Research in Social Stratification and Mobility 38 (2014) 18–31

Contents lists available at ScienceDirect

Research in Social Stratification and Mobility

j ourna l h om epa ge: ht tp : / /www.e lsev ier .com/ locate / rssm

Unequal returns to academic credentials as a hiddendimension of race and class inequality in American collegeenrollments

Tina Wildhagen ∗

Department of Sociology, Smith College, Northampton, MA, USA

a r t i c l e i n f o

Article history:Received 12 December 2013Received in revised form 6 March 2014Accepted 29 April 2014Available online 9 May 2014

Keywords:AP coursesCollege enrollmentSocial closureCredentialing theoryEducational inequality

a b s t r a c t

This study asks whether growing access to academic credentials for students from disad-vantaged groups will lead to a decrease in the value of those credentials for these groupsin college enrollments. Drawing on credentialing theory and the concept of adaptive socialclosure, I argue that as certain academic credentials become democratized (i.e., more acces-sible to disadvantaged students), their value decreases for students from disadvantagedrace and class groups at the same time as it increases for students from privileged raceand class groups. To test this idea, I use data from two cohorts of American high schoolgraduates to estimate changes in the educational payoff of participation in the AdvancedPlacement (AP) program for students across racial and social class groups. The results showthat at the same time as students from disadvantaged groups gained wider access to theAP program, its effect on their rates of college enrollment declined. During the same timeperiod, the AP effect on the rates of college enrollment for students from privileged groupsincreased. I conclude that unequal returns to academic credentials for privileged and disad-vantaged students represent a hidden dimension of race and class inequality in American

college enrollments. Moreover, the results demonstrate the possibility that as access to anacademic credential democratizes, as is the case with the AP program, privileged groupsare better able to insulate themselves from the negative effects of credential inflation.

© 2014 International Sociological Association Research Com- Stratifi

mittee 28 on Social

1. Introduction

The contemporary college admissions regime in the U.S.values applicants who can demonstrate achievement inmultiple dimensions, including academics, extracurricu-lar activities, and community service. This focus on the

“whole student” is seen widely as a democratizing forcein higher education because it decreases the emphasison standardized tests, which have been criticized roundly

∗ Correspondence to: Smith College, Department of Sociology, 215Wright Hall, Northampton, MA 01063, USA. Tel.: +1 413 585 3313.

E-mail address: twildhag@smith.edu

http://dx.doi.org/10.1016/j.rssm.2014.04.0020276-5624/© 2014 International Sociological Association Research Committee 28reserved.

cation and Mobility. Published by Elsevier Ltd. All rights reserved.

for their racial and class biases. Though standardized testscores still loom large in admissions decisions (Alon, 2009;Alon & Tienda, 2007), a number of colleges and universi-ties have moved to deemphasize their importance, withsome schools even opting to make the submission of testscores voluntary (NCFOT, 2012). Students who traditionallyperform worse on standardized tests (i.e., those from raceand class groups that have been historically marginalized inthe education system) are thought to benefit from the cur-rent admissions regime because they have opportunities

to demonstrate academic promise outside of standardizedtest scores.

Whereas standardized test scores tend to affect postsec-ondary school attendance similarly for all students (Alon,

on Social Stratification and Mobility. Published by Elsevier Ltd. All rights

tratificat

2di(itstaaptthgwedfttho

sPrAtdTt2dgWsos

gtAdctnitsens

2

2

tt

T. Wildhagen / Research in Social S

009; Alon & Tienda, 2007), many of the indicators of aca-emic promise that are becoming increasingly important

n the admissions process are non-standardized credentialsNACAC, 2009), which means that they do not necessar-ly carry the same value in the admissions process (or,herefore, yield the same returns) for applicants acrosstatus groups. I develop the argument that democratiza-ion of non-standardized academic credentials can lead ton increase in the value of those credentials in collegedmissions for students from privileged groups becauserivileged groups are able to employ adaptive social closureechniques surrounding those non-standardized creden-ials. If non-standardized admissions credentials can yieldigher educational returns for students from privilegedroups than for those from disadvantaged groups, thisould suggest that the current admissions regime in higher

ducation, with its use of a variety of indicators of aca-emic achievement, may not be as beneficial to studentsrom disadvantaged groups as is widely thought. Differen-ial educational returns to non-standardized credentials,hen, may represent an invisible dimension of inequality inigher education, compared with the more visible problemf unequal access to academic credentials across groups.

This study examines whether one prominent non-tandardized credential—participation in the Advancedlacement (AP) program—yields differential educationaleturns across racial and social class groups. I focus onP coursework because it is a non-standardized credential

hat has become increasingly accessible to high school stu-ents from disadvantaged groups over the last few decades.he analyses examine whether the effects of AP course-aking on college enrollment changed between 1992 and004—years of huge growth for the AP program—for stu-ents from privileged and disadvantaged race and classroups. The main question addressed by this study is:hen access to a non-standardized admissions credential,

uch as the AP program, is democratized, does the valuef the credential decline for disadvantaged students at theame time as it increases for their privileged peers?

This is an important question because if privilegedroups are able to preserve the educational advantageshat accrue to them from a given academic credential (e.g.,P courses) even when their monopolization of the cre-ential itself has declined, then growing access to theseredentials for disadvantaged groups would lack the powero reduce inequality in higher education enrollment. Withon-standardized admissions credentials likely to become

ncreasingly important in university admissions as selec-ive schools continue to deemphasize the importance oftandardized test scores, working toward equality in higherducation enrollment depends on developing an aware-ess of how these non-standardized indicators pay off fortudents from privileged and disadvantaged groups.

. Background

.1. Academic credentials

Credentials, such as academic degrees and certifica-ions, purportedly convey standardized information abouthe skills and competencies of their bearers to third parties.

ion and Mobility 38 (2014) 18–31 19

Beginning with Weber (1916/1951) and further developedby Collins (1979), sociologists have developed a creden-tialing theory that conceives of credentialing systems asformalized ways of maintaining inequality between statusgroups by regulating access to resources and opportunities.A key tenet of credentialing theory is that ‘the content andoccupational significance of credentials are more culturaland exclusionary than technical and efficacious’ (Brown,2001, p. 20). Thus, the credential bearer need not actuallypossess the set of skills and dispositions that the credentialsupposedly indicates in order to benefit from that creden-tial. However, by moving away from regulating access toresources primarily on the basis of ascriptive character-istics, credentialing systems are seen as legitimate andmeritocratic arbiters of opportunity. Thus, credentialingsystems serve to legitimate inequalities between statusgroups by formalizing and justifying these inequalities asthe result of perceived meritocratic differences betweenstatus groups.

While credentials typically refer to one’s qualifica-tions in the job market, this study examines qualificationsfor admission to colleges and universities. This studycontributes to the sociological literature on credential-ism by distinguishing between standardized and non-standardized credentials, where the standardized/non-standardized distinction refers to whether the social valueof the credential is equal across its bearers. If creden-tials serve as cultural markers more than indicators ofactual competencies, as credentialing theory holds, thenthis allows room for the social meaning of a credential tobe manipulated such that the credential becomes morevaluable for some groups than others. In this view, cre-dentials have the potential to perpetuate inequality notonly because members of privileged groups are more likelyto possess them, but also because credentials may beperceived as more valuable when they are possessed bymembers of privileged groups.

