rrsquos

educ

atio

n0

027

002

60

026

001

50

026

(00

02)

(0

002

)

(00

02)

(0

003

)

(00

03)

d[

com

plet

e in

fo

003

50

094

002

20

029

on a

ll si

blin

gs]

(00

13)

(0

012

)

(00

13)

(00

15)

d[bo

th p

aren

ts r

epor

t]0

090

094

009

00

097

(00

12)

(0

012

)

(00

12)

(0

016

)

The Journal of Human Resources768

Tabl

e 5

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

Fath

errsquos

edu

catio

n0

011

(00

02)

A

ge o

f fa

ther

minus0

001

at c

hild

birt

h(0

002

)C

onst

ant

010

50

104

028

80

241

012

9(0

089

)(0

089

)(0

094

)

(01

35)

(01

41)

R2

008

790

0943

010

180

0885

008

55N

umbe

r of

obs

erva

tions

792

87

928

792

84

766

454

1N

umbe

r of

fam

ily

311

23

112

311

21

869

173

2cl

uste

rs

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Kantarevic and Mechoulan 769

Tabl

e 6

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

003

80

084

009

003

10

122

(00

18)

(0

025

)

(00

36)

(0

053

)(0

049

)

d[se

cond

-bor

n]0

071

008

2times

10minus4

002

2(0

022

)

(00

31)

(0

051

)(0

044

)d[

thir

d-bo

rn]

003

10

012

003

5(0

028

)(0

041

)(0

037

)d[

four

th-b

orn]

006

10

041

(00

37)

(00

33)

d[fif

th-b

orn]

000

7(0

028

)A

ge o

f th

e m

othe

r at

chi

ldbi

rth

000

30

004

000

90

008

001

1(0

002

)(0

002

)(0

003

)

(00

04)

(0

003

)

d[M

ale]

minus00

3minus0

041

000

7minus0

094

minus00

82(0

016

)(0

016

)

(00

2)(0

026

)

(00

18)

A

ge0

028

002

80

051

001

20

037

(00

07)

(0

009

)

(00

1)

(0

008

)(0

011

)

Age

23

times10

minus43

times10

minus45

times10

minus48

times10

minus54times

10minus4

(9times

10minus5

)

(10minus5

)

(10minus5

)

(9times

10minus5

)(1

0minus4)

d[

Whi

te]

001

20

124

004

90

033

002

2(0

021

)(0

023

)(0

026

)(0

034

)(0

028

)M

othe

rrsquos

educ

atio

n0

024

002

50

027

001

80

027

(00

04)

(0

004

)

(00

05)

(0

005

)

(00

04)

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 2: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

correlated with a childrsquos birth order The omitted variable bias results in a clear offsetof the birth-order effect and represents a simple yet unrecognized source of modelmisspecification

A causal interpretation of the previous analysis would be premature Total numberof siblings the age of the mother at childbirth and other covariates such as parentaleducation are likely correlated with unobservable socioeconomic characteristics Inparticular the precise causal determination of early motherhood on childrenrsquos aca-demic outcomes has received considerable attention (for example GeronimusKorenman and Hillemeier 1994 Hofferth and Reid 2002 Lopez-Turley 2003) fol-lowing an even larger debate on the consequences of early pregnancy on mothers them-selves1 Yet even if early motherhood does not cause lower educational attainment fora child it is still possible that first-borns perform relatively better conditional on earlymotherhood

It would be very difficult to find compelling instrumental variables for all ourpotentially endogenous regressors Therefore to provide additional credibility to ourresults we use a fixed effects (FE) model which by construction removes variablesthat are constant within a family As such we take care of unobserved family-levelheterogeneity The results on birth order are broadly consistent with our initial ones

The PSID enables us to check whether those patterns vary by ethnicity and whetherthe effect we find is in the higher educational realm where financing matters In partic-ular we investigate whether birth order influences secondary or postsecondary educationWe find that birth-order effects are relatively stronger for White families Furthermoreboth ordinary least squares (OLS) and FE estimations show that the first-born lead isalready revealed at the high school stage2 Yet the exact mechanism through which first-borns appear to be advantaged is not fully identifiable from our data

Lastly the PSID gives us an opportunity to track outcomes over a longer periodthan just school years Therefore as a final check of the robustness of the results weestimate the impact of birth order on hourly earnings The same patterns emerge sothat when we omit the age of the mother at birth we find no effect whereas when weinclude it we find a strong positive influence of birth order on hourly earnings We donot find compelling evidence of differential birth-order effects on earnings betweenWhite and Black families

Our work relates to an active literature in the economics of the family that is fun-damental to our understanding of the intra-household allocation of resources3 Ourresults are consistent with those found by Black Devereux and Salvanes (2005) inNorway Booth and Kee (2005) in the United Kingdom and Conley and Glauber(2004) in the United States Yet unlike Conley and Glauber (2004) we are able to gobeyond grade for age

The Journal of Human Resources756

1 Obviously the age of the mother at childbirth is linked to a number of variables that should affect a childrsquoseducational attainment Younger mothers are more likely to be single have less human capital etc Alsoadverse effects of unplanned motherhood may dissipate over time (Bronars and Grogger 1993)2 Specifically birth-order effects are significant for large Black families at the high school level only andfor White families of any size these effects are significant at both the high school and college level We alsolook at the probability of repeating a grade conditional on high school completion which does not seem sig-nificantly influenced by birth order3 Birdsall (1979) Behrman (1986) Behrman and Taubman (1986) Kessler (1991) We elaborate on someother studies in the paper

Our contribution is perhaps most closely related to the work of Hanushek (1992)who used a sample of school children from low-income Black families in early 1970sIndiana Hanushekrsquos paper advances that while being early in the birth order impliesa distinct advantage it is entirely due to the higher probability of coming from a smallfamily Following Lindert (1977) the paper also highlights within large families adistinct and sizeable pattern of falling and then rising attainment with respect to birthorder to the point that it becomes best to be last-born In contrast to Hanushek oursample is more representative We are thus able to examine longer-run outcomes andwe also look at racial differences Our empirical results first present a close versionof Hanushekrsquos findings before challenging their robustness by introducing age ofmother at birth in the model and running fixed effects estimations

The rest of the paper is organized as follows Section II presents our data SectionIII shows how birth order affects various educational outcomes through different esti-mation strategies Section IV then extends the analysis to earnings Finally Section Vconcludes

II Description of the Data

Our data come from the Childbirth and Adoption History File(CAHF) a special supplemental file of the PSID The CAHF covers eligible people4

living in a PSID family at the time of the interview in any wave from 1985 through2001

The population examined here (henceforth the ldquoindex personsrdquo) consists of allthose for whom the CAHF sample contains records of the childbirth histories of atleast one of their parents The CAHF allows us to compile information on their birthorder and the total number of children that their parent(s) report(s)

The index persons with missing information on their birth order or for whom thenumber of siblings is not ascertained are necessarily excluded from the sample Toensure that all mothers have completed their fertility so that we correctly identify thetotal number of siblings we further limit the sample to those index persons whosemother was older than 44 in the last year she reported

Siblings are defined based on the childbirth histories of mothers5 In addition tothe birth order and the number of siblings of the index persons we have obtainedadditional demographic information on them and their parents from other PSID filesusing the unique individual identifiers that are present in the PSID main and supple-mental files

Notably the PSID suffers from an important attrition bias More educated peopletend to stick with the questionnaire over longer periods of time thus it appears that

4 Eligible persons are defined as heads or wives of any age and other members of the family unit aged12ndash44 at the time of the interview These individuals are asked retrospective questions about their birth andadoption histories at the time of their first interview In each succeeding wave these histories are updated5 The sample shrinks by 30 percent if siblings are defined based on the childbirth histories of fathers Incase both parents report we were able to identify between siblings and half siblings however this distinc-tion did not change any of our results We include a variable expressing whether both parents report in ourregressions

Kantarevic and Mechoulan 757

Tabl

e 1

Des

crip

tive

Sta

tist

ics

Var

iabl

eSt

anda

rd

Inde

x in

divi

dual

Obs

erva

tions

Mea

nD

evia

tion

Min

imum

Max

imum

Yea

rs o

f co

mpl

eted

edu

catio

n8

147

126

22

131

17Pe

rcen

t com

plet

ed h

igh

scho

ol8

147

082

mdash0

1L

og h

ourl

y ea

rnin

gs in

200

13

028

267

071

minus12

599

Age

(in

200

1)8

318

385

78

7625

89Pe

rcen

t mal

e8

318

05

mdash0

1Pe

rcen

t Whi

te8

318

047

mdash0

1N

umbe

r of

sib

lings

831

84

862

722

16Pe

rcen

t firs

t-bo

rn8

318

027

mdash0

1Pe

rcen

t sec

ond-

born

831

80

27mdash

01

Perc

ent t

hird

-bor

n8

318

017

mdash0

1Pe

rcen

t fou

rth-

born

831

80

11mdash

01

Perc

ent fi

fth-

born

831

80

07mdash

01

Info

rmat

ion

on a

ll si

blin

gs8

318

056

mdash0

1B

oth

pare

nts

repo

rt th

e ch

ildbi

rth

831

80

60mdash

01

Fam

ily in

com

eA

ge 1

ndash62

695

136

399

283

508

976

60A

ge 7

ndash14

457

619

993

173

601

173

255

393

Age

1ndash1

43

474

188

5014

026

109

217

848

0M

othe

r co

ntin

uous

ly m

arri

edA

ge 1

ndash68

318

025

mdash0

1A

ge 7

ndash14

831

80

36mdash

01

Age

1ndash1

48

318

021

mdash0

1M

othe

rA

ge a

t bir

th8

292

261

85

8815

48Y

ears

of

com

plet

ed e

duca

tion

810

211

03

302

120

Fath

erA

ge a

t bir

th5

000

293

26

5417

60Y

ears

of

com

plet

ed e

duca

tion

489

211

13

367

117

Incl

udes

ind

ex p

erso

ns w

ho h

ave

at l

east

one

sib

ling

who

are

25

year

s or

old

er i

n 20

01 a

nd w

hose

mot

her

is a

t le

ast

44 y

ears

old

in

the

last

yea

r sh

e re

port

ed T

he n

umbe

r of

dis

tinct

fam

ilies

is

311

2

The Journal of Human Resources758

Kantarevic and Mechoulan 759

education is decreasing over cohorts which is of course untrue according to the USCensus Because a first-born is older than other siblings by definition this alonecould in theory produce a spurious positive impact of being first-born on educationWe have checked that this problem is of no consequence for our results6

The summary statistics of our sample are presented in Table 1 The detaileddescription of variables is relegated to Appendix 1 We found more than 8000 indexpersons (from more than 3100 distinct families) older than 24 in 2001 with at leastone other sibling and whose mother has completed her fertility This is to be con-trasted with Conley and Glauber (2004) who use larger sample from the US Censusbut focus only on children under the age of 20 living at home

The main dependent variablemdashyears of completed educationmdashhas an average of1262 years7 About 82 percent of those selected index persons have at least 12 years ofeducation that is to say have completed high school8 Our measure of earningsmdashloghourly wage in 2001mdashshows an average corresponding to $145hour However theinformation on hourly wages or salary income is not available for most index persons

The average age in our sample is 39 About half of respondents are male and 47percent are White The average number of siblings is 486 This high number is con-sistent with the PSID oversampling minorities and low-income populations Fifty-sixpercent of the index persons have all their siblings reporting and 60 percent have boththeir parents reporting their childbirth history

We improved our analysis by including important observed family level-effects thatvary across parity family income and whether the mother is married Wheneveravailable we constructed the corresponding information for each of the first 14 yearsof life of each index person9 The main limitation is that information on familyincome cannot be recovered for many index individuals

Lastly we include two variables describing characteristics of index individualsrsquo par-ents namely education (11 years for both mother and father on average) and the age ofthe parents at birth of index persons (26 for mothers and 29 for fathers on average)

III Methods and Results

A The First-Born Effect

We first use an OLS estimation with robust standard errors clustered by family unit(identified by the mother) which relaxes the independence assumption between the

6 The regressions presented in this article contain age controls that separate cohort from birth order effects 7 A more appropriate variable would be education at age 25 However the number of observations avail-able would drop considerably and we would not be able to run estimations by sibling size All of our otherresults when running estimations that control for siblings size hold when replacing education with educationat age 25 for those where such information can be traced8 A negligible fraction of those who declare a certain education level at some point in their life declare lesseducation later Removing such observations did not alter our results We therefore choose the latest educa-tion level reported as our variable of interest9 We then ran our regressions using the average of these variables for three age groups 1ndash6 7ndash14 and1ndash14 For each age group the averages are respectively calculated only if the information is present in atleast half of the years of the age group

The Journal of Human Resources760

error terms and requires only that the observations be independent across clusters10

In Table 2 we first test the hypothesis that being early in the birth order implies a dis-tinct advantage that is entirely due to the higher probability of coming from a smallfamily

Columns 1 and 2 of Table 2 reject that claim but help us understand why it mayhave been made In Column 1 we omit the number of siblings therefore the signif-icant coefficient on first-born reflects not only the birth order effect but also the prob-ability of coming from a small family In Column 2 the inclusion of the number ofsiblings leaves the coefficient on first-born insignificant

In Column 3 we include age of the mother at childbirth and find a positive andhighly significant effect of being first-born on years of education This effect is con-firmed when including the fatherrsquos characteristics in Column 4 where both a father anda mother report Often not all siblings report this is especially the case for large fam-ilies To check if we are biasing our results by including such families in Column 5 werestrict our attention to families with complete information on all siblings The findingsare similar there too further showing that they could not be driven by selective attri-tion within families by birth order In all specifications we find a stronger effect amongWhite families

The results presented here are for the impact of being first-born in families of morethan one child This particular procedure takes advantage of the full size of our sam-ple and can be useful when there are not enough observations to run separate estima-tions for different siblingsrsquo sizes11 We see that it allows us to reveal a significant androbust birth order effect

However we still obtain similar results when looking at individual birth ordereffects by siblingsrsquo size in Table 3 where all specifications include age of the motherat childbirth Although large families show higher birth order point estimates theeffect is present in two-sibling families as well Note that the coefficient on first-bornis only weakly significant in Column 4mdashfamilies of five siblingsmdashlikely because ofthe small sample size

The reason why the inclusion of the age of the mother at childbirth makes the coef-ficient on first-born larger in magnitude and more significant is clear The age of themother at childbirth is mechanically positively correlated with birth order and evenmore strongly across large families Conversely we see that it is positively correlatedwith a childrsquos education Then if having a high birth order carries a negative impacton education the two effects of birth order and age of the mother at childbirth com-pete against one another Therefore the coefficient on first-born in Table 2 Column 2reflects an omitted variable bias The results hold for both males and females formothers with or without more than a high school education We also found that spac-ing between births be it that of the first-born child with respect to the second-born orto the last-born child does not alter the conclusions either Additional OLS regres-sions confirmed that the effect is significantly present among White families of any

10 We also used random effects procedures but because they yield almost identical results as those withthe family clustered standard errors those are not reported11 For example we found a significant birth-order effect when considering separately mothers who firstgave birth early and mothers who first gave birth ldquolaterdquo (using various definitions of ldquoearlyrdquo and ldquolaterdquo) yetwe do not have enough observations to split the sample by maternal age and individual siblings size

Kantarevic and Mechoulan 761Ta

ble

2O

LS

Reg

ress

ion

wit

h D

epen

dent

Var

iabl

e E

duca

tion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

016

20

012

028

80

239

026

8(0

068

)

(00

69)

(00

72)

(0

098

)

(00

84)

To

tal n

umbe

r of

sib

lings

minus01

04minus0

121

minus01

08minus0

125

(00

15)

(0

015

)

(00

19)

(0

029

)

Age

of

mot

her

at c

hild

birt

h0

058

006

10

065

(00

05)

(0

011

)

(00

08)

d[

Mal

e]minus0

205

minus02

1minus0

226

minus01

86minus0

159

(00

47)

(0

046

)

(00

46)

(0

058

)

(00

57)

A

ge0

121

015

90

162

014

30

087

(00

27)

(0

027

)

(00

27)

(0

039

)

(00

40)

A

ge2

minus00

01minus0

001

minus00

01minus0

001

minus00

03(0

000

3)

(0

000

3)

(0

000

3)

(0

001

)

(00

001)

d[W

hite

]0

119

minus00

03minus0

099

minus02

95minus0

107

(00

72)

(00

73)

(00

71)

(00

87)

(0

098

)d[

first

-bor

n] times

d[W

hite

]0

126

019

30

184

028

60

172

(00

92)

(00

92)

(0

092

)

(01

15)

(0

104

)M

othe

rrsquos

educ

atio

n0

213

019

80

199

011

40

236

(00

12)

(0

012

)

(00

12)

(0

017

)

(00

18)

d[

all s

iblin

gs r

epor

t]0

261

009

60

181

014

4(0

068

)

(00

69)

(00

69)

(0

089

)d[

both

par

ents

rep

ort]

052

30

553

052

00

533

(00

63)

(0

062

)

(00

61)

(0

081

)

Fath

errsquos

edu

catio

n0

125

(00

12)

A

ge o

f fa

ther

at c

hild

birt

h0

006

(00

09)

Con

stan

t6

713

671

75

073

516

25

848

(05

28)

(05

31)

(0

541

)

(07

75)

(0

776

)

R2

015

780

1691

018

890

2009

019

82N

umbe

r of

obs

erva

tions

792

87

928

792

87

766

454

1N

umbe

r of

fam

ily c

lust

ers

311

23

112

311

21

869

173

2

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources762

Tabl

e 3

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Edu

cati

on

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

027

50

698

071

50

534

091

5(0

104

)

(01

36)

(0

185

)

(03

16)

(02

36)

d[

seco

nd-b

orn]

043

40

463

033

40

578

(01

17)

(0

16)

(02

82)

(01

95)

d[

thir

d-bo

rn]

017

10

212

036

8(0

14)

(02

47)

(01

75)

d[

four

th-b

orn]

minus00

120

466

(02

14)

(01

52)

d[

fifth

-bor

n]0

221

(01

33)

Age

of

the

mot

her

at c

hild

birt

h0

048

007

90

067

006

80

07(0

012

)

(00

13)

(0

013

)

(00

23)

(0

013

)

d[M

ale]

minus02

87minus0

115

minus00

35minus0

34

minus00

33(0

102

)

(00

91)

(01

07)

(01

31)

(0

087

)

Age

021

70

190

156

007

50

167

(00

56)

(0

057

)

(00

49)

(0

044

)(0

057

)

Kantarevic and Mechoulan 763

Age

20

002

000

20

002

3times

10minus4

000

2(7

times10

minus4)

(7

times10

minus4)

(6

times10

minus4)

(5

times10

minus4)

(8times

10minus4

)

d[W

hite

]0

119

minus01

244

times10

minus5minus0

03

minus01

14(0

12)

(01

3)(0

13)

(01

81)

(01

37)

Mot

herrsquo

s ed

ucat

ion

025

70

250

179

012

40

182

(00

25)

(0

026

)

(00

23)

(0

035

)

(00

22)

d[

all s

iblin

gs r

epor

t]0

147

021

2minus0

066

035

10

21(0

154

)(0

136

)(0

128

)(0

178

)

(01

49)

d[bo

th p

aren

ts r

epor

t]0

241

061

20

755

054

20

371

(01

18)

(0

118

)

(01

26)

(0

189

)

(01

27)

C

onst

ant

333

92

694

425

86

607

412

(11

48)

(1

184

)

(10

11)

(1

293

)

(14

15)

R

20

185

021

019

011

40

127

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources764

size greater than one whereas it is only present within large families among BlacksTherefore the ethnic differential disappears when considering large families only

As noted earlier a causal interpretation of the age of the mother at childbirth wouldhinge on the assumption of its exogeneity Without instrumental variables or a treat-ment versus control quasi experiment it is difficult to draw conclusions12 The age ofthe mother at childbirth itself positively correlated with birth order could easilyproxy for other unobserved variables such as level of human capital and parentalresources13

To address this problem our fixed-effect estimation (Table 4a and b) removes fam-ily characteristics and unobserved family-level heterogeneity14 Family fixed effectsaddress family unobservables to the extent that they are constant over time While wetry to incorporate observables that vary across birth order to affirm the robustness ofour results we are constrained by the availability of such variables in the data set

Unfortunately the coefficients on age or on age of the mother at childbirth are unin-formative in those fixed effects regressions Deviations from family means for ageconvey the same information as deviations from family means for age of mother atchildbirth We thus do not provide separate estimations with age and age of motherat childbirth The age of the mother at first childbirth is certainly relevant but here itis differenced out Still to assess this issue better we ran separate regressions split-ting the sample by maternal age15 To summarize the following results are robust toexcluding mothers who first gave birth as teens but we do not have enough observa-tions to meaningfully run the fixed effects regressions by sibling size on that lattergroup

In Tables 4a and 4b the fixed effects estimations that only control for age and gen-der confirm the previous results for the most part16 We also provide estimations includ-ing the marital status of the mother17 during the childrsquos first 14 years18 The results donot change by much19 Clearly the suggestion that first-borns are most likely to live

12 Also while it is possible in the context of larger samples to instrument for siblings sizemdashusing twinbirths (Black Devereux and Salvanes 2005) or using the fact that parents of two same sex children are morelikely to have a third child (Conley and Glauber 2004)mdashone cannot instrument for birth order per se13 This is further evidenced by the fact that including a dummy variable indicating whether the mother wasmarried at the time of childbirth instead of (and obviously also along with) age of mother at childbirth alsoresults in highly significant birth order coefficients Results are available upon request14 However it is worth noting that those fixed effects do not solve all endogeneity issues For example itmay be the case that first-born ldquoqualityrdquo is affecting subsequent fertility We thank an anonymous referee forpointing out this caveat15 We thank an anonymous referee for this suggestion16 In Column 1mdashfamilies of two siblingsmdashthe coefficient on first-born becomes significant at the 10 per-cent level when including index persons of 23 and 24 years of age suggesting the nonsignificance whenrestricting at age 25 stems from a small sample size Separate FE regressions for White and Black familiesconfirmed the ethnic differentials found earlier17 We tried two definitions number of years marriednumber of years considered in the age group and = 1if continuously married over the years considered in the age group 0 otherwise Since the results on birthorder do not change qualitatively with either of those we only report the results with the latter definition18 We also ran similar estimations with different age ranges and obtained similar results19 Because the information on marital status is retrospective we do not lose any observation by includingthis covariate Unfortunately this is not the case when adding average family income or average employmentstatus of the mother during the childrsquos first years

Kantarevic and Mechoulan 765

Table 4aFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Not Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0216 0437 0683 0594 0477(0181) (0206) (0265) (0379) (0224)

d[second- 0282 0412 0421 0199born] (0141) (0214) (0323) (0195)

d[third- 0138 0361 0055born] (0159) (0261) (0172)

d[fourth- minus0031 0181born] (0207) (0149)

d[fifth- minus0012born] (0125)

Observations 1422 1743 1477 922 2487groups 934 835 562 388 573

Table 4bFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0224 0436 0696 0585 0492(0181) (0206) (0267) (0379) (0226)

d[second-born] 028 0423 0413 0212(0141) (0216) (0323) (0197)

d[third-born] 0146 0363 0063(0159) (0261) (0173)

d[fourth-born] minus0025 0188(0207) (0149)

d[fifth-born] minus0016(0125)

Mother minus0196 minus0167 0112 0381 minus0105Continuously (0315) (0221) (0221) (0274) (0178)married aged 1ndash14

Number of 1422 1743 1477 999 2487observations

Number of 934 835 562 328 573groups

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources766

their critical development years in a stable household as opposed to later-borns whomay experience the divorce of their parents cannot entirely explain the first-bornadvantage At the same time the persistence of birth-order effects naturally poses theproblem of their origin

The literature is not able to distinguish between different theories on the topic ofbirth order For example schooling circumstances play a large role in educational out-comes and may be related to birth order20 Our analysis is limited in the sense that itcannot discriminate between many competing hypotheses on why birth order appearsto be important21

Nevertheless we checked whether what appears as a first-born advantage predom-inantly comes from financial constraints for example parents sending their first-bornto college and running out of money for the following siblings Conley offers the fol-lowing argument ldquo[I]n terms of parental investment the cup starts to run dry as wego down the line Parental resources it appears are allotted on a first come first-served basisrdquo22 Yet if it turns out that first-borns perform better beforehand then atheory based on budget constraints cannot fully account for our results In Tables 5ndash7we estimate the probability of completing high school following the same methodol-ogy as in Tables 2ndash4 We find that first-borns have a higher probability of completinghigh school than later born siblings

