Did Legalized Abortion Lower Crime

8/17/2019 Did Legalized Abortion Lower Crime

1/29

Did Legalized Abortion Lower Crime?Author(s): Ted JoyceSource: The Journal of Human Resources, Vol. 39, No. 1 (Winter, 2004), pp. 1-28Published by: University of Wisconsin PressStable URL: http://www.jstor.org/stable/3559003

Accessed: 25/08/2010 18:59

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at

http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless

you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you

may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at

http://www.jstor.org/action/showPublisher?publisherCode=uwisc.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed

page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of

content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms

of scholarship. For more information about JSTOR, please contact [email protected].

University of Wisconsin Press is collaborating with JSTOR to digitize, preserve and extend access to The

Journal of Human Resources.

http://www.jstor.org

http://www.jstor.org/action/showPublisher?publisherCode=uwischttp://www.jstor.org/stable/3559003?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/action/showPublisher?publisherCode=uwischttp://www.jstor.org/action/showPublisher?publisherCode=uwischttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/3559003?origin=JSTOR-pdfhttp://www.jstor.org/action/showPublisher?publisherCode=uwisc


2/29

Did Legalized Abortion

Lower

Crime?

Ted

Joyce

ABSTRACT

In

this

paper

I

compare changes

in

homicide

and arrest

rates

among

co-

horts

born

before

and

after

the

legalization of

abortion to

changes

in

crime

in

the same

years

among

similar

cohorts

who were

unexposed

to

le-

galized

abortion. I

find

little consistent

evidence

that the

legalization of

abortion

in

selected states around

1970,

and

then

in

the

remaining

states

following

Roe

v.

Wade,

had an

effect

on

recent

crime

rates.

I

conclude

that

the

dramatic association as

reported

in

a

recent

study

is most

likely

the result

of

unmeasured

period effects

such

as

changes

in crack

cocaine use.

I. Introduction

In a recentandcontroversial

rticle,

Donohue

and Levitt

(2001)

pre-

sent

evidence hat

he

legalization

of

abortionn

1973

explains

over half

of

the recent

decline n crimeacross

he

UnitedStates.

A

50

percent

ncrease

n

the mean

abortion

ratio s

associated

with

an 11

percent

decrease n violent

crime,

an 8

percent

decrease

in

property

rime and

a

12

percent

decrease n murder.These effects are

generally

largerandmorepreciselyestimated han he effectsof incarcerationndpoliceman-

power.

Moreover,

hey

conclude that

the

full

impact

on crime

of

Roe

v.

Wadewill

not

be felt for

another

20

years.

To

quote,

"Ourresults

suggest

that all

else

equal,

Ted

Joyce

is

a

professor

of

economicsat Baruch

College

and a

researcher

with the National Bureau

of

EconomicResearch.

Thisworkwas

supported

by

a

grant rom

the

Open

Society

Institute.John

Donohue

III

and Steven

Levitt

graciously

shared

their

data and

programs,

which

greatly acilitated

the

author's

analysis. They

also

providedhelpful

comments n

earlier

drafts.

The

author

also

thanks

Greg

Colman

or

researchassistanceand

Robert

Kaestner,

Michael

Grossman,

Sanders

Korenman,

Philip

Cook,

Phillip

Levine,

John

Lott,

and numerous eminar

participants

or helpful

comments.He states

that the views and errors

in

this

manuscript

re

his

and not

those

of

the

Open Society

Institute,

Baruch

College,or the NationalBureauof EconomicResearch.The data usedin this article can be obtained

beginningAugust

2004

throughJuly 2007from

Dr.

Ted

Joyce,

NationalBureau

of

EconomicRe-

search,

365

Fifth

Avenue,

th

Floor,

New

York,

NY

10016-4309.

[Submitted

May

2002;

acceptedJuly

2002]

ISSN

022-166X

?

2004

by

the

Boardof

Regents

of

the

University

of

Wisconsin

System

THE

JOURNAL OF

HUMAN

RESOURCES

*

XXXIX

*

1


3/29

2 The

Journal

of

HumanResources

legalized

abortion

will

account

or

persistent

declines of

1

percent

a

year

in

crime

over

the next two

decades"

p.

415).

Given the social costs associated

with crime

and the controversy surrounding bortion,a causal link between abortion and

crime

has

profound

mplications

or social

policy.

The

purpose

of this

paper

s to

analyze

he associationbetween

egal

abortion nd

crime.

The

primary

differencebetween

my analysis

of abortionand

crime

and that

of

Donohue and

Levitt is the identification

trategy.

Donohue and

Levitt

regress

crime

ratesbetween 1985 and 1996

on abortion atios

agged

15

to 25

yearsadjusted

for

state

and

year

fixed effects.

However,

the

study period

coincides

with

the rise

and

decline

of the

crack

cocaine

epidemic,

which

many

observers

ink to the

spread

of

guns

and the

unprecedented

ncrease

n

youth

violence

(Cook

and

Laub

1998;

Blumstein

1995; Blumstein,Rivara,

and Rosenfeld

2000).

Moreover,

datafrom

po-

lice surveys,emergency ooms,and fromurinesamplesof arresteesn majormetro-

politan

areas

suggests

that the

timing

of the

arrival,

diffusion,

and decline

in

crack

use

varied

significantly

by city

(Golub

and Johnson

1997;

Cork

1999;

Grogger

and

Willis

2000).

Thus,

even

in models with stateand

year

fixed

effects,

the

relationship

between abortion

and crime

may

be

biased

by

differences

n

within-state

rowth

n

cocaine

marketsover

time,

a classic

problem

of omitted

variables.

A

crudesolution

is

to include

controlsfor

state-specific

inear or

quadratic

rends.

However,

this is

not

possible

in

the

contextof Donohueand Levitt's

model,

becausethe

trend erms

remove

all variation n the abortion atio.

I take

a different

approach

o the

identification

f an

abortion-crime

exus. I use

the

early legalization

of abortion n selected states

prior

to Roe v. Wadeand then

national

egalization

afterRoe

in

the

remaining

tates

to

identify

exogenous

shifts

in

unintended

hildbearing. pecifically,

estimate

a

reduced-form

quation

n which

changes

in arrest

and

homicide rates

among

cohorts before and after

exposure

o

legalized

abortion re

compared

o

changesamong

cohorts hatare

unexposed.1

his

is similar

to Donohue

and Levitt's fixed effect

specification,

since

identification

comes

from

changes

n

crimeand abortion cross

states.

However,

I show that hese

estimates

are sensitiveto the

years

thatare

analyzed,

which I

interpret

s an omitted

variable

problem

relatedto

unobserved,

tate-specificperiod

effects.

I then use a

difference-in-differencestimatorbased

on

a within-state omparison roup

o net

out

changes

n

crime

associated

with

hard-to-measureactors

hat

vary

by

state and

year,

such

as

the

spread

of crackcocaine and its

spillover

effects.

In

these

analyses

I

find no

effect

of abortion

egalization

on crime

regardless

of

the

years

analyzed.

The difference-in-difference

trategy

has two other

advantages

n

an

analysis

of

abortion

nd

crime.

First,

Donohueand Levittuse the ratioof abortions

o

birthsas

an inverse

proxy

for unwantedbirths.

However,

abortion s

endogenous

o sexual

activity,

contraception

nd

childbearing.

A rise in abortion

may

have

relatively

ittle

effect on

unwanted

childbearing.

t is

noteworthy,

hat the abortion ate

rose from

16.3 abortions

per

1,000

women

ages

15

to 44 in

1973

to

29.3

in

1980,

an increase

of 79 percent.Overthesameperiod,however, he number f birthsper1,000women

1.

See Levine

et al.

(1999),

Gruber,

Levine,

and

Staiger

1999),

Angrist

and Evans

(1999)

for a

similar

approach pplied

o

fertility,

child

well-being,

and teen

pregnancy, espectively.