In her audit study of the effects of the “negative” creden-tial of a criminal record, Pager (2007) demonstrates thatnot only are African American men more likely to hold thecredential of a criminal record than white men, but alsothat this credential is more damaging to African Ameri-can men than to white men in the unskilled labor market.Mass incarceration, then, perpetuates racial inequality notonly because more African American men than white menare “marked” as criminals, but also because African Amer-ican men are disproportionately damaged by holding thecriminal credential. Applying this logic to the institutionof education, I argue that the education system perpetu-ates inequality between groups not only because privilegedgroups have greater access to academic credentials that arevalued in the college admissions process, but also becausesome of these credentials carry greater value for membersof privileged groups than disadvantaged groups. The lat-ter inequality is an important topic of study because it isless visible, and therefore more insidious, than the morevisible inequality in access to credentials. In the current

study, I concentrate on the non-standardized credential ofAP coursework, which has become increasingly accessibleto students from disadvantaged groups over the last fewdecades.

tratificat

20 T. Wildhagen / Research in Social S

In the next section, I explore how AP courseworkcould have become a non-standardized academic creden-tial in the college admissions process, with the value ofAP coursework varying based on a student’s status groupmembership.

2.2. Adaptive social closure

Privilege and disadvantage in the American educa-tion system have come to be preserved largely throughadaptive social closure mechanisms (Alon, 2009). Unlikeexclusionary social closure, which blocks disadvantagedgroups’ access to valued resources based on their mem-bership in disadvantaged groups, adaptive social closuregives privileged group members an advantage in meetingthe universal criteria that are used to determine accessto resources. Adaptive closure mechanisms become nec-essary for preserving privilege when exclusionary socialclosure mechanisms are no longer effective or sociallyacceptable. According to Parkin (1979, p. 47), in modernsocieties “exclusionary rules and institutions must alwaysbe justified by universal criteria that are indifferent to thepretensions or stigmata of birth” (emphasis added). In thepurportedly meritocratic institution of education, parentsfrom privileged groups focus on building the actual orostensible academic merit of their children, frequently atthe expense of students from disadvantaged groups (Alon,2009). The strategies that parents employ to enhance theirchildren’s merit at the expense of others are adaptive socialclosure strategies.1

The use of adaptive closure strategies intensifies when aparticular level of schooling becomes universally accessibleto all students, as did high school in the United States dur-ing the 20th century. As Lucas (2001) posits with his theoryof effectively maintained inequality, once a level of school-ing becomes universal, privileged parents focus on securingqualitative advantages at that level of education. Seekingout qualitative advantages when the playing field has beenleveled in terms of quantitative access requires the use ofadaptive closure strategies that make students from priv-ileged groups appear more deserving of these advantages.An important qualitative advantage at the high school levelis access to advanced coursework, such as AP courses.

In order for AP coursework to function as a non-standardized academic credential, some AP courses mustbe seen as higher-quality than others. Indeed, AP coursesdo vary widely across schools in both actual and perceivedquality. The actual academic quality and rigor of AP courses

varies across schools because some schools that participatein the AP program lack the resources to build strong APcourses (The College Board, 2001, 2011, 2012). Also, the

1 It is important to note that adaptive strategies for preserving privilegeare not normally undertaken with the explicit goal of preserving exist-ing institutional arrangements that perpetuate inequality. Instead, theseefforts are fueled by an ideology of entitlement, which legitimates theprivileged status of advantaged students’ place in the education system byusing culturally biased markers as purportedly objective indicators of stu-dents’ intelligence (Wells & Serna, 1996). Typically, both advantaged anddisadvantaged parents accept this ideology because it successfully masksthe biases inherent in the ways in which academic aptitude is measuredand assessed in schools.

ion and Mobility 38 (2014) 18–31

perceived quality of AP courses varies across high schoolsbecause high schools themselves vary greatly in theirperceived quality. The status hierarchy that exists in Amer-ican higher education, and the resultant fierce statuscompetition that has resulted among colleges and univer-sities, has filtered down to American secondary educationin recent years. As an example of the status competitionamong high schools, US News & World Report, which beganits influential annual ranking of colleges and universitiesin 1983, began ranking high schools in 2007. It is thisperceived, if not always actual, variation in the quality ofschools and their course offerings that opens the door for APcourses to become non-standardized academic credentials.

The strategies employed by contemporary parents fromprivileged groups to ensure that their children attendschools that are perceived as high-quality and that theyare enrolled in high-quality coursework serve as examplesof adaptive social closure. Even though families from priv-ileged groups “will certainly be better equipped to copewith a closure system on its children’s behalf,” when uni-versal selection criteria are in place, such families “muststill approach the task more in the manner of a challengewith serious risks attached than as a foregone conclusion”(Parkin, 1979, p. 63). Indeed, research suggests that par-ents from privileged groups tend to exert an enormousamount of effort, first, to ensure that their children areplaced in courses that are perceived as high-quality and,second, to ensure that their children’s schools continue tostratify their course offerings according to level of difficultyto maintain the hierarchy that provides their children withrelative advantages (Klugman, 2013; Useem, 1992; Wells& Oakes, 1996; Wells & Serna, 1996). In their efforts tofind high-quality schools and courses for their children,privileged parents devise and execute carefully craftedstrategies about where to live, where to enroll their chil-dren in school, and how to make sure that their children’sschools offer opportunities for their children to distinguishthemselves from their peers.

Recent research also suggests that the use of adaptiveclosure strategies may intensify in response to increas-ing access to qualitative advantages for disadvantagedstudents. Klugman (2013) finds that when the state ofCalifornia undertook several concerted efforts to increaseaccess to the AP program for disadvantaged students,schools in districts with high concentrations of disadvan-taged students did increase their AP offerings. However,schools in districts with high concentrations of upper-middle class students also increased their AP courseofferings, leading to stable or growing social class gaps inaccess to the AP program. Interviews with school officialssuggested that much of this increase in AP courses at moreadvantaged schools was driven by parent demand, suggest-ing that these parents were “motivated to accumulate moreopportunities for distinction” for their children (Klugman,2013, p. 26).

2.3. Usurpationary closure: challenges to AP domination

from below

Students from groups that have traditionally beenunderrepresented in higher education (e.g., African

tratificat

AvtpfbBmGfb2h&

nd[duvcaccestmpd1htws&rtEop(

ctwnfhct&ecssctet

to the credential democratizes. As the AP program hasexpanded, the actual and perceived academic rigor of APcourses has come to vary widely across schools (The CollegeBoard, 2001). With such wide variation, college admissions

T. Wildhagen / Research in Social S

mericans, Latinos, Native Americans, students from disad-antaged socioeconomic backgrounds) have limited accesso AP coursework compared with their more advantagedeers. For example, African American students accountedor 14.7% of the graduating public high school class of 2011,ut only 9% of AP exam takers in these schools (The Collegeoard, 2012). In many states, this underrepresentation isuch more severe than the national average indicates. Ineorgia, for example, African American students accounted

or 35.1% of all graduates from public high schools in 2011,ut they accounted for only 12.6% of AP exam takers. In005, 61% of AP exam takers had at least one parent whoseighest level of education was a university degree (Sawtell

Gillie, 2007)Whereas exclusionary and adaptive closure mecha-

isms serve to preserve privileges and resources forominant groups, usurpationary closure mechanisms “aimto] bit[e] into the resources and benefits accruing toominant groups in society” (Parkin, 1979, p. 74). Thenderrepresentation of students from historically disad-antaged racial, ethnic, and social class groups in APourses has not escaped public attention, and groups thatre underrepresented in the AP program have successfullyhanneled this attention in order to gain wider access to APoursework. For example, in 1999 the American Civil Lib-rties Union (ACLU) filed a class action lawsuit against thetate of California, in which it argued that “the State’s failureo assure ‘equal and adequate access’ to Advanced Place-

ent courses further perpetuates educational inequalities,articularly for students enrolled in lower income, pre-ominantly African-American and Latino schools” (ACLU,999). Several similar legal challenges over the last decadeave prompted the federal government to provide fundingo states to increase the number of low-income studentsho take part in the AP program, and many states offer

imilar programs of their own (Gollub, Bertenthal, Labov, Curtis, 2002). In 2008, Stephanie Monroe, Assistant Sec-

etary of Education for Civil Rights, issued a statement fromhe Office for Civil Rights at the United States Department ofducation warning schools that any discrimination basedn ‘race, color, or national origin’ in AP course admissionrocesses was a violation of Title VI of the Civil Rights ActMonroe, 2008).