Specifically Table 5 shows again that the first-born effect is not an artifact of fam-ily size that it increases in magnitude and significance when including age of motherat birth that it is robust to including characteristics of father and to restricting thesample to families in which all siblings report Table 6 shows that these results alsohold when regressions are estimated separately by family size (except for families offive presumably because of the small sample size) Tables 7a and 7b are the counter-parts of Tables 4a and 4b Fixed effects regressions controlling for age and gendersupport the results found in Tables 5 and 623

Also we estimated education at age 18 conditional on high school completion (at18 or older) to see if later-borns are more likely to repeat grades which would sup-port a theory of birth-order effects based on cognitive development differences Wedid not find any evidence for this However the small sample size resulting fromselecting index persons with available information at age 18 warrants some caution

Finally we found evidence among White families only that conditional on com-pleting high school first-borns are more likely to receive postsecondary educationYet for all races conditional on postsecondary education we found no clear advan-tage to being first-born24 In summary financial constraints do seem to play a role butsome factors early in life contribute to the first-born premium puzzle

20 We thank an anonymous referee for this point21 We refer the reader to the survey of those theories presented in Black Devereux and Salvanes (2005)22 Conley (2004) p 6923 The results held when including motherrsquos marital status We ran into the same small size problems whenadding family income and motherrsquos employment status for different age ranges of each child The ethnic dif-ferential disappears when considering all families (but persists when focusing on smaller families) suggest-ing that some of the birth order effect among White families comes from postsecondary education24 Results are available from the authors

Kantarevic and Mechoulan 767

Tabl

e 5

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

004

40

024

005

50

039

004

4(0

014

)

(00

14)

(00

14)

(0

018

)

(00

17)

To

tal n

umbe

r of

0

014

001

60

011

001

2si

blin

gs(0

003

)

(00

03)

(0

004

)

(00

05)

A

ge o

f th

e m

othe

r 0

006

000

70

005

at c

hild

birt

h(0

000

1)

(0

002

)

(00

01)

d[

Mal

e]0

047

004

80

049

004

70

037

(00

09)

(0

009

)

(00

09)

(0

009

)

(00

10)

A

ge0

024

002

90

029

003

20

023

(00

04)

(0

004

)

(00

04)

(0

007

)

(00

07)

A

ge2

000

020

0003

000

020

0003

000

02(5

times10

minus5)

(5

times10

minus5)

(5

times10

minus5)

(8

times10

minus5)

(8

times10

minus5)

d[

Whi

te]

000

50

012

002

30

042

001

9(0

014

)(0

014

)(0

014

)(0

017

)

(00

19)

d[fir

st-b

orn]

timesd

001

5minus0

005

minus00

060

011

000

6[W

hite

](0

017

5)(0

017

)(0

017

)(0

021

)(0

021

)M

othe

rrsquos

educ

atio

n0

027

002

60

026

001

50

026

(00

02)

(0

002

)

(00

02)

(0

003

)

(00

03)

d[

com

plet

e in

fo

003

50

094

002

20

029

on a

ll si

blin

gs]

(00

13)

(0

012

)

(00

13)

(00

15)

d[bo

th p

aren

ts r

epor

t]0

090

094

009

00

097

(00

12)

(0

012

)

(00

12)

(0

016

)

The Journal of Human Resources768

Tabl

e 5

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

Fath

errsquos

edu

catio

n0

011

(00

02)

A

ge o

f fa

ther

minus0

001

at c

hild

birt

h(0

002

)C

onst

ant

010

50

104

028

80

241

012

9(0

089

)(0

089

)(0

094

)

(01

35)

(01

41)

R2

008

790

0943

010

180

0885

008

55N

umbe

r of

obs

erva

tions

792

87

928

792

84

766

454

1N

umbe

r of

fam

ily

311

23

112

311

21

869

173

2cl

uste

rs

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Kantarevic and Mechoulan 769

Tabl

e 6

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

003

80

084

009

003

10

122

(00

18)

(0

025

)

(00

36)

(0

053

)(0

049

)

d[se

cond

-bor

n]0

071

008

2times

10minus4

002

2(0

022

)

(00

31)

(0

051

)(0

044

)d[

thir

d-bo

rn]

003

10

012

003

5(0

028

)(0

041

)(0

037

)d[

four

th-b

orn]

006

10

041

(00

37)

(00

33)

d[fif

th-b

orn]

000

7(0

028

)A

ge o

f th

e m

othe

r at

chi

ldbi

rth

000

30

004

000

90

008

001

1(0

002

)(0

002

)(0

003

)

(00

04)

(0

003

)

d[M

ale]

minus00

3minus0

041

000

7minus0

094

minus00

82(0

016

)(0

016

)

(00

2)(0

026

)

(00

18)

A

ge0

028

002

80

051

001

20

037

(00

07)

(0

009

)

(00

1)

(0

008

)(0

011

)

Age

23

times10

minus43

times10

minus45

times10

minus48

times10

minus54times

10minus4

(9times

10minus5

)

(10minus5

)

(10minus5

)

(9times

10minus5

)(1

0minus4)

d[

Whi

te]

001

20

124

004

90

033

002

2(0

021

)(0

023

)(0

026

)(0

034

)(0

028

)M

othe

rrsquos

educ

atio

n0

024

002

50

027

001

80

027

(00

04)

(0

004

)

(00

05)

(0

005

)

(00

04)

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 3: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

Our contribution is perhaps most closely related to the work of Hanushek (1992)who used a sample of school children from low-income Black families in early 1970sIndiana Hanushekrsquos paper advances that while being early in the birth order impliesa distinct advantage it is entirely due to the higher probability of coming from a smallfamily Following Lindert (1977) the paper also highlights within large families adistinct and sizeable pattern of falling and then rising attainment with respect to birthorder to the point that it becomes best to be last-born In contrast to Hanushek oursample is more representative We are thus able to examine longer-run outcomes andwe also look at racial differences Our empirical results first present a close versionof Hanushekrsquos findings before challenging their robustness by introducing age ofmother at birth in the model and running fixed effects estimations

The rest of the paper is organized as follows Section II presents our data SectionIII shows how birth order affects various educational outcomes through different esti-mation strategies Section IV then extends the analysis to earnings Finally Section Vconcludes

II Description of the Data

Our data come from the Childbirth and Adoption History File(CAHF) a special supplemental file of the PSID The CAHF covers eligible people4

living in a PSID family at the time of the interview in any wave from 1985 through2001

The population examined here (henceforth the ldquoindex personsrdquo) consists of allthose for whom the CAHF sample contains records of the childbirth histories of atleast one of their parents The CAHF allows us to compile information on their birthorder and the total number of children that their parent(s) report(s)

The index persons with missing information on their birth order or for whom thenumber of siblings is not ascertained are necessarily excluded from the sample Toensure that all mothers have completed their fertility so that we correctly identify thetotal number of siblings we further limit the sample to those index persons whosemother was older than 44 in the last year she reported

Siblings are defined based on the childbirth histories of mothers5 In addition tothe birth order and the number of siblings of the index persons we have obtainedadditional demographic information on them and their parents from other PSID filesusing the unique individual identifiers that are present in the PSID main and supple-mental files

Notably the PSID suffers from an important attrition bias More educated peopletend to stick with the questionnaire over longer periods of time thus it appears that

4 Eligible persons are defined as heads or wives of any age and other members of the family unit aged12ndash44 at the time of the interview These individuals are asked retrospective questions about their birth andadoption histories at the time of their first interview In each succeeding wave these histories are updated5 The sample shrinks by 30 percent if siblings are defined based on the childbirth histories of fathers Incase both parents report we were able to identify between siblings and half siblings however this distinc-tion did not change any of our results We include a variable expressing whether both parents report in ourregressions

Kantarevic and Mechoulan 757

Tabl

e 1

Des

crip

tive

Sta

tist

ics

Var

iabl

eSt

anda

rd

Inde

x in

divi

dual

Obs

erva

tions

Mea

nD

evia

tion

Min

imum

Max

imum

Yea

rs o

f co

mpl

eted

edu

catio

n8

147

126

22

131

17Pe

rcen

t com

plet

ed h

igh

scho

ol8

147

082

mdash0

1L

og h

ourl

y ea

rnin

gs in

200

13

028

267

071

minus12

599

Age

(in

200

1)8

318

385

78

7625

89Pe

rcen

t mal

e8

318

05

mdash0

1Pe

rcen

t Whi

te8

318

047

mdash0

1N

umbe

r of

sib

lings

831

84

862

722

16Pe

rcen

t firs

t-bo

rn8

318

027

mdash0

1Pe

rcen

t sec

ond-

born

831

80

27mdash

01

Perc

ent t

hird

-bor

n8

318

017

mdash0

1Pe

rcen

t fou

rth-

born

831

80

11mdash

01

Perc

ent fi

fth-

born

831

80

07mdash

01

Info

rmat

ion

on a

ll si

blin

gs8

318

056

mdash0

1B

oth

pare

nts

repo

rt th

e ch

ildbi

rth

831

80

60mdash

01

Fam

ily in

com

eA

ge 1

ndash62

695

136

399

283

508

976

60A

ge 7

ndash14

457

619

993

173

601

173

255

393

Age

1ndash1

43

474

188

5014

026

109

217

848

0M

othe

r co

ntin

uous

ly m

arri

edA

ge 1

ndash68

318

025

mdash0

1A

ge 7

ndash14

831

80

36mdash

01

Age

1ndash1

48

318

021

mdash0

1M

othe

rA

ge a

t bir

th8

292

261

85

8815

48Y

ears

of

com

plet

ed e

duca

tion

810

211

03

302

120

Fath

erA

ge a

t bir

th5

000

293

26

5417

60Y

ears

of

com

plet

ed e

duca

tion

489

211

13

367

117

Incl

udes

ind

ex p

erso

ns w

ho h

ave

at l

east

one

sib

ling

who

are

25

year

s or

old

er i

n 20

01 a

nd w

hose

mot

her

is a

t le

ast

44 y

ears

old

in

the

last

yea

r sh

e re

port

ed T

he n

umbe

r of

dis

tinct

fam

ilies

is

311

2

The Journal of Human Resources758

Kantarevic and Mechoulan 759

education is decreasing over cohorts which is of course untrue according to the USCensus Because a first-born is older than other siblings by definition this alonecould in theory produce a spurious positive impact of being first-born on educationWe have checked that this problem is of no consequence for our results6

The summary statistics of our sample are presented in Table 1 The detaileddescription of variables is relegated to Appendix 1 We found more than 8000 indexpersons (from more than 3100 distinct families) older than 24 in 2001 with at leastone other sibling and whose mother has completed her fertility This is to be con-trasted with Conley and Glauber (2004) who use larger sample from the US Censusbut focus only on children under the age of 20 living at home

The main dependent variablemdashyears of completed educationmdashhas an average of1262 years7 About 82 percent of those selected index persons have at least 12 years ofeducation that is to say have completed high school8 Our measure of earningsmdashloghourly wage in 2001mdashshows an average corresponding to $145hour However theinformation on hourly wages or salary income is not available for most index persons

The average age in our sample is 39 About half of respondents are male and 47percent are White The average number of siblings is 486 This high number is con-sistent with the PSID oversampling minorities and low-income populations Fifty-sixpercent of the index persons have all their siblings reporting and 60 percent have boththeir parents reporting their childbirth history

We improved our analysis by including important observed family level-effects thatvary across parity family income and whether the mother is married Wheneveravailable we constructed the corresponding information for each of the first 14 yearsof life of each index person9 The main limitation is that information on familyincome cannot be recovered for many index individuals

Lastly we include two variables describing characteristics of index individualsrsquo par-ents namely education (11 years for both mother and father on average) and the age ofthe parents at birth of index persons (26 for mothers and 29 for fathers on average)

III Methods and Results

A The First-Born Effect

We first use an OLS estimation with robust standard errors clustered by family unit(identified by the mother) which relaxes the independence assumption between the

6 The regressions presented in this article contain age controls that separate cohort from birth order effects 7 A more appropriate variable would be education at age 25 However the number of observations avail-able would drop considerably and we would not be able to run estimations by sibling size All of our otherresults when running estimations that control for siblings size hold when replacing education with educationat age 25 for those where such information can be traced8 A negligible fraction of those who declare a certain education level at some point in their life declare lesseducation later Removing such observations did not alter our results We therefore choose the latest educa-tion level reported as our variable of interest9 We then ran our regressions using the average of these variables for three age groups 1ndash6 7ndash14 and1ndash14 For each age group the averages are respectively calculated only if the information is present in atleast half of the years of the age group

The Journal of Human Resources760

error terms and requires only that the observations be independent across clusters10

In Table 2 we first test the hypothesis that being early in the birth order implies a dis-tinct advantage that is entirely due to the higher probability of coming from a smallfamily

Columns 1 and 2 of Table 2 reject that claim but help us understand why it mayhave been made In Column 1 we omit the number of siblings therefore the signif-icant coefficient on first-born reflects not only the birth order effect but also the prob-ability of coming from a small family In Column 2 the inclusion of the number ofsiblings leaves the coefficient on first-born insignificant

In Column 3 we include age of the mother at childbirth and find a positive andhighly significant effect of being first-born on years of education This effect is con-firmed when including the fatherrsquos characteristics in Column 4 where both a father anda mother report Often not all siblings report this is especially the case for large fam-ilies To check if we are biasing our results by including such families in Column 5 werestrict our attention to families with complete information on all siblings The findingsare similar there too further showing that they could not be driven by selective attri-tion within families by birth order In all specifications we find a stronger effect amongWhite families

The results presented here are for the impact of being first-born in families of morethan one child This particular procedure takes advantage of the full size of our sam-ple and can be useful when there are not enough observations to run separate estima-tions for different siblingsrsquo sizes11 We see that it allows us to reveal a significant androbust birth order effect

However we still obtain similar results when looking at individual birth ordereffects by siblingsrsquo size in Table 3 where all specifications include age of the motherat childbirth Although large families show higher birth order point estimates theeffect is present in two-sibling families as well Note that the coefficient on first-bornis only weakly significant in Column 4mdashfamilies of five siblingsmdashlikely because ofthe small sample size

The reason why the inclusion of the age of the mother at childbirth makes the coef-ficient on first-born larger in magnitude and more significant is clear The age of themother at childbirth is mechanically positively correlated with birth order and evenmore strongly across large families Conversely we see that it is positively correlatedwith a childrsquos education Then if having a high birth order carries a negative impacton education the two effects of birth order and age of the mother at childbirth com-pete against one another Therefore the coefficient on first-born in Table 2 Column 2reflects an omitted variable bias The results hold for both males and females formothers with or without more than a high school education We also found that spac-ing between births be it that of the first-born child with respect to the second-born orto the last-born child does not alter the conclusions either Additional OLS regres-sions confirmed that the effect is significantly present among White families of any

10 We also used random effects procedures but because they yield almost identical results as those withthe family clustered standard errors those are not reported11 For example we found a significant birth-order effect when considering separately mothers who firstgave birth early and mothers who first gave birth ldquolaterdquo (using various definitions of ldquoearlyrdquo and ldquolaterdquo) yetwe do not have enough observations to split the sample by maternal age and individual siblings size

Kantarevic and Mechoulan 761Ta

ble

2O

LS

Reg

ress

ion

wit

h D

epen

dent

Var

iabl

e E

duca

tion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

016

20

012

028

80

239

026

8(0

068

)

(00

69)

(00

72)

(0

098

)

(00

84)

To

tal n

umbe

r of

sib

lings

minus01

04minus0

121

minus01

08minus0

125

(00

15)

(0

015

)

(00

19)

(0

029

)

Age

of

mot

her

at c

hild

birt

h0

058

006

10

065

(00

05)

(0

011

)

(00

08)

d[

Mal

e]minus0

205

minus02

1minus0

226

minus01

86minus0

159

(00

47)

(0

046

)

(00

46)

(0

058

)

(00

57)

A

ge0

121

015

90

162

014

30

087

(00

27)

(0

027

)

(00

27)

(0

039

)

(00

40)

A

ge2

minus00

01minus0

001

minus00

01minus0

001

minus00

03(0

000

3)

(0

000

3)

(0

000

3)

(0

001

)

(00

001)

d[W

hite

]0

119

minus00

03minus0

099

minus02

95minus0

107

(00

72)

(00

73)

(00

71)

(00

87)

(0

098

)d[

first

-bor

n] times

d[W

hite

]0

126

019

30

184

028

60

172

(00

92)

(00

92)

(0

092

)

(01

15)

(0

104

)M

othe

rrsquos

educ

atio

n0

213

019

80

199

011

40

236

(00

12)

(0

012

)

(00

12)

(0

017

)

(00

18)

d[

all s

iblin

gs r

epor

t]0

261

009

60

181

014

4(0

068

)

(00

69)

(00

69)

(0

089

)d[

both

par

ents

rep

ort]

052

30

553

052

00

533

(00

63)

(0

062

)

(00

61)

(0

081

)

Fath

errsquos

edu

catio

n0

125

(00

12)

A

ge o

f fa

ther

at c

hild

birt

h0

006

(00

09)

Con

stan

t6

713

671

75

073

516

25

848

(05

28)

(05

31)

(0

541

)

(07

75)

(0

776

)

R2

015

780

1691

018

890

2009

019

82N

umbe

r of

obs

erva

tions

792

87

928

792

87

766

454

1N

umbe

r of

fam

ily c

lust

ers

311

23

112

311

21

869

173

2

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources762

Tabl

e 3

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Edu

cati

on

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

027

50

698

071

50

534

091

5(0

104

)

(01

36)

(0

185

)

(03

16)

(02

36)

d[

seco

nd-b

orn]

043

40

463

033

40

578

(01

17)

(0

16)

(02

82)

(01

95)

d[

thir

d-bo

rn]

017

10

212

036

8(0

14)

(02

47)

(01

75)

d[

four

th-b

orn]

minus00

120

466

(02

14)

(01

52)

d[

fifth

-bor

n]0

221

(01

33)

Age

of

the

mot

her

at c

hild

birt

h0

048

007

90

067

006

80

07(0

012

)

(00

13)

(0

013

)

(00

23)

(0

013

)

d[M

ale]

minus02

87minus0

115

minus00

35minus0

34

minus00

33(0

102

)

(00

91)

(01

07)

(01

31)

(0

087

)

Age

021

70

190

156

007

50

167

(00

56)

(0

057

)

(00

49)

(0

044

)(0

057

)

Kantarevic and Mechoulan 763

Age

20

002

000

20

002

3times

10minus4

000

2(7

times10

minus4)

(7

times10

minus4)

(6

times10

minus4)

(5

times10

minus4)

(8times

10minus4

)

d[W

hite

]0

119

minus01

244

times10

minus5minus0

03

minus01

14(0

12)

(01

3)(0

13)

(01

81)

(01

37)

Mot

herrsquo

s ed

ucat

ion

025

70

250

179

012

40

182

(00

25)

(0

026

)

(00

23)

(0

035

)

(00

22)

d[

all s

iblin

gs r

epor

t]0

147

021

2minus0

066

035

10

21(0

154

)(0

136

)(0

128

)(0

178

)

(01

49)

d[bo

th p

aren

ts r

epor

t]0

241

061

20

755

054

20

371

(01

18)

(0

118

)

(01

26)

(0

189

)

(01

27)

C

onst

ant

333

92

694

425

86

607

412

(11

48)

(1

184

)

(10

11)

(1

293

)

(14

15)

R

20

185

021

019

011

40

127

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources764

size greater than one whereas it is only present within large families among BlacksTherefore the ethnic differential disappears when considering large families only

As noted earlier a causal interpretation of the age of the mother at childbirth wouldhinge on the assumption of its exogeneity Without instrumental variables or a treat-ment versus control quasi experiment it is difficult to draw conclusions12 The age ofthe mother at childbirth itself positively correlated with birth order could easilyproxy for other unobserved variables such as level of human capital and parentalresources13

To address this problem our fixed-effect estimation (Table 4a and b) removes fam-ily characteristics and unobserved family-level heterogeneity14 Family fixed effectsaddress family unobservables to the extent that they are constant over time While wetry to incorporate observables that vary across birth order to affirm the robustness ofour results we are constrained by the availability of such variables in the data set

Unfortunately the coefficients on age or on age of the mother at childbirth are unin-formative in those fixed effects regressions Deviations from family means for ageconvey the same information as deviations from family means for age of mother atchildbirth We thus do not provide separate estimations with age and age of motherat childbirth The age of the mother at first childbirth is certainly relevant but here itis differenced out Still to assess this issue better we ran separate regressions split-ting the sample by maternal age15 To summarize the following results are robust toexcluding mothers who first gave birth as teens but we do not have enough observa-tions to meaningfully run the fixed effects regressions by sibling size on that lattergroup

In Tables 4a and 4b the fixed effects estimations that only control for age and gen-der confirm the previous results for the most part16 We also provide estimations includ-ing the marital status of the mother17 during the childrsquos first 14 years18 The results donot change by much19 Clearly the suggestion that first-borns are most likely to live

12 Also while it is possible in the context of larger samples to instrument for siblings sizemdashusing twinbirths (Black Devereux and Salvanes 2005) or using the fact that parents of two same sex children are morelikely to have a third child (Conley and Glauber 2004)mdashone cannot instrument for birth order per se13 This is further evidenced by the fact that including a dummy variable indicating whether the mother wasmarried at the time of childbirth instead of (and obviously also along with) age of mother at childbirth alsoresults in highly significant birth order coefficients Results are available upon request14 However it is worth noting that those fixed effects do not solve all endogeneity issues For example itmay be the case that first-born ldquoqualityrdquo is affecting subsequent fertility We thank an anonymous referee forpointing out this caveat15 We thank an anonymous referee for this suggestion16 In Column 1mdashfamilies of two siblingsmdashthe coefficient on first-born becomes significant at the 10 per-cent level when including index persons of 23 and 24 years of age suggesting the nonsignificance whenrestricting at age 25 stems from a small sample size Separate FE regressions for White and Black familiesconfirmed the ethnic differentials found earlier17 We tried two definitions number of years marriednumber of years considered in the age group and = 1if continuously married over the years considered in the age group 0 otherwise Since the results on birthorder do not change qualitatively with either of those we only report the results with the latter definition18 We also ran similar estimations with different age ranges and obtained similar results19 Because the information on marital status is retrospective we do not lose any observation by includingthis covariate Unfortunately this is not the case when adding average family income or average employmentstatus of the mother during the childrsquos first years

Kantarevic and Mechoulan 765

Table 4aFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Not Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0216 0437 0683 0594 0477(0181) (0206) (0265) (0379) (0224)

d[second- 0282 0412 0421 0199born] (0141) (0214) (0323) (0195)

d[third- 0138 0361 0055born] (0159) (0261) (0172)

d[fourth- minus0031 0181born] (0207) (0149)

d[fifth- minus0012born] (0125)

Observations 1422 1743 1477 922 2487groups 934 835 562 388 573

Table 4bFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0224 0436 0696 0585 0492(0181) (0206) (0267) (0379) (0226)

d[second-born] 028 0423 0413 0212(0141) (0216) (0323) (0197)

d[third-born] 0146 0363 0063(0159) (0261) (0173)

d[fourth-born] minus0025 0188(0207) (0149)

d[fifth-born] minus0016(0125)

Mother minus0196 minus0167 0112 0381 minus0105Continuously (0315) (0221) (0221) (0274) (0178)married aged 1ndash14

Number of 1422 1743 1477 999 2487observations

Number of 934 835 562 328 573groups

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources766

their critical development years in a stable household as opposed to later-borns whomay experience the divorce of their parents cannot entirely explain the first-bornadvantage At the same time the persistence of birth-order effects naturally poses theproblem of their origin

The literature is not able to distinguish between different theories on the topic ofbirth order For example schooling circumstances play a large role in educational out-comes and may be related to birth order20 Our analysis is limited in the sense that itcannot discriminate between many competing hypotheses on why birth order appearsto be important21

Nevertheless we checked whether what appears as a first-born advantage predom-inantly comes from financial constraints for example parents sending their first-bornto college and running out of money for the following siblings Conley offers the fol-lowing argument ldquo[I]n terms of parental investment the cup starts to run dry as wego down the line Parental resources it appears are allotted on a first come first-served basisrdquo22 Yet if it turns out that first-borns perform better beforehand then atheory based on budget constraints cannot fully account for our results In Tables 5ndash7we estimate the probability of completing high school following the same methodol-ogy as in Tables 2ndash4 We find that first-borns have a higher probability of completinghigh school than later born siblings

Specifically Table 5 shows again that the first-born effect is not an artifact of fam-ily size that it increases in magnitude and significance when including age of motherat birth that it is robust to including characteristics of father and to restricting thesample to families in which all siblings report Table 6 shows that these results alsohold when regressions are estimated separately by family size (except for families offive presumably because of the small sample size) Tables 7a and 7b are the counter-parts of Tables 4a and 4b Fixed effects regressions controlling for age and gendersupport the results found in Tables 5 and 623