A recent

manuscript

y

Lott and

Whitley

(2001)

also

focuses on a

comparison

f cohorts

exposed

and

unexposed

o

legalized

abortion.

They report

a

positive

but

relatively

small

associationbetween

legalized

abortionand murder

rates.


4/29

Joyce

3

ages

15

to

44

was

essentially unchanged,

rom

69.2

to 68.4.

By

contrast,

here

is

substantial

vidence

thatthe

early

egalization

of abortion

n

selected states

nduced

a significantdecline in fertilitybetween 1971 and 1973 (Sklarand Berkov 1974;

Joyce

and Mocan

1990;

Levine et al.

1999;

Angrist

and Evans

1999).

This

change

in

fertility

s a more

plausible

source

of

exogenous

variationwith

which to

identify

a decline

in

unwanted

births

than

within-state

hanges

in

reported egal

abortions

between

1973 and

1985.

The

other

advantage

of the

difference-in-difference

pproach

s that

it

obviates

the need

to

measure

llegal

or

unreported

bortion

n

the

years

before

legalization.

Donohueand Levitt use

no data

on

abortion

prior

o 1973. Their

analysis

of

arrests

by single

year

of

age,

for

instance,

pertains

o

birthcohortsborn

between 1961 and

1981

where

approximately

0

percent

of the

state/age/cohort

cells

are

assigned

an

abortion atioof zero.However,demographersaveconcluded hatmostlegalabor-

tions in

the

early

1970s

replaced llegal

abortions

Tietze

1973;

Sklarand Berkov

1974).

If

the

underreporting

f

abortion

were

random

among

states,

theirestimates

would be biased

downward.As

I

show

below,

however,

the

measurement rror

s

negatively

correlated

with

the true abortion ate

in

1972

and thus the directionof

the bias

is

unknown.

II.

Conceptual

and

Empirical

Issues

A.

Abortionand

Unintended

Childbearing

As

outlined

by

Donohue

and

Levitt,

thereare

several

ways

in

which

legal

abortion

can

affectcrime.Cohort

ize is one. Fewerbirthsmean

ewercriminals n

subsequent

years.

Second,

egal

abortion

may

also

affect crimerates

hrough

a relative

decrease

in

fertility

ates

among

poor,

young,

and

minority

women.Since children

rom

disad-

vantaged

backgrounds

re

more

ikely

to

commit

crimesas

teens

or

adults,

he

result

of a selective reduction

n

childbearing

s a

drop

in crime

rates

approximately

5

to 25

years

later.

Third,

even if the decline in

fertility

rates caused

by legalized

abortion

were

distributed

qually

among

all

women,

a

fall

in

unintended

hildbearing

could

bring

abouta fall in crime if thosebornfromunintended

pregnancies

were

more

likely

to

commit

crime than ndividuals rom

pregnancies

hat

were

intended.

DonohueandLevitt's

dentification

trategy

s to correlate rime

ratesand

arrests

to

lagged

abortion

atios

adjusted

or

state

and

year

fixed

effects.

Abortionratios

serve as

an

inverse

proxy

for unwanted

hildbearing.

n their

analysis

of arrests

of

youths

15

to 24

years

of

age,

they

regress

arrests

by single

year

of

age

on the

abortion

ratio

in the

year

before a cohortwas born.

Thus,

arrests

of

18-year-olds

n

1988

in

state are

correlatedwith

the

abortion

atio

in

state

in

1969

(t-18-1).

B. Periodand CohortEffects

The

biggest

challenge

o

identifying

a

cohort

effect associatedwith

egalized

abortion

is the

potential

confounding

rom

strongperiod

effects such as the

spread

of

crack

cocaine.

Therewas an

unprecedented

ise

in

youth

homicide

between

1985 and

1993.

The

rise

among

blacks

greatly

exceeded

that of whites and almost all

the

growth

n


5/29

4

The Journal

of HumanResources

homicide

nvolved

handguns

Blumstein

1995; 2000;

Cook and

Laub

1998).

Crimi-

nologists

have

largely

attributedhe

growth

n

youth

homicide

o the violent

develop-

mentof crackcocainemarketsn poorurban enters Blumstein,Rivara,andRosen-

feld

2000).

The lack

of consistentdata on the extent of cocaine

use or the

spread

of

illegal

handguns,

however,

has limited

empirical

work.

Despite

the lack of

data,

several

sources

suggest

that the

introduction f crack

occurred

n

the

mid-to-late

1980s

(Cork

1999;

Grogger

and Willis

2000;

Caulkins

2001).

Grogger

and Willis

(2000)

surveyed

police

departments

n

27

metropolitan

areas as

to the

year

in

which crack was first noted and

compared

responses

rom

the

survey

with

changes

n

indications

of

drug

use from

emergency

room ncidents

as collected

by

the

Drug

Abuse

Warning

Network

(DAWN).

Arrivaldates tended

to

be earliest

n

East and West Coast cities

and

later

for cites in

the

Midwest.Cork

(1999) used dataon drugarrestsandgun homicides to associatechanges n crack

market

activity

and

youth

murder

ates. He also

found

that clusters

of

drug

arrests

began

first

in the West and Northeastbefore

moving

inland.

The

peak

in crackuse and ts declinefollowed

a

similar

pattern.

Analyses

of

urine

among

arrestees

rom

the

Drug

Use

Forecasting

DUF)

program

uggest

thatcrack

use

began

to fall around

1989

in

New

York,

Philadelphia,

nd

Los

Angeles

but later

andmore

slowly

in

Cleveland,

Chicago,

and

Indianapolis

Golub

andJohnson

1997).

For

example,

the

proportion

f arrestees hat tested

positive

for crack/cocaine

n

1989

exceeded

70

percent

n New Yorkand

Philadelphia,

0

percent

n

Washington

D.C. and 56

percent

n

Los

Angeles.

In

Cleveland,

Chicago,

Dallas, Denver,

Hous-

ton,

Indianapolis,

Kansas

City,

San

Antonio,

andSt.

Louis,

the

prevalence

f crack/

cocaine

among

arrestees

anged

rom

approximately

0 to 55

percent

n

1989

and

in

several cities

actually

rose in the

early

1990s.

Several

points

from this discussionare relevant.

First,

data on

crackuse

by

state

and

year

are

too

incomplete

o

apply empirically.

Second,

what

is known

suggests

thatcrack

markets

developed

n different ities at different imes

and thus

represent

a

state-year

eriod

effect that s not

captured y

national rends.

Third,

he dataalso

suggest

that

New York

City

and Los

Angeles

were

early

sites of crack

markets.Not

only

are these

the

largest

cities

in

the

two

largest

states,

but abortion

became

egal

in both statesroughly hreeyearsbeforeRoe.Thus,DonohueandLevitt'sevidence

that

crime fell

earlierand faster

in

the

early legalizing

states

may

be

spurious,

a

resultof the

differential

iming

in the evolution of crackmarkets.

The

potential

confounding

rom

time-varyingperiod

effects

is

illustrated

y

the

time-seriesof

age-

and

race-specific

homiciderates.

Figure

a shows

homiciderates

for

white

teens

(ages

15 to

19)

and

young

adults

ages

20

to

24)

in

repeal

and

nonre-

peal

states

rom 1985

to

1997;

Figure

b

presents

he

corresponding

eries

or blacks.

Repeal

states

arethose that

egalized

abortion etween

1969

and

1970:

Alaska,

Cali-

fornia,

Hawaii,

New

York,

and

Washington.

also include

Washington

D.C.

among

the

early egalizers.2

Abortionbecame

egal

in

the

nonrepeal

tates

n

1973

with

the

SupremeCourtdecisionin Roe v. Wade.

2.