Though the usurpationary claims on access to APoursework have helped both low-income students andhose from underrepresented racial and ethnic groups gainider access to AP courses, this increased access mayot increase the chances of university admissions equally

or all of these students. Problems of equal access toigher education in the US have been much more suc-essfully framed in terms of racial and ethnic inequalityhan in terms of social class inequality (Bowen, Kurzweil,

Tobin, 2005; Schmidt, 2007). Disadvantaged racial andthnic groups have successfully employed usurpationarylosure in higher education by “manipulat[ing] the beliefystem of the dominant group [and] pointing up incon-istencies between its advertised doctrines and its actual

onduct” (Parkin, 1979, p. 85). The inconsistency betweenhe “advertised doctrine” of equal opportunity and thempirical reality of racial and ethnic inequality in educa-ional opportunities has more successfully captured elites’

ion and Mobility 38 (2014) 18–31 21

attention than has the inconsistency between the ideal ofequal opportunity and the reality of socioeconomic inequal-ity in educational opportunities. For example, the SupremeCourt has ruled that race can be considered as a factor inuniversity admissions, but no such decision has been madeon social class (Bowen & Bok, 1998; Schmidt, 2007). It ispossible, then, that AP course gains made by students fromunderrepresented racial and ethnic groups more effectivelytranslate into university admissions benefits than have thegains made by socioeconomically disadvantaged students.

2.4. AP courses in college admissions

The College Board’s AP program has grown steadilyeach decade since it was established in 1955. Since itsinception, the AP program has emerged as a “nationallyrecognized program of academic excellence” intended toprepare high school students for university-level academicwork (The College Board, 2001, p. v). Between 1980 and2004, which includes the timespan for the data used inthis study, the number of high school students takingan AP exam increased by 660%, and the total number ofAP exams taken increased by 750% (Geiser & Santelices,2004).

The popularity of the AP program lies in the fact thatit claims to prepare high school students for the aca-demic rigors of college, thereby making students moreattractive applicants for university admissions officers. Asthe program has proliferated, it has become increasinglyimportant in the college admissions process, particularlyfor selective universities (Camara & Michalis, 2005; Geiser& Santelices, 2004; Gollub et al., 2002). Using a sampleof 133 deans of admission at colleges and universities,Gollub et al. (2002, p. 55) found that admissions officers“generally view the presence of AP or IB2 courses on atranscript as an indicator of the applicant’s willingness toconfront academic challenges” and that these courses are“of greatest importance for admission to highly selectiveschools.”

3. The current study

The question driving this study is how increasinglydemocratized access to a non-standardized academic cre-dential affects the educational payoff of that credentialfor students from privileged and disadvantaged groups.Whereas the college enrollment payoff to a given standard-ized test score does not appear to differ across groups (Alon,2009; Alon & Tienda, 2007), the payoff to non-standardizedcredentials like AP coursework may be greater for stu-dents from privileged status groups, and differences inreturns to such credentials may actually increase as access

2 The International Baccalaureate (IB) program offers advanced course-work to students during the last two years of high school. As with the APprogram, many colleges offer college credit for IB coursework.

tratificat

The key independent variable is whether the studentparticipated in the AP program during high school. This

22 T. Wildhagen / Research in Social S

officers cannot assume that an AP course taken at one highschool signifies the same academic rigor as an AP course atanother high school, particularly when those high schoolsthemselves differ in perceived or actual quality. This allowsfor the possibility that AP coursework will be seen as morevaluable for some students than others. With privilegedstudents more likely to attend high schools that are seenas high-quality, admissions officers are likely to perceivetheir AP coursework as more valuable than the AP course-work of disadvantaged students. Thus, this study positsthat disadvantaged students’ underrepresentation in APcourses is not the only inequality faced by these studentsvis-a-vis the AP program. With a non-standardized admis-sions credential like AP courses, socioeconomic, racial, andethnic disparities in returns to the credential may be animportant mechanism for maintaining inequality in highereducation.

I investigate how the impact of AP participation on col-lege enrollment has changed over time and across groups. Ifprivileged groups have successfully monopolized access toAP courses that are perceived as higher-quality (or actuallyare higher-quality), then the effect of AP participation oncollege enrollment should have increased over time for stu-dents from privileged groups and decreased for those fromdisadvantaged groups. That is, as the credential has democ-ratized, I predict that its value has increased for studentsfrom privileged race and class groups and decreased forthose from disadvantaged groups. However, as discussedabove, given that issues of inequality in the higher educa-tion system have been more successfully framed in termsof race and ethnicity than social class, I predict larger lossesin the value of the AP credential for students who aredisadvantaged by social class relative to their more advan-taged peers than for students from underrepresented racialand ethnic groups relative to those from overrepresentedgroups.

Because a large proportion of applicants to selective col-leges complete AP coursework in high school, the effectof participating in the AP program on selective collegeenrollment may have decreased over time for all groups.Still, to the extent that privileged groups have been ableto protect access to high-quality AP courses, they mayhave more effectively insulated themselves from the conse-quences of AP credential inflation than have disadvantagedgroups. This would suggest a smaller decline in the effect ofAP coursework on selective college enrollment over timefor students from privileged groups than disadvantagedgroups.

4. Data

The data for this study are drawn from the NationalEducation Longitudinal Study of 1988 (NELS) and the Edu-cation Longitudinal Study of 2002 (ELS)—two nationallyrepresentative longitudinal samples of US high school stu-dents graduating in 1992 and 2004. Both NELS and ELS wereconducted by the National Center for Education Statistics

(NCES). They are well-suited to investigating the ques-tions raised by this study because they cover a criticaltimespan for the AP program. Between 1992 and 2004,the AP program experienced magnificent growth, which

ion and Mobility 38 (2014) 18–31

resulted in intensified efforts on the part of privileged par-ents to protect access to high-quality AP coursework andon the part of advocates and parents of students fromless-privileged groups to fight for wider access to the APprogram.

Both the NELS and ELS employed a two-stage proba-bility design to select nationally representative samples ofstudents and schools. In the NELS, students were first sur-veyed in eighth grade, while the ELS first surveyed studentsin tenth grade. In both studies, schools were first sampledas the primary sampling unit, and students were then sam-pled from schools as the second-stage sampling unit. Bothstudies include data from student surveys, cognitive testsadministered to students by the NCES, surveys of one of thestudents’ parents, surveys of two of the students’ teachers,and high school transcripts. This study uses data from eachof those sources of information.

All independent variables with missing data used in theanalyses were multiply imputed using the software NORM(Schafer, 2000). NORM employs an algorithm based on theassumptions that the data come from a multivariate nor-mal distribution and are missing at random. Allison (2000)assessed the ability of NORM to produce unbiased esti-mates of missing data by creating a hypothetical datasetand then causing about half of the values on one variableto be missing. He found that NORM “does an excellent job”of producing unbiased estimates of the desired parameters(Allison, 2000, p. 302).

For the analyses presented in this study, I generatedten sets of multiply-imputed NELS and ELS data. All analy-ses were conducted on all ten multiply-imputed samples.The parameter estimates presented in the results of thestudy are the averages of each of the five parameter esti-mates produced by the five analyses. Standard errors arenot averages of the estimated standard errors, but are cal-culated using the formula given by Rubin (1987), whichinflates the standard error to account for increased uncer-tainty.