Also we estimated education at age 18 conditional on high school completion (at18 or older) to see if later-borns are more likely to repeat grades which would sup-port a theory of birth-order effects based on cognitive development differences Wedid not find any evidence for this However the small sample size resulting fromselecting index persons with available information at age 18 warrants some caution

Finally we found evidence among White families only that conditional on com-pleting high school first-borns are more likely to receive postsecondary educationYet for all races conditional on postsecondary education we found no clear advan-tage to being first-born24 In summary financial constraints do seem to play a role butsome factors early in life contribute to the first-born premium puzzle

20 We thank an anonymous referee for this point21 We refer the reader to the survey of those theories presented in Black Devereux and Salvanes (2005)22 Conley (2004) p 6923 The results held when including motherrsquos marital status We ran into the same small size problems whenadding family income and motherrsquos employment status for different age ranges of each child The ethnic dif-ferential disappears when considering all families (but persists when focusing on smaller families) suggest-ing that some of the birth order effect among White families comes from postsecondary education24 Results are available from the authors

Kantarevic and Mechoulan 767

Tabl

e 5

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

004

40

024

005

50

039

004

4(0

014

)

(00

14)

(00

14)

(0

018

)

(00

17)

To

tal n

umbe

r of

0

014

001

60

011

001

2si

blin

gs(0

003

)

(00

03)

(0

004

)

(00

05)

A

ge o

f th

e m

othe

r 0

006

000

70

005

at c

hild

birt

h(0

000

1)

(0

002

)

(00

01)

d[

Mal

e]0

047

004

80

049

004

70

037

(00

09)

(0

009

)

(00

09)

(0

009

)

(00

10)

A

ge0

024

002

90

029

003

20

023

(00

04)

(0

004

)

(00

04)

(0

007

)

(00

07)

A

ge2

000

020

0003

000

020

0003

000

02(5

times10

minus5)

(5

times10

minus5)

(5

times10

minus5)

(8

times10

minus5)

(8

times10

minus5)

d[

Whi

te]

000

50

012

002

30

042

001

9(0

014

)(0

014

)(0

014

)(0

017

)

(00

19)

d[fir

st-b

orn]

timesd

001

5minus0

005

minus00

060

011

000

6[W

hite

](0

017

5)(0

017

)(0

017

)(0

021

)(0

021

)M

othe

rrsquos

educ

atio

n0

027

002

60

026

001

50

026

(00

02)

(0

002

)

(00

02)

(0

003

)

(00

03)

d[

com

plet

e in

fo

003

50

094

002

20

029

on a

ll si

blin

gs]

(00

13)

(0

012

)

(00

13)

(00

15)

d[bo

th p

aren

ts r

epor

t]0

090

094

009

00

097

(00

12)

(0

012

)

(00

12)

(0

016

)

The Journal of Human Resources768

Tabl

e 5

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

Fath

errsquos

edu

catio

n0

011

(00

02)

A

ge o

f fa

ther

minus0

001

at c

hild

birt

h(0

002

)C

onst

ant

010

50

104

028

80

241

012

9(0

089

)(0

089

)(0

094

)

(01

35)

(01

41)

R2

008

790

0943

010

180

0885

008

55N

umbe

r of

obs

erva

tions

792

87

928

792

84

766

454

1N

umbe

r of

fam

ily

311

23

112

311

21

869

173

2cl

uste

rs

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Kantarevic and Mechoulan 769

Tabl

e 6

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

003

80

084

009

003

10

122

(00

18)

(0

025

)

(00

36)

(0

053

)(0

049

)

d[se

cond

-bor

n]0

071

008

2times

10minus4

002

2(0

022

)

(00

31)

(0

051

)(0

044

)d[

thir

d-bo

rn]

003

10

012

003

5(0

028

)(0

041

)(0

037

)d[

four

th-b

orn]

006

10

041

(00

37)

(00

33)

d[fif

th-b

orn]

000

7(0

028

)A

ge o

f th

e m

othe

r at

chi

ldbi

rth

000

30

004

000

90

008

001

1(0

002

)(0

002

)(0

003

)

(00

04)

(0

003

)

d[M

ale]

minus00

3minus0

041

000

7minus0

094

minus00

82(0

016

)(0

016

)

(00

2)(0

026

)

(00

18)

A

ge0

028

002

80

051

001

20

037

(00

07)

(0

009

)

(00

1)

(0

008

)(0

011

)

Age

23

times10

minus43

times10

minus45

times10

minus48

times10

minus54times

10minus4

(9times

10minus5

)

(10minus5

)

(10minus5

)

(9times

10minus5

)(1

0minus4)

d[

Whi

te]

001

20

124

004

90

033

002

2(0

021

)(0

023

)(0

026

)(0

034

)(0

028

)M

othe

rrsquos

educ

atio

n0

024

002

50

027

001

80

027

(00

04)

(0

004

)

(00

05)

(0

005

)

(00

04)

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 4: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

Tabl

e 1

Des

crip

tive

Sta

tist

ics

Var

iabl

eSt

anda

rd

Inde

x in

divi

dual

Obs

erva

tions

Mea

nD

evia

tion

Min

imum

Max

imum

Yea

rs o

f co

mpl

eted

edu

catio

n8

147

126

22

131

17Pe

rcen

t com

plet

ed h

igh

scho

ol8

147

082

mdash0

1L

og h

ourl

y ea

rnin

gs in

200

13

028

267

071

minus12

599

Age

(in

200

1)8

318

385

78

7625

89Pe

rcen

t mal

e8

318

05

mdash0

1Pe

rcen

t Whi

te8

318

047

mdash0

1N

umbe

r of

sib

lings

831

84

862

722

16Pe

rcen

t firs

t-bo

rn8

318

027

mdash0

1Pe

rcen

t sec

ond-

born

831

80

27mdash

01

Perc

ent t

hird

-bor

n8

318

017

mdash0

1Pe

rcen

t fou

rth-

born

831

80

11mdash

01

Perc

ent fi

fth-

born

831

80

07mdash

01

Info

rmat

ion

on a

ll si

blin

gs8

318

056

mdash0

1B

oth

pare

nts

repo

rt th

e ch

ildbi

rth

831

80

60mdash

01

Fam

ily in

com

eA

ge 1

ndash62

695

136

399

283

508

976

60A

ge 7

ndash14

457

619

993

173

601

173

255

393

Age

1ndash1

43

474

188

5014

026

109

217

848

0M

othe

r co

ntin

uous

ly m

arri

edA

ge 1

ndash68

318

025

mdash0

1A

ge 7

ndash14

831

80

36mdash

01

Age

1ndash1

48

318

021

mdash0

1M

othe

rA

ge a

t bir

th8

292

261

85

8815

48Y

ears

of

com

plet

ed e

duca

tion

810

211

03

302

120

Fath

erA

ge a

t bir

th5

000

293

26

5417

60Y

ears

of

com

plet

ed e

duca

tion

489

211

13

367

117

Incl

udes

ind

ex p

erso

ns w

ho h

ave

at l

east

one

sib

ling

who

are

25

year

s or

old

er i

n 20

01 a

nd w

hose

mot

her

is a

t le

ast

44 y

ears

old

in

the

last

yea

r sh

e re

port

ed T

he n

umbe

r of

dis

tinct

fam

ilies

is

311

2

The Journal of Human Resources758

Kantarevic and Mechoulan 759

education is decreasing over cohorts which is of course untrue according to the USCensus Because a first-born is older than other siblings by definition this alonecould in theory produce a spurious positive impact of being first-born on educationWe have checked that this problem is of no consequence for our results6

The summary statistics of our sample are presented in Table 1 The detaileddescription of variables is relegated to Appendix 1 We found more than 8000 indexpersons (from more than 3100 distinct families) older than 24 in 2001 with at leastone other sibling and whose mother has completed her fertility This is to be con-trasted with Conley and Glauber (2004) who use larger sample from the US Censusbut focus only on children under the age of 20 living at home

The main dependent variablemdashyears of completed educationmdashhas an average of1262 years7 About 82 percent of those selected index persons have at least 12 years ofeducation that is to say have completed high school8 Our measure of earningsmdashloghourly wage in 2001mdashshows an average corresponding to $145hour However theinformation on hourly wages or salary income is not available for most index persons

The average age in our sample is 39 About half of respondents are male and 47percent are White The average number of siblings is 486 This high number is con-sistent with the PSID oversampling minorities and low-income populations Fifty-sixpercent of the index persons have all their siblings reporting and 60 percent have boththeir parents reporting their childbirth history

We improved our analysis by including important observed family level-effects thatvary across parity family income and whether the mother is married Wheneveravailable we constructed the corresponding information for each of the first 14 yearsof life of each index person9 The main limitation is that information on familyincome cannot be recovered for many index individuals

Lastly we include two variables describing characteristics of index individualsrsquo par-ents namely education (11 years for both mother and father on average) and the age ofthe parents at birth of index persons (26 for mothers and 29 for fathers on average)

III Methods and Results

A The First-Born Effect

We first use an OLS estimation with robust standard errors clustered by family unit(identified by the mother) which relaxes the independence assumption between the

6 The regressions presented in this article contain age controls that separate cohort from birth order effects 7 A more appropriate variable would be education at age 25 However the number of observations avail-able would drop considerably and we would not be able to run estimations by sibling size All of our otherresults when running estimations that control for siblings size hold when replacing education with educationat age 25 for those where such information can be traced8 A negligible fraction of those who declare a certain education level at some point in their life declare lesseducation later Removing such observations did not alter our results We therefore choose the latest educa-tion level reported as our variable of interest9 We then ran our regressions using the average of these variables for three age groups 1ndash6 7ndash14 and1ndash14 For each age group the averages are respectively calculated only if the information is present in atleast half of the years of the age group

The Journal of Human Resources760

error terms and requires only that the observations be independent across clusters10

In Table 2 we first test the hypothesis that being early in the birth order implies a dis-tinct advantage that is entirely due to the higher probability of coming from a smallfamily

Columns 1 and 2 of Table 2 reject that claim but help us understand why it mayhave been made In Column 1 we omit the number of siblings therefore the signif-icant coefficient on first-born reflects not only the birth order effect but also the prob-ability of coming from a small family In Column 2 the inclusion of the number ofsiblings leaves the coefficient on first-born insignificant

In Column 3 we include age of the mother at childbirth and find a positive andhighly significant effect of being first-born on years of education This effect is con-firmed when including the fatherrsquos characteristics in Column 4 where both a father anda mother report Often not all siblings report this is especially the case for large fam-ilies To check if we are biasing our results by including such families in Column 5 werestrict our attention to families with complete information on all siblings The findingsare similar there too further showing that they could not be driven by selective attri-tion within families by birth order In all specifications we find a stronger effect amongWhite families

The results presented here are for the impact of being first-born in families of morethan one child This particular procedure takes advantage of the full size of our sam-ple and can be useful when there are not enough observations to run separate estima-tions for different siblingsrsquo sizes11 We see that it allows us to reveal a significant androbust birth order effect

However we still obtain similar results when looking at individual birth ordereffects by siblingsrsquo size in Table 3 where all specifications include age of the motherat childbirth Although large families show higher birth order point estimates theeffect is present in two-sibling families as well Note that the coefficient on first-bornis only weakly significant in Column 4mdashfamilies of five siblingsmdashlikely because ofthe small sample size

The reason why the inclusion of the age of the mother at childbirth makes the coef-ficient on first-born larger in magnitude and more significant is clear The age of themother at childbirth is mechanically positively correlated with birth order and evenmore strongly across large families Conversely we see that it is positively correlatedwith a childrsquos education Then if having a high birth order carries a negative impacton education the two effects of birth order and age of the mother at childbirth com-pete against one another Therefore the coefficient on first-born in Table 2 Column 2reflects an omitted variable bias The results hold for both males and females formothers with or without more than a high school education We also found that spac-ing between births be it that of the first-born child with respect to the second-born orto the last-born child does not alter the conclusions either Additional OLS regres-sions confirmed that the effect is significantly present among White families of any

10 We also used random effects procedures but because they yield almost identical results as those withthe family clustered standard errors those are not reported11 For example we found a significant birth-order effect when considering separately mothers who firstgave birth early and mothers who first gave birth ldquolaterdquo (using various definitions of ldquoearlyrdquo and ldquolaterdquo) yetwe do not have enough observations to split the sample by maternal age and individual siblings size

Kantarevic and Mechoulan 761Ta

ble

2O

LS

Reg

ress

ion

wit

h D

epen

dent

Var

iabl

e E

duca

tion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

016

20

012

028

80

239

026

8(0

068

)

(00

69)

(00

72)

(0

098

)

(00

84)

To

tal n

umbe

r of

sib

lings

minus01

04minus0

121

minus01

08minus0

125

(00

15)

(0

015

)

(00

19)

(0

029

)

Age

of

mot

her

at c

hild

birt

h0

058

006

10

065

(00

05)

(0

011

)

(00

08)

d[

Mal

e]minus0

205

minus02

1minus0

226

minus01

86minus0

159

(00

47)

(0

046

)

(00

46)

(0

058

)

(00

57)

A

ge0

121

015

90

162

014

30

087

(00

27)

(0

027

)

(00

27)

(0

039

)

(00

40)

A

ge2

minus00

01minus0

001

minus00

01minus0

001

minus00

03(0

000

3)

(0

000

3)

(0

000

3)

(0

001

)

(00

001)

d[W

hite

]0

119

minus00

03minus0

099

minus02

95minus0

107

(00

72)

(00

73)

(00

71)

(00

87)

(0

098

)d[

first

-bor

n] times

d[W

hite

]0

126

019

30

184

028

60

172

(00

92)

(00

92)

(0

092

)

(01

15)

(0

104

)M

othe

rrsquos

educ

atio

n0

213

019

80

199

011

40

236

(00

12)

(0

012

)

(00

12)

(0

017

)

(00

18)

d[

all s

iblin

gs r

epor

t]0

261

009

60

181

014

4(0

068

)

(00

69)

(00

69)

(0

089

)d[

both

par

ents

rep

ort]

052

30

553

052

00

533

(00

63)

(0

062

)

(00

61)

(0

081

)

Fath

errsquos

edu

catio

n0

125

(00

12)

A

ge o

f fa

ther

at c

hild

birt

h0

006

(00

09)

Con

stan

t6

713

671

75

073

516

25

848

(05

28)

(05

31)

(0

541

)

(07

75)

(0

776

)

R2

015

780

1691

018

890

2009

019

82N

umbe

r of

obs

erva

tions

792

87

928

792

87

766

454

1N

umbe

r of

fam

ily c

lust

ers

311

23

112

311

21

869

173

2

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources762

Tabl

e 3

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Edu

cati

on

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

027

50

698

071

50

534

091

5(0

104

)

(01

36)

(0

185

)

(03

16)

(02

36)

d[

seco

nd-b

orn]

043

40

463

033

40

578

(01

17)

(0

16)

(02

82)

(01

95)

d[

thir

d-bo

rn]

017

10

212

036

8(0

14)

(02

47)

(01

75)

d[

four

th-b

orn]

minus00

120

466

(02

14)

(01

52)

d[

fifth

-bor

n]0

221

(01

33)

Age

of

the

mot

her

at c

hild

birt

h0

048

007

90

067

006

80

07(0

012

)

(00

13)

(0

013

)

(00

23)

(0

013

)

d[M

ale]

minus02

87minus0

115

minus00

35minus0

34

minus00

33(0

102

)

(00

91)

(01

07)

(01

31)

(0

087

)

Age

021

70

190

156

007

50

167

(00

56)

(0

057

)

(00

49)

(0

044

)(0

057

)

Kantarevic and Mechoulan 763

Age

20

002

000

20

002

3times

10minus4

000

2(7

times10

minus4)

(7

times10

minus4)

(6

times10

minus4)

(5

times10

minus4)

(8times

10minus4

)

d[W

hite

]0

119

minus01

244

times10

minus5minus0

03

minus01

14(0

12)

(01

3)(0

13)

(01

81)

(01

37)

Mot

herrsquo

s ed

ucat

ion

025

70

250

179

012

40

182

(00

25)

(0

026

)

(00

23)

(0

035

)

(00

22)

d[

all s

iblin

gs r

epor

t]0

147

021

2minus0

066

035

10

21(0

154

)(0

136

)(0

128

)(0

178

)

(01

49)

d[bo

th p

aren

ts r

epor

t]0

241

061

20

755

054

20

371

(01

18)

(0

118

)

(01

26)

(0

189

)

(01

27)

C

onst

ant

333

92

694

425

86

607

412

(11

48)

(1

184

)

(10

11)

(1

293

)

(14

15)

R

20

185

021

019

011

40

127

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources764

size greater than one whereas it is only present within large families among BlacksTherefore the ethnic differential disappears when considering large families only

As noted earlier a causal interpretation of the age of the mother at childbirth wouldhinge on the assumption of its exogeneity Without instrumental variables or a treat-ment versus control quasi experiment it is difficult to draw conclusions12 The age ofthe mother at childbirth itself positively correlated with birth order could easilyproxy for other unobserved variables such as level of human capital and parentalresources13

To address this problem our fixed-effect estimation (Table 4a and b) removes fam-ily characteristics and unobserved family-level heterogeneity14 Family fixed effectsaddress family unobservables to the extent that they are constant over time While wetry to incorporate observables that vary across birth order to affirm the robustness ofour results we are constrained by the availability of such variables in the data set

Unfortunately the coefficients on age or on age of the mother at childbirth are unin-formative in those fixed effects regressions Deviations from family means for ageconvey the same information as deviations from family means for age of mother atchildbirth We thus do not provide separate estimations with age and age of motherat childbirth The age of the mother at first childbirth is certainly relevant but here itis differenced out Still to assess this issue better we ran separate regressions split-ting the sample by maternal age15 To summarize the following results are robust toexcluding mothers who first gave birth as teens but we do not have enough observa-tions to meaningfully run the fixed effects regressions by sibling size on that lattergroup

In Tables 4a and 4b the fixed effects estimations that only control for age and gen-der confirm the previous results for the most part16 We also provide estimations includ-ing the marital status of the mother17 during the childrsquos first 14 years18 The results donot change by much19 Clearly the suggestion that first-borns are most likely to live

12 Also while it is possible in the context of larger samples to instrument for siblings sizemdashusing twinbirths (Black Devereux and Salvanes 2005) or using the fact that parents of two same sex children are morelikely to have a third child (Conley and Glauber 2004)mdashone cannot instrument for birth order per se13 This is further evidenced by the fact that including a dummy variable indicating whether the mother wasmarried at the time of childbirth instead of (and obviously also along with) age of mother at childbirth alsoresults in highly significant birth order coefficients Results are available upon request14 However it is worth noting that those fixed effects do not solve all endogeneity issues For example itmay be the case that first-born ldquoqualityrdquo is affecting subsequent fertility We thank an anonymous referee forpointing out this caveat15 We thank an anonymous referee for this suggestion16 In Column 1mdashfamilies of two siblingsmdashthe coefficient on first-born becomes significant at the 10 per-cent level when including index persons of 23 and 24 years of age suggesting the nonsignificance whenrestricting at age 25 stems from a small sample size Separate FE regressions for White and Black familiesconfirmed the ethnic differentials found earlier17 We tried two definitions number of years marriednumber of years considered in the age group and = 1if continuously married over the years considered in the age group 0 otherwise Since the results on birthorder do not change qualitatively with either of those we only report the results with the latter definition18 We also ran similar estimations with different age ranges and obtained similar results19 Because the information on marital status is retrospective we do not lose any observation by includingthis covariate Unfortunately this is not the case when adding average family income or average employmentstatus of the mother during the childrsquos first years

Kantarevic and Mechoulan 765

Table 4aFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Not Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0216 0437 0683 0594 0477(0181) (0206) (0265) (0379) (0224)

d[second- 0282 0412 0421 0199born] (0141) (0214) (0323) (0195)

d[third- 0138 0361 0055born] (0159) (0261) (0172)

d[fourth- minus0031 0181born] (0207) (0149)

d[fifth- minus0012born] (0125)

Observations 1422 1743 1477 922 2487groups 934 835 562 388 573

Table 4bFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0224 0436 0696 0585 0492(0181) (0206) (0267) (0379) (0226)

d[second-born] 028 0423 0413 0212(0141) (0216) (0323) (0197)

d[third-born] 0146 0363 0063(0159) (0261) (0173)

d[fourth-born] minus0025 0188(0207) (0149)

d[fifth-born] minus0016(0125)

Mother minus0196 minus0167 0112 0381 minus0105Continuously (0315) (0221) (0221) (0274) (0178)married aged 1ndash14

Number of 1422 1743 1477 999 2487observations

Number of 934 835 562 328 573groups

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources766

their critical development years in a stable household as opposed to later-borns whomay experience the divorce of their parents cannot entirely explain the first-bornadvantage At the same time the persistence of birth-order effects naturally poses theproblem of their origin

The literature is not able to distinguish between different theories on the topic ofbirth order For example schooling circumstances play a large role in educational out-comes and may be related to birth order20 Our analysis is limited in the sense that itcannot discriminate between many competing hypotheses on why birth order appearsto be important21

Nevertheless we checked whether what appears as a first-born advantage predom-inantly comes from financial constraints for example parents sending their first-bornto college and running out of money for the following siblings Conley offers the fol-lowing argument ldquo[I]n terms of parental investment the cup starts to run dry as wego down the line Parental resources it appears are allotted on a first come first-served basisrdquo22 Yet if it turns out that first-borns perform better beforehand then atheory based on budget constraints cannot fully account for our results In Tables 5ndash7we estimate the probability of completing high school following the same methodol-ogy as in Tables 2ndash4 We find that first-borns have a higher probability of completinghigh school than later born siblings

Specifically Table 5 shows again that the first-born effect is not an artifact of fam-ily size that it increases in magnitude and significance when including age of motherat birth that it is robust to including characteristics of father and to restricting thesample to families in which all siblings report Table 6 shows that these results alsohold when regressions are estimated separately by family size (except for families offive presumably because of the small sample size) Tables 7a and 7b are the counter-parts of Tables 4a and 4b Fixed effects regressions controlling for age and gendersupport the results found in Tables 5 and 623

Also we estimated education at age 18 conditional on high school completion (at18 or older) to see if later-borns are more likely to repeat grades which would sup-port a theory of birth-order effects based on cognitive development differences Wedid not find any evidence for this However the small sample size resulting fromselecting index persons with available information at age 18 warrants some caution

Finally we found evidence among White families only that conditional on com-pleting high school first-borns are more likely to receive postsecondary educationYet for all races conditional on postsecondary education we found no clear advan-tage to being first-born24 In summary financial constraints do seem to play a role butsome factors early in life contribute to the first-born premium puzzle

20 We thank an anonymous referee for this point21 We refer the reader to the survey of those theories presented in Black Devereux and Salvanes (2005)22 Conley (2004) p 6923 The results held when including motherrsquos marital status We ran into the same small size problems whenadding family income and motherrsquos employment status for different age ranges of each child The ethnic dif-ferential disappears when considering all families (but persists when focusing on smaller families) suggest-ing that some of the birth order effect among White families comes from postsecondary education24 Results are available from the authors

Kantarevic and Mechoulan 767

Tabl

e 5

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

004

40

024

005

50

039

004

4(0

014

)

(00

14)

(00

14)

(0

018

)

(00

17)

To

tal n

umbe

r of

0

014

001

60

011

001

2si

blin

gs(0

003

)

(00

03)

(0

004

)

(00

05)

A

ge o

f th

e m

othe

r 0

006

000

70

005

at c

hild

birt

h(0

000

1)

(0

002

)

(00

01)

d[

Mal

e]0

047

004

80

049

004

70

037

(00

09)

(0

009

)

(00

09)

(0

009

)

(00

10)

A

ge0

024

002

90

029

003

20

023

(00

04)

(0

004

)

(00

04)

(0

007

)

(00

07)

A

ge2

000

020

0003

000

020

0003

000

02(5

times10

minus5)

(5

times10

minus5)

(5

times10

minus5)

(8

times10

minus5)

(8

times10

minus5)

d[

Whi

te]

000

50

012

002

30

042

001

9(0

014

)(0

014

)(0

014

)(0

017

)

(00

19)

d[fir

st-b

orn]

timesd

001

5minus0

005

minus00

060

011

000

6[W

hite

](0

017

5)(0

017

)(0

017

)(0

021

)(0

021

)M

othe

rrsquos

educ

atio

n0

027

002

60

026

001

50

026

(00

02)

(0

002

)

(00

02)

(0

003

)

(00

03)

d[

com

plet

e in

fo

003

50

094

002

20

029

on a

ll si

blin

gs]