Washington

D.C. has

not

been treated

as an

"early

egalizer"

n

previous

analyses.

However,

he

1969

decision n United

Statesv. Vuitch endered

he District'sabortionaw unconstitutional.

s a

result,

writes

Lader,

"Washington's

bortion acilitiessoonranked

mong

he busiest n the

country,

with

20,000

patients

in

1971"

(Lader

1974,

p.

115).

Data on abortion

n

1971

from the

Center

or Disease Control

1972)

support

Lader'sobservation.

The residentabortion

atio

(abortions

er

1,000

live

births)

n D.C. in

1971


6/29

Joyce

5

Two

points

are

noteworthy.

First,

homicide rates

rise

earlier

n

repeal

than in

nonrepeal

tatesconsistentwith

the

earlierarrival

f

crack

n

New York

and

Califor-

nia.Second, hecurvilinearrendnhomiciderates s similaramong eens andyoung

adults within

repeal

and

nonrepeal

tates and is

inconsistent

with

a

strong

cohort

affect associatedwith

legalized

abortion.Most

teens

in

1985

were bornbefore 1970

and

thus

were

unexposed

o

legalized

abortion n utero.

By

1990, however,

eens

in

repeal

stateshad

been

bornafter 1970 and were thus

exposed.

Put

differently,

eens

in

repeal

states

in

1990

represent

he first

cohort

of

"more wanted"births.

Thus,

evidenceof

a

cohorteffect associated

with

the

pre-Roe

egalization

f abortionwould

be a relative

decrease

n

teen homiciderates

n

repeal

states

beginning

around

1988,

followed

five

years

later

by

a

similardecline

amongyoung

adults.There s

no

evi-

dence

of

such

a

pattern mong

eitherblacks or whites. In

fact,

the coincidentmove-

ment in homicide ratesby teens and young adults s more consistentwith strong

period

effects.

In

order o

isolate

a

cohorteffect associated

with

the

legalization

of

abortion,

esearchersmust

adjust

or

these dramatic rends

n

crime

within-states.

C.

Mismeasurement

nd

Endogeneity

of

Abortion

Anotherdrawbacko

Donohue

and

Levitt'

empirical trategy

s the

mismeasurement

of

abortion nd ts

endogeneity

n

the

years

after

egalization.

Demographers

stimate

that

approximately

wo-thirds

f all

legal

abortions

eplaced

llegal

ones

in

the first

year

after

egalization.

Estimates

are based on

the

change

in

births

between 1970

and 1971

compared

o the numberof

reported

bortions n 1971

(Sklar

and Berkov

1974;

Tietze

1973).

As

noted

above,

Donohue

and Levitt

have no dataon

abortion

for

cohorts

born

before

1974

and thus assume

a zero abortion atio

for

more than

half

their

observations.

A

facile

argument

s to

assume hat

any

error s

likely

random

and

estimatesare biased downward.But this

assumption

s

decisively

contradicted

by

the

data. As a

simple example,

Kansashad an abortionratio of

414

per

1,000

live births

n

1973. Donohueand

Levitt

assume he abortion

atio n

Kansas

s

zero

in 1972.

However,

datacollected

by

the Centers or

Disease

Control

CDC)(Centers

for

Disease Control

1974)

indicate

that Kansashad an observedabortionratio of

369

per 1,000

live births

n

1972

Going further,

estimated

he

residentabortion

rate

n

1972

using

published

CDC

dataand the

algorithm

sed

by

AGI for

assigning

abortions

by

state of residence

n

1973. The correlation

between

residentabortion

rates

or

ratios

in

1972

and

1973

is

0.95.

In

other

words,

states

with

the

greatest

abortion

ratios

in

1973

had the

greatest

abortion atios in 1972.

By

assuming

he

abortion

atiowas

zero

in

the

45

nonrepeal

tates and

Washington,

D.C.,

Donohue

and

Levitt build

in

an error hat

s

negatively

correlatedwith the trueabortion ate.

As

a

result,

the directionof the

bias

is unknown.3

was

793,

more han

double

hatof

New

Yorkor

California.

hus,

include

Washington,

.C. in all

analyses

as

a

repeal

state.

However,

my

results

are not sensitiveto its inclusion

as

a

repeal

state.

3. To illustrate,LetA72be the observedabortion atio n 1972,a72 he actualabortion atio and

u72

the

error.

Thus

A72

=

a72

+

U72

Recall

that

A72

=

0 in

their

analysis;

hus,

a72

>

0 and

u72

<

0 and

the

true abortion atioand the

error

are

negatively

correlated;moreover,

given

the

strongpositive

correlation

etween he observedabortion

ratios

n

1972

and

1973 noted

above,

the correlation etween

a72

and

u72

is

undoubtedly

obust.

Now

let


7/29

6

TheJournalof Human

Resources

The other

difficulty

with the abortion atioas a measure

of unwanted

hildbearing

is

that abortion s

endogenous

o sexual

activity,

contraception

nd

fertility.

Some

pregnancieshatwereaborted n the mid- to late 1970smaynothave been conceived

hadabortion emained

llegal.

This weakens he

linkbetweenabortion ndunwanted

childbearing.

n

addition,

Donohue and Levitt use

the abortion

ratio

(abortions/

births)

and referto

it as the abortion ate

(abortions/women).

This exacerbates he

endogeneityproblem

and makes the abortion atio a

less clear

proxy

for unwanted

births.

The

growth

n AFDC and Medicaid

n

the

1970s,

for

instance,

changed

he

price

of

a

birth or

manypoor

women.

Thus,

the abortion

atio

may vary

for reasons

unrelated o unwanted

hildbearing.

D. Selectedreplicationof Donohue and Levitt's indings

To illustrate

ome of

the difficultieswith Donohueand

Levitt's identification trat-

egy,

I have

replicated

heir

key findings

and

presented

hem

n

Table 1. Their

primary

evidence of an association

between

abortion

and crime

comes from two sets of re-

gressions.

In the

first,

rates of violent

crime,

property

rime,

and murder

by

state

and

year

are

regressed

n what he authors

erm,

he effective

abortion ate.The latter

is an

average

of state

abortion atios

rom

1970 to

1985

weightedby

the

proportion

f

arrestees

"exposed"

o

legalized

abortion.4

n

the

second

set of

regressions,

he

loga-

rithmof arrests or violent

and

property

rime

by single

year

of

age

is

regressed

on

the state abortion atio

the

year

beforethe

cohortwas

born.

Arrests

pertain

o teens

and

young

adults 15 to 24

years

of

age

between1985 and

1996,

which

correspond

to

birth cohorts rom 1961

to

1981.

Donohue and Levitt

assume that the abortion

ratio is zero for cohorts

bornbefore 1974.

Row

1 of

Table

1

replicates

he

key

index crime

regressions

rom Donohue

and

Levitt

(2001,

Table

4).

Only

the coefficienton the effective

abortion ate s shown.

As Donohueand Levitt

note,

an

increaseof one standard

eviation

n

the effective

abortion

atio,

an increase

of

approximately

00

abortions

er

1,000

live

births,

ow-

ers crime between 9

and 13

percent.

As Donohue and

Levitt

demonstrate,

hese

estimatesare

quite

robust

o

changes

n the

set of included

variables.5

However,

he

estimates are very

sensitive

to the

period analyzed,

as

shown

in

Rows 2

and 3.

Specifically,

f

the same

specification

s

in

Row 1 is estimated

or the

years

1985

to

C be the crime

rate and

following

Maddala

1992)

write the

simple

relationship

etweencrime and

the

observedabortion

atio as used

by

Donohueand Levitt as follows:

(2)

C

=

A

+

e

p

<

0

Substitute

a

+

u)

for A in

Equation

. It is

straightforward

o show

that

plim

b

=

P(oaa

oau)/(aa

+

2oa,

+

oua)

where

aij

is the relevantcovariance.