After imputation of independent variables the samplesizes for the Class of 1992 (NELS) were 2380 for stu-dents from underrepresented racial and ethnic groups,8120 for students from overrepresented racial and eth-nic groups, 7040 for first-generation students, and 3030for continuing-generation students.3 For the Class of 2004(ELS), the sample sizes were 4980 for students fromunderrepresented racial and ethnic groups, 9830 for stu-dents from overrepresented racial and ethnic groups, 8660for first-generation students, and 6140 for continuing-generation students. The samples were restricted tostudents who graduated from high school.

5. Measures

5.1. AP participation

3 Sample sizes are rounded to the nearest ten, as required by the NCES,as a way to protect anonymity.

tratificat

iaewptoovbttnacgpas

5

cg&rmgsrAnoigiApifp

dpsvoriioti

g1(g

T. Wildhagen / Research in Social S

nformation is taken from high school transcripts. The vari-ble is divided into two categories: those students whonrolled in at least one AP course in high school and thoseho took no AP courses. Modeling participation in the AProgram instead of the number of AP courses that studentsook in high school is necessary because the proportionf students in any one status group who took more thanne AP course is very low, meaning that modeling theariable as the number of AP courses taken would haveeen inappropriate.4 College admissions officers adjustheir expectations for how many AP courses they would likeo see on students’ transcripts depending on the extensive-ess of the AP program at students’ high schools. To guardgainst the possibility that differences in the number of APourses offered by students’ high schools are driving anyroup differences in the relationship between AP coursearticipation and college enrollment (Klugman, 2012), thenalyses account for the number of AP courses offered bytudents’ high schools.

.2. Race, ethnicity, and social class

Research suggests that in the American context, racean shape students’ access to resources and institutionalatekeepers’ perceptions of them independent of class (Kao

Thompson, 2003). Therefore, the analyses test whetherelationships between AP course-taking and college enroll-ent differ across social class groups and racial/ethnic

roups. Grouping students according to race/ethnicity andocial class allows for a comparison of the magnitudes ofacial/ethnic and social class inequality in the value of theP credential. I use a dichotomous variable for race and eth-icity that indicates whether the student belongs to a racialr ethnic group that has been historically overrepresentedn American higher education (whites and East Asians) or aroup that has been historically underrepresented in Amer-can higher education (African Americans, Latino/as, Nativemericans, and Southeast Asians). It would have beenreferable not to aggregate the racial and ethnic groups

nto those broader categories, but the numbers of studentsrom some of the minority groups participating in the AProgram were too small to allow for further disaggregation.

I use parents’ education to measure social class. Stu-ents are divided into two groups: those with at least onearent who has a university degree (continuing-generationtudents) and those whose parents did not obtain a uni-ersity degree (first-generation students). I employ thisperationalization of social class for three reasons. First,esearch suggests that parents’ education may be morenfluential in educational outcomes and experiences than

s parents’ income (Mayer, 1997). Second, the proportionf Americans attending college has steadily increased overhe last few decades, leading to “first-generation” statusncreasingly becoming a mark of class disadvantage. For

4 The percentage of students who took more than one AP course, byroup and cohort, are as follows: (1) underrepresented students: 22% in992, 11% in 2004; (2) overrepresented students: 2% in 1992, 22% in 2004;3) first-generation students: 3% in 1992, 10% in 2004; and (4) continuingeneration students: 5% in 1992, 18% in 2004.

ion and Mobility 38 (2014) 18–31 23

example, in the NELS data (Class of 1992), 70% of the highschool graduates were first-generation students comparedto only 56% of the high school graduates in the ELS data(Class of 2004). Whereas the family income gap betweenstudents with at least one college graduate parent andthose without a college graduate parent was $38,000 in1992, the gap was $43,000 in 2004, which shows a sub-stantial and increasing difference in financial resourcesacross the two groups. Third, as first-generation collegestudents have come to represent a smaller proportion ofthe college-going population in recent decades, many col-leges and universities themselves have come to recognize“first-generation” status as an important marker of socialclass.

5.3. Dependent variables

This study examines enrollment in postsecondary edu-cation institutions as the outcome. I use two dependentvariables that measure students’ postsecondary enroll-ment. Four-year college is a dichotomous variable measur-ing whether students had enrolled in a four-year collegeor university within two years of graduating high school.Selective four-year college is a dichotomous variable mea-suring whether students attended a selective four-yearcollege or university within two years of graduating highschool, given four-year college attendance.5

5.4. Family and educational background and schoolcharacteristics

I selected a number of variables to generate the propen-sity score for participating in the AP program. Thesevariables include demographics and family background;academic preparation, achievement, and effort; behaviorin and engagement with school; educational expectationsand attitudes about education; peers’ attitudes towardeducation; and high school information. Demographic andbackground variables include whether the student is maleor female, whether the student’s first language is English,family income, whether the student lives in a two-parentfamily, and years of educational attainment for the stu-dents’ parents. Academic preparation variables includewhether the student was ever placed in special educa-tion or remedial classes in math or English and whetherthe student had plans early in high school to take the SATor ACT. Academic achievement variables include students’scores on NCES-administered tests of math and readingskills and GPA. Academic effort, school behavior, and engage-ment variables include time spent on homework outside

of school, the number of times the student skipped classor was late to school, the number of times the studentwas suspended from school or received in-school suspen-sion, the number of times the student was disciplined

5 To determine the selectivity of four-year institutions, the NCES usedthe Carnegie institutional selectivity measure, which measures selectivityby using the institution’s reported 25th percentile scores—the top scorefor the bottom quarter of the distribution of scores at that institution—onthe SAT and ACT for incoming students (Carnegie Foundation, 2012).

tratificat

24 T. Wildhagen / Research in Social S

for breaking school rules, whether the student regularlybrought required materials to class (e.g., writing utensils,books), and teachers’ perceptions of the student’s academicengagement. Variables measuring educational expectationsand attitudes about education include the student’s aca-demic self-concept, a composite measure of the student’sattitudes about school; the student’s expectations for edu-cational attainment; and the parent’s expectations for thestudent’s educational attainment. Peers’ attitudes towardeducation is an average of the student’s three closestfriends’ attitudes about the importance of school. Finally,the following school characteristics are included in thepropensity score model: school sector, the percent of stu-dents in the school taking AP courses, and the percent ofstudents in the school who are white.

Descriptive statistics by race/ethnicity and parentaleducation are presented in Appendix for all variables usedin the analyses.

6. Methods

The research questions investigated in this studydepend on isolating the relationship between participatingin the AP program and the likelihood of college enroll-ment. It is important that any differences found betweengroups in the relationship between AP participation andcollege enrollment reflect the effect of AP participation,not some other variable that is associated with AP course-taking. Randomized field trials are the ‘gold standard’for estimating causal effects of social programs, becausethey balance control and treatment groups on observedand unobserved variables (Schneider, Carnoy, Kilpatrick,Schmidt, & Shavelson, 2007). However, because the suc-cessful completion of AP courses requires some subject areapreparation and depends to some degree on student pre-ferences, conducting random field trials to determine theeffects of AP courses on educational outcomes is not feasi-ble. To approximate random assignment to AP courses, I usepropensity score matching (PSM), a method developed byRosenbaum and Rubin (1983a, 1983b). PSM allows the ana-lyst to estimate causal effects using observational data bymatching individuals who have received the “treatment”(i.e., AP courses) with those who have not received thetreatment but are otherwise very similar. Matching allowsthe analyst to estimate the outcomes for individuals whohave experienced the treatment as if they had not expe-rienced the treatment (Morgan, 2001; Winship & Morgan,1999), yielding the causal effect of the treatment.