(00

13)

(0

012

)

(00

13)

(00

15)

d[bo

th p

aren

ts r

epor

t]0

090

094

009

00

097

(00

12)

(0

012

)

(00

12)

(0

016

)

The Journal of Human Resources768

Tabl

e 5

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

Fath

errsquos

edu

catio

n0

011

(00

02)

A

ge o

f fa

ther

minus0

001

at c

hild

birt

h(0

002

)C

onst

ant

010

50

104

028

80

241

012

9(0

089

)(0

089

)(0

094

)

(01

35)

(01

41)

R2

008

790

0943

010

180

0885

008

55N

umbe

r of

obs

erva

tions

792

87

928

792

84

766

454

1N

umbe

r of

fam

ily

311

23

112

311

21

869

173

2cl

uste

rs

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Kantarevic and Mechoulan 769

Tabl

e 6

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

003

80

084

009

003

10

122

(00

18)

(0

025

)

(00

36)

(0

053

)(0

049

)

d[se

cond

-bor

n]0

071

008

2times

10minus4

002

2(0

022

)

(00

31)

(0

051

)(0

044

)d[

thir

d-bo

rn]

003

10

012

003

5(0

028

)(0

041

)(0

037

)d[

four

th-b

orn]

006

10

041

(00

37)

(00

33)

d[fif

th-b

orn]

000

7(0

028

)A

ge o

f th

e m

othe

r at

chi

ldbi

rth

000

30

004

000

90

008

001

1(0

002

)(0

002

)(0

003

)

(00

04)

(0

003

)

d[M

ale]

minus00

3minus0

041

000

7minus0

094

minus00

82(0

016

)(0

016

)

(00

2)(0

026

)

(00

18)

A

ge0

028

002

80

051

001

20

037

(00

07)

(0

009

)

(00

1)

(0

008

)(0

011

)

Age

23

times10

minus43

times10

minus45

times10

minus48

times10

minus54times

10minus4

(9times

10minus5

)

(10minus5

)

(10minus5

)

(9times

10minus5

)(1

0minus4)

d[

Whi

te]

001

20

124

004

90

033

002

2(0

021

)(0

023

)(0

026

)(0

034

)(0

028

)M

othe

rrsquos

educ

atio

n0

024

002

50

027

001

80

027

(00

04)

(0

004

)

(00

05)

(0

005

)

(00

04)

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 5: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

Kantarevic and Mechoulan 759

education is decreasing over cohorts which is of course untrue according to the USCensus Because a first-born is older than other siblings by definition this alonecould in theory produce a spurious positive impact of being first-born on educationWe have checked that this problem is of no consequence for our results6

The summary statistics of our sample are presented in Table 1 The detaileddescription of variables is relegated to Appendix 1 We found more than 8000 indexpersons (from more than 3100 distinct families) older than 24 in 2001 with at leastone other sibling and whose mother has completed her fertility This is to be con-trasted with Conley and Glauber (2004) who use larger sample from the US Censusbut focus only on children under the age of 20 living at home

The main dependent variablemdashyears of completed educationmdashhas an average of1262 years7 About 82 percent of those selected index persons have at least 12 years ofeducation that is to say have completed high school8 Our measure of earningsmdashloghourly wage in 2001mdashshows an average corresponding to $145hour However theinformation on hourly wages or salary income is not available for most index persons

The average age in our sample is 39 About half of respondents are male and 47percent are White The average number of siblings is 486 This high number is con-sistent with the PSID oversampling minorities and low-income populations Fifty-sixpercent of the index persons have all their siblings reporting and 60 percent have boththeir parents reporting their childbirth history

We improved our analysis by including important observed family level-effects thatvary across parity family income and whether the mother is married Wheneveravailable we constructed the corresponding information for each of the first 14 yearsof life of each index person9 The main limitation is that information on familyincome cannot be recovered for many index individuals

Lastly we include two variables describing characteristics of index individualsrsquo par-ents namely education (11 years for both mother and father on average) and the age ofthe parents at birth of index persons (26 for mothers and 29 for fathers on average)

III Methods and Results

A The First-Born Effect

We first use an OLS estimation with robust standard errors clustered by family unit(identified by the mother) which relaxes the independence assumption between the

6 The regressions presented in this article contain age controls that separate cohort from birth order effects 7 A more appropriate variable would be education at age 25 However the number of observations avail-able would drop considerably and we would not be able to run estimations by sibling size All of our otherresults when running estimations that control for siblings size hold when replacing education with educationat age 25 for those where such information can be traced8 A negligible fraction of those who declare a certain education level at some point in their life declare lesseducation later Removing such observations did not alter our results We therefore choose the latest educa-tion level reported as our variable of interest9 We then ran our regressions using the average of these variables for three age groups 1ndash6 7ndash14 and1ndash14 For each age group the averages are respectively calculated only if the information is present in atleast half of the years of the age group

The Journal of Human Resources760

error terms and requires only that the observations be independent across clusters10

In Table 2 we first test the hypothesis that being early in the birth order implies a dis-tinct advantage that is entirely due to the higher probability of coming from a smallfamily

Columns 1 and 2 of Table 2 reject that claim but help us understand why it mayhave been made In Column 1 we omit the number of siblings therefore the signif-icant coefficient on first-born reflects not only the birth order effect but also the prob-ability of coming from a small family In Column 2 the inclusion of the number ofsiblings leaves the coefficient on first-born insignificant

In Column 3 we include age of the mother at childbirth and find a positive andhighly significant effect of being first-born on years of education This effect is con-firmed when including the fatherrsquos characteristics in Column 4 where both a father anda mother report Often not all siblings report this is especially the case for large fam-ilies To check if we are biasing our results by including such families in Column 5 werestrict our attention to families with complete information on all siblings The findingsare similar there too further showing that they could not be driven by selective attri-tion within families by birth order In all specifications we find a stronger effect amongWhite families

The results presented here are for the impact of being first-born in families of morethan one child This particular procedure takes advantage of the full size of our sam-ple and can be useful when there are not enough observations to run separate estima-tions for different siblingsrsquo sizes11 We see that it allows us to reveal a significant androbust birth order effect

However we still obtain similar results when looking at individual birth ordereffects by siblingsrsquo size in Table 3 where all specifications include age of the motherat childbirth Although large families show higher birth order point estimates theeffect is present in two-sibling families as well Note that the coefficient on first-bornis only weakly significant in Column 4mdashfamilies of five siblingsmdashlikely because ofthe small sample size

The reason why the inclusion of the age of the mother at childbirth makes the coef-ficient on first-born larger in magnitude and more significant is clear The age of themother at childbirth is mechanically positively correlated with birth order and evenmore strongly across large families Conversely we see that it is positively correlatedwith a childrsquos education Then if having a high birth order carries a negative impacton education the two effects of birth order and age of the mother at childbirth com-pete against one another Therefore the coefficient on first-born in Table 2 Column 2reflects an omitted variable bias The results hold for both males and females formothers with or without more than a high school education We also found that spac-ing between births be it that of the first-born child with respect to the second-born orto the last-born child does not alter the conclusions either Additional OLS regres-sions confirmed that the effect is significantly present among White families of any

10 We also used random effects procedures but because they yield almost identical results as those withthe family clustered standard errors those are not reported11 For example we found a significant birth-order effect when considering separately mothers who firstgave birth early and mothers who first gave birth ldquolaterdquo (using various definitions of ldquoearlyrdquo and ldquolaterdquo) yetwe do not have enough observations to split the sample by maternal age and individual siblings size

Kantarevic and Mechoulan 761Ta

ble

2O

LS

Reg

ress

ion

wit

h D

epen

dent

Var

iabl

e E

duca

tion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

016

20

012

028

80

239

026

8(0

068

)

(00

69)

(00

72)

(0

098

)

(00

84)

To

tal n

umbe

r of

sib

lings

minus01

04minus0

121

minus01

08minus0

125

(00

15)

(0

015

)

(00

19)

(0

029

)

Age

of

mot

her

at c

hild

birt

h0

058

006

10

065

(00

05)

(0

011

)

(00

08)

d[

Mal

e]minus0

205

minus02

1minus0

226

minus01

86minus0

159

(00

47)

(0

046

)

(00

46)

(0

058

)

(00

57)

A

ge0

121

015

90

162

014

30

087

(00

27)

(0

027

)

(00

27)

(0

039

)

(00

40)

A

ge2

minus00

01minus0

001

minus00

01minus0

001

minus00

03(0

000

3)

(0

000

3)

(0

000

3)

(0

001

)

(00

001)

d[W

hite

]0

119

minus00

03minus0

099

minus02

95minus0

107

(00

72)

(00

73)

(00

71)

(00

87)

(0

098

)d[

first

-bor

n] times

d[W

hite

]0

126

019

30

184

028

60

172

(00

92)

(00

92)

(0

092

)

(01

15)

(0

104

)M

othe

rrsquos

educ

atio

n0

213

019

80

199

011

40

236

(00

12)

(0

012

)

(00

12)

(0

017

)

(00

18)

d[

all s

iblin

gs r

epor

t]0

261

009

60

181

014

4(0

068

)

(00

69)

(00

69)

(0

089

)d[

both

par

ents

rep

ort]

052

30

553

052

00

533

(00

63)

(0

062

)

(00

61)

(0

081

)

Fath

errsquos

edu

catio

n0

125

(00

12)

A

ge o

f fa

ther

at c

hild

birt

h0

006

(00

09)

Con

stan

t6

713

671

75

073

516

25

848

(05

28)

(05

31)

(0

541

)

(07

75)

(0

776

)

R2

015

780

1691

018

890

2009

019

82N

umbe

r of

obs

erva

tions

792

87

928

792

87

766

454

1N

umbe

r of

fam

ily c

lust

ers

311

23

112

311

21

869

173

2

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources762

Tabl

e 3

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Edu

cati

on

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

027

50

698

071

50

534

091

5(0

104

)

(01

36)

(0

185

)

(03

16)

(02

36)

d[

seco

nd-b

orn]

043

40

463

033

40

578

(01

17)

(0

16)

(02

82)

(01

95)

d[

thir

d-bo

rn]

017

10

212

036

8(0

14)

(02

47)

(01

75)

d[

four

th-b

orn]

minus00

120

466

(02

14)

(01

52)

d[

fifth

-bor

n]0

221

(01

33)

Age

of

the

mot

her

at c

hild

birt

h0

048

007

90

067

006

80

07(0

012

)

(00

13)

(0

013

)

(00

23)

(0

013

)

d[M

ale]

minus02

87minus0

115

minus00

35minus0

34

minus00

33(0

102

)

(00

91)

(01

07)

(01

31)

(0

087

)

Age

021

70

190

156

007

50

167

(00

56)

(0

057

)

(00

49)

(0

044

)(0

057

)

Kantarevic and Mechoulan 763

Age

20

002

000

20

002

3times

10minus4

000

2(7

times10

minus4)

(7

times10

minus4)

(6

times10

minus4)

(5

times10

minus4)

(8times

10minus4

)

d[W

hite

]0

119

minus01

244

times10

minus5minus0

03

minus01

14(0

12)

(01

3)(0

13)

(01

81)

(01

37)

Mot

herrsquo

s ed

ucat

ion

025

70

250

179

012

40

182

(00

25)

(0

026

)

(00

23)

(0

035

)

(00

22)

d[

all s

iblin

gs r

epor

t]0

147

021

2minus0

066

035

10

21(0

154

)(0

136

)(0

128

)(0

178

)

(01

49)

d[bo

th p

aren

ts r

epor

t]0

241

061

20

755

054

20

371

(01

18)

(0

118

)

(01

26)

(0

189

)

(01

27)

C

onst

ant

333

92

694

425

86

607

412

(11

48)

(1

184

)

(10

11)

(1

293

)

(14

15)

R

20

185

021

019

011

40

127

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources764

size greater than one whereas it is only present within large families among BlacksTherefore the ethnic differential disappears when considering large families only

As noted earlier a causal interpretation of the age of the mother at childbirth wouldhinge on the assumption of its exogeneity Without instrumental variables or a treat-ment versus control quasi experiment it is difficult to draw conclusions12 The age ofthe mother at childbirth itself positively correlated with birth order could easilyproxy for other unobserved variables such as level of human capital and parentalresources13

To address this problem our fixed-effect estimation (Table 4a and b) removes fam-ily characteristics and unobserved family-level heterogeneity14 Family fixed effectsaddress family unobservables to the extent that they are constant over time While wetry to incorporate observables that vary across birth order to affirm the robustness ofour results we are constrained by the availability of such variables in the data set

Unfortunately the coefficients on age or on age of the mother at childbirth are unin-formative in those fixed effects regressions Deviations from family means for ageconvey the same information as deviations from family means for age of mother atchildbirth We thus do not provide separate estimations with age and age of motherat childbirth The age of the mother at first childbirth is certainly relevant but here itis differenced out Still to assess this issue better we ran separate regressions split-ting the sample by maternal age15 To summarize the following results are robust toexcluding mothers who first gave birth as teens but we do not have enough observa-tions to meaningfully run the fixed effects regressions by sibling size on that lattergroup

In Tables 4a and 4b the fixed effects estimations that only control for age and gen-der confirm the previous results for the most part16 We also provide estimations includ-ing the marital status of the mother17 during the childrsquos first 14 years18 The results donot change by much19 Clearly the suggestion that first-borns are most likely to live

12 Also while it is possible in the context of larger samples to instrument for siblings sizemdashusing twinbirths (Black Devereux and Salvanes 2005) or using the fact that parents of two same sex children are morelikely to have a third child (Conley and Glauber 2004)mdashone cannot instrument for birth order per se13 This is further evidenced by the fact that including a dummy variable indicating whether the mother wasmarried at the time of childbirth instead of (and obviously also along with) age of mother at childbirth alsoresults in highly significant birth order coefficients Results are available upon request14 However it is worth noting that those fixed effects do not solve all endogeneity issues For example itmay be the case that first-born ldquoqualityrdquo is affecting subsequent fertility We thank an anonymous referee forpointing out this caveat15 We thank an anonymous referee for this suggestion16 In Column 1mdashfamilies of two siblingsmdashthe coefficient on first-born becomes significant at the 10 per-cent level when including index persons of 23 and 24 years of age suggesting the nonsignificance whenrestricting at age 25 stems from a small sample size Separate FE regressions for White and Black familiesconfirmed the ethnic differentials found earlier17 We tried two definitions number of years marriednumber of years considered in the age group and = 1if continuously married over the years considered in the age group 0 otherwise Since the results on birthorder do not change qualitatively with either of those we only report the results with the latter definition18 We also ran similar estimations with different age ranges and obtained similar results19 Because the information on marital status is retrospective we do not lose any observation by includingthis covariate Unfortunately this is not the case when adding average family income or average employmentstatus of the mother during the childrsquos first years

Kantarevic and Mechoulan 765

Table 4aFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Not Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0216 0437 0683 0594 0477(0181) (0206) (0265) (0379) (0224)

d[second- 0282 0412 0421 0199born] (0141) (0214) (0323) (0195)

d[third- 0138 0361 0055born] (0159) (0261) (0172)

d[fourth- minus0031 0181born] (0207) (0149)

d[fifth- minus0012born] (0125)

Observations 1422 1743 1477 922 2487groups 934 835 562 388 573

Table 4bFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0224 0436 0696 0585 0492(0181) (0206) (0267) (0379) (0226)

d[second-born] 028 0423 0413 0212(0141) (0216) (0323) (0197)

d[third-born] 0146 0363 0063(0159) (0261) (0173)

d[fourth-born] minus0025 0188(0207) (0149)

d[fifth-born] minus0016(0125)

Mother minus0196 minus0167 0112 0381 minus0105Continuously (0315) (0221) (0221) (0274) (0178)married aged 1ndash14

Number of 1422 1743 1477 999 2487observations

Number of 934 835 562 328 573groups

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources766

their critical development years in a stable household as opposed to later-borns whomay experience the divorce of their parents cannot entirely explain the first-bornadvantage At the same time the persistence of birth-order effects naturally poses theproblem of their origin

The literature is not able to distinguish between different theories on the topic ofbirth order For example schooling circumstances play a large role in educational out-comes and may be related to birth order20 Our analysis is limited in the sense that itcannot discriminate between many competing hypotheses on why birth order appearsto be important21

Nevertheless we checked whether what appears as a first-born advantage predom-inantly comes from financial constraints for example parents sending their first-bornto college and running out of money for the following siblings Conley offers the fol-lowing argument ldquo[I]n terms of parental investment the cup starts to run dry as wego down the line Parental resources it appears are allotted on a first come first-served basisrdquo22 Yet if it turns out that first-borns perform better beforehand then atheory based on budget constraints cannot fully account for our results In Tables 5ndash7we estimate the probability of completing high school following the same methodol-ogy as in Tables 2ndash4 We find that first-borns have a higher probability of completinghigh school than later born siblings

Specifically Table 5 shows again that the first-born effect is not an artifact of fam-ily size that it increases in magnitude and significance when including age of motherat birth that it is robust to including characteristics of father and to restricting thesample to families in which all siblings report Table 6 shows that these results alsohold when regressions are estimated separately by family size (except for families offive presumably because of the small sample size) Tables 7a and 7b are the counter-parts of Tables 4a and 4b Fixed effects regressions controlling for age and gendersupport the results found in Tables 5 and 623

Also we estimated education at age 18 conditional on high school completion (at18 or older) to see if later-borns are more likely to repeat grades which would sup-port a theory of birth-order effects based on cognitive development differences Wedid not find any evidence for this However the small sample size resulting fromselecting index persons with available information at age 18 warrants some caution

Finally we found evidence among White families only that conditional on com-pleting high school first-borns are more likely to receive postsecondary educationYet for all races conditional on postsecondary education we found no clear advan-tage to being first-born24 In summary financial constraints do seem to play a role butsome factors early in life contribute to the first-born premium puzzle

20 We thank an anonymous referee for this point21 We refer the reader to the survey of those theories presented in Black Devereux and Salvanes (2005)22 Conley (2004) p 6923 The results held when including motherrsquos marital status We ran into the same small size problems whenadding family income and motherrsquos employment status for different age ranges of each child The ethnic dif-ferential disappears when considering all families (but persists when focusing on smaller families) suggest-ing that some of the birth order effect among White families comes from postsecondary education24 Results are available from the authors

Kantarevic and Mechoulan 767

Tabl

e 5

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

004

40

024

005

50

039

004

4(0

014

)

(00

14)

(00

14)

(0

018

)

(00

17)

To

tal n

umbe

r of

0

014

001

60

011

001

2si

blin

gs(0

003

)

(00

03)

(0

004

)

(00

05)

A

ge o

f th

e m

othe

r 0

006

000

70

005

at c

hild

birt

h(0

000

1)

(0

002

)

(00

01)

d[

Mal

e]0

047

004

80

049

004

70

037

(00

09)

(0

009

)

(00

09)

(0

009

)

(00

10)

A

ge0

024

002

90

029

003

20

023

(00

04)

(0

004

)

(00

04)

(0

007

)

(00

07)

A

ge2

000

020

0003

000

020

0003

000

02(5

times10

minus5)

(5

times10

minus5)

(5

times10

minus5)

(8

times10

minus5)

(8

times10

minus5)

d[

Whi

te]

000

50

012

002

30

042

001

9(0

014

)(0

014

)(0

014

)(0

017

)

(00

19)

d[fir

st-b

orn]

timesd

001

5minus0

005

minus00

060

011

000

6[W

hite

](0

017

5)(0

017

)(0

017

)(0

021

)(0

021

)M

othe

rrsquos

educ

atio

n0

027

002

60

026

001

50

026

(00

02)

(0

002

)

(00

02)

(0

003

)

(00

03)

d[

com

plet

e in

fo

003

50

094

002

20

029

on a

ll si

blin

gs]

(00

13)

(0

012

)

(00

13)

(00

15)

d[bo

th p

aren

ts r

epor

t]0

090

094

009

00

097

(00

12)

(0

012

)

(00

12)

(0

016

)

The Journal of Human Resources768

Tabl

e 5

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

Fath

errsquos

edu

catio

n0

011

(00

02)

A

ge o

f fa

ther

minus0

001

at c

hild

birt

h(0

002

)C

onst

ant

010

50

104

028

80

241

012

9(0

089

)(0

089

)(0

094

)

(01

35)

(01

41)

R2

008

790

0943

010

180

0885

008

55N

umbe

r of

obs

erva

tions

792

87

928

792

84

766

454

1N

umbe

r of

fam

ily

311

23

112

311

21

869

173

2cl

uste

rs

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Kantarevic and Mechoulan 769

Tabl

e 6

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

003

80

084

009

003

10

122

(00

18)

(0

025

)

(00

36)

(0

053

)(0

049

)

d[se

cond

-bor

n]0

071

008

2times

10minus4

002

2(0

022

)

(00

31)

(0

051

)(0

044

)d[

thir

d-bo

rn]

003

10

012

003

5(0

028

)(0

041

)(0

037

)d[

four

th-b

orn]

006

10

041

(00

37)

(00

33)

d[fif

th-b

orn]

000

7(0

028

)A

ge o

f th

e m

othe

r at

chi

ldbi

rth

000

30

004

000

90

008

001

1(0

002

)(0

002

)(0

003

)

(00

04)

(0

003

)

d[M

ale]

minus00

3minus0

041

000

7minus0

094

minus00

82(0

016

)(0

016

)

(00

2)(0

026

)

(00

18)

A

ge0

028

002

80

051

001

20

037

(00

07)

(0

009

)

(00

1)

(0

008

)(0

011

)

Age

23

times10

minus43

times10

minus45

times10

minus48

times10

minus54times

10minus4

(9times

10minus5

)

(10minus5

)

(10minus5

)

(9times

10minus5

)(1

0minus4)

d[

Whi

te]

001

20

124

004

90

033

002

2(0

021

)(0

023

)(0

026

)(0

034

)(0

028

)M

othe

rrsquos

educ

atio

n0

024

002

50

027

001

80

027

(00

04)

(0

004

)

(00

05)

(0

005

)

(00

04)

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 6: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

The Journal of Human Resources760

error terms and requires only that the observations be independent across clusters10

In Table 2 we first test the hypothesis that being early in the birth order implies a dis-tinct advantage that is entirely due to the higher probability of coming from a smallfamily

Columns 1 and 2 of Table 2 reject that claim but help us understand why it mayhave been made In Column 1 we omit the number of siblings therefore the signif-icant coefficient on first-born reflects not only the birth order effect but also the prob-ability of coming from a small family In Column 2 the inclusion of the number ofsiblings leaves the coefficient on first-born insignificant

In Column 3 we include age of the mother at childbirth and find a positive andhighly significant effect of being first-born on years of education This effect is con-firmed when including the fatherrsquos characteristics in Column 4 where both a father anda mother report Often not all siblings report this is especially the case for large fam-ilies To check if we are biasing our results by including such families in Column 5 werestrict our attention to families with complete information on all siblings The findingsare similar there too further showing that they could not be driven by selective attri-tion within families by birth order In all specifications we find a stronger effect amongWhite families

The results presented here are for the impact of being first-born in families of morethan one child This particular procedure takes advantage of the full size of our sam-ple and can be useful when there are not enough observations to run separate estima-tions for different siblingsrsquo sizes11 We see that it allows us to reveal a significant androbust birth order effect

However we still obtain similar results when looking at individual birth ordereffects by siblingsrsquo size in Table 3 where all specifications include age of the motherat childbirth Although large families show higher birth order point estimates theeffect is present in two-sibling families as well Note that the coefficient on first-bornis only weakly significant in Column 4mdashfamilies of five siblingsmdashlikely because ofthe small sample size

The reason why the inclusion of the age of the mother at childbirth makes the coef-ficient on first-born larger in magnitude and more significant is clear The age of themother at childbirth is mechanically positively correlated with birth order and evenmore strongly across large families Conversely we see that it is positively correlatedwith a childrsquos education Then if having a high birth order carries a negative impacton education the two effects of birth order and age of the mother at childbirth com-pete against one another Therefore the coefficient on first-born in Table 2 Column 2reflects an omitted variable bias The results hold for both males and females formothers with or without more than a high school education We also found that spac-ing between births be it that of the first-born child with respect to the second-born orto the last-born child does not alter the conclusions either Additional OLS regres-sions confirmed that the effect is significantly present among White families of any

10 We also used random effects procedures but because they yield almost identical results as those withthe family clustered standard errors those are not reported11 For example we found a significant birth-order effect when considering separately mothers who firstgave birth early and mothers who first gave birth ldquolaterdquo (using various definitions of ldquoearlyrdquo and ldquolaterdquo) yetwe do not have enough observations to split the sample by maternal age and individual siblings size