Because

a,,.

and

6,,

are

both

positive

and

o,au

s

negative,

he effect

of the systematic rroron theplim of b is unknown n this simplecontext.

4. In 45 states

plus

the District

of Columbia

hey

assumethe abortion

atio was zero between

1961

and

1972. For

the other five states

they

estimateabortions or 1970-72

by backcasting inearly

rom

1973

totals and then assumea zero

abortion atiofrom 1961 to

1969.

5.

The

important xception

s when

they

include a

state-specific

rend

erm

(Donohue

and Levitt

2001,

Table

5).


8/29

Joyce

7

Table

1

The

Relationship

between Abortion

and

Crime:

Regressions

of

Total Index

Crime

Rates and Log Arrests by Single Year of Age for 15- to 24-Year-Olds

Panel

A: Index Crime Rates on Effective

Abortion Ratio

Violent

Crime

Property

Crime Murder

Row/period

1.

1985-97

-0.129 -0.091 -0.121

(0.024) (0.018) (0.047)

2.

1985-90

0.017 -0.033

0.276

(0.045)

(0.018)

(0.066)

3. 1991-97

-0.209

-0.186 -0.338

(0.035)

(0.034)

(0.053)

Panel B:

Log

Arrests

on

Lagged

Abortion

Ratio

Violent Crime

Property

Crime Murder

Arrest

Arrest Arrest

4.

1985-96 -0.015

-0.040

-0.028

(0.003) (0.004) (0.006)

5. 1985-90

0.020

-0.028 0.041

(0.006)

(0.006)

(0.013)

6.

1991-96

-0.011

-0.041 -0.013

(0.007) (0.006)

(0.007)

7. Birth

cohorts

-0.009 0.011 0.009

1974-81

(0.008) (0.008) (0.022)

Figures

standard

rrors)

re he

coefficients

on the

effective

abortion

atio

PanelA)

or

the

lagged

abortion

ratio

Panel

B).

Rows

1

and4

replicate

he

regressions

rom

Tables

4

and 7 in DonohueandLevitt

(2001).

Rows

2, 3, 5,

and

6

estimate he

same

specifications

ut

for

the

designated ubperiods.

Row

7

limits the

regressions

f

log

arrests o cohorts or

which

abortion ata

are available.This

sample

ncludesarrests

f

individuals

15 to

22

years

of

age

and

years

1989

to 1996.

Following

Donohueand

Levitt,

the abortion

ratio

has

been

multipliedby

100

in

all

regressions.


9/29

8

The Journal

of

Human Resources

45

-

40-> . TTeen Repeal

?35

-

'~

30-

Young

Adults

Repeal

?*

25-

1

20-

15

-

Young

Adults

Nonrepeal

10-

5

-

Teen

Nonrepeal

0 I

85 86 87

88 89 90 91 92 93 94

95 96

97

Year

Figure

la

White

Homicide Rates

for

Teens

(15-19)

and

Young

Adults

(20-24)

by

Repeal*

and

Nonrepeal

States,

1985-97

1990,

the coefficient

on

the effective abortion ratio becomes

positive

and

statistically

insignificant

in

the case

of

violent

crime,

negative

but

greatly

reduced

in

the

case

of

property

crime

(p

<

.10),

and

positive,

very

large,

and

statistically significant

in

the case of murder. When

I estimate

the model for the

years

1991

to

1997,

the results

are

largely

reversed. For

each

crime,

the

coefficient

on the

effective

abortion

ratio

is

negative

and

statistically

significant.

Indeed,

the

change

in the effective

abortion

ratio between

1991

and 1997

multiplied by

its coefficient

in

the murder

regression

explains

the entire fall in

homicide

between

1991

and

1997.6

Estimates

in

Panel

B

are

from the

same exercise

as

in

Panel A

but

applied

to

age-

specific

arrests. In these

regressions,

the natural

logarithm

of arrests for

15-

to

24-

year-olds by single year

of

age

are

regressed

on the

abortion ratio in

the

year

before

each

cohort

was born. The unit of observation is the

cohort/state/age

cell. Estimates

in Row 4

again replicate

the results

in

Donohue

and

Levitt

(2001,

Table

7);

estimates

in Rows 5 and 6 are for

the

designated subperiods.

The

pattern

observed with the

index

crimes

in

Panel

A is

repeated

in

Panel

B: abortion

is

inversely

related to arrests

(p

<

.01)

over the full

period,

but the

association

reverses

sign

for

violent crime

and murder arrests between 1985

and

1990,

and is

consistently negative

when

esti-

mated for

years

1991 and

1996.

The lack of

temporal homogeneity

in

the abortion-crime association

points

to

problems

of omitted variables.7

As

shown in

Figure

1,

murder rates

among

teens

6. The murder ate ell from

9.8

to

6.8

per

100,000

between1991

and

1997,

a

declineof 31

percent.

The

effective abortion ate

for

murder

ose from 33 to 142

per

1,000

live

birthsover

the same

period.

Thus,

the

predictedchange

in the

log

murderrate

based on the

regression

result

for

murder n Row 3 is

-0.00338*(142

-

33)

=

-0.368

or

36.8

percent.

7. Donohue and Levitt

(2003)

argue

hat tests of

abortionand total

crime are

weak between 1985 and

1990

becausea

relatively

mall

proportion

f all

criminalswere

exposed

o

legalized

abortion efore 1990.


10/29

Joyce

9

300-

Teen

Repeal

>250-

20000

Young

Adults

Repeal

.~

150-

0

100

-

_

^^[

Young

Adult

Nonrepeal

50

Teen

Nonrepeal

85

86

87

88 89

90 91 92

93

94

95

96 97

Year

Figure

lb

Black

Homicide

Rates

for

Teens

(15-19)

and

Young

Adults

(20-24)

by Repeal

and

Nonrepeal

States,

1985-97

*Repeal

States:

AK,

CA,

DC, HI, NY,

WA

Source:

FBI's

Supplemental

Homicide

Reports

and

young

adults

rise

rapidly

between 1985

and

1992 and then fall

precipitously.

Lagged

abortion

ratios are also

rising during

this

time.

Year

fixed effects remove

national trends in both abortion

and

crime,

but

they

do not

eliminate

confounding

from

state-specific

shocks

associated

with

say,

the diffusion of

crack cocaine.

One

solution

is

to include

controls for

state-specific

linear or

quadratic

trends

but such

terms remove all variation

in the abortion

ratio.8

The

other

notable

result

in

Table 1 is the

lack

of

any

association between abortion

and arrests when the

analysis

is limited to

cohorts

for

which data on

abortion

exist

(Table

1,

Row

7).

These

regressions

associate

arrests between

1989

and

1996

to

abor-

tion between 1974 and 1981. This is a

period

of

rapid

growth

in

reported

legal

abortion

and

there

is

substantial

variation both

within and between states.

Moreover,

the

As

evidence,

hey

point

to their

relatively

ow

effective abortion atioover this

period.

However,

he

low

figure

results from

their

inappropriate

ssumption

hat

there were

no

abortions

prior

to

1973

in the

45

nonrepeal

tates.

Early

surveillance

y

the CDC

found hat here

were

175,508

reported

bortions

n

1970,

480,259

in

1971,

and

586,760

in 1972 n the

UnitedStates

Centers

or

Disease

Control

1971,

1972,

1973).

Moreover,

he residentabortion

atio n the

repeal

states:

Alaska, California,

Washington

D.C.,

Hawaii,

New

York,

and

Washington,

was 340 in 1971

and370 in 1972

(Author's

alculations

based

on data rom

CDC

(1972,

Table

4)

and

CDC

(1974,

Table

5).

According

o

CDC

data,

he abortion atio or the entire

US

peaked

n

1981 at 358

(Koonin

et al.

1997).