The statistical analyses proceeded in two stages. First,logistic regression was used to predict the propensity ofeach student to be assigned to the treatment (i.e., partici-pate in the AP program). I included a number of variables

Table 1Percent of students participating in the AP program, by race/ethnicity and parent

Class of 1992 Class

Underrepresented 2.13 18.42Overrepresented 8.65 31.45First generation 2.77 17.49Continuing generation 18.36 40.58

ion and Mobility 38 (2014) 18–31

(listed in Section 5.2) in the propensity score equationthat could affect both selection into the treatment andthe dependent variables being measured. Following Hongand Raudenbush (2005, 2006) and Attewell and Domina(2008), I estimated a logit model that predicts the propen-sity of students to participate in the AP program using bothstudent and school characteristics, since both sets of factorscan affect whether students take AP courses. I estimatedpropensity scores separately for each of the four groups ofstudents based on parent education and race/ethnicity.

In the next step, students with very similar propen-sity scores who did receive the treatment (students whodid take AP courses) and who did not (students who didnot take AP courses) were matched in a process known asradius matching. Each student who did take AP courses wasmatched with students who fell within .1 standard devi-ations of the propensity score of the treated student butwho did not complete AP coursework. The matched stu-dents who did not complete AP coursework functioned asa control group. I used the common support match algo-rithm in Stata, which drops individuals who have receivedthe treatment but lack similar counterparts who have notreceived the treatment (Becker & Ichino, 2002).

To ensure that matched groups did not differ in impor-tant ways, I confirmed that all predictors of the propensityscore were balanced across control and treatment groups(i.e., those who took AP courses and those who did not takethem). I used t-tests to compare the means of each pre-dictor across matched groups and examined standardizedbiases for all predictors (Rosenbaum and Rubin, 1985). Theresults suggested that all control variables were reasonablybalanced across control and treatment groups. Standard-ized biases for matched groups were less than .05 for allpredictors of the propensity score, and t-tests revealed thatmean differences between AP takers and non-AP takers didnot differ at the p < .05 level for any of the predictors of thepropensity score.

Finally, matched treatment and control groups werecompared on each of the dependent variables. Standarderrors for treatment effects were estimated using boot-strapping to minimize bias. Differences between matchedgroups on the dependent variables indicated differences inthe relationship between AP coursework and the depend-ent variables. Regression models were then run to checkthe robustness of the PSM results.

7. Results

7.1. Democratization of the AP program

Table 1 presents AP participation for students from thefour subgroups: underrepresented and overrepresented

education.

of 2004 Difference Percent change

16.29 764.79 22.8 263.58 14.72 531.41 22.22 121.02

T. Wildhagen / Research in Social Stratification and Mobility 38 (2014) 18–31 25

Table 2Differences in college enrollment for matched AP course-takers and non-AP course-takers, by race/ethnicity and parent education.

Class of 1992 Class of 2004

Four-year Selective four-year Four-year Selective four-year

AP ∼AP Difference AP ∼AP Difference AP ∼AP Difference AP ∼AP Difference

Underrepresented .90 .68 .22*** ,a .62 .32 .30*** .68 .51 .16*** .41 .26 .15*** ,c

Overrepresented .93 .81 .12*** .68 .50 .19*** .82 .68 .14*** .54 .39 .15***

First generation .90 .69 .22*** ,b .51 .27 .24*** .66 .51 .15*** ,c .38 .23 .15*** ,c

Continuing generation .94 .88 .06* .74 .55 .19*** .87 .74 .14*** ,c .59 .43 .16***

Sample sizes for 1992: UR = 2280 for four-year, 720 for selective four-year; OR = 7950 for four-year, 3680 for selective four-year; FG = 5140 for four-year,2160 for selective four-year; CG = 2900 for four-year, 2160 for selective four-year.Sample sizes for 2004: UR = 4370 for four-year, 1450 for selective four-year; OR = 8830 for four-year, 4520 for selective four-year; FG = 7650 for four-year,2430 for selective four-year; CG = 5580 for four-year, 3570 for selective four-year. (All sample sizes rounded to the nearest ten, as required by the NCES, tohelp protect anonymity.)

a Racial/ethnic difference between AP course effects significant at p < .05.b Social class difference between AP course effects significant at p < .05.c Change in size of AP course effect over time significant at p < .05.

course-P cours

rcocdrgA7gfs34aTta

7

teatptpftsorleagbtc

* p < .05 for differences between matched AP course-takers and non-AP*** p < .001 for differences between matched AP course-takers and non-A

acial and ethnic groups and first-generation andontinuing-generation students. The data show anverall expansion of the AP program, with access to APourses increasing more dramatically for students fromisadvantaged groups. In 1992, only 2.13% of underrep-esented students and 2.77% of first-generation studentsraduating from high school had enrolled in at least oneP course. By 2004, those percentages had increased by65% for underrepresented students and 531% for first-eneration students. Enrollment of students in AP coursesrom privileged groups also increased, but not at suchtaggering rates. AP participation increased from 8.65% to1.45% for overrepresented students and from 18.36% to0.58% for continuing-generation students between 1992nd 2004 (growth rates of 264% and 121%, respectively).he usurpationary claims in recent years for greater accesso the AP program for students from disadvantaged groupsppear to have been successful.

.2. Value of the AP credential

Moving to the results of the PSM analyses, I examinehe effects of participating in the AP program on collegenrollment for the four groups of students in the 1992nd 2004 cohorts. First I address the question of whetherhe AP course credential is worth more for students fromrivileged than disadvantaged groups. Table 2 presentshe proportions of AP course-takers and their matchedeers from the Classes of 1992 and 2004 who enrolled inour-year colleges within two years of high school gradua-ion. Because the groups are matched based on propensitycores that predict AP participation, differences in the ratesf college attendance between matched groups shouldeflect the relationship between AP participation and col-ege attendance. The results show statistically significantffects of AP participation on college enrollment in generalnd on enrollment at selective colleges, in particular, for all

roups in both cohorts. As Table 2 shows, for all groups inoth cohorts, significantly larger proportions of AP course-akers enrolled in four-year colleges and selective collegesompared to their matched peers who did not participate

takers.e-takers.

in the AP program. Thus, for all students in both cohorts theAP credential was a valuable asset in college admissions.

However, for the 1992 cohort, the effects of AP par-ticipation on college enrollment were larger for studentsfrom underrepresented racial and ethnic groups and firstgeneration students compared to their peers from over-represented racial and ethnic groups and continuinggeneration students, respectively. In this earlier cohort,22% more underrepresented and first-generation studentsenrolled in a four-year college if they had participatedin the AP program compared to their very similar peerswho had not participated in the AP program. Only 12%and 6% more overrepresented and continuing-generationstudents, respectively, enrolled in a four-year college ifthey had participated in the AP program compared totheir similar peers who had not participated in the pro-gram. These differences in AP course effects on four-yearcollege attendance between underrepresented and over-represented students, and between first-generation andcontinuing-generation students, were both significant atthe p < .05 level. Differences across groups in the effect ofAP participation on enrollment in a selective college did notdiffer significantly in this cohort.

By the time the Class of 2004 was making collegeenrollment decisions, the effects of AP participation hadconverged across groups, with its effect roughly equal forall students. By this time, between 14% and 16% more stu-dents in all groups enrolled in a university, or in a selectivecollege given four-year college attendance, if they had par-ticipated in the AP program compared to their matchedpeers who had not participated in the program. Thus, thegroup differences in the value of the AP credential thatfavored disadvantaged groups in 1992 had disappeared by2004.