Kantarevic and Mechoulan 761Ta

ble

2O

LS

Reg

ress

ion

wit

h D

epen

dent

Var

iabl

e E

duca

tion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

016

20

012

028

80

239

026

8(0

068

)

(00

69)

(00

72)

(0

098

)

(00

84)

To

tal n

umbe

r of

sib

lings

minus01

04minus0

121

minus01

08minus0

125

(00

15)

(0

015

)

(00

19)

(0

029

)

Age

of

mot

her

at c

hild

birt

h0

058

006

10

065

(00

05)

(0

011

)

(00

08)

d[

Mal

e]minus0

205

minus02

1minus0

226

minus01

86minus0

159

(00

47)

(0

046

)

(00

46)

(0

058

)

(00

57)

A

ge0

121

015

90

162

014

30

087

(00

27)

(0

027

)

(00

27)

(0

039

)

(00

40)

A

ge2

minus00

01minus0

001

minus00

01minus0

001

minus00

03(0

000

3)

(0

000

3)

(0

000

3)

(0

001

)

(00

001)

d[W

hite

]0

119

minus00

03minus0

099

minus02

95minus0

107

(00

72)

(00

73)

(00

71)

(00

87)

(0

098

)d[

first

-bor

n] times

d[W

hite

]0

126

019

30

184

028

60

172

(00

92)

(00

92)

(0

092

)

(01

15)

(0

104

)M

othe

rrsquos

educ

atio

n0

213

019

80

199

011

40

236

(00

12)

(0

012

)

(00

12)

(0

017

)

(00

18)

d[

all s

iblin

gs r

epor

t]0

261

009

60

181

014

4(0

068

)

(00

69)

(00

69)

(0

089

)d[

both

par

ents

rep

ort]

052

30

553

052

00

533

(00

63)

(0

062

)

(00

61)

(0

081

)

Fath

errsquos

edu

catio

n0

125

(00

12)

A

ge o

f fa

ther

at c

hild

birt

h0

006

(00

09)

Con

stan

t6

713

671

75

073

516

25

848

(05

28)

(05

31)

(0

541

)

(07

75)

(0

776

)

R2

015

780

1691

018

890

2009

019

82N

umbe

r of

obs

erva

tions

792

87

928

792

87

766

454

1N

umbe

r of

fam

ily c

lust

ers

311

23

112

311

21

869

173

2

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources762

Tabl

e 3

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Edu

cati

on

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

027

50

698

071

50

534

091

5(0

104

)

(01

36)

(0

185

)

(03

16)

(02

36)

d[

seco

nd-b

orn]

043

40

463

033

40

578

(01

17)

(0

16)

(02

82)

(01

95)

d[

thir

d-bo

rn]

017

10

212

036

8(0

14)

(02

47)

(01

75)

d[

four

th-b

orn]

minus00

120

466

(02

14)

(01

52)

d[

fifth

-bor

n]0

221

(01

33)

Age

of

the

mot

her

at c

hild

birt

h0

048

007

90

067

006

80

07(0

012

)

(00

13)

(0

013

)

(00

23)

(0

013

)

d[M

ale]

minus02

87minus0

115

minus00

35minus0

34

minus00

33(0

102

)

(00

91)

(01

07)

(01

31)

(0

087

)

Age

021

70

190

156

007

50

167

(00

56)

(0

057

)

(00

49)

(0

044

)(0

057

)

Kantarevic and Mechoulan 763

Age

20

002

000

20

002

3times

10minus4

000

2(7

times10

minus4)

(7

times10

minus4)

(6

times10

minus4)

(5

times10

minus4)

(8times

10minus4

)

d[W

hite

]0

119

minus01

244

times10

minus5minus0

03

minus01

14(0

12)

(01

3)(0

13)

(01

81)

(01

37)

Mot

herrsquo

s ed

ucat

ion

025

70

250

179

012

40

182

(00

25)

(0

026

)

(00

23)

(0

035

)

(00

22)

d[

all s

iblin

gs r

epor

t]0

147

021

2minus0

066

035

10

21(0

154

)(0

136

)(0

128

)(0

178

)

(01

49)

d[bo

th p

aren

ts r

epor

t]0

241

061

20

755

054

20

371

(01

18)

(0

118

)

(01

26)

(0

189

)

(01

27)

C

onst

ant

333

92

694

425

86

607

412

(11

48)

(1

184

)

(10

11)

(1

293

)

(14

15)

R

20

185

021

019

011

40

127

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources764

size greater than one whereas it is only present within large families among BlacksTherefore the ethnic differential disappears when considering large families only

As noted earlier a causal interpretation of the age of the mother at childbirth wouldhinge on the assumption of its exogeneity Without instrumental variables or a treat-ment versus control quasi experiment it is difficult to draw conclusions12 The age ofthe mother at childbirth itself positively correlated with birth order could easilyproxy for other unobserved variables such as level of human capital and parentalresources13

To address this problem our fixed-effect estimation (Table 4a and b) removes fam-ily characteristics and unobserved family-level heterogeneity14 Family fixed effectsaddress family unobservables to the extent that they are constant over time While wetry to incorporate observables that vary across birth order to affirm the robustness ofour results we are constrained by the availability of such variables in the data set

Unfortunately the coefficients on age or on age of the mother at childbirth are unin-formative in those fixed effects regressions Deviations from family means for ageconvey the same information as deviations from family means for age of mother atchildbirth We thus do not provide separate estimations with age and age of motherat childbirth The age of the mother at first childbirth is certainly relevant but here itis differenced out Still to assess this issue better we ran separate regressions split-ting the sample by maternal age15 To summarize the following results are robust toexcluding mothers who first gave birth as teens but we do not have enough observa-tions to meaningfully run the fixed effects regressions by sibling size on that lattergroup

In Tables 4a and 4b the fixed effects estimations that only control for age and gen-der confirm the previous results for the most part16 We also provide estimations includ-ing the marital status of the mother17 during the childrsquos first 14 years18 The results donot change by much19 Clearly the suggestion that first-borns are most likely to live

12 Also while it is possible in the context of larger samples to instrument for siblings sizemdashusing twinbirths (Black Devereux and Salvanes 2005) or using the fact that parents of two same sex children are morelikely to have a third child (Conley and Glauber 2004)mdashone cannot instrument for birth order per se13 This is further evidenced by the fact that including a dummy variable indicating whether the mother wasmarried at the time of childbirth instead of (and obviously also along with) age of mother at childbirth alsoresults in highly significant birth order coefficients Results are available upon request14 However it is worth noting that those fixed effects do not solve all endogeneity issues For example itmay be the case that first-born ldquoqualityrdquo is affecting subsequent fertility We thank an anonymous referee forpointing out this caveat15 We thank an anonymous referee for this suggestion16 In Column 1mdashfamilies of two siblingsmdashthe coefficient on first-born becomes significant at the 10 per-cent level when including index persons of 23 and 24 years of age suggesting the nonsignificance whenrestricting at age 25 stems from a small sample size Separate FE regressions for White and Black familiesconfirmed the ethnic differentials found earlier17 We tried two definitions number of years marriednumber of years considered in the age group and = 1if continuously married over the years considered in the age group 0 otherwise Since the results on birthorder do not change qualitatively with either of those we only report the results with the latter definition18 We also ran similar estimations with different age ranges and obtained similar results19 Because the information on marital status is retrospective we do not lose any observation by includingthis covariate Unfortunately this is not the case when adding average family income or average employmentstatus of the mother during the childrsquos first years

Kantarevic and Mechoulan 765

Table 4aFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Not Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0216 0437 0683 0594 0477(0181) (0206) (0265) (0379) (0224)

d[second- 0282 0412 0421 0199born] (0141) (0214) (0323) (0195)

d[third- 0138 0361 0055born] (0159) (0261) (0172)

d[fourth- minus0031 0181born] (0207) (0149)

d[fifth- minus0012born] (0125)

Observations 1422 1743 1477 922 2487groups 934 835 562 388 573

Table 4bFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0224 0436 0696 0585 0492(0181) (0206) (0267) (0379) (0226)

d[second-born] 028 0423 0413 0212(0141) (0216) (0323) (0197)

d[third-born] 0146 0363 0063(0159) (0261) (0173)

d[fourth-born] minus0025 0188(0207) (0149)

d[fifth-born] minus0016(0125)

Mother minus0196 minus0167 0112 0381 minus0105Continuously (0315) (0221) (0221) (0274) (0178)married aged 1ndash14

Number of 1422 1743 1477 999 2487observations

Number of 934 835 562 328 573groups

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources766

their critical development years in a stable household as opposed to later-borns whomay experience the divorce of their parents cannot entirely explain the first-bornadvantage At the same time the persistence of birth-order effects naturally poses theproblem of their origin

The literature is not able to distinguish between different theories on the topic ofbirth order For example schooling circumstances play a large role in educational out-comes and may be related to birth order20 Our analysis is limited in the sense that itcannot discriminate between many competing hypotheses on why birth order appearsto be important21

Nevertheless we checked whether what appears as a first-born advantage predom-inantly comes from financial constraints for example parents sending their first-bornto college and running out of money for the following siblings Conley offers the fol-lowing argument ldquo[I]n terms of parental investment the cup starts to run dry as wego down the line Parental resources it appears are allotted on a first come first-served basisrdquo22 Yet if it turns out that first-borns perform better beforehand then atheory based on budget constraints cannot fully account for our results In Tables 5ndash7we estimate the probability of completing high school following the same methodol-ogy as in Tables 2ndash4 We find that first-borns have a higher probability of completinghigh school than later born siblings

Specifically Table 5 shows again that the first-born effect is not an artifact of fam-ily size that it increases in magnitude and significance when including age of motherat birth that it is robust to including characteristics of father and to restricting thesample to families in which all siblings report Table 6 shows that these results alsohold when regressions are estimated separately by family size (except for families offive presumably because of the small sample size) Tables 7a and 7b are the counter-parts of Tables 4a and 4b Fixed effects regressions controlling for age and gendersupport the results found in Tables 5 and 623

Also we estimated education at age 18 conditional on high school completion (at18 or older) to see if later-borns are more likely to repeat grades which would sup-port a theory of birth-order effects based on cognitive development differences Wedid not find any evidence for this However the small sample size resulting fromselecting index persons with available information at age 18 warrants some caution

Finally we found evidence among White families only that conditional on com-pleting high school first-borns are more likely to receive postsecondary educationYet for all races conditional on postsecondary education we found no clear advan-tage to being first-born24 In summary financial constraints do seem to play a role butsome factors early in life contribute to the first-born premium puzzle

20 We thank an anonymous referee for this point21 We refer the reader to the survey of those theories presented in Black Devereux and Salvanes (2005)22 Conley (2004) p 6923 The results held when including motherrsquos marital status We ran into the same small size problems whenadding family income and motherrsquos employment status for different age ranges of each child The ethnic dif-ferential disappears when considering all families (but persists when focusing on smaller families) suggest-ing that some of the birth order effect among White families comes from postsecondary education24 Results are available from the authors

Kantarevic and Mechoulan 767

Tabl

e 5

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

004

40

024

005

50

039

004

4(0

014

)

(00

14)

(00

14)

(0

018

)

(00

17)

To

tal n

umbe

r of

0

014

001

60

011

001

2si

blin

gs(0

003

)

(00

03)

(0

004

)

(00

05)

A

ge o

f th

e m

othe

r 0

006

000

70

005

at c

hild

birt

h(0

000

1)

(0

002

)

(00

01)

d[

Mal

e]0

047

004

80

049

004

70

037

(00

09)

(0

009

)

(00

09)

(0

009

)

(00

10)

A

ge0

024

002

90

029

003

20

023

(00

04)

(0

004

)

(00

04)

(0

007

)

(00

07)

A

ge2

000

020

0003

000

020

0003

000

02(5

times10

minus5)

(5

times10

minus5)

(5

times10

minus5)

(8

times10

minus5)

(8

times10

minus5)

d[

Whi

te]

000

50

012

002

30

042

001

9(0

014

)(0

014

)(0

014

)(0

017

)

(00

19)

d[fir

st-b

orn]

timesd

001

5minus0

005

minus00

060

011

000

6[W

hite

](0

017

5)(0

017

)(0

017

)(0

021

)(0

021

)M

othe

rrsquos

educ

atio

n0

027

002

60

026

001

50

026

(00

02)

(0

002

)

(00

02)

(0

003

)

(00

03)

d[

com

plet

e in

fo

003

50

094

002

20

029

on a

ll si

blin

gs]

(00

13)

(0

012

)

(00

13)

(00

15)

d[bo

th p

aren

ts r

epor

t]0

090

094

009

00

097

(00

12)

(0

012

)

(00

12)

(0

016

)

The Journal of Human Resources768

Tabl

e 5

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

Fath

errsquos

edu

catio

n0

011

(00

02)

A

ge o

f fa

ther

minus0

001

at c

hild

birt

h(0

002

)C

onst

ant

010

50

104

028

80

241

012

9(0

089

)(0

089

)(0

094

)

(01

35)

(01

41)

R2

008

790

0943

010

180

0885

008

55N

umbe

r of

obs

erva

tions

792

87

928

792

84

766

454

1N

umbe

r of

fam

ily

311

23

112

311

21

869

173

2cl

uste

rs

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Kantarevic and Mechoulan 769

Tabl

e 6

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

003

80

084

009

003

10

122

(00

18)

(0

025

)

(00

36)

(0

053

)(0

049

)

d[se

cond

-bor

n]0

071

008

2times

10minus4

002

2(0

022

)

(00

31)

(0

051

)(0

044

)d[

thir

d-bo

rn]

003

10

012

003

5(0

028

)(0

041

)(0

037

)d[

four

th-b

orn]

006

10

041

(00

37)

(00

33)

d[fif

th-b

orn]

000

7(0

028

)A

ge o

f th

e m

othe

r at

chi

ldbi

rth

000

30

004

000

90

008

001

1(0

002

)(0

002

)(0

003

)

(00

04)

(0

003

)

d[M

ale]

minus00

3minus0

041

000

7minus0

094

minus00

82(0

016

)(0

016

)

(00

2)(0

026

)

(00

18)

A

ge0

028

002

80

051

001

20

037

(00

07)

(0

009

)

(00

1)

(0

008

)(0

011

)

Age

23

times10

minus43

times10

minus45

times10

minus48

times10

minus54times

10minus4

(9times

10minus5

)

(10minus5

)

(10minus5

)

(9times

10minus5

)(1

0minus4)

d[

Whi

te]

001

20

124

004

90

033

002

2(0

021

)(0

023

)(0

026

)(0

034

)(0

028

)M

othe

rrsquos

educ

atio

n0

024

002

50

027

001

80

027

(00

04)

(0

004

)

(00

05)

(0

005

)

(00

04)

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 7: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

Kantarevic and Mechoulan 761Ta

ble

2O

LS

Reg

ress

ion

wit

h D

epen

dent

Var

iabl

e E

duca

tion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

016

20

012

028

80

239

026

8(0

068

)

(00

69)

(00

72)

(0

098

)

(00

84)

To

tal n

umbe

r of

sib

lings

minus01

04minus0

121

minus01

08minus0

125

(00

15)

(0

015

)

(00

19)

(0

029

)

Age

of

mot

her

at c

hild

birt

h0

058

006

10

065

(00

05)

(0

011

)

(00

08)

d[

Mal

e]minus0

205

minus02

1minus0

226

minus01

86minus0

159

(00

47)

(0

046

)

(00

46)

(0

058

)

(00

57)

A

ge0

121

015

90

162

014

30

087

(00

27)

(0

027

)

(00

27)

(0

039

)

(00

40)

A

ge2

minus00

01minus0

001

minus00

01minus0

001

minus00

03(0

000

3)

(0

000

3)

(0

000

3)

(0

001

)

(00

001)

d[W

hite

]0

119

minus00

03minus0

099

minus02

95minus0

107

(00

72)

(00

73)

(00

71)

(00

87)

(0

098

)d[

first

-bor

n] times

d[W

hite

]0

126

019

30

184

028

60

172

(00

92)

(00

92)

(0

092

)

(01

15)

(0

104

)M

othe

rrsquos

educ

atio

n0

213

019

80

199

011

40

236

(00

12)

(0

012

)

(00

12)

(0

017

)

(00

18)

d[

all s

iblin

gs r

epor

t]0

261

009

60

181

014

4(0

068

)

(00

69)

(00

69)

(0

089

)d[

both

par

ents

rep

ort]

052

30

553

052

00

533

(00

63)

(0

062

)

(00

61)

(0

081

)

Fath

errsquos

edu

catio

n0

125

(00

12)

A

ge o

f fa

ther

at c

hild

birt

h0

006

(00

09)

Con

stan

t6

713

671

75

073

516

25

848

(05

28)

(05

31)

(0

541

)

(07

75)

(0

776

)

R2

015

780

1691

018

890

2009

019

82N

umbe

r of

obs

erva

tions

792

87

928

792

87

766

454

1N

umbe

r of

fam

ily c

lust

ers

311

23

112

311

21

869

173

2

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources762

Tabl

e 3

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Edu

cati

on

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

027

50

698

071

50

534

091

5(0

104

)

(01

36)

(0

185

)

(03

16)

(02

36)

d[

seco

nd-b

orn]

043

40

463

033

40

578

(01

17)

(0

16)

(02

82)

(01

95)

d[

thir

d-bo

rn]

017

10

212

036

8(0

14)

(02

47)

(01

75)

d[

four

th-b

orn]

minus00

120

466

(02

14)

(01

52)

d[

fifth

-bor

n]0

221

(01

33)

Age

of

the

mot

her

at c

hild

birt

h0

048

007

90

067

006

80

07(0

012

)

(00

13)

(0

013

)

(00

23)

(0

013

)

d[M

ale]

minus02

87minus0

115

minus00

35minus0

34

minus00

33(0

102

)

(00

91)

(01

07)

(01

31)

(0

087

)

Age

021

70

190

156

007

50

167

(00

56)

(0

057

)

(00

49)

(0

044

)(0

057

)

Kantarevic and Mechoulan 763

Age

20

002

000

20

002

3times

10minus4

000

2(7

times10

minus4)

(7

times10

minus4)

(6

times10

minus4)

(5

times10

minus4)

(8times

10minus4

)

d[W

hite

]0

119

minus01

244

times10

minus5minus0

03

minus01

14(0

12)

(01

3)(0

13)

(01

81)

(01

37)

Mot

herrsquo

s ed

ucat

ion

025

70

250

179

012

40

182

(00

25)

(0

026

)

(00

23)

(0

035

)

(00

22)

d[

all s

iblin

gs r

epor

t]0

147

021

2minus0

066

035

10

21(0

154

)(0

136

)(0

128

)(0

178

)

(01

49)

d[bo

th p

aren

ts r

epor

t]0

241

061

20

755

054

20

371

(01

18)

(0

118

)

(01

26)

(0

189

)

(01

27)

C

onst

ant

333

92

694

425

86

607

412

(11

48)

(1

184

)

(10

11)

(1

293

)

(14

15)

R

20

185

021

019

011

40

127

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources764

size greater than one whereas it is only present within large families among BlacksTherefore the ethnic differential disappears when considering large families only

As noted earlier a causal interpretation of the age of the mother at childbirth wouldhinge on the assumption of its exogeneity Without instrumental variables or a treat-ment versus control quasi experiment it is difficult to draw conclusions12 The age ofthe mother at childbirth itself positively correlated with birth order could easilyproxy for other unobserved variables such as level of human capital and parentalresources13

To address this problem our fixed-effect estimation (Table 4a and b) removes fam-ily characteristics and unobserved family-level heterogeneity14 Family fixed effectsaddress family unobservables to the extent that they are constant over time While wetry to incorporate observables that vary across birth order to affirm the robustness ofour results we are constrained by the availability of such variables in the data set

Unfortunately the coefficients on age or on age of the mother at childbirth are unin-formative in those fixed effects regressions Deviations from family means for ageconvey the same information as deviations from family means for age of mother atchildbirth We thus do not provide separate estimations with age and age of motherat childbirth The age of the mother at first childbirth is certainly relevant but here itis differenced out Still to assess this issue better we ran separate regressions split-ting the sample by maternal age15 To summarize the following results are robust toexcluding mothers who first gave birth as teens but we do not have enough observa-tions to meaningfully run the fixed effects regressions by sibling size on that lattergroup

In Tables 4a and 4b the fixed effects estimations that only control for age and gen-der confirm the previous results for the most part16 We also provide estimations includ-ing the marital status of the mother17 during the childrsquos first 14 years18 The results donot change by much19 Clearly the suggestion that first-borns are most likely to live

12 Also while it is possible in the context of larger samples to instrument for siblings sizemdashusing twinbirths (Black Devereux and Salvanes 2005) or using the fact that parents of two same sex children are morelikely to have a third child (Conley and Glauber 2004)mdashone cannot instrument for birth order per se13 This is further evidenced by the fact that including a dummy variable indicating whether the mother wasmarried at the time of childbirth instead of (and obviously also along with) age of mother at childbirth alsoresults in highly significant birth order coefficients Results are available upon request14 However it is worth noting that those fixed effects do not solve all endogeneity issues For example itmay be the case that first-born ldquoqualityrdquo is affecting subsequent fertility We thank an anonymous referee forpointing out this caveat15 We thank an anonymous referee for this suggestion16 In Column 1mdashfamilies of two siblingsmdashthe coefficient on first-born becomes significant at the 10 per-cent level when including index persons of 23 and 24 years of age suggesting the nonsignificance whenrestricting at age 25 stems from a small sample size Separate FE regressions for White and Black familiesconfirmed the ethnic differentials found earlier17 We tried two definitions number of years marriednumber of years considered in the age group and = 1if continuously married over the years considered in the age group 0 otherwise Since the results on birthorder do not change qualitatively with either of those we only report the results with the latter definition18 We also ran similar estimations with different age ranges and obtained similar results19 Because the information on marital status is retrospective we do not lose any observation by includingthis covariate Unfortunately this is not the case when adding average family income or average employmentstatus of the mother during the childrsquos first years

Kantarevic and Mechoulan 765

Table 4aFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Not Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0216 0437 0683 0594 0477(0181) (0206) (0265) (0379) (0224)

d[second- 0282 0412 0421 0199born] (0141) (0214) (0323) (0195)

d[third- 0138 0361 0055born] (0159) (0261) (0172)

d[fourth- minus0031 0181born] (0207) (0149)

d[fifth- minus0012born] (0125)

Observations 1422 1743 1477 922 2487groups 934 835 562 388 573

Table 4bFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0224 0436 0696 0585 0492(0181) (0206) (0267) (0379) (0226)

d[second-born] 028 0423 0413 0212(0141) (0216) (0323) (0197)

d[third-born] 0146 0363 0063(0159) (0261) (0173)

d[fourth-born] minus0025 0188(0207) (0149)

d[fifth-born] minus0016(0125)

Mother minus0196 minus0167 0112 0381 minus0105Continuously (0315) (0221) (0221) (0274) (0178)married aged 1ndash14

Number of 1422 1743 1477 999 2487observations

Number of 934 835 562 328 573groups

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources766

their critical development years in a stable household as opposed to later-borns whomay experience the divorce of their parents cannot entirely explain the first-bornadvantage At the same time the persistence of birth-order effects naturally poses theproblem of their origin

The literature is not able to distinguish between different theories on the topic ofbirth order For example schooling circumstances play a large role in educational out-comes and may be related to birth order20 Our analysis is limited in the sense that itcannot discriminate between many competing hypotheses on why birth order appearsto be important21

Nevertheless we checked whether what appears as a first-born advantage predom-inantly comes from financial constraints for example parents sending their first-bornto college and running out of money for the following siblings Conley offers the fol-lowing argument ldquo[I]n terms of parental investment the cup starts to run dry as wego down the line Parental resources it appears are allotted on a first come first-served basisrdquo22 Yet if it turns out that first-borns perform better beforehand then atheory based on budget constraints cannot fully account for our results In Tables 5ndash7we estimate the probability of completing high school following the same methodol-ogy as in Tables 2ndash4 We find that first-borns have a higher probability of completinghigh school than later born siblings

Specifically Table 5 shows again that the first-born effect is not an artifact of fam-ily size that it increases in magnitude and significance when including age of motherat birth that it is robust to including characteristics of father and to restricting thesample to families in which all siblings report Table 6 shows that these results alsohold when regressions are estimated separately by family size (except for families offive presumably because of the small sample size) Tables 7a and 7b are the counter-parts of Tables 4a and 4b Fixed effects regressions controlling for age and gendersupport the results found in Tables 5 and 623

Also we estimated education at age 18 conditional on high school completion (at18 or older) to see if later-borns are more likely to repeat grades which would sup-port a theory of birth-order effects based on cognitive development differences Wedid not find any evidence for this However the small sample size resulting fromselecting index persons with available information at age 18 warrants some caution