In

other

words,

cohorts

born

in

repeal

states between

1971and 1973 wereexposedto a level of abortionhatexceededthemaximum verageexposure or the

entire

country

at

any

time

since abortionbecame

egal.

8.

The

adjusted

R-squared

n

a

regression

f

the

effective

abortion atio

on state

dummies,

year

dummies,

and

state-specific

inear rends

s over

0.99,

which

explains

the

sensitivity

of Donohue

and

Levitt's esti-

mates to the inclusionof

state-specific

inear

rend erms

Donohue

andLevitt

2001,

Table

5).

Moreover,

quadratic

rendsare more

appropriate

iven

the

curvilinear

rajectory

f

crime

rates,

but theirestimates

become nonsensicalwhen

such termsare

included.


11/29

10

The

Journal

of

Human Resources

matching

of

arrest rates

by single year

of

age

to

the abortion

ratio in the

year

the

cohort was

"in

utero"

is

a more direct

means

of

linking

the

exposure

to the

outcome

than is the analysis of the total crime rateregressed on an "effective abortionratio,"

a

highly aggregated

measure of

exposure.

The absence of a correlation between

abor-

tion and

arrests

n

this

subsample

suggests

that Donohue and Levitt's decision

to

code

the abortion

ratio as zero

prior

to

legalization

may

be

driving

their

results.

Alterna-

tively,

the

endogeneity

of

abortion

may

explain

the

lack

of an association with

arrests.

States

in which

the cost of abortion is

lower

may

have

greater

sexual

activity,

lower

use of

contraception

and

higher

abortion

rates than states in which

the

cost and

stigma

associated

with

abortion are

greater.

If

true,

then

variation in

abortion

may

be

only

weakly

associated

with differences

in

unintended

childbearing.9

In

the

empirical

analysis

that

follows,

I

attempt

to

address

each

of the

identification

issues

just

discussed. The

advantage

of the difference-in-differences

strategy

is that

by staying

close

to the

"experiment"

made available

by

the

legalization

of

abortion,

I associate

changes

in crime

with

plausibly exogenous

changes

in

unintended fertil-

ity.

At

the same

time,

I

avoid

problems

with

poorly

measured

abortion. What

I

lose

is

any

dose-response

effect associated with

variation

in

unwanted

childbearing.

How-

ever,

in some

analyses

I

estimate

models

separately

for

states

with

abortion rates

above and below

the median abortion

rate

in

1973.

If

abortion

rates

were

essentially

zero

in

1972

in

the

nonrepeal

states,

as

Donohue and

Levitt

assume,

then

the

effects

should be more

negative

for the

states with

greater

post-Roe

abortion rates.

III.

Empirical

Specification

and

Results

A.

Comparison

by

Year

of

Birth in

Repeal

and

Nonrepeal

States

Abortion laws

in

Alaska,

California, Hawaii,

New

York,

Washington,

and the

District

of

Columbia,

what I have

referred

to

as

the

"repeal

states,"

changed

dramatically

between late 1969

and 1970. The

result

was

de

jure

or de

facto

legalization

in

repeal

states almost three

years prior

to

national

legalization

in

1973.

Thus,

there are

two

major policy

changes

that

I

use to

identify

effects

of

abortion

on crime:

early legaliza-

tion

among

cohorts from

repeal

states

and national

legalization following

Roe.

I

limit

the

analysis

to 15- to 24-years-olds because the Uniform Crime

Reports

record arrests

by single

year

of

age

for this

group only.

These

are

the same

data

used

by

Donohue

and

Levitt.

In

addition,

I

analyze

homicide offenses

as

recorded on

the

FBI's

Supplemental

Homicide

Reports

(SHR)

[Fox 2000].

These are

also available

by

single

year

of

age.10

I

further

limit

this

sample

to

cohorts

born

between 1967 and

1979.1"

9.

Joyce

(2001)

shows that the

resident

abortion

ate

n

repeal

states

s

almostdoublethat of

nonrepeal

statesbetween 1975

and

1985,

but that the

fertility

rate

s

the

same

in

both

groups

of

states.The

higher

pregnancy

ate but similar

ertility

rate

in

repeal

states

s

consistent

with

greater

exual

activity

and/or

less

contraception

nduced,

n

part,

by

the

protection

gainst

unwanted

hildbearing

fforded

by

the

rela-

tively

greater

accessibility

of

abortion ervices.

10. Thebiggestdrawback o the SHR is theirreporting eficiencies. nformationn the age andrace of

the

offender

when

missing

is

imputed

based

on

the known distribution

y

age/race/sex

of victims

and

offenders

by

state and

year

(Maltz

1999).

Nevertheless,

Supplemental

Homicide

Reports

are

widely

used

to

track

crime

by age

and race

(Maltz

1999;

Cook and Laub

1998;

Fox

and

Zawitz

2000).

Moreover,

use

them in

conjunction

with

murderarrestrates.

Thus,

a

consistent

relationship

etween abortion

nd

crime

across

these two

measures

of homicide

provides

an

important

heck of these

data.

11.

Cohort

s

equal

to

year

minus

age.


12/29

Joyce

11

I

structure

he


DD)

analysis

n

two

ways.

In

the

first,

I

comparechanges

in

crime

by

birth

cohorts

before and after

exposure

o

legalized

abortion.This is closest to what Donohueand Levittdo, but theyuse a continuous

measureof

abortion o

proxy

unwanted

childbearing.

The

identifying

variation s

based on

cross-state

changes

n

crime

among

cohortsof

the same

age.

In

the

other

set of

DDs,

changes

n

crime

among

cohorts

before and

after

exposure

o

legalized

abortion

n

utero

are

compared

o

changes

among

older

cohorts who are

close in

age,

but who

were

unexposed

o

legalized

abortion.The

identifying

variation

omes

from

within-state

hanges

in

crime.

In

the cross-stateDDs

exposure

s

based on

state/year-of-birth

nteractions.

pe-

cifically,

I

definebirth

years

1967-69 as the

pre-exposure

ears

and

1971-73 as the

post-exposure

ear

in

repeal

states.I

subtract

hanges

n

crime

among

cohorts

born

between1967-69 and 1971-73 in nonrepeal tates fromchangesobserved or the

same

cohorts

n

repeal

states. The

identifying

assumption

s that

changes

n

crime

among

cohorts n

nonrepeal

tates are a

good

counterfactualor

changes

in

repeal

states.

A

potentialproblem

with this

strategy

s that

hard o

measure

period

effects,

such as the

spread

of

crack,

may

affect crime n

repeal

and

nonrepeal

tatesat differ-

ent

times and with different

ntensity.

f

so,

then

nonrepeal

tates

do

not

provide

an

adequate

ounterfactual

see

Figures

la

and

lb).

To

improve

the

counterfactual,

estimate

models limited to a

subsetof states n

which

therewas

evidence of

crack/

cocaine use

in

their

major

cities

between

1984

and

1989 as

reported y

Grogger

and

Willis

(2000).

These nclude

Colorado,

Florida,

Georgia,

llinois,

Indiana,

Louisiana,

Maryland,

Massachusetts,

Michigan,

Missouri,

New

Jersey,

Ohio,

Pennsylvania,

Texas,

and

Virginia.

referto

these as the

comparison

tates. The

purpose

s to

pair

repeal

states o a subsetof

nonrepeal

tates hat

may

have

experienced

imilar

period

effects. The

relevant

regression

s

as follows:

(1)

LnCajy

=

Po

+

f,(Repealj

*

Y70y)

+

2(Repealj

*

Y7173y)

+

3(Repealy

*

Y7476y)

+

-4(Repealy

*

Y7779y)

+

Uaj

+

Vay

+

ajy

whereLnCajys the natural ogarithmof arrestsor homicides for age group,a, in

state,

,

and

year

of

birth,

y.