The theory presented above suggests that as a cre-dential democratizes, its value will decrease for studentsfrom disadvantaged groups, while privileged groups will

be able to protect the value of democratized creden-tials for members of their own groups. Fig. 1 presentsthe percent change in AP participation effects on col-lege enrollment across groups during this period of rapid

26 T. Wildhagen / Research in Social Stratification and Mobility 38 (2014) 18–31

llment b

Fig. 1. Percent change in AP participation effects on college enro

expansion for the program. As anticipated, for overrepre-sented and continuing-generation students, the effects ofAP participation on four-year college enrollment increasedbetween 1992 and 2004, but to a greater extent forcontinuing-generation students (133%) than for thosefrom overrepresented racial and ethnic groups (17%). Theincrease was statistically significant at p < .05 only forcontinuing-generation students. For their disadvantagedcounterparts, the effects of AP participation on four-yearcollege enrollment decreased over the same period, butonly the decline for first-generation students (32%) wasstatistically significant at p < .05. Consistent with expecta-tions about credential inflation at selective colleges, theeffect of AP participation on selective college enrollmentdiminished between 1992 and 2004 for students from allfour groups. However, the declines were smaller for stu-dents from privileged than disadvantaged groups. In factthe declines for students from overrepresented racial andethnic groups (21%) and continuing-generation students(16%) were not statistically significant. The value of theAP credential decreased more steeply and by statisticallysignificant margins for those from underrepresented racialand ethnic groups (50%) and for first-generation students(38%).

To check the robustness of the results of the PSM analy-sis, I modeled regressions that predicted each dependentvariable (four-year college attendance and selective col-lege attendance) for each of the four groups of students.Each regression controlled for all of the variables used aspredictors of the propensity score as well as a dummy vari-able for AP participation and the propensity score itself.The coefficients from each set of regressions were usedto generate predicted values for the dependent variables

for students who had taken AP courses and those who hadnot taken any AP courses. The difference in the predictedvalue of the dependent variable between AP course-takersand non-AP course-takers represents the impact of AP

etween 1992 and 2004, by race/ethnicity and parent education.

participation on the dependent variable as measured bythe regression model.

Fig. 2 presents the percent change in the impact ofAP course-taking on college attendance from 1994 to2004 from the regression analyses. (Full regression resultsavailable upon request from the author.) The results ofthe regression models produced patterns similar to thoseproduced by the PSM analyses. Similar to the resultsfrom the PSM analysis, the value of AP participationin college admissions dropped for first-generation andunderrepresented students and increased for continuinggeneration students. (It did not change for overrepre-sented students.) As with the PSM results, the effect ofthe AP credential diminished in selective college admis-sions for all students, but to a greater extent for studentsfrom disadvantaged than privileged groups. The similarpattern of results for the PSM and regression analysesboost confidence that the results do indicate real changesin the value of the AP credential over time and acrossgroups.

8. Discussion and conclusion

The results presented here indicate that the collegeenrollment benefits of participating in the AP programdropped significantly for students from disadvantagedgroups between 1992 and 2004, even as they gained greateraccess to the program. Except in the case of selective collegeadmissions, the benefits of being marked as an AP studentincreased for students from privileged groups during thesame time period, particularly for continuing-generationstudents. These results highlight a heretofore unrecognizedmechanism of social inequality in college admissions: non-

standardized academic credentials perpetuate inequalitynot only because they are less available to students fromdisadvantaged groups, but also because they become lessvaluable to students from disadvantaged groups as the

T. Wildhagen / Research in Social Stratification and Mobility 38 (2014) 18–31 27

Fig. 2. Percent change in AP participation effects on college enrollment between 1992 and 2004, by race/ethnicity and parent education (results of PSMr

cidcatvigcag

gyeirWagpeat

daepsccmtE

obustness check).

redential becomes more widely available. Ironically, then,ncreased access to a non-standardized credential forisadvantaged students can drive down the value of theredential for these students, while privileged groups areble to protect the academic payoffs that they receive fromhe credential. The results also suggest that even when thealue of a credential is diluted for the entire population, asn the case of AP courses at selective colleges, privilegedroups are better situated to insulate themselves from theonsequences of this inflation, leading to a disproportion-te loss in the value of the credential for disadvantagedroups.

Fig. 3 illustrates this phenomenon, presenting projectedroup differences in the value of the AP credential for four-ear college enrollment (Panel A) and selective collegenrollment (Panel B), assuming that the percent changen the credential’s value between 1992 and 2004 wouldemain the same for each group between 2004 and 2016.

hereas the value of the AP credential had convergedcross all groups in 2004, by 2016 the racial and social classaps are projected to reemerge, but with students fromrivileged race and class groups enjoying larger positiveffects of the AP credential on four-year college enrollmentnd selective college enrollment, the opposite direction ofhese gaps in 1992.

The benefits of AP courses in college enrollmentiverged much more sharply along social class than racialnd ethnic lines, with continuing-generation studentsxperiencing more than a 100% increase in the effect of AParticipation on college enrollment, and first-generationtudents experiencing significant drops in the value of theredential for both general college enrollment and selective

ollege enrollment. It is possible that usurpationary claimsade in terms of race and ethnicity have helped to stem

he inflation of the value of the AP credential for whites andast Asians. Therefore, the advantages of AP course-taking

may have been more effectively protected along social classlines than racial and ethnic lines.

Strategies of adaptive closure on the part of privilegedparents and status-seeking high schools have likely con-tributed to this increase in the value of the AP credentialfor privileged students and the concomitant drop in itsvalue for disadvantaged students. When social closurepractices surrounding a non-standardized credential shiftfrom attempts to monopolize access to the credential toinfluencing how institutional gatekeepers perceive the cre-dential, the differences in the meaning of the credentialcrystallize along status group lines. Because status groupdifferences in the returns to non-standardized credentialsare largely invisible, adaptive closure strategies that resultin these differential returns may also go largely undetectedas such.

One limitation of this study is that the data lack directmeasures of adaptive closure strategies. Thus, we donot know what parents from privileged groups in thesesamples actually did to gain and protect access to high-quality AP courses for their children. For example, wedo not know how many of these parents made residen-tial moves to gain access to a high school with betterAP course offerings, lobbied their children’s teachers fortheir children to be placed in AP courses, or opposedefforts to increase access to AP courses at their children’sschools. It is important to keep in mind that adaptivesocial closure strategies employed by privileged groupsare not necessarily devices consciously evolved for thepurpose of subordinating other groups. In the educationsystem, adaptive closure strategies are deeply embeddedin ideologies of meritocracy and entitlement, encourag-

ing both privileged and disadvantaged families to see thecompetition for educational resources as fair and just.Measuring closure strategies when they are not con-scious, and when employing them conforms to normative

28 T. Wildhagen / Research in Social Stratification and Mobility 38 (2014) 18–31

Fig. 3. Actual and projected changes in the value of the AP credential across groups. Note: Projected changes between 2004 and 2016 assume the samerate of change in the value of the AP credential that occurred between 1992 and 2004 for each group.

expectations about how parents should behave regardingtheir children’s education, does present some challenges.Qualitative research is better suited to the task of uncov-ering exactly how contemporary parents from privilegedgroups are protecting their children’s access to high-qualityAP courses, and this is a promising direction for futureresearch.

In order to capture an understanding of the full pro-cess by which adaptive social closure can lead to a drop inthe value of democratized academic credentials, researchalso needs to investigate how college admissions officers

construct the meaning of non-standardized credentialsfor different kinds of applicants. Understanding how andwhether adaptive closure strategies affect admissions offi-cers’ perceptions of the meaning of these credentials could

hold clues about how democratized credentials come to bedevalued for members of disadvantaged groups.

Acknowledgements

This research was supported by a grant from the Ameri-can Educational Research Association which receives fundsfor its ‘AERA Grants Program’ from the National ScienceFoundation under NSF Grant #DRL-0941014. Opinionsreflect those of the author and do not necessarily reflectthose of the granting agencies.