Finally we found evidence among White families only that conditional on com-pleting high school first-borns are more likely to receive postsecondary educationYet for all races conditional on postsecondary education we found no clear advan-tage to being first-born24 In summary financial constraints do seem to play a role butsome factors early in life contribute to the first-born premium puzzle

20 We thank an anonymous referee for this point21 We refer the reader to the survey of those theories presented in Black Devereux and Salvanes (2005)22 Conley (2004) p 6923 The results held when including motherrsquos marital status We ran into the same small size problems whenadding family income and motherrsquos employment status for different age ranges of each child The ethnic dif-ferential disappears when considering all families (but persists when focusing on smaller families) suggest-ing that some of the birth order effect among White families comes from postsecondary education24 Results are available from the authors

Kantarevic and Mechoulan 767

Tabl

e 5

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

004

40

024

005

50

039

004

4(0

014

)

(00

14)

(00

14)

(0

018

)

(00

17)

To

tal n

umbe

r of

0

014

001

60

011

001

2si

blin

gs(0

003

)

(00

03)

(0

004

)

(00

05)

A

ge o

f th

e m

othe

r 0

006

000

70

005

at c

hild

birt

h(0

000

1)

(0

002

)

(00

01)

d[

Mal

e]0

047

004

80

049

004

70

037

(00

09)

(0

009

)

(00

09)

(0

009

)

(00

10)

A

ge0

024

002

90

029

003

20

023

(00

04)

(0

004

)

(00

04)

(0

007

)

(00

07)

A

ge2

000

020

0003

000

020

0003

000

02(5

times10

minus5)

(5

times10

minus5)

(5

times10

minus5)

(8

times10

minus5)

(8

times10

minus5)

d[

Whi

te]

000

50

012

002

30

042

001

9(0

014

)(0

014

)(0

014

)(0

017

)

(00

19)

d[fir

st-b

orn]

timesd

001

5minus0

005

minus00

060

011

000

6[W

hite

](0

017

5)(0

017

)(0

017

)(0

021

)(0

021

)M

othe

rrsquos

educ

atio

n0

027

002

60

026

001

50

026

(00

02)

(0

002

)

(00

02)

(0

003

)

(00

03)

d[

com

plet

e in

fo

003

50

094

002

20

029

on a

ll si

blin

gs]

(00

13)

(0

012

)

(00

13)

(00

15)

d[bo

th p

aren

ts r

epor

t]0

090

094

009

00

097

(00

12)

(0

012

)

(00

12)

(0

016

)

The Journal of Human Resources768

Tabl

e 5

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

Fath

errsquos

edu

catio

n0

011

(00

02)

A

ge o

f fa

ther

minus0

001

at c

hild

birt

h(0

002

)C

onst

ant

010

50

104

028

80

241

012

9(0

089

)(0

089

)(0

094

)

(01

35)

(01

41)

R2

008

790

0943

010

180

0885

008

55N

umbe

r of

obs

erva

tions

792

87

928

792

84

766

454

1N

umbe

r of

fam

ily

311

23

112

311

21

869

173

2cl

uste

rs

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Kantarevic and Mechoulan 769

Tabl

e 6

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

003

80

084

009

003

10

122

(00

18)

(0

025

)

(00

36)

(0

053

)(0

049

)

d[se

cond

-bor

n]0

071

008

2times

10minus4

002

2(0

022

)

(00

31)

(0

051

)(0

044

)d[

thir

d-bo

rn]

003

10

012

003

5(0

028

)(0

041

)(0

037

)d[

four

th-b

orn]

006

10

041

(00

37)

(00

33)

d[fif

th-b

orn]

000

7(0

028

)A

ge o

f th

e m

othe

r at

chi

ldbi

rth

000

30

004

000

90

008

001

1(0

002

)(0

002

)(0

003

)

(00

04)

(0

003

)

d[M

ale]

minus00

3minus0

041

000

7minus0

094

minus00

82(0

016

)(0

016

)

(00

2)(0

026

)

(00

18)

A

ge0

028

002

80

051

001

20

037

(00

07)

(0

009

)

(00

1)

(0

008

)(0

011

)

Age

23

times10

minus43

times10

minus45

times10

minus48

times10

minus54times

10minus4

(9times

10minus5

)

(10minus5

)

(10minus5

)

(9times

10minus5

)(1

0minus4)

d[

Whi

te]

001

20

124

004

90

033

002

2(0

021

)(0

023

)(0

026

)(0

034

)(0

028

)M

othe

rrsquos

educ

atio

n0

024

002

50

027

001

80

027

(00

04)

(0

004

)

(00

05)

(0

005

)

(00

04)

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 8: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

The Journal of Human Resources762

Tabl

e 3

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Edu

cati

on

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

027

50

698

071

50

534

091

5(0

104

)

(01

36)

(0

185

)

(03

16)

(02

36)

d[

seco

nd-b

orn]

043

40

463

033

40

578

(01

17)

(0

16)

(02

82)

(01

95)

d[

thir

d-bo

rn]

017

10

212

036

8(0

14)

(02

47)

(01

75)

d[

four

th-b

orn]

minus00

120

466

(02

14)

(01

52)

d[

fifth

-bor

n]0

221

(01

33)

Age

of

the

mot

her

at c

hild

birt

h0

048

007

90

067

006

80

07(0

012

)

(00

13)

(0

013

)

(00

23)

(0

013

)

d[M

ale]

minus02

87minus0

115

minus00

35minus0

34

minus00

33(0

102

)

(00

91)

(01

07)

(01

31)

(0

087

)

Age

021

70

190

156

007

50

167

(00

56)

(0

057

)

(00

49)

(0

044

)(0

057

)

Kantarevic and Mechoulan 763

Age

20

002

000

20

002

3times

10minus4

000

2(7

times10

minus4)

(7

times10

minus4)

(6

times10

minus4)

(5

times10

minus4)

(8times

10minus4

)

d[W

hite

]0

119

minus01

244

times10

minus5minus0

03

minus01

14(0

12)

(01

3)(0

13)

(01

81)

(01

37)

Mot

herrsquo

s ed

ucat

ion

025

70

250

179

012

40

182

(00

25)

(0

026

)

(00

23)

(0

035

)

(00

22)

d[

all s

iblin

gs r

epor

t]0

147

021

2minus0

066

035

10

21(0

154

)(0

136

)(0

128

)(0

178

)

(01

49)

d[bo

th p

aren

ts r

epor

t]0

241

061

20

755

054

20

371

(01

18)

(0

118

)

(01

26)

(0

189

)

(01

27)

C

onst

ant

333

92

694

425

86

607

412

(11

48)

(1

184

)

(10

11)

(1

293

)

(14

15)

R

20

185

021

019

011

40

127

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources764

size greater than one whereas it is only present within large families among BlacksTherefore the ethnic differential disappears when considering large families only

As noted earlier a causal interpretation of the age of the mother at childbirth wouldhinge on the assumption of its exogeneity Without instrumental variables or a treat-ment versus control quasi experiment it is difficult to draw conclusions12 The age ofthe mother at childbirth itself positively correlated with birth order could easilyproxy for other unobserved variables such as level of human capital and parentalresources13

To address this problem our fixed-effect estimation (Table 4a and b) removes fam-ily characteristics and unobserved family-level heterogeneity14 Family fixed effectsaddress family unobservables to the extent that they are constant over time While wetry to incorporate observables that vary across birth order to affirm the robustness ofour results we are constrained by the availability of such variables in the data set

Unfortunately the coefficients on age or on age of the mother at childbirth are unin-formative in those fixed effects regressions Deviations from family means for ageconvey the same information as deviations from family means for age of mother atchildbirth We thus do not provide separate estimations with age and age of motherat childbirth The age of the mother at first childbirth is certainly relevant but here itis differenced out Still to assess this issue better we ran separate regressions split-ting the sample by maternal age15 To summarize the following results are robust toexcluding mothers who first gave birth as teens but we do not have enough observa-tions to meaningfully run the fixed effects regressions by sibling size on that lattergroup

In Tables 4a and 4b the fixed effects estimations that only control for age and gen-der confirm the previous results for the most part16 We also provide estimations includ-ing the marital status of the mother17 during the childrsquos first 14 years18 The results donot change by much19 Clearly the suggestion that first-borns are most likely to live

12 Also while it is possible in the context of larger samples to instrument for siblings sizemdashusing twinbirths (Black Devereux and Salvanes 2005) or using the fact that parents of two same sex children are morelikely to have a third child (Conley and Glauber 2004)mdashone cannot instrument for birth order per se13 This is further evidenced by the fact that including a dummy variable indicating whether the mother wasmarried at the time of childbirth instead of (and obviously also along with) age of mother at childbirth alsoresults in highly significant birth order coefficients Results are available upon request14 However it is worth noting that those fixed effects do not solve all endogeneity issues For example itmay be the case that first-born ldquoqualityrdquo is affecting subsequent fertility We thank an anonymous referee forpointing out this caveat15 We thank an anonymous referee for this suggestion16 In Column 1mdashfamilies of two siblingsmdashthe coefficient on first-born becomes significant at the 10 per-cent level when including index persons of 23 and 24 years of age suggesting the nonsignificance whenrestricting at age 25 stems from a small sample size Separate FE regressions for White and Black familiesconfirmed the ethnic differentials found earlier17 We tried two definitions number of years marriednumber of years considered in the age group and = 1if continuously married over the years considered in the age group 0 otherwise Since the results on birthorder do not change qualitatively with either of those we only report the results with the latter definition18 We also ran similar estimations with different age ranges and obtained similar results19 Because the information on marital status is retrospective we do not lose any observation by includingthis covariate Unfortunately this is not the case when adding average family income or average employmentstatus of the mother during the childrsquos first years

Kantarevic and Mechoulan 765

Table 4aFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Not Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0216 0437 0683 0594 0477(0181) (0206) (0265) (0379) (0224)

d[second- 0282 0412 0421 0199born] (0141) (0214) (0323) (0195)

d[third- 0138 0361 0055born] (0159) (0261) (0172)

d[fourth- minus0031 0181born] (0207) (0149)

d[fifth- minus0012born] (0125)

Observations 1422 1743 1477 922 2487groups 934 835 562 388 573

Table 4bFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0224 0436 0696 0585 0492(0181) (0206) (0267) (0379) (0226)

d[second-born] 028 0423 0413 0212(0141) (0216) (0323) (0197)

d[third-born] 0146 0363 0063(0159) (0261) (0173)

d[fourth-born] minus0025 0188(0207) (0149)

d[fifth-born] minus0016(0125)

Mother minus0196 minus0167 0112 0381 minus0105Continuously (0315) (0221) (0221) (0274) (0178)married aged 1ndash14

Number of 1422 1743 1477 999 2487observations

Number of 934 835 562 328 573groups

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources766

their critical development years in a stable household as opposed to later-borns whomay experience the divorce of their parents cannot entirely explain the first-bornadvantage At the same time the persistence of birth-order effects naturally poses theproblem of their origin

The literature is not able to distinguish between different theories on the topic ofbirth order For example schooling circumstances play a large role in educational out-comes and may be related to birth order20 Our analysis is limited in the sense that itcannot discriminate between many competing hypotheses on why birth order appearsto be important21

Nevertheless we checked whether what appears as a first-born advantage predom-inantly comes from financial constraints for example parents sending their first-bornto college and running out of money for the following siblings Conley offers the fol-lowing argument ldquo[I]n terms of parental investment the cup starts to run dry as wego down the line Parental resources it appears are allotted on a first come first-served basisrdquo22 Yet if it turns out that first-borns perform better beforehand then atheory based on budget constraints cannot fully account for our results In Tables 5ndash7we estimate the probability of completing high school following the same methodol-ogy as in Tables 2ndash4 We find that first-borns have a higher probability of completinghigh school than later born siblings

Specifically Table 5 shows again that the first-born effect is not an artifact of fam-ily size that it increases in magnitude and significance when including age of motherat birth that it is robust to including characteristics of father and to restricting thesample to families in which all siblings report Table 6 shows that these results alsohold when regressions are estimated separately by family size (except for families offive presumably because of the small sample size) Tables 7a and 7b are the counter-parts of Tables 4a and 4b Fixed effects regressions controlling for age and gendersupport the results found in Tables 5 and 623

Also we estimated education at age 18 conditional on high school completion (at18 or older) to see if later-borns are more likely to repeat grades which would sup-port a theory of birth-order effects based on cognitive development differences Wedid not find any evidence for this However the small sample size resulting fromselecting index persons with available information at age 18 warrants some caution

Finally we found evidence among White families only that conditional on com-pleting high school first-borns are more likely to receive postsecondary educationYet for all races conditional on postsecondary education we found no clear advan-tage to being first-born24 In summary financial constraints do seem to play a role butsome factors early in life contribute to the first-born premium puzzle

20 We thank an anonymous referee for this point21 We refer the reader to the survey of those theories presented in Black Devereux and Salvanes (2005)22 Conley (2004) p 6923 The results held when including motherrsquos marital status We ran into the same small size problems whenadding family income and motherrsquos employment status for different age ranges of each child The ethnic dif-ferential disappears when considering all families (but persists when focusing on smaller families) suggest-ing that some of the birth order effect among White families comes from postsecondary education24 Results are available from the authors

Kantarevic and Mechoulan 767

Tabl

e 5

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

004

40

024

005

50

039

004

4(0

014

)

(00

14)

(00

14)

(0

018

)

(00

17)

To

tal n

umbe

r of

0

014

001

60

011

001

2si

blin

gs(0

003

)

(00

03)

(0

004

)

(00

05)

A

ge o

f th

e m

othe

r 0

006

000

70

005

at c

hild

birt

h(0

000

1)

(0

002

)

(00

01)

d[

Mal

e]0

047

004

80

049

004

70

037

(00

09)

(0

009

)

(00

09)

(0

009

)

(00

10)

A

ge0

024

002

90

029

003

20

023

(00

04)

(0

004

)

(00

04)

(0

007

)

(00

07)

A

ge2

000

020

0003

000

020

0003

000

02(5

times10

minus5)

(5

times10

minus5)

(5

times10

minus5)

(8

times10

minus5)

(8

times10

minus5)

d[

Whi

te]

000

50

012

002

30

042

001

9(0

014

)(0

014

)(0

014

)(0

017

)

(00

19)

d[fir

st-b

orn]

timesd

001

5minus0

005

minus00

060

011

000

6[W

hite

](0

017

5)(0

017

)(0

017

)(0

021

)(0

021

)M

othe

rrsquos

educ

atio

n0

027

002

60

026

001

50

026

(00

02)

(0

002

)

(00

02)

(0

003

)

(00

03)

d[

com

plet

e in

fo

003

50

094

002

20

029

on a

ll si

blin

gs]

(00

13)

(0

012

)

(00

13)

(00

15)

d[bo

th p

aren

ts r

epor

t]0

090

094

009

00

097

(00

12)

(0

012

)

(00

12)

(0

016

)

The Journal of Human Resources768

Tabl

e 5

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

Fath

errsquos

edu

catio

n0

011

(00

02)

A

ge o

f fa

ther

minus0

001

at c

hild

birt

h(0

002

)C

onst

ant

010

50

104

028

80

241

012

9(0

089

)(0

089

)(0

094

)

(01

35)

(01

41)

R2

008

790

0943

010

180

0885

008

55N

umbe

r of

obs

erva

tions

792

87

928

792

84

766

454

1N

umbe

r of

fam

ily

311

23

112

311

21

869

173

2cl

uste

rs

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Kantarevic and Mechoulan 769

Tabl

e 6

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

003

80

084

009

003

10

122

(00

18)

(0

025

)

(00

36)

(0

053

)(0

049

)

d[se

cond

-bor

n]0

071

008

2times

10minus4

002

2(0

022

)

(00

31)

(0

051

)(0

044

)d[

thir

d-bo

rn]

003

10

012

003

5(0

028

)(0

041

)(0

037

)d[

four

th-b

orn]

006

10

041

(00

37)

(00

33)

d[fif

th-b

orn]

000

7(0

028

)A

ge o

f th

e m

othe

r at

chi

ldbi

rth

000

30

004

000

90

008

001

1(0

002

)(0

002

)(0

003

)

(00

04)

(0

003

)

d[M

ale]

minus00

3minus0

041

000

7minus0

094

minus00

82(0

016

)(0

016

)

(00

2)(0

026

)

(00

18)

A

ge0

028

002

80

051

001

20

037

(00

07)

(0

009

)

(00

1)

(0

008

)(0

011

)

Age

23

times10

minus43

times10

minus45

times10

minus48

times10

minus54times

10minus4

(9times

10minus5

)

(10minus5

)

(10minus5

)

(9times

10minus5

)(1

0minus4)

d[

Whi

te]

001

20

124

004

90

033

002

2(0

021

)(0

023

)(0

026

)(0

034

)(0

028

)M

othe

rrsquos

educ

atio

n0

024

002

50

027

001

80

027

(00

04)

(0

004

)

(00

05)

(0

005

)

(00

04)

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 9: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

Kantarevic and Mechoulan 763

Age

20

002

000

20

002

3times

10minus4

000

2(7

times10

minus4)

(7

times10

minus4)

(6

times10

minus4)

(5

times10

minus4)

(8times

10minus4

)

d[W

hite

]0

119

minus01

244

times10

minus5minus0

03

minus01

14(0

12)

(01

3)(0

13)

(01

81)

(01

37)

Mot

herrsquo

s ed

ucat

ion

025

70

250

179

012

40

182

(00

25)

(0

026

)

(00

23)

(0

035

)

(00

22)

d[

all s

iblin

gs r

epor

t]0

147

021

2minus0

066

035

10

21(0

154

)(0

136

)(0

128

)(0

178

)

(01

49)

d[bo

th p

aren

ts r

epor

t]0

241

061

20

755

054

20

371

(01

18)

(0

118

)

(01

26)

(0

189

)

(01

27)

C

onst

ant

333

92

694

425

86

607

412

(11

48)

(1

184

)

(10

11)

(1

293

)

(14

15)

R

20

185

021

019

011

40

127

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources764

size greater than one whereas it is only present within large families among BlacksTherefore the ethnic differential disappears when considering large families only

As noted earlier a causal interpretation of the age of the mother at childbirth wouldhinge on the assumption of its exogeneity Without instrumental variables or a treat-ment versus control quasi experiment it is difficult to draw conclusions12 The age ofthe mother at childbirth itself positively correlated with birth order could easilyproxy for other unobserved variables such as level of human capital and parentalresources13

To address this problem our fixed-effect estimation (Table 4a and b) removes fam-ily characteristics and unobserved family-level heterogeneity14 Family fixed effectsaddress family unobservables to the extent that they are constant over time While wetry to incorporate observables that vary across birth order to affirm the robustness ofour results we are constrained by the availability of such variables in the data set

Unfortunately the coefficients on age or on age of the mother at childbirth are unin-formative in those fixed effects regressions Deviations from family means for ageconvey the same information as deviations from family means for age of mother atchildbirth We thus do not provide separate estimations with age and age of motherat childbirth The age of the mother at first childbirth is certainly relevant but here itis differenced out Still to assess this issue better we ran separate regressions split-ting the sample by maternal age15 To summarize the following results are robust toexcluding mothers who first gave birth as teens but we do not have enough observa-tions to meaningfully run the fixed effects regressions by sibling size on that lattergroup

In Tables 4a and 4b the fixed effects estimations that only control for age and gen-der confirm the previous results for the most part16 We also provide estimations includ-ing the marital status of the mother17 during the childrsquos first 14 years18 The results donot change by much19 Clearly the suggestion that first-borns are most likely to live

12 Also while it is possible in the context of larger samples to instrument for siblings sizemdashusing twinbirths (Black Devereux and Salvanes 2005) or using the fact that parents of two same sex children are morelikely to have a third child (Conley and Glauber 2004)mdashone cannot instrument for birth order per se13 This is further evidenced by the fact that including a dummy variable indicating whether the mother wasmarried at the time of childbirth instead of (and obviously also along with) age of mother at childbirth alsoresults in highly significant birth order coefficients Results are available upon request14 However it is worth noting that those fixed effects do not solve all endogeneity issues For example itmay be the case that first-born ldquoqualityrdquo is affecting subsequent fertility We thank an anonymous referee forpointing out this caveat15 We thank an anonymous referee for this suggestion16 In Column 1mdashfamilies of two siblingsmdashthe coefficient on first-born becomes significant at the 10 per-cent level when including index persons of 23 and 24 years of age suggesting the nonsignificance whenrestricting at age 25 stems from a small sample size Separate FE regressions for White and Black familiesconfirmed the ethnic differentials found earlier17 We tried two definitions number of years marriednumber of years considered in the age group and = 1if continuously married over the years considered in the age group 0 otherwise Since the results on birthorder do not change qualitatively with either of those we only report the results with the latter definition18 We also ran similar estimations with different age ranges and obtained similar results19 Because the information on marital status is retrospective we do not lose any observation by includingthis covariate Unfortunately this is not the case when adding average family income or average employmentstatus of the mother during the childrsquos first years

Kantarevic and Mechoulan 765

Table 4aFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Not Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0216 0437 0683 0594 0477(0181) (0206) (0265) (0379) (0224)

d[second- 0282 0412 0421 0199born] (0141) (0214) (0323) (0195)

d[third- 0138 0361 0055born] (0159) (0261) (0172)

d[fourth- minus0031 0181born] (0207) (0149)

d[fifth- minus0012born] (0125)

Observations 1422 1743 1477 922 2487groups 934 835 562 388 573

Table 4bFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0224 0436 0696 0585 0492(0181) (0206) (0267) (0379) (0226)

d[second-born] 028 0423 0413 0212(0141) (0216) (0323) (0197)

d[third-born] 0146 0363 0063(0159) (0261) (0173)

d[fourth-born] minus0025 0188(0207) (0149)

d[fifth-born] minus0016(0125)

Mother minus0196 minus0167 0112 0381 minus0105Continuously (0315) (0221) (0221) (0274) (0178)married aged 1ndash14

Number of 1422 1743 1477 999 2487observations

Number of 934 835 562 328 573groups

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources766

their critical development years in a stable household as opposed to later-borns whomay experience the divorce of their parents cannot entirely explain the first-bornadvantage At the same time the persistence of birth-order effects naturally poses theproblem of their origin

The literature is not able to distinguish between different theories on the topic ofbirth order For example schooling circumstances play a large role in educational out-comes and may be related to birth order20 Our analysis is limited in the sense that itcannot discriminate between many competing hypotheses on why birth order appearsto be important21

Nevertheless we checked whether what appears as a first-born advantage predom-inantly comes from financial constraints for example parents sending their first-bornto college and running out of money for the following siblings Conley offers the fol-lowing argument ldquo[I]n terms of parental investment the cup starts to run dry as wego down the line Parental resources it appears are allotted on a first come first-served basisrdquo22 Yet if it turns out that first-borns perform better beforehand then atheory based on budget constraints cannot fully account for our results In Tables 5ndash7we estimate the probability of completing high school following the same methodol-ogy as in Tables 2ndash4 We find that first-borns have a higher probability of completinghigh school than later born siblings

Specifically Table 5 shows again that the first-born effect is not an artifact of fam-ily size that it increases in magnitude and significance when including age of motherat birth that it is robust to including characteristics of father and to restricting thesample to families in which all siblings report Table 6 shows that these results alsohold when regressions are estimated separately by family size (except for families offive presumably because of the small sample size) Tables 7a and 7b are the counter-parts of Tables 4a and 4b Fixed effects regressions controlling for age and gendersupport the results found in Tables 5 and 623

Also we estimated education at age 18 conditional on high school completion (at18 or older) to see if later-borns are more likely to repeat grades which would sup-port a theory of birth-order effects based on cognitive development differences Wedid not find any evidence for this However the small sample size resulting fromselecting index persons with available information at age 18 warrants some caution

Finally we found evidence among White families only that conditional on com-pleting high school first-borns are more likely to receive postsecondary educationYet for all races conditional on postsecondary education we found no clear advan-tage to being first-born24 In summary financial constraints do seem to play a role butsome factors early in life contribute to the first-born premium puzzle

20 We thank an anonymous referee for this point21 We refer the reader to the survey of those theories presented in Black Devereux and Salvanes (2005)22 Conley (2004) p 6923 The results held when including motherrsquos marital status We ran into the same small size problems whenadding family income and motherrsquos employment status for different age ranges of each child The ethnic dif-ferential disappears when considering all families (but persists when focusing on smaller families) suggest-ing that some of the birth order effect among White families comes from postsecondary education24 Results are available from the authors

Kantarevic and Mechoulan 767

Tabl

e 5

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

004

40

024

005

50

039

004

4(0

014

)

(00

14)