This is the

same

dependent

ariableused

by

Donohue

and Levitt

(2001).

Repeal

is a

dummy

variable

hat is one

for

repeal

states;

Y70,

Y7173,

Y7476,

and Y7779 are

dummy

variables or cohorts

born

n

the

designated

years.

The

omitted

category

ncludesthe birth

years

1967-69.

Equation

1

also in-

cludes fixed

effects for

age-state

(Uaj)

and

age-year

(Vay)

nteractions.

Thus

P2,

the

coefficient on the

interactionof

Repeal

and

Y7173,

measures the

proportionate

change

in

crime

between the

1971-73 and

1967-69

birth

cohorts n

repeal

states

relative o

nonrepeal

tates.'2

The coefficienton

the other

nteraction

erm,

P3,

mea-

12.

Significant

ourt decisions in the

fall of

1969 affected

abortion aws in

California

nd

Washington,

DC.

Legalization

ccurred

n

Alaska in

July

of

1970,

Hawaii n

Marchof

1970,

New York in

July

of

1970 and

Washington

n

Novemberof

1970. Given

that the full

impact

of

these reformson

unintended

childbearing

would not be evident until

1971,

I

treat

Repeal

*1970 as a

separate

nteractionn

order o

compareperiods

clearlypre

and

post

the

change

n

the

legalization

see

Sklar and Berkov

1974;

Gruber,

Levine and

Staiger

1999).

Including

1970

in

the

prelawperiod

does

not affect

my

results.


13/29

12

The Journalof HumanResources

sures

the effect of national

egalization.

f

abortion

owers

crime,

then

Roe v. Wade

should

bring

about a relative

mprovement

n

crime rates

among

nonrepeal

tates.

As a result,P3shouldapproach ero dependingon the speedof adjustment;nd[4

should

unambiguously

qual

zero as

adjustment

o national

egalization

s

completed

(Gruber,

Levine,

and

Staiger

1999).

Results

from the estimateof

Equation

1 are shown in

Table 2.

I

display only

the coefficientson the interaction

erms

[2,

P3

and

P4

n

Equation

1. There

are two

specifications

or each

measureof crime. The first

includes all states

and contrasts

changes

n crime

between

repeal

relative o

nonrepeal

tates.13 he

second imits the

sample

o

repeal

and

15

comparison

tates.

I have

also includedestimatesof the reduced-form

egressionusing

the natural

log

of state

fertility

rates as the

dependent

ariable

Columns

1

and

2).

These esti-

mates are almost identical o those of Levineet al. (1999).14They show that

early

legalization

n

the

repeal

states was

associatedwith

approximately

6

percent

ela-

tive decline

in

fertility

rates

regardless

of

whether use all 51

states

(Column1)

or

only

repeal

and

comparison

tates

(Column

2).

National

egalization

ollowing

Roe

v. Wadehad no additional

mpact

on

fertility

rates

in

repeal

states.

Estimates n

the first

row

of Table

2

indicate hat

arrestsand

homicidesfell for

cohorts

born

between

1971 and 1973 relative o those

born

between

1967 and

1969

in

repeal

relative to

nonrepeal

tates. These estimates

are

largely

consistentwith

resultsobtained

by

Donohueand Levitt

(2001).

Violent crime

arrests,

or

instance,

declined5.0

percent

more n

repeal

relative o

nonrepeal

tatesover this

period Col-umn

3).

This decline is similar n

magnitude

o the effect obtained

by

Donohue

and

Levittwith a continuous

measureof

abortion.l5

owever,

wo other

patterns

merge

from these results

that are less

supportive

of the

Donohue and Levitt

hypothesis.

First,

estimatesbasedon the

subsample

f

repeal

and

comparison

tatesare

relatively

small in

magnitude

and

statistically nsignificant.

The coefficient on

violent crime

arrests,

or

instance,

is

-0.026,

half as

large

as when

all states are

included.

A

distinguishing

haracteristicf the

comparison

tates s that

hey

all have

large

urban

centerswith a sizeable

African-American

opulation.

As

such,

the

comparison

tates

may provide

a more

crediblecounterfactualor

changes

n

crime

among

the

repeal

states,which are dominatedby Californiaand New York. The otherinconsistent

13. The unit of observation s the

cohort/state/age

cell. There are

potentially

,896

observations

iven

10

age

groups,

51 states and various

years.

Threehundred nd

forty-one

observations n arrests

nd 157

on homicideare

missing

because

some states did not

report

arrestsor

homicides

n

selected

years.

There

are3 cells with zerosfor

violent

arrests,

66 for murder rrests nd

739

for

homicides.The

model ncludes

dummy

variables or all

age

and

state nteractions s well as

age

and

year

of birth

nteractions,

s

repre-

sented

by

the last two termsof

Equation

1. The

specification

s

identical o that of Donohue

and Levitt

(2001)

with the

important

difference hat

I

have included

categorical

variables o

measuredifferential

exposure

o

legalized

abortion nsteadof the actual

abortion atio.

14.

Unlike

Levine et al.

(1999),

I include

Washington,

D.C.

as

a

repeal

state.

15. Donohueand Levitt

multiply

he

coefficienton abortion

by

350,

which is

the difference n abortion

ratiosbetweenstates n thetop thirdversus bottom hirdof abortion atios.Using the results n Row 4,

Column

1

of Table

1,

this

yields

an effect of

-5.3

percent

-0.015

*

350).

The

precision

of theirestimates

andmine differbecause

I

allow for a more

general

covariance tructure

mong

states

following

Betrand,

Duflo,

andMullainathan

2002).

This s

implemented

n Stata

by

clustering

n

state.WhenI redo

Donohue

and Levitt's

regressions

of

log

arrests

and allow for a more

general

covariance

tructure,

he

standard

errors

double. This is not

surprising

ince

60

percent

of their observations

ssume an abortion atio of

zero,

which

probably

nducessubstantial

erial

autocorrelation.


14/29

Table

2

Reduced-Form

Estimates

of

Fertility

Rates and

Log

Arrests and

Murders

among

15- to

24-Yea

and

Nonrepeal

States

1985-96

Ln

Fertility

Rate Ln

Violent

Crime Ln

Property

L

Women 15-44 Arrests CrimeArrests

(1)

(2)

(3)

(4)

(5)

(6) (7)

Repeal

71-73

-0.065* -0.063*

-0.050

-0.026

-0.066+

-0.030 -0.108

(0.011)

(0.014)

(0.066)

(0.070)

(0.032) (0.039)

(0.061

Repeal

74-76

-0.015 -0.002

-0.021

0.027

-0.069

-0.014

-0.167*

(0.021) (0.025)

(0.103)

(0.116)

(0.063) (0.082)

(0.063

Repeal

77-79 0.004 0.010

-0.066

0.022

-0.089 -0.010

-0.409*

(0.033) (0.037) (0.123) (0.145) (0.089) (0.126) (0.135

Only repeal

and

No Yes

No

Yes

No

Yes

No

comparison

states?

R-squared

0.963 0.974

0.983

0.979

0.982 0.976

0.929

N

969 399

4,552

1,890

4,555

1,890 3,889

Except

for

Columns 1

and

2,

coefficients

standard

rrors

below)

are

relative

changes

n arrestsand

homicides n

repea

birthcohorts

1971-73, 1974-76,

and

1977-79)

relative o the

1967-69

birth

cohorts.

Columns

1

and 2 show

the redu

each

outcome

here are

two

specifications:

Columns

1,

3, 5, 7,

and

9 use all

states;

Columns

2, 4, 6, 8,

and 10 use

onl

list of comparisontates).All specificationsorarrests ndhomicides nclude ixedeffectsfor interactions f ageandstat

1 in

the

text).