Appendix.

See Tables A1 and A2.

T. W

ildhagen /

Research

in Social

Stratification and

Mobility

38 (2014)

18–31

29

Table A1Descriptions of variables used in analyses, by race/ethnicity and parent education (Class of 2004, ELS).

Variable Metric Underrepresented Overrepresented First generation Continuing generation

Mean SD Mean SD Mean SD Mean SD

Dependent variablesFour-year college 0/1 .327 .469 .511 .500 .312 .463 .636 .481Selective four-year collegea 0/1 .231 .422 .409 .492 .227 .419 .458 .498Math score, 12th grade 15.200–82.540 41.099 14.035 51.762 14.812 43.979 14.255 53.910 15.099SAT score 400–1600 894.490 196.395 1043.954 197.611 936.043 186.301 1068.529 206.869

Independent variableAP course 0/1 .184 .388 .315 .464 .175 .380 .406 .491

Propensity score predictorsb

Demographic variablesFemale 0/1 .501 .500 .500 .500 .502 .500 .499 .500Two parents 0/1 .655 .475 .822 .382 .746 .435 .791 .406English second language 0/1 .267 .442 .130 .336 .195 .396 .152 .359Parent education 1–8 4.020 2.099 4.750 2.061 2.999 1.310 6.621 .754Family income 0–200,000 47,654 40,661 70,951 49,751 45,322 33,570 87,873 54,195Academic preparationSpecial education 0/1 .094 .284 .068 .252 .090 .279 .058 .239Remedial English 0/1 .097 .291 .083 .277 .093 .288 .080 .273Remedial math 0/1 .115 .313 .094 .294 .108 .307 .091 .291Plans to take SAT/ACT 0/1 .651 .468 .733 .441 .632 .474 .807 .398Academic achievementMath score, 10th grade 12.523–69.719 32.284 10.910 40.963 11.309 34.958 11.167 42.244 11.620Reading score, 10th grade 10.199–49.085 25.843 9.127 31.989 9.332 27.503 9.249 33.225 9.356GPA, 9th grade .0–4.0 2.253 .891 2.755 .868 2.395 .901 2.848 .850Academic effort and school behaviorHours of homework 0–26 5.383 5.846 6.339 6.083 5.188 5.549 7.173 6.453Preparation for class 1–4 3.006 .867 3.204 .757 3.097 .816 3.190 .779Teachers’ perceptions of studentengagement

−2.868 to 1.385 (standardized index,0 = sample mean)

−.243 .757 .104 .737 −.165 .752 .196 .726

Skip class 0–11 1.435 2.794 .895 2.218 1.230 2.596 .870 2.198Late for school 0–11 3.565 3.437 2.650 3.091 3.071 3.299 2.815 3.159Trouble in school 0–11 1.496 2.499 1.254 2.280 1.412 2.421 1.233 2.269In-school suspension 0–11 .510 1.525 .242 1.090 .427 1.414 .202 .998Suspension 0–11 .327 1.243 .146 .791 .254 1.067 .143 .820Attitudes toward educationAcademic self-concept 1–4 2.623 .663 2.683 .659 2.585 .658 2.773 .650Attitudes toward school −4.353 to 1.818 (standardized index,

0 = sample mean).089 .683 −.048 .719 −.022 .719 .028 .696

Educational expectations 1–7 4.983 1.507 5.214 1.400 4.864 1.506 5.519 1.250Parents’ educational expectations 1–7 5.465 1.354 5.339 1.229 5.175 1.367 5.677 1.064Friends’ attitudes toward school 1–3 2.413 .508 2.414 .506 2.374 .516 2.469 .487School characteristicsPrivate school (non-Catholic) 0/1 .048 .214 .112 .315 .047 .211 .152 .359Catholic school 0/1 .092 .288 .138 .345 .083 .276 .176 .381Percent students in AP 0–81 13.902 12.177 15.822 14.881 12.692 11.832 18.473 16.049Percent students white 0–100 43.518 31.570 77.199 23.922 62.550 35.552 70.062 28.618a Given four-year college enrollment.b Propensity score predictors come from the 10th-grade wave of the survey.

30

T. W

ildhagen /

Research

in Social

Stratification and

Mobility

38 (2014)

18–31Table A2Descriptions of variables used in analyses, by race/ethnicity and parent education (Class of 1992, NELS).

Variable Metric Underrepresented Overrepresented First generation Continuing generation

Mean SD Mean SD Mean SD Mean SD

Dependent variablesFour-year college 0/1 .420 .494 .562 .496 .429 .495 .744 .436Selective four-year collegea 0/1 .206 .405 .310 .462 .172 .378 .410 .492Math score, 12th grade 16.97–78.1 40.198 13.135 50.403 14.337 44.794 13.760 57.164 12.843SAT score 410–1539 786.543 190.242 941.174 211.176 844.920 187.714 1021.121 208.429Independent variableAP course 0/1 .021 .144 .087 .281 .028 .164 .184 .387Propensity score predictorsb

Demographic variablesFemale 0/1 .521 .500 .504 .500 .519 .500 .498 .500Two parents 0/1 .692 .459 .835 .372 .767 .422 .882 .322English second language 0/1 .278 .443 .076 .263 .137 .343 .103 .304Parent education 1–6 2.506 1.172 3.171 1.238 2.404 .752 4.660 .763Family income 0–200,000 25,396 25,624 43,101 36,567 28,515 23,180 66,552 44,900Academic preparationSpecial education 0/1 .026 .151 .023 .143 .023 .141 .014 .114Remedial English 0/1 .218 .405 .188 .386 .216 .405 .113 .317Remedial math 0/1 .260 .425 .193 .391 .239 .416 .108 .316Plans to take SAT/ACT 0/1 .579 .492 .653 .479 .565 .496 .841 .373Academic achievementMath score, 10th grade 16.37–72.76 35.659 12.439 45.021 13.995 39.632 13.233 51.867 12.668Reading score, 10th grade 10.15–48.8 25.617 9.201 31.291 9.982 27.928 9.609 35.843 8.988GPA, 9th grade .5–4.0 2.744 .714 2.981 .752 2.810 .744 3.265 .647Academic effort and school behaviorHours of homework 0–16 3.668 3.854 4.535 4.459 3.739 3.954 5.971 4.864Preparation for class 1–4 1.816 .629 1.747 .563 1.780 .586 1.713 .534Teachers’ perceptions ofstudent engagement

−1.607 to 1.478 (standardizedindex, 0 = sample mean)

−.171 .499 .016 .529 −.125 .499 .261 .490

Skip class 0–11 1.731 2.904 1.353 2.617 1.539 2.776 1.129 2.351Late for school 0–11 3.362 3.313 2.822 3.159 2.956 3.200 2.866 3.149Trouble in school 0–11 1.362 2.225 1.369 2.393 1.447 2.417 1.101 2.034In-school suspension 0–11 .436 1.157 .299 1.141 .385 1.212 .137 .713Suspension 0–11 .265 .910 .163 .806 .222 .906 .060 .428Attitudes toward educationAcademic self-concept 1–6 4.138 .996 4.264 1.019 4.132 1.008 4.547 .953Attitudes toward school −3.770 to 1.487 (standardized

index, 0 = sample mean).051 .745 -.053 .764 -.068 .761 .104 .706

Educational expectations 1–9 5.802 2.336 6.197 2.189 5.699 2.266 7.366 1.499Parents’ educationalexpectations

0/1 .700 .452 .709 .453 .657 .470 .866 .345

Friends’ attitudes towardschool

1–3 2.451 .492 2.408 .494 2.389 .498 2.503 .471

School characteristicsPrivate school(non-Catholic)