(00

14)

(0

018

)

(00

17)

To

tal n

umbe

r of

0

014

001

60

011

001

2si

blin

gs(0

003

)

(00

03)

(0

004

)

(00

05)

A

ge o

f th

e m

othe

r 0

006

000

70

005

at c

hild

birt

h(0

000

1)

(0

002

)

(00

01)

d[

Mal

e]0

047

004

80

049

004

70

037

(00

09)

(0

009

)

(00

09)

(0

009

)

(00

10)

A

ge0

024

002

90

029

003

20

023

(00

04)

(0

004

)

(00

04)

(0

007

)

(00

07)

A

ge2

000

020

0003

000

020

0003

000

02(5

times10

minus5)

(5

times10

minus5)

(5

times10

minus5)

(8

times10

minus5)

(8

times10

minus5)

d[

Whi

te]

000

50

012

002

30

042

001

9(0

014

)(0

014

)(0

014

)(0

017

)

(00

19)

d[fir

st-b

orn]

timesd

001

5minus0

005

minus00

060

011

000

6[W

hite

](0

017

5)(0

017

)(0

017

)(0

021

)(0

021

)M

othe

rrsquos

educ

atio

n0

027

002

60

026

001

50

026

(00

02)

(0

002

)

(00

02)

(0

003

)

(00

03)

d[

com

plet

e in

fo

003

50

094

002

20

029

on a

ll si

blin

gs]

(00

13)

(0

012

)

(00

13)

(00

15)

d[bo

th p

aren

ts r

epor

t]0

090

094

009

00

097

(00

12)

(0

012

)

(00

12)

(0

016

)

The Journal of Human Resources768

Tabl

e 5

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

Fath

errsquos

edu

catio

n0

011

(00

02)

A

ge o

f fa

ther

minus0

001

at c

hild

birt

h(0

002

)C

onst

ant

010

50

104

028

80

241

012

9(0

089

)(0

089

)(0

094

)

(01

35)

(01

41)

R2

008

790

0943

010

180

0885

008

55N

umbe

r of

obs

erva

tions

792

87

928

792

84

766

454

1N

umbe

r of

fam

ily

311

23

112

311

21

869

173

2cl

uste

rs

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Kantarevic and Mechoulan 769

Tabl

e 6

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

003

80

084

009

003

10

122

(00

18)

(0

025

)

(00

36)

(0

053

)(0

049

)

d[se

cond

-bor

n]0

071

008

2times

10minus4

002

2(0

022

)

(00

31)

(0

051

)(0

044

)d[

thir

d-bo

rn]

003

10

012

003

5(0

028

)(0

041

)(0

037

)d[

four

th-b

orn]

006

10

041

(00

37)

(00

33)

d[fif

th-b

orn]

000

7(0

028

)A

ge o

f th

e m

othe

r at

chi

ldbi

rth

000

30

004

000

90

008

001

1(0

002

)(0

002

)(0

003

)

(00

04)

(0

003

)

d[M

ale]

minus00

3minus0

041

000

7minus0

094

minus00

82(0

016

)(0

016

)

(00

2)(0

026

)

(00

18)

A

ge0

028

002

80

051

001

20

037

(00

07)

(0

009

)

(00

1)

(0

008

)(0

011

)

Age

23

times10

minus43

times10

minus45

times10

minus48

times10

minus54times

10minus4

(9times

10minus5

)

(10minus5

)

(10minus5

)

(9times

10minus5

)(1

0minus4)

d[

Whi

te]

001

20

124

004

90

033

002

2(0

021

)(0

023

)(0

026

)(0

034

)(0

028

)M

othe

rrsquos

educ

atio

n0

024

002

50

027

001

80

027

(00

04)

(0

004

)

(00

05)

(0

005

)

(00

04)

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 10: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

The Journal of Human Resources764

size greater than one whereas it is only present within large families among BlacksTherefore the ethnic differential disappears when considering large families only

As noted earlier a causal interpretation of the age of the mother at childbirth wouldhinge on the assumption of its exogeneity Without instrumental variables or a treat-ment versus control quasi experiment it is difficult to draw conclusions12 The age ofthe mother at childbirth itself positively correlated with birth order could easilyproxy for other unobserved variables such as level of human capital and parentalresources13

To address this problem our fixed-effect estimation (Table 4a and b) removes fam-ily characteristics and unobserved family-level heterogeneity14 Family fixed effectsaddress family unobservables to the extent that they are constant over time While wetry to incorporate observables that vary across birth order to affirm the robustness ofour results we are constrained by the availability of such variables in the data set

Unfortunately the coefficients on age or on age of the mother at childbirth are unin-formative in those fixed effects regressions Deviations from family means for ageconvey the same information as deviations from family means for age of mother atchildbirth We thus do not provide separate estimations with age and age of motherat childbirth The age of the mother at first childbirth is certainly relevant but here itis differenced out Still to assess this issue better we ran separate regressions split-ting the sample by maternal age15 To summarize the following results are robust toexcluding mothers who first gave birth as teens but we do not have enough observa-tions to meaningfully run the fixed effects regressions by sibling size on that lattergroup

In Tables 4a and 4b the fixed effects estimations that only control for age and gen-der confirm the previous results for the most part16 We also provide estimations includ-ing the marital status of the mother17 during the childrsquos first 14 years18 The results donot change by much19 Clearly the suggestion that first-borns are most likely to live

12 Also while it is possible in the context of larger samples to instrument for siblings sizemdashusing twinbirths (Black Devereux and Salvanes 2005) or using the fact that parents of two same sex children are morelikely to have a third child (Conley and Glauber 2004)mdashone cannot instrument for birth order per se13 This is further evidenced by the fact that including a dummy variable indicating whether the mother wasmarried at the time of childbirth instead of (and obviously also along with) age of mother at childbirth alsoresults in highly significant birth order coefficients Results are available upon request14 However it is worth noting that those fixed effects do not solve all endogeneity issues For example itmay be the case that first-born ldquoqualityrdquo is affecting subsequent fertility We thank an anonymous referee forpointing out this caveat15 We thank an anonymous referee for this suggestion16 In Column 1mdashfamilies of two siblingsmdashthe coefficient on first-born becomes significant at the 10 per-cent level when including index persons of 23 and 24 years of age suggesting the nonsignificance whenrestricting at age 25 stems from a small sample size Separate FE regressions for White and Black familiesconfirmed the ethnic differentials found earlier17 We tried two definitions number of years marriednumber of years considered in the age group and = 1if continuously married over the years considered in the age group 0 otherwise Since the results on birthorder do not change qualitatively with either of those we only report the results with the latter definition18 We also ran similar estimations with different age ranges and obtained similar results19 Because the information on marital status is retrospective we do not lose any observation by includingthis covariate Unfortunately this is not the case when adding average family income or average employmentstatus of the mother during the childrsquos first years

Kantarevic and Mechoulan 765

Table 4aFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Not Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0216 0437 0683 0594 0477(0181) (0206) (0265) (0379) (0224)

d[second- 0282 0412 0421 0199born] (0141) (0214) (0323) (0195)

d[third- 0138 0361 0055born] (0159) (0261) (0172)

d[fourth- minus0031 0181born] (0207) (0149)

d[fifth- minus0012born] (0125)

Observations 1422 1743 1477 922 2487groups 934 835 562 388 573

Table 4bFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0224 0436 0696 0585 0492(0181) (0206) (0267) (0379) (0226)

d[second-born] 028 0423 0413 0212(0141) (0216) (0323) (0197)

d[third-born] 0146 0363 0063(0159) (0261) (0173)

d[fourth-born] minus0025 0188(0207) (0149)

d[fifth-born] minus0016(0125)

Mother minus0196 minus0167 0112 0381 minus0105Continuously (0315) (0221) (0221) (0274) (0178)married aged 1ndash14

Number of 1422 1743 1477 999 2487observations

Number of 934 835 562 328 573groups

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources766

their critical development years in a stable household as opposed to later-borns whomay experience the divorce of their parents cannot entirely explain the first-bornadvantage At the same time the persistence of birth-order effects naturally poses theproblem of their origin

The literature is not able to distinguish between different theories on the topic ofbirth order For example schooling circumstances play a large role in educational out-comes and may be related to birth order20 Our analysis is limited in the sense that itcannot discriminate between many competing hypotheses on why birth order appearsto be important21

Nevertheless we checked whether what appears as a first-born advantage predom-inantly comes from financial constraints for example parents sending their first-bornto college and running out of money for the following siblings Conley offers the fol-lowing argument ldquo[I]n terms of parental investment the cup starts to run dry as wego down the line Parental resources it appears are allotted on a first come first-served basisrdquo22 Yet if it turns out that first-borns perform better beforehand then atheory based on budget constraints cannot fully account for our results In Tables 5ndash7we estimate the probability of completing high school following the same methodol-ogy as in Tables 2ndash4 We find that first-borns have a higher probability of completinghigh school than later born siblings

Specifically Table 5 shows again that the first-born effect is not an artifact of fam-ily size that it increases in magnitude and significance when including age of motherat birth that it is robust to including characteristics of father and to restricting thesample to families in which all siblings report Table 6 shows that these results alsohold when regressions are estimated separately by family size (except for families offive presumably because of the small sample size) Tables 7a and 7b are the counter-parts of Tables 4a and 4b Fixed effects regressions controlling for age and gendersupport the results found in Tables 5 and 623

Also we estimated education at age 18 conditional on high school completion (at18 or older) to see if later-borns are more likely to repeat grades which would sup-port a theory of birth-order effects based on cognitive development differences Wedid not find any evidence for this However the small sample size resulting fromselecting index persons with available information at age 18 warrants some caution

Finally we found evidence among White families only that conditional on com-pleting high school first-borns are more likely to receive postsecondary educationYet for all races conditional on postsecondary education we found no clear advan-tage to being first-born24 In summary financial constraints do seem to play a role butsome factors early in life contribute to the first-born premium puzzle

20 We thank an anonymous referee for this point21 We refer the reader to the survey of those theories presented in Black Devereux and Salvanes (2005)22 Conley (2004) p 6923 The results held when including motherrsquos marital status We ran into the same small size problems whenadding family income and motherrsquos employment status for different age ranges of each child The ethnic dif-ferential disappears when considering all families (but persists when focusing on smaller families) suggest-ing that some of the birth order effect among White families comes from postsecondary education24 Results are available from the authors

Kantarevic and Mechoulan 767

Tabl

e 5

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

004

40

024

005

50

039

004

4(0

014

)

(00

14)

(00

14)

(0

018

)

(00

17)

To

tal n

umbe

r of

0

014

001

60

011

001

2si

blin

gs(0

003

)

(00

03)

(0

004

)

(00

05)

A

ge o

f th

e m

othe

r 0

006

000

70

005

at c

hild

birt

h(0

000

1)

(0

002

)

(00

01)

d[

Mal

e]0

047

004

80

049

004

70

037

(00

09)

(0

009

)

(00

09)

(0

009

)

(00

10)

A

ge0

024

002

90

029

003

20

023

(00

04)

(0

004

)

(00

04)

(0

007

)

(00

07)

A

ge2

000

020

0003

000

020

0003

000

02(5

times10

minus5)

(5

times10

minus5)

(5

times10

minus5)

(8

times10

minus5)

(8

times10

minus5)

d[

Whi

te]

000

50

012

002

30

042

001

9(0

014

)(0

014

)(0

014

)(0

017

)

(00

19)

d[fir

st-b

orn]

timesd

001

5minus0

005

minus00

060

011

000

6[W

hite

](0

017

5)(0

017

)(0

017

)(0

021

)(0

021

)M

othe

rrsquos

educ

atio

n0

027

002

60

026

001

50

026

(00

02)

(0

002

)

(00

02)

(0

003

)

(00

03)

d[

com

plet

e in

fo

003

50

094

002

20

029

on a

ll si

blin

gs]

(00

13)

(0

012

)

(00

13)

(00

15)

d[bo

th p

aren

ts r

epor

t]0

090

094

009

00

097

(00

12)

(0

012

)

(00

12)

(0

016

)

The Journal of Human Resources768

Tabl

e 5

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

Fath

errsquos

edu

catio

n0

011

(00

02)

A

ge o

f fa

ther

minus0

001

at c

hild

birt

h(0

002

)C

onst

ant

010

50

104

028

80

241

012

9(0

089

)(0

089

)(0

094

)

(01

35)

(01

41)

R2

008

790

0943

010

180

0885

008

55N

umbe

r of

obs

erva

tions

792

87

928

792

84

766

454

1N

umbe

r of

fam

ily

311

23

112

311

21

869

173

2cl

uste

rs

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Kantarevic and Mechoulan 769

Tabl

e 6

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

003

80

084

009

003

10

122

(00

18)

(0

025

)

(00

36)

(0

053

)(0

049

)

d[se

cond

-bor

n]0

071

008

2times

10minus4

002

2(0

022

)

(00

31)

(0

051

)(0

044

)d[

thir

d-bo

rn]

003

10

012

003

5(0

028

)(0

041

)(0

037

)d[

four

th-b

orn]

006

10

041

(00

37)

(00

33)

d[fif

th-b

orn]

000

7(0

028

)A

ge o

f th

e m

othe

r at

chi

ldbi

rth

000

30

004

000

90

008

001

1(0

002

)(0

002

)(0

003

)

(00

04)

(0

003

)

d[M

ale]

minus00

3minus0

041

000

7minus0

094

minus00

82(0

016

)(0

016

)

(00

2)(0

026

)

(00

18)

A

ge0

028

002

80

051

001

20

037

(00

07)

(0

009

)

(00

1)

(0

008

)(0

011

)

Age

23

times10

minus43

times10

minus45

times10

minus48

times10

minus54times

10minus4

(9times

10minus5

)

(10minus5

)

(10minus5

)

(9times

10minus5

)(1

0minus4)

d[

Whi

te]

001

20

124

004

90

033

002

2(0

021

)(0

023

)(0

026

)(0

034

)(0

028

)M

othe

rrsquos

educ

atio

n0

024

002

50

027

001

80

027

(00

04)

(0

004

)

(00

05)

(0

005

)

(00

04)

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 11: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

Kantarevic and Mechoulan 765

Table 4aFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Not Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0216 0437 0683 0594 0477(0181) (0206) (0265) (0379) (0224)

d[second- 0282 0412 0421 0199born] (0141) (0214) (0323) (0195)

d[third- 0138 0361 0055born] (0159) (0261) (0172)

d[fourth- minus0031 0181born] (0207) (0149)

d[fifth- minus0012born] (0125)

Observations 1422 1743 1477 922 2487groups 934 835 562 388 573

Table 4bFixed Effects Linear Regression with Dependent Variable Education (AllCoefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0224 0436 0696 0585 0492(0181) (0206) (0267) (0379) (0226)

d[second-born] 028 0423 0413 0212(0141) (0216) (0323) (0197)

d[third-born] 0146 0363 0063(0159) (0261) (0173)

d[fourth-born] minus0025 0188(0207) (0149)

d[fifth-born] minus0016(0125)

Mother minus0196 minus0167 0112 0381 minus0105Continuously (0315) (0221) (0221) (0274) (0178)married aged 1ndash14

Number of 1422 1743 1477 999 2487observations

Number of 934 835 562 328 573groups

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources766

their critical development years in a stable household as opposed to later-borns whomay experience the divorce of their parents cannot entirely explain the first-bornadvantage At the same time the persistence of birth-order effects naturally poses theproblem of their origin

The literature is not able to distinguish between different theories on the topic ofbirth order For example schooling circumstances play a large role in educational out-comes and may be related to birth order20 Our analysis is limited in the sense that itcannot discriminate between many competing hypotheses on why birth order appearsto be important21

Nevertheless we checked whether what appears as a first-born advantage predom-inantly comes from financial constraints for example parents sending their first-bornto college and running out of money for the following siblings Conley offers the fol-lowing argument ldquo[I]n terms of parental investment the cup starts to run dry as wego down the line Parental resources it appears are allotted on a first come first-served basisrdquo22 Yet if it turns out that first-borns perform better beforehand then atheory based on budget constraints cannot fully account for our results In Tables 5ndash7we estimate the probability of completing high school following the same methodol-ogy as in Tables 2ndash4 We find that first-borns have a higher probability of completinghigh school than later born siblings

Specifically Table 5 shows again that the first-born effect is not an artifact of fam-ily size that it increases in magnitude and significance when including age of motherat birth that it is robust to including characteristics of father and to restricting thesample to families in which all siblings report Table 6 shows that these results alsohold when regressions are estimated separately by family size (except for families offive presumably because of the small sample size) Tables 7a and 7b are the counter-parts of Tables 4a and 4b Fixed effects regressions controlling for age and gendersupport the results found in Tables 5 and 623

Also we estimated education at age 18 conditional on high school completion (at18 or older) to see if later-borns are more likely to repeat grades which would sup-port a theory of birth-order effects based on cognitive development differences Wedid not find any evidence for this However the small sample size resulting fromselecting index persons with available information at age 18 warrants some caution

Finally we found evidence among White families only that conditional on com-pleting high school first-borns are more likely to receive postsecondary educationYet for all races conditional on postsecondary education we found no clear advan-tage to being first-born24 In summary financial constraints do seem to play a role butsome factors early in life contribute to the first-born premium puzzle

20 We thank an anonymous referee for this point21 We refer the reader to the survey of those theories presented in Black Devereux and Salvanes (2005)22 Conley (2004) p 6923 The results held when including motherrsquos marital status We ran into the same small size problems whenadding family income and motherrsquos employment status for different age ranges of each child The ethnic dif-ferential disappears when considering all families (but persists when focusing on smaller families) suggest-ing that some of the birth order effect among White families comes from postsecondary education24 Results are available from the authors

Kantarevic and Mechoulan 767

Tabl

e 5

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

004

40

024

005

50

039

004

4(0

014

)

(00

14)

(00

14)

(0

018

)

(00

17)

To

tal n

umbe

r of

0

014

001

60

011

001

2si

blin

gs(0

003

)

(00

03)

(0

004

)

(00

05)

A

ge o

f th

e m

othe

r 0

006

000

70

005

at c

hild

birt

h(0

000

1)

(0

002

)

(00

01)

d[

Mal

e]0

047

004

80

049

004

70

037

(00

09)

(0

009

)

(00

09)

(0

009

)

(00

10)

A

ge0

024

002

90

029

003

20

023

(00

04)

(0

004

)

(00

04)

(0

007

)

(00

07)

A

ge2

000

020

0003

000

020

0003

000

02(5

times10

minus5)

(5

times10

minus5)

(5

times10

minus5)

(8

times10

minus5)

(8

times10

minus5)

d[

Whi

te]

000

50

012

002

30

042

001

9(0

014

)(0

014

)(0

014

)(0

017

)

(00

19)

d[fir

st-b

orn]

timesd

001

5minus0

005

minus00

060

011

000

6[W

hite

](0

017

5)(0

017

)(0

017

)(0

021

)(0

021

)M

othe

rrsquos

educ

atio

n0

027

002

60

026

001

50

026

(00

02)

(0

002

)

(00

02)

(0

003

)

(00

03)

d[

com

plet

e in

fo

003

50

094

002

20

029

on a

ll si

blin

gs]

(00

13)

(0

012

)

(00

13)

(00

15)

d[bo

th p

aren

ts r

epor

t]0

090

094

009

00

097

(00

12)

(0

012

)

(00

12)

(0

016

)

The Journal of Human Resources768

Tabl

e 5

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

Fath

errsquos

edu

catio

n0

011

(00

02)

A

ge o

f fa

ther

minus0

001

at c

hild

birt

h(0

002

)C

onst

ant

010

50

104

028

80

241

012

9(0

089

)(0

089

)(0

094

)

(01

35)

(01

41)

R2

008

790

0943

010

180

0885

008

55N

umbe

r of

obs

erva

tions

792

87

928

792

84

766

454

1N

umbe

r of

fam

ily

311

23

112

311

21

869

173

2cl

uste

rs

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Kantarevic and Mechoulan 769

Tabl

e 6

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

003

80

084

009

003

10

122

(00

18)

(0

025

)

(00

36)

(0

053

)(0

049

)

d[se

cond

-bor

n]0

071

008

2times

10minus4

002

2(0

022

)

(00

31)

(0

051

)(0

044

)d[

thir

d-bo

rn]

003

10

012

003

5(0

028

)(0

041

)(0

037

)d[

four

th-b

orn]

006

10

041

(00

37)

(00

33)

d[fif

th-b

orn]

000

7(0

028

)A

ge o

f th

e m

othe

r at

chi

ldbi

rth

000

30

004

000

90

008

001

1(0

002

)(0

002

)(0

003

)

(00

04)

(0

003

)

d[M

ale]

minus00

3minus0

041

000

7minus0

094

minus00

82(0

016

)(0

016

)

(00

2)(0

026

)

(00

18)

A

ge0

028

002

80

051

001

20

037

(00

07)

(0

009

)

(00

1)

(0

008

)(0

011

)

Age

23

times10

minus43

times10

minus45

times10

minus48

times10

minus54times

10minus4

(9times

10minus5

)

(10minus5

)

(10minus5

)

(9times

10minus5

)(1

0minus4)

d[

Whi

te]

001

20

124

004

90

033

002

2(0

021

)(0

023

)(0

026

)(0

034

)(0

028

)M

othe

rrsquos

educ

atio

n0

024

002

50

027

001

80

027

(00

04)

(0

004

)

(00

05)

(0

005

)

(00

04)

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 12: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

The Journal of Human Resources766

their critical development years in a stable household as opposed to later-borns whomay experience the divorce of their parents cannot entirely explain the first-bornadvantage At the same time the persistence of birth-order effects naturally poses theproblem of their origin

The literature is not able to distinguish between different theories on the topic ofbirth order For example schooling circumstances play a large role in educational out-comes and may be related to birth order20 Our analysis is limited in the sense that itcannot discriminate between many competing hypotheses on why birth order appearsto be important21

Nevertheless we checked whether what appears as a first-born advantage predom-inantly comes from financial constraints for example parents sending their first-bornto college and running out of money for the following siblings Conley offers the fol-lowing argument ldquo[I]n terms of parental investment the cup starts to run dry as wego down the line Parental resources it appears are allotted on a first come first-served basisrdquo22 Yet if it turns out that first-borns perform better beforehand then atheory based on budget constraints cannot fully account for our results In Tables 5ndash7we estimate the probability of completing high school following the same methodol-ogy as in Tables 2ndash4 We find that first-borns have a higher probability of completinghigh school than later born siblings

Specifically Table 5 shows again that the first-born effect is not an artifact of fam-ily size that it increases in magnitude and significance when including age of motherat birth that it is robust to including characteristics of father and to restricting thesample to families in which all siblings report Table 6 shows that these results alsohold when regressions are estimated separately by family size (except for families offive presumably because of the small sample size) Tables 7a and 7b are the counter-parts of Tables 4a and 4b Fixed effects regressions controlling for age and gendersupport the results found in Tables 5 and 623

Also we estimated education at age 18 conditional on high school completion (at18 or older) to see if later-borns are more likely to repeat grades which would sup-port a theory of birth-order effects based on cognitive development differences Wedid not find any evidence for this However the small sample size resulting fromselecting index persons with available information at age 18 warrants some caution

Finally we found evidence among White families only that conditional on com-pleting high school first-borns are more likely to receive postsecondary educationYet for all races conditional on postsecondary education we found no clear advan-tage to being first-born24 In summary financial constraints do seem to play a role butsome factors early in life contribute to the first-born premium puzzle

20 We thank an anonymous referee for this point21 We refer the reader to the survey of those theories presented in Black Devereux and Salvanes (2005)22 Conley (2004) p 6923 The results held when including motherrsquos marital status We ran into the same small size problems whenadding family income and motherrsquos employment status for different age ranges of each child The ethnic dif-ferential disappears when considering all families (but persists when focusing on smaller families) suggest-ing that some of the birth order effect among White families comes from postsecondary education24 Results are available from the authors

Kantarevic and Mechoulan 767

Tabl

e 5

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

004

40

024

005

50

039

004

4(0

014

)

(00

14)

(00

14)

(0

018

)

(00

17)

To

tal n

umbe

r of

0

014

001

60

011

001

2si

blin

gs(0

003

)

(00

03)

(0

004

)

(00

05)

A

ge o

f th

e m

othe

r 0

006

000

70

005

at c

hild

birt

h(0

000

1)

(0

002

)

(00

01)

d[

Mal

e]0

047

004

80

049

004

70

037

(00

09)

(0

009

)

(00

09)

(0

009

)

(00

10)

A

ge0

024

002

90

029

003

20

023

(00

04)

(0

004

)

(00

04)

(0

007

)

(00

07)

A

ge2

000

020

0003

000

020

0003

000

02(5

times10

minus5)