Standard

rrorshave been

adjusted

or

intra-class

orrelation

within state

by clustering

on state with

Stat

cells in the

full

sample

of

arrestand homicides:10

age groups,

51

states and a

variednumberof

age/year

cells since

t

1967 and

1979. Cells

are lost due

to

nonreporting

y

states and

zero

crimes

(see

Footnote 13 in

text).

All

regression

+p

<

.05;

*p

<

.01


15/29

14

The Journalof HumanResources

pattern

s that estimates

of

P2

and

33

n

Equation

1 exceed

those of

P1

n

absolute

value

for

murderarrestsand murders.

As

Gruber,Levine,

and

Staiger

(1999)

have

argued,one wouldexpecta relativedecrease n adverseoutcomes n nonrepeal tates

following

national

egalization,

which

shoulddrive

52

and

[3

to zero.

Instead,

find

thatcohorts

born

between 1977

and

1979

n

repeal

states

experience

elativedeclines

in

murdersand murderarrestsof

between 28 to

41

percent,

much

larger

than

the

declines

experienced mong

he 1971-73

birth

cohorts.

mportantly,

here s no

rela-

tive decrease

n

fertility

or cohorts

bornbetween 1977 and

1979,

which

undermines

a link between abortion

and

crime.16

Lastly,

I

divide

the

sample

between

1985-90

and

1991-96 and reestimate

Equa-

tion

1

separately

or the two

subperiods.

The results

or

all states

are shown

n

Table

3. The

pattern

s

similar o what occurredwhen

I

split

the

sample

in Table 1. If I

restrict he sample o thosearrestedbetween1985 and 1990,I find thatexposure o

legalized

abortion n

the

repeal

states is

positively

related

o arrests

and

murders.

By

contrast,

nalyses

of arrestsand

murders rom 1991 to

1996 reveal

the

opposite.

Moreover,

he

coefficients

n each

subperiod

are

large

in

absolutevalue and

they

are

unexpectedly

arger

for

cohorts

born after 1973 relative to

those

born

before

1973.

The

temporal nconsistency

alls into

question

he

DD

strategy

based on

changes

in similar cohorts

across states.

Changes

n

crime

in

nonrepeal

tates

will

be an

inappropriate

ounterfactual,

f

crack

markets

developed

earlier

and

had a

greater

impact

on state

crime rates

in

repeal

relativeto

nonrepeal

tates.

I

turn,

therefore,

to

my

alternative

trategy

of

using

a within-state

omparison

group

to

adjust

for

hard o measure

period

effects.

I focus

firston the

1985-90

period,

which

provides

a broad

comparison

of

aggregated

rime and

arrestrates

of

teen and

young

adults.

Donohue

and

Levitt have criticized

my

use of this

period,

since I

fail

to

use data

from he 1990s.

However,

1985-90

is a useful

period

because can createa

plausible

within-state

omparison roup

hat

was

clearly

affected

by

the

upsurge

n

crime,

but

thatwas

unexposed

o

legalized

abortion.

econd,

I

can

analyze

he same

experiment

by

race

given

the

availability

of

population

data

by

state,

year,

and race for five-

year age groups.

This

adds an

important

imension

o

the test

since the

legalization

of

abortion

had a

muchlargereffect on black relative to white fertility(Levineet

al.

1999;

Angrist

and

Evans

1999).

I then

turn

o a test of

abortionand crime

using

arrest

and homicide rates in

1990s.

I

have

to narrowthe

age groups analyzed

n

order o

isolate

those

exposed

and

unexposed

o national

egalization

ollowing

Roe.

However,

I

use some of

the

most

crime-prone ge groups

and

the narrow

age

bands

have the

advantage

f

minimizing

differences

n

age-crime

profiles

between he ex-

posed

and

comparison

roups.

16. There

s

virtually

no

difference

n

fertility

rates

between

repeal

and

nonrepeal

tates

for the

years

1977-79

despite

he fact

that he abortion ate s 76

percent

greater

n

repeal

states

see

Figure

1 in

Joyce

2001). For abortion o lower crime, therefore, t must be argued hatabortion mproved he timingof

births,

which

n

turn

had an

enormous

ffect on

the

well-being

of

the effected

cohorts.

ndeed,

he

effects

of better-timed

irths

on

homicidehave

to

be an

orderof

magnitude

greater

han the effects associated

with an

actual

decrease

n

fertility

or this

story

to hold.

This

seems

implausible

n

light

of

the recent

literature

n the

effects

of

delayedchildbearing mong

teens

(Geronimous

nd Korenman

1992;

Hotz,

McElroy,

and Sanders

1999).


16/29

Table 3

Split-Sample Reduced-form

Estimates

of Log

Arrests and Murders

among

15- to 24-Year-Olds

and Nonrepeal States: 1985-90 and 1991-96

Ln

Violent

Crime Ln

Property

Crime

Arrests

Arrests

Ln

Murder

Ar

(1)

(2) (3)

(4)

(5)

1985-90

1991-96 1985-90

1991-96

1985-90 19

Repeal

71-73 0.054 -0.045 0.009

-0.097+

0.106 -0

(0.061) (0.054) (0.027) (0.044) (0.059) (

Repeal

74-76 0.194

-0.012 0.055

-0.139

0.162 -0

(0.118)

(0.076)

(0.060) (0.081) (0.208)

(

Repeal

77-79

-0.054

-0.183

-0

na

(0.101)

na

(0.123)

na

(

R-squared

0.990 0.987 0.995

0.987

0.929

N

1,917

2,635 1,919

2,636

1,901

2

See

note

to Table 2.

"na,"

not

applicable

ince

15-year-olds

orn

n

1977 would be arrested fter 1990.


17/29

16

The Journal

of Human Resources

14-

~~~~~12

-~-

-TeenRepealYoung

Adult

Repeal

10

-

8

X

6

TeenComparison

6

,

YoungAdultComparison

4

2-

0

85

86 87 88

89

90 91

92 93 94

95 96

Year

Figure

2a

Violent

Crime

Arrest Rates

for

Teens and

Young

Adults

by

Repeal

and

Compari-

son States*

B.

Comparisons

by

Year

of

Crime within

Repeal

and

Nonrepeal

States

In this

analysis

I

compare

the

change

in arrests

and homicide rates

among

teens

between

1985

and

1990

to

the

change among young

adults. Teens and

young

adults

in

1985

were born

prior

to

1971

and thus

unexposed

to

legalized

abortion

in utero.

By

1990,

almost all

teens had

have been born after 1970

but

few of the

young

adults.

Thus,

teens

in

repeal

states

go

from

unexposed

to

exposed

between 1985

and 1990

and

young

adults

remain

essentially unexposed.

A

limitation of

using

a within-state

comparison

group

is

that the

age-crime profile

of

teens and

young

adults

may

differ.

Thus,

I allow

for a third set

of

differences

(DDD)

in

which

I

subtract

the

DD in

nonrepeal

states from

the DD

in

repeal

states. Since

few teens

in

the

nonrepeal

states

were exposed to legalized abortion during this period, the DD in nonrepeal states

measures

age

effects under the

assumption

of common

period

and cohort

effects.

Figures

2

and 3

present

time-series

of arrests

and homicide rates

stratified

by

repeal

and

comparison

states.

A

key

observation

is that the level and

pattern

of crime

among

teens

and

young

adults

is more similar within

states

than across.

This

provides

visual

support

for the use

of a

within-state

DD. To test for a cohort

effect more

formally,

I

estimate

the

following regression.

(2)

LnCRajt

=

Po

+

plTeena

+

2(Teena*Repealj)

+

3(Repealj

Y8788,)

+

P4(Teena*Y8788,)

+

[5(Repealj

*

Y8990t)

+

P6(Teena*

Y8990,)

+

37(Teen*Repeal*Y8990)

+

Xj,t

+

Uj

+

V,

+

eajt

where

LnCRajt

s the natural

logarithm

of the arrest

or homicide rate

for

age group

a

(teen

or

young

adult),

in

state

j,

and

year

t.

Repeal

is a

dummy

variable

that is


18/29

Joyce

17

40-

Teens

Repeal

350. -

30

Comparison

825

X

15

Young

Adults

Compaison

Comparison

10

5

0"

^

i

r

1

a

r ~

l

l

i

85

86

87 88 89 90 91 92

93

94 95 96

Year

Figure

2b

Property

Crime Arrest

Rates

for

Teens

and

Young

Adults

by

Repeal

and

Compari-

son States*

*Repeal

tates nclude:

AK,CA,DC,HI,NY,WA;

omparison

tates nclude:

CO,FL,GA,IL,IN,LA,MD,

MA,MI,MO,NJ,OH,PA,TX,VA

50-

45

-

5

/-

Teen

Repeal

40-

35-

?

Young

Adult

Repeal

30

Y25

20

~

YoungAdult

Comparison

_.

85

86

87

88 89 90 91 92

93

94 95 96

Year

Figure

3a

Murder Arrest Rates

for

Teens and

Young

Adults

by Repeal

and

Comparison

States*


19/29

18

The

Journal

of

Human

Resources

70-

60 -