(5

times10

minus5)

(5

times10

minus5)

(8

times10

minus5)

(8

times10

minus5)

d[

Whi

te]

000

50

012

002

30

042

001

9(0

014

)(0

014

)(0

014

)(0

017

)

(00

19)

d[fir

st-b

orn]

timesd

001

5minus0

005

minus00

060

011

000

6[W

hite

](0

017

5)(0

017

)(0

017

)(0

021

)(0

021

)M

othe

rrsquos

educ

atio

n0

027

002

60

026

001

50

026

(00

02)

(0

002

)

(00

02)

(0

003

)

(00

03)

d[

com

plet

e in

fo

003

50

094

002

20

029

on a

ll si

blin

gs]

(00

13)

(0

012

)

(00

13)

(00

15)

d[bo

th p

aren

ts r

epor

t]0

090

094

009

00

097

(00

12)

(0

012

)

(00

12)

(0

016

)

The Journal of Human Resources768

Tabl

e 5

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

Fath

errsquos

edu

catio

n0

011

(00

02)

A

ge o

f fa

ther

minus0

001

at c

hild

birt

h(0

002

)C

onst

ant

010

50

104

028

80

241

012

9(0

089

)(0

089

)(0

094

)

(01

35)

(01

41)

R2

008

790

0943

010

180

0885

008

55N

umbe

r of

obs

erva

tions

792

87

928

792

84

766

454

1N

umbe

r of

fam

ily

311

23

112

311

21

869

173

2cl

uste

rs

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Kantarevic and Mechoulan 769

Tabl

e 6

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

003

80

084

009

003

10

122

(00

18)

(0

025

)

(00

36)

(0

053

)(0

049

)

d[se

cond

-bor

n]0

071

008

2times

10minus4

002

2(0

022

)

(00

31)

(0

051

)(0

044

)d[

thir

d-bo

rn]

003

10

012

003

5(0

028

)(0

041

)(0

037

)d[

four

th-b

orn]

006

10

041

(00

37)

(00

33)

d[fif

th-b

orn]

000

7(0

028

)A

ge o

f th

e m

othe

r at

chi

ldbi

rth

000

30

004

000

90

008

001

1(0

002

)(0

002

)(0

003

)

(00

04)

(0

003

)

d[M

ale]

minus00

3minus0

041

000

7minus0

094

minus00

82(0

016

)(0

016

)

(00

2)(0

026

)

(00

18)

A

ge0

028

002

80

051

001

20

037

(00

07)

(0

009

)

(00

1)

(0

008

)(0

011

)

Age

23

times10

minus43

times10

minus45

times10

minus48

times10

minus54times

10minus4

(9times

10minus5

)

(10minus5

)

(10minus5

)

(9times

10minus5

)(1

0minus4)

d[

Whi

te]

001

20

124

004

90

033

002

2(0

021

)(0

023

)(0

026

)(0

034

)(0

028

)M

othe

rrsquos

educ

atio

n0

024

002

50

027

001

80

027

(00

04)

(0

004

)

(00

05)

(0

005

)

(00

04)

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 13: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

Kantarevic and Mechoulan 767

Tabl

e 5

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

004

40

024

005

50

039

004

4(0

014

)

(00

14)

(00

14)

(0

018

)

(00

17)

To

tal n

umbe

r of

0

014

001

60

011

001

2si

blin

gs(0

003

)

(00

03)

(0

004

)

(00

05)

A

ge o

f th

e m

othe

r 0

006

000

70

005

at c

hild

birt

h(0

000

1)

(0

002

)

(00

01)

d[

Mal

e]0

047

004

80

049

004

70

037

(00

09)

(0

009

)

(00

09)

(0

009

)

(00

10)

A

ge0

024

002

90

029

003

20

023

(00

04)

(0

004

)

(00

04)

(0

007

)

(00

07)

A

ge2

000

020

0003

000

020

0003

000

02(5

times10

minus5)

(5

times10

minus5)

(5

times10

minus5)

(8

times10

minus5)

(8

times10

minus5)

d[

Whi

te]

000

50

012

002

30

042

001

9(0

014

)(0

014

)(0

014

)(0

017

)

(00

19)

d[fir

st-b

orn]

timesd

001

5minus0

005

minus00

060

011

000

6[W

hite

](0

017

5)(0

017

)(0

017

)(0

021

)(0

021

)M

othe

rrsquos

educ

atio

n0

027

002

60

026

001

50

026

(00

02)

(0

002

)

(00

02)

(0

003

)

(00

03)

d[

com

plet

e in

fo

003

50

094

002

20

029

on a

ll si

blin

gs]

(00

13)

(0

012

)

(00

13)

(00

15)

d[bo

th p

aren

ts r

epor

t]0

090

094

009

00

097

(00

12)

(0

012

)

(00

12)

(0

016

)

The Journal of Human Resources768

Tabl

e 5

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

Fath

errsquos

edu

catio

n0

011

(00

02)

A

ge o

f fa

ther

minus0

001

at c

hild

birt

h(0

002

)C

onst

ant

010

50

104

028

80

241

012

9(0

089

)(0

089

)(0

094

)

(01

35)

(01

41)

R2

008

790

0943

010

180

0885

008

55N

umbe

r of

obs

erva

tions

792

87

928

792

84

766

454

1N

umbe

r of

fam

ily

311

23

112

311

21

869

173

2cl

uste

rs

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Kantarevic and Mechoulan 769

Tabl

e 6

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

003

80

084

009

003

10

122

(00

18)

(0

025

)

(00

36)

(0

053

)(0

049

)

d[se

cond

-bor

n]0

071

008

2times

10minus4

002

2(0

022

)

(00

31)

(0

051

)(0

044

)d[

thir

d-bo

rn]

003

10

012

003

5(0

028

)(0

041

)(0

037

)d[

four

th-b

orn]

006

10

041

(00

37)

(00

33)

d[fif

th-b

orn]

000

7(0

028

)A

ge o

f th

e m

othe

r at

chi

ldbi

rth

000

30

004

000

90

008

001

1(0

002

)(0

002

)(0

003

)

(00

04)

(0

003

)

d[M

ale]

minus00

3minus0

041

000

7minus0

094

minus00

82(0

016

)(0

016

)

(00

2)(0

026

)

(00

18)

A

ge0

028

002

80

051

001

20

037

(00

07)

(0

009

)

(00

1)

(0

008

)(0

011

)

Age

23

times10

minus43

times10

minus45

times10

minus48

times10

minus54times

10minus4

(9times

10minus5

)

(10minus5

)

(10minus5

)

(9times

10minus5

)(1

0minus4)

d[

Whi

te]

001

20

124

004

90

033

002

2(0

021

)(0

023

)(0

026

)(0

034

)(0

028

)M

othe

rrsquos

educ

atio

n0

024

002

50

027

001

80

027

(00

04)

(0

004

)

(00

05)

(0

005

)

(00

04)

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 14: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

The Journal of Human Resources768

Tabl

e 5

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

Fath

errsquos

edu

catio

n0

011

(00

02)

A

ge o

f fa

ther

minus0

001

at c

hild

birt

h(0

002

)C

onst

ant

010

50

104

028

80

241

012

9(0

089

)(0

089

)(0

094

)

(01

35)

(01

41)

R2

008

790

0943

010

180

0885

008

55N

umbe

r of

obs

erva

tions

792

87

928

792

84

766

454

1N

umbe

r of

fam

ily

311

23

112

311

21

869

173

2cl

uste

rs

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Kantarevic and Mechoulan 769

Tabl

e 6

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

003

80

084

009

003

10

122

(00

18)

(0

025

)

(00

36)

(0

053

)(0

049

)

d[se

cond

-bor

n]0

071

008

2times

10minus4

002

2(0

022

)

(00

31)

(0

051

)(0

044

)d[

thir

d-bo

rn]

003

10

012

003

5(0

028

)(0

041

)(0

037

)d[

four

th-b

orn]

006

10

041

(00

37)

(00

33)

d[fif

th-b

orn]

000

7(0

028

)A

ge o

f th

e m

othe

r at

chi

ldbi

rth

000

30

004

000

90

008

001

1(0

002

)(0

002

)(0

003

)

(00

04)

(0

003

)

d[M

ale]

minus00

3minus0

041

000

7minus0

094

minus00

82(0

016

)(0

016

)

(00

2)(0

026

)

(00

18)

A

ge0

028

002

80

051

001

20

037

(00

07)

(0

009

)

(00

1)

(0

008

)(0

011

)

Age

23

times10

minus43

times10

minus45

times10

minus48

times10

minus54times

10minus4

(9times

10minus5

)

(10minus5

)

(10minus5

)

(9times

10minus5

)(1

0minus4)

d[

Whi

te]

001

20

124

004

90

033

002

2(0

021

)(0

023

)(0

026

)(0

034

)(0

028

)M

othe

rrsquos

educ

atio

n0

024

002

50

027

001

80

027

(00

04)

(0

004

)

(00

05)

(0

005

)

(00

04)

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 15: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

Kantarevic and Mechoulan 769

Tabl

e 6

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Hig

h Sc

hool

Com

plet

ion

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

003

80

084

009

003

10

122

(00

18)

(0

025

)

(00

36)

(0

053

)(0

049

)

d[se

cond

-bor

n]0

071

008

2times

10minus4

002

2(0

022

)

(00

31)

(0

051

)(0

044

)d[

thir

d-bo

rn]

003

10

012

003

5(0

028

)(0

041

)(0

037

)d[

four

th-b

orn]

006

10

041

(00

37)

(00

33)

d[fif

th-b

orn]

000

7(0

028

)A

ge o

f th

e m

othe

r at

chi

ldbi

rth

000

30

004

000

90

008

001

1(0

002

)(0

002

)(0

003

)

(00

04)

(0

003

)

d[M

ale]

minus00

3minus0

041

000

7minus0

094

minus00

82(0

016

)(0

016

)

(00

2)(0

026

)

(00

18)

A

ge0

028

002

80

051

001

20

037

(00

07)

(0

009

)

(00

1)

(0

008

)(0

011

)

Age

23

times10

minus43

times10

minus45

times10

minus48

times10

minus54times

10minus4

(9times

10minus5

)

(10minus5

)

(10minus5

)

(9times

10minus5

)(1

0minus4)

d[

Whi

te]

001

20

124

004

90

033

002

2(0

021

)(0

023

)(0

026

)(0

034

)(0

028

)M

othe

rrsquos

educ

atio

n0

024

002

50

027

001

80

027

(00

04)

(0

004

)

(00

05)

(0

005

)

(00

04)

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 16: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

The Journal of Human Resources770

Tabl

e 6

(con

tinu

ed)

(1)

(2)

(3)

(4)

(5)

d[al

l sib

lings

rep

ort]

minus00

110

03minus0

034

005

90

037

(00

25)

(00

24)

(00

24)

(00

34)

(00

3)d[

both

par

ents

rep

ort]

004

50

076

012

50

088

009

8(0

019

)

(00

22)

(0

027

)

(00

36)

(0

026

)

Con

stan

t0

115

025

40

927

002

60

644

(01

6)(0

2)

(02

)

(02

34)

(02

61)

R

20

076

008

40

138

007

80

088

Num

ber

of o

bser

vatio

ns1

398

170

51

444

959

242

2N

umbe

r of

fam

ily c

lust

ers

913

811

542

308

538

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

)(1

)ndash(5

) F

amili

es o

f 2

3 4

5 a

nd 6

and

abo

ve s

iblin

gs r

espe

ctiv

ely

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 17: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

Table 7aFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Not Including Marital Status ofMother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0132 0076 0124(0029) (0037) (0051) (0073)

(0051)d[second-born] 0064 0107 0039 0029

(0026) (0041) (0062) (0045)d[third-born] 0048 0061 0047

(003) (005) (0039)d[fourth-born] minus0054 0049

(004) (0034)d[fifth-born] 0004

(0029)Observations 1422 1743 1477 999 2487

groups 934 835 562 328 573

Table 7bFixed Effects Linear Regression with Dependent Variable High School Completion(All Coefficients Deviation from Family Means) Including Marital Status of Mother

(1) (2) (3) (4) (5)

d[first-born] 0033 0089 0134 0075 0129(0029) (0037) (0051) (0073) (0052)

d[second-born] 0063 0109 0039 0033(0026) (0041) (0062) (0045)

d[third-born] 0049 0061 0049(003) (0051) (0039)

d[fourth-born] minus0053 0051(004) (0034)

d[fifth-born] 0004(0028)

Mother minus0001 minus0058 0019 0025 minus0035continuously (0051) (004) (0042) (0053) (0041)married age 1ndash14

Observations 1422 1743 1477 999 2487Groups 934 835 562 328 573

(1)ndash(5) all mothers have completed their fertility (age gt 44) all respondents assumed to have completed theireducation (age gt 24) (1)ndash(5) Families of 2 3 4 5 and 6 and above siblings respectively 10 percent sig-nificance 5 percent significance 1 percent significance Robust standard errors clustered by familyNo estimation was run with age because including both age and age of mother at birth is redundant in a fam-ily fixed effects regression The regressions also include age of mother at birth and an indicator for males

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 18: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

The Journal of Human Resources772

B The ldquoLast-Born Effectrdquo in Large Families

We now test the hypothesis that within large (more than five siblings) families the last-borns do better than the middle-borns who in turn do worst There is some support in theliterature for a so-called ldquocrunch in the middlerdquo effect25 This nonlinear pattern was alsoadvanced by Hanushek (1992) and in the context of time allocation by Lindert (1977)

There are many ways to replicate these findings Short of enough observations for eachfamily size when family size is very large the variables of interest chosen in Table 8are dummies indicating whether a child is first-born and whether a child is last-born

When omitting the age of mother at childbirth in Column 1 the first-born coeffi-cient is insignificant as earlier but the coefficient on last born is positive and signifi-cant Notice that this does not happen when we run the same regression on smallerfamilies However in Column 2 once the age of the mother at childbirth is factoredin we find that being last-born confers no advantage but that being first-born does

The fixed effects estimation (Column 3) confirms the absence of any upward trendfrom middle-born to last-born26 The interpretation of those results is similar to theones presented earlier and the same qualifications apply27

IV Birth Order and Earnings

Because education is a key in determining earnings we should like-wise find a similar birth order effect on earnings Our sample is more limited becausewe only have information on earnings for heads or wives who declare working (about36 percent of our initial sample) Nonetheless Table 9 shows that the results on hourlyearnings display the same patterns as for education namely a nonsignificant effect ofbirth order when age of the mother at birth is omitted and a significant effect when itis included Curiously for non-White non-Black families we noticed the persistenceof a robust first-born effect on earnings after controlling for education This deservesfuture research Note that with only a few hundred observations on hourly earningsfor each sibling size we cannot run meaningful OLS or fixed effects estimations bysiblings size28

V Conclusion

We have shown how the omission of the age of the mother at child-birth effect results in an underestimation of the impact of being first-born and an over-

25 ldquoFor almost as long as sociologists have been studying who gets ahead they have found that kids from largefamilies do more poorly than those from small ones There is however one exception to this Last-born childrenfrom very large families seem to fare quite well the middle kids do worstrdquo Conley (2004) p 66 and p 6926 Adding whether the mother was continuously married during the childrenrsquos first years does not changethe results qualitatively and similar small size problems arose when adding family income and employmentstatus of the mother for different age ranges of each child Results are available from the authors27 Restricting the sample to both parents present at childbirth yields similar results However we cannotdo the same with large families with complete information on all siblings The sample becomes too small28 Fixed effects estimations for the entire sample confirm the OLS results though

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 19: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

Kantarevic and Mechoulan 773

Table 8Regression with Dependent Variable Completed Education (Large Families)

Family OLS Fixed Effects

(1) (2) (3)

d[first-born] 0078 0427 0084(0187) (0195) (0039)

d[lastborn] 0343 minus0047 0008(0126) (0138) (0032)

d[White] minus0059 minus0164(0159) (0149)

d[first-born] times d[White] 0099 0098 minus0001(0322) (0321) (0059)

d[last-born] times d[White] minus0052 minus0039 minus0057(0198) (0195) (0047)

Total number of siblings minus0059 minus009(0031) (0033)

Age of the mother at 0055 0002childbirth (0011) (0002)

d[Male] minus0322 minus0331 minus0085(0089) (0087) (0017)

Age 0168 0172(0066) (0068)

Age2 minus0002 minus0002(00008) (00001)

Motherrsquos education 0175 0180(0022) (0021)

d[all siblings report] 0112 0176(0151) (0151)

d[both parents report] 0449 0429(0129) (0126)

Constant 6755 5207 0707(1317) (1329) (0066)

R2 01114 01292 001Number of observations 2422 2422 2487Number of family clusters 538 538 573

(1)ndash(3) all mothers have completed their fertility (age gt 44) all respondents assumed to have completedtheir education (age gt 24) all respondent from families gt 5 siblings Robust standard errors clustered byfamily 10 percent significance 5 percent significance 1 percent significance

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 20: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

The Journal of Human Resources774

Tabl

e 9

OL

S R

egre

ssio

n w

ith

Dep

ende

nt V

aria

ble

Log

Hou

rly

Wag

e in

200

1

(1)

(2)

(3)

(4)

(5)

d[fir

st-b

orn]

010

20

067

010

60

126

014

3(0

043

)

(00

44)

(00

45)

(0

057

)

(00

51)

To

tal n

umbe

r of

sib

lings

minus00

21minus0

024

minus00

24minus0

04

(00

07)

(0

007

)

(00

09)

(0

014

)

Age

of

the

mot

her

at c

hild

birt

h0

008

000

80

009

(00

02)

(0

005

)(0

003

)

d[M

ale]

030

00

299

029

80

346

029

(00

245)

(00

24)

(0

024

)

(00

29)

(0

03)

Age

005

90

067

006

50

068

007

8(0

013

)

(00

13)

(0

013

)

(00

17)

(0

021

)

Age

2minus0

000

5minus0

000

6minus0

001

minus00

005

minus00

01(0

000

2)

(0

000

2)

(0

000

2)

(0

000

2)

(00

002)

d[W

hite

]0

144

010

90

096

001

90

117

(00

35)

(0

036

)

(00

36)

(0

043

)(0

046

)

d[fir

st-b

orn]

timesd[

Whi

te]

minus00

86minus0

062

minus00

65minus0

094

minus01

28(0

054

)(0

054

)(0

054

)(0

064

)(0

061

)

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 21: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

Kantarevic and Mechoulan 775

Mot

herrsquo

s ed

ucat

ion

004

80

044

004

30

021

004

5(0

006

)

(00

06)

(0

006

)

(00

08)

(0

009

)

d[co

mpl

ete

info

on

all s

iblin

gs]

010

20

063

007

10

095

(00

32)

(0

035

)(0

036

)

(00

44)

d[

both

par

ents

rep

ort]

009

20

098

009

70

169

(00

29)

(0

029

)

(00

29)

(0

037

)

Fath

errsquos

edu

catio

n0

045

(00

06)

Age

of

fath

er a

t chi

ldbi

rth

000

2(0

004

)C

onst

ant

023

50

254

010

5minus0

262

minus01

62(0

263

)(0

261

)(0

264

)(0

342

)(0

400

)R

20

1594

016

290

1658

022

060

1759

Num

ber

of o

bser

vatio

ns3

000

300

03

000

205

91

962

Num

ber

of f

amily

clu

ster

s1

575

157

51

575

107

51

015

(1)ndash

(5)

all

mot

hers

hav

e co

mpl

eted

thei

r fe

rtili

ty (

age

gt44

) a

ll re

spon

dent

s as

sum

ed to

hav

e co

mpl

eted

thei

r ed

ucat

ion

(age

gt24

) an

d ha

ve a

t lea

st o

ne o

ther

sib

ling

(4)

d[b

oth

pare

nts

repo

rt]

=1

and

(5)

d[c

ompl

ete

info

on

all s

iblin

gs]

=1

1

0 pe

rcen

t sig

nific

ance

5 p

erce

nt s

igni

fican

ce

1

per

cent

sig

nific

ance

Rob

ust s

tand

ard

erro

rs c

lust

ered

by

fam

ily

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 22: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

The Journal of Human Resources776

Appendix 1Description of the Variables

Variable Description

Years of completed educationa Years of education reported in the most recent yearCompleted high schoola = 1 if years of completed education greater than

or equal to 12Hourly earningse = hourly earnings in 2001 if the index person is

a head or a wife of a household and missing otherwise

Agea = the age of index person based on the year of birth

Malea = 1 if the gender of index person is maleWhiteb = 1 if the race of mother of index person is White

or if the race of mother of index person is missing but the race of father is White

Number of siblingsc The total number of childbirths reported by the mother of index person if mother older than 44 in the last year in which she reported otherwise set to missing

Birth orderc (first-born etc) The birth order of index personFamily incomeb = Total income of the household The average

family income is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+ years for ages 7ndash14 and 7+ years for ages 1ndash14)

Mother marriedd = 1 the mother is continuously married during the relevant period and 0 if not The average marital status of mother is calculated only if it is available for at least 50 percent of years in the relevant time period (3+ years for ages 1ndash6 4+years for ages 7ndash14 and 7+ years for ages 1ndash14)

Information on all siblings = 1 if all siblings present in the sample and reportBoth parents report the = 1 if both the mother and father of index person

childbirth report the birth of index person

Data Sourcesa Individual PSID fileb Family PSID filec Childbirth and Adoption History Filed Marriage History Filee Hours of Work and Wages File

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74

Page 23: Birth Order, Educational Attainment, and EarningsI. Introduction Whether birth order affects performance has been an open empirical question for decades. In this study, we examine

Kantarevic and Mechoulan 777

estimation of the impact of being last-born At this point however the age of themother at childbirth should be interpreted broadly as a proxy for a set of maternalinputs Most importantly fixed-effects estimations confirmed the presence of a sig-nificant positive first-born effect and the absence of either specific middle-born orlast-born effects among large families First-born children on average retain an advan-tage acquired early on throughout both their educational and professional life Thiseffect is enhanced within White families

Our data do not permit to contribute to the recent debate over the impact of familysize on educational attainment Therefore while we tentatively agree with Hanushek(1992) that smaller family sizes may be responsible for a rise in scholastic perfor-mances over cohorts in the United States we would like to emphasize that this effectis compounded by a corresponding increase in the proportion of first-born children

References

Birdsall Nancy 1979 ldquoSiblings and Schooling in Urban Columbiardquo Dissertation YaleUniversity Department of Economics

Behrman Jere R 1986 ldquoBirth Order Nutrition and Health Intra-Household Allocation inRural Indiardquo University of Pennsylvania Unpublished

Behrman Jere R and Paul Taubman 1986 ldquoBirth Order Schooling and Earningsrdquo Journalof Labor Economics 4(3)S121ndashS45

Black Sandra E Paul J Devereux and Kjell G Salvanes 2005 ldquoThe More the Merrier TheEffect of Family Composition on Childrenrsquos Educationrdquo Quarterly Journal of Economics120(2)669ndash700

Booth Alison L and Hiau Joo Kee 2005 ldquoBirth Order Matters The Effect of Family Sizeand Birth Order on Educational Attainmentrdquo IZA working paper No 1713

Bronars Stephen G and Jeff Grogger 1994 ldquoThe Economic Consequences of UnwedMotherhood Using Twin Births as a Natural Experimentrdquo American Economic Review84(5)1141ndash56

Conley Dalton 2004 The Pecking Order Pantheon Books New YorkConley Dalton and Rebecca Glauber 2004 ldquoParental Educational Investment and Childrenrsquos

Academic Risk Estimates of the Impact of Sibship Size and Birth Order from ExogenousVariation in Fertilityrdquo Journal of Human Resources This issue

Geronimus Arline T Sanders Korenman and Marianne M Hillemeier 1994 ldquoDoes YoungMaternal Age Affect Child Development Evidence from Cousin Comparisons in theUnited Statesrdquo Population and Development Review 20(3)585ndash609

Hanushek Eric A 1992 ldquoThe Trade-off between Child Quantity and Qualityrdquo Journal ofPolitical Economy 100(1)84ndash117

Hofferth Sandra L and Lori Reid 2002 ldquoEarly Childbearing and Childrenrsquos Achievementand Behavior over Timerdquo Perspectives on Sexual and Reproductive Health 34(1)41ndash49

Kessler Daniel 1991 ldquoBirth Order Family Size and Achievement Family Structure andWage Determinationrdquo Journal of Labor Economics 9(4)413ndash26

Lindert Peter H 1977 ldquoSibling Position and Achievementrdquo Journal of Human Resources12(2)198ndash219

Lopez-Turley Ruth N 2003 ldquoAre Children of Young Mothers Disadvantaged Because ofTheir Motherrsquos Age or Family Backgroundrdquo Child Development 74(2)465ndash74