~~~~~~~~~~60

^^^

^

"^~~Teen

epeal

50

40

-

Young

Adult

Repeal

$

30

Young

Adult

Comparison

20

Teen

Comparison

10-

0

85

86 87 88 89

90 91 92 93

94 95 96

Year

Figure

3b

Murder Rates

for

Teens

and

Young

Adults

by Repeal

and

Comparison

States*

*Repeal

states

nclude:

AK,CA,DC,HI,NY,WA;

omparison

tates nclude:

CO,FL,GA,IL,IN,LA,MD,

MA,MI,MO,NJ,OH,PA,TX,VA

one for

repeal

states;

Y8788 and

Y8990 are

dummy

variables for the

designated years

and Teen

is an indicator of those

15

to

19

years

of

age

as

compared

to

young

adults

ages

20

to

24. The

omitted

category

includes

the

years

1985-86.

State and

year

effects are

represented

by

Uj

and

Vt,

and

Xjt

is

the

matrix

of

control variables used

by

Donohue and Levitt

(2001)

in

their

regressions

of index

crime

rates. The DDD

estimate

is

17,

which

measures the

proportionate change

in

arrest

or homicide rates

before and

after

exposure

to

legalized

abortion

(years

1985-86

versus

1989-90)

among

teens relative

to

young

adults

in

repeal

relative to

nonrepeal

states.

If

abortion

lowers crime, then P7 should be negative.17

Results

are

displayed

in

Table 4.

The

first three

columns show

estimates for

arrest

rates;

the next

three

columns

present

estimates for homicide rates

for

all

perpetrators,

then

separately

for

whites and blacks. The

sample

in Panel A includes all

available

states whereas

Panel

B

is

limited to

repeal

and

comparison

states

only.

The

figures

in

Row

1

represent

the


of

arrest and homicide rates

(in

logs)

between

teens

and

young

adults for the

years

1989-90

and

1985-86

in

repeal

states.18

Thus,

the

natural

logarithm

of violent

crime arrest rates rose

0.02

or

2.0

percent

17. There

are

several

differences

etween

Equations

and2 thatmeritnote. In

Equation

I

analyze

arrest

andhomicide ates, nsteadof levels;I also aggregatearrests ndhomicidesby age forteens(ages 15 to

19)

and

young

adults

ages

20 to

24).

Aggregation

lso

lessens the loss

of

cells due

to zero

homicides

n

a

semi-logarithmic

pecification.

n

addition,

he

regressions

re

by year

of

arrestor homicide

and

not

by

year

of

birth

explicitly.

This makes the

structure f the

DDD

more

transparent.

inally,

the

analysis

s

limited

to

the

years

1985-90.

18. The

DD

estimates

are obtained rom the

regressions.

Using

the notation rom

Equation

2,

the

DD

estimate or

repeal

states

is

P6

+

V7.


20/29

Table

4

Changes

in

Log

Arrest and Homicide Rates

for

Teens

(15-19)

Relative

to

Young

Adults

(20-24

by Exposure

to

Legalized

Abortion,

1985-90

ArrestRate

for

Violent

Property

Murder

Changes

n

Arrestsand Homicide

(90-89)-(86-85):

Teens Panel

A:

newly

exposed,

young

adults

unexposed

1.

DD,

teens-adults,

epeal

states 0.020

-0.127* 0.379*

(0.031)

(0.011)

(0.050)

2.

DD,

teens-adults,

onrepeal

tates -0.010 -0.098*

0.210*

(0.036) (0.022)

(0.044)

3. DDD

(Row

1-Row

2)

0.030 -0.029 0.169+

(0.047)

(0.025)

(0.066)

R-squared 0.934 0.917 0.867

N 594

594

576

Panel

B:

Repeal

and

C

4.

DD,

teens-adults,

epeal

states

0.019

-0.126* 0.380*

(0.032) (0.011)

(0.053)

5.

DD, teens-adults,

omparison

tates

-0.043 -0.127* 0.243*

(0.052)

(0.035) (0.056)

6.

DDD

(Row

4-Row

5)

0.063 0.000

0.137

(0.060) (0.036) (0.076)

R-squared

0.948

0.938

0.919

N

242

242

241

Difference-in-difference-in-difference

DDD)

estimates how relative

changes

n arrest

and

homicideratesbetween hose

in

repeal

and

nonrepeal

tates

[Equation

in

the

text].

There

are 612

possible

state/age/year

cells

in the full

sample

(5

cells

are

due to a

nonreporting y

statesand/or zero crimes.Standard rrorsare in

parentheses.

Models

ncludecontrols

generosity,

oncealed

gun

aws,

andbeer ax as in DonohueandLevitt

2001).

All

models nclude tateand

year

ixed

effect

or

race-specificpopulation,

15 to

24

years

of

age.

+

p

<

.05;

*

p

<

.01.


21/29

20 The Journalof

HumanResources

more

among

teens relative to

young

adults n

repeal

states

between 1989-90 and

1985-86.

In

nonrepeal

tatesviolent crime arrest

ates

fell 1.0

percent

more

among

teens relative o youngadults(Row 2). The DDD estimates n Row 3 indicate hat

violent

crime

arrestrates

increased

3.0

percent

more

among

teens

in

repeal

states

relative o teens

in

nonrepeal

tates

adjusted

or within-state rends

n

arrests.

The last threecolumnscontrast otal

homicideratesand

then

separately

or

whites

and blacks.

When limited to

repeal

and

comparison

tates

only

(Panel B),

we see

that white teen

homicide ratesrose 56

percent

more than homiciderates

of

young

adults

n

repeal

states and

44

percent

more than in

comparison

tates. The corre-

sponding

changes

among

blacks

are

43 and 42

percentrespectively.Clearly,

age

and

period

effects are

huge

during

his

period.

Not

only

is there

a dramatic elative

increase

n

teen

homicide

rates,

but

it

occurs in both

repeal

and

nonrepeal

tates.

As a result, heDDD estimatesprovideno evidencethatexposure o legalizedabor-

tion

among

eens

in

repeal

stateshad

any

dampening

ffect

on

the

rise

in homicide.

The

lack

of

an

effect on black

homicide

rates s

particularly

oteworthy,

ince

legal-

ized abortion

had a

greater

mpact

on

black relativeto white

fertility.

C.

Within-state

Comparisons

n

Nonrepeal

states:

The

Effect

of

Roe v.

Wadeon Crime

The

next

set of

analyses

takes

advantage

of

the second "natural

xperiment,"

he

national

egalization

of

Did Legalized Abortion Lower Crime

Documents

Transcript of Did Legalized Abortion Lower Crime