0/1 .023 .149 .075 .263 .021 .145 .168 .374

Catholic school 0/1 .051 .220 .069 .254 .050 .218 .110 .312Percent students in AP 0–100 10.058 12.377 11.655 15.746 9.320 12.270 15.865 19.458Percent students white 0–100 41.940 32.018 81.897 22.969 70.348 32.026 78.624 24.972

a Given four-year college enrollment.b Propensity score predictors come from the 8th-grade and 10th-grade waves of the survey.

tratificat

R

A

A

A

A

A

B

B

B

B

C

C

C

G

G

H

H

K

K

K

L

M

T. Wildhagen / Research in Social S

eferences

merican Civil Liberties Union (ACLU). (1999). In class-action lawsuit, ACLUsays CA students are denied equal access to advanced placement courses.Retrieved from http://www.aclu.org/racial-justice/class-action-lawsuit-aclu-says-ca-students-are-denied-equal-access-advanced-placement

llison, P. D. (2000). Multiple imputation for missing data: A cautionarytale. Sociological Methods & Research, 28, 301–309.

lon, S. (2009). The evolution of class inequality in higher education: Com-petition, exclusion, and adaptation. American Sociological Review, 74,731–755.

lon, S., & Tienda, M. (2007). Diversity opportunity, and the shiftingmeritocracy in higher education. American Sociological Review, 72,487–511.

ttewell, P., & Domina, T. (2008). Raising the bar: Curricular intensity andacademic performance. Educational Evaluation and Policy Analysis, 30,51–71.

ecker, S. O., & Ichino, A. (2002). Estimation of average treatment effectsbased on propensity scores. Stata Journal, 2, 358–377.

owen, W. G., & Bok, D. (1998). The shape of the river: Long-term con-sequences of considering race in college and university admissions.Princeton, NJ: Princeton University Press.

owen, W. G., Kurzweil, M. A., & Tobin, E. M. (2005). Equity and excellence inAmerican higher education. Charlottesville, VA: University of VirginiaPress.

rown, D. K. (2001). The social sources of educational credentialism: Sta-tus cultures, labor markets, and organizations. Sociology of Education,Extra Issue, 19–34.

amara, W., & Michalis, M. (2005). AP use in admissions: A response to Geiserand Santelices. New York: The College Board.

arnegie Foundation (2012). Retrieved from http://classifications.carnegiefoundation.org/methodology/ugrad profile.php

ollins, R. (1979). The credential society: An historical sociology of educationand stratification. New York: Academic Press.

eiser, S., & Santelices, V. (2004). The role of advanced placement and hon-ors courses in college admissions. Berkeley: University of California(Research & Occasional Paper Series No. CSHE 4.04)

ollub, J. P., Bertenthal, M. W., Labov, J. B., & Curtis, P. C. (2002). Learn-ing and understanding: Improving advanced study of mathematics andscience in high schools. Washington, DC: National Academies Press.

ong, G., & Raudenbush, S. W. (2006). Evaluating kindergarten retentionpolicy: A case study of causal inference for multilevel obser-vational data. Journal of the American Statistical Association, 101,901–910.

ong, G., & Raudenbush, S. W. (2005). Effects of kindergarten retentionpolicy on children’s cognitive growth in reading and mathematics.Educational Evaluation and Policy Analysis, 27, 205–224.

ao, G., & Thompson, J. S. (2003). Racial and ethnic stratification in edu-cational achievement and attainment. Annual Review of Sociology, 29,417–442.

lugman, J. (2013). The advanced placement arms race and the repro-duction of educational inequality. Teachers College Record, 115,1–34.

lugman, J. (2012). How resource inequalities among high schools repro-duce class advantages in college destinations. Research in HigherEducation, 53, 803–830.

ucas, S. R. (2001). Effectively maintained inequality: Education transi-tions track mobility, and social background effects. American Journalof Sociology, 106, 1642–1690.

ayer, S. E. (1997). What money can’t buy: Family income and children’s lifechances. Cambridge, MA: Harvard University Press.

ion and Mobility 38 (2014) 18–31 31

Monroe, S. J. (2008). Letter from the assistant secretary of the Office forCivil Rights in the United States Department of Education. Retrievedfrom http://www2.ed.gov/about/offices/list/ocr/letters/colleague-20080522.pdf

Morgan, S. L. (2001). Counterfactuals, causal effect heterogeneity, and theCatholic school effect on learning? Sociology of Education, 74, 341–374.

National Association for College Admission Counseling (NACAC). (2009).Factors in the admission decision. Retrieved from http://www.nacacnet.org/research/PublicationsResources/Marketplace/Documents/RTPBrief Factors.pdf

The National Center for Fair and Open Testing (NCFOT). (2012). SAT/ACToptional 4-year universities test score optional list. Retrieved fromhttp://www.fairtest.org/university/optional

Pager, D. (2007). Marked: Race, crime and finding work in an era of massincarceration. Chicago: The University of Chicago Press.

Parkin, F. (1979). Marxism and class theory: A bourgeois critique. New York:Columbia University Press.

Rosenbaum, P. R., & Rubin., D. B. (1983). The bias due to incomplete match-ing? Biometrics, 41, 103–116.

Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensityscore in observational studies for causal effects? Biometrika, 70, 41–55.

Rosenbaum, P. R., & Rubin, D. B. (1985). Constructing a control groupusing multivariate matched sampling methods that incorporate thepropensity score. The American Statistician, 39, 33–38.

Rubin, D. B. (1987). Multiple imputation for nonresponse in surveys. NewYork: John Wiley.

Sawtell, E. A., & Gillie, J. M. (2007). What we know about AP students.Presentation at the AP annual conference. Retrieved from http://research.collegeboard.org/sites/default/files/publications/2012/7/presentation-2007-2-what-we-know-ap-students.pdf

Schafer, J. L. (2000). NORM: Multiple imputation of incomplete multivariatedata under a normal model, Version 2.03, software for Windows 95/98/NT.http://www.stat.psu.edu/∼jls/misoftwa.html

Schmidt, P. (2007). Color and money: How rich white kids are winning thewar over college affirmative action. New York: Palgrave Macmillan.

Schneider, B., Carnoy, M., Kilpatrick, J., Schmidt, W. H., & Shavelson, R. J.(2007). Estimating causal effects using experimental and observationaldesigns (report from the Governing Board of the American EducationalResearch Association Grants Program). Washington, DC: American Edu-cational Research Association.

The College Board. (2012). The 8th annual AP report to the nation. Retrievedhttp://apreport.collegeboard.org/sites/default/files/downloads/pdfs/AP Main Report Final.pdf

The College Board. (2011). The college completion agenda? Retrieved fromhttp://completionagenda.collegeboard.org/percentage-public-high-schools-offering-ap%C2%AE-or-ib-courses-four-core-subject-areas

The College Board. (2001). Access to excellence: A report of the commissionon the future of the advanced placement program. Retrieved fromhttp://research.collegeboard.org/sites/default/files/publications/2012/7/misc2001-2-future-advanced-placement.pdf

Useem, E. (1992). Middle schools and math groups: Parents’ involvementin children’s placement. Sociology of Education, 65, 263–279.

Weber, M. (1951). The religion of China. New York: Free Press (originalwork published 1916).

Wells, A. S., & Oakes, J. (1996). Potential pitfalls of systemic reform:Early lessons from detracking research. Sociology of Education, 69,E135–E143.

Wells, A. S., & Serna, I. (1996). The politics of culture: Understanding localpolitical resistance to detracking in racially mixed schools. HarvardEducational Review, 66, 93–118.

Winship, C., & Morgan, S. L. (1999). The estimation of causal effects fromobservational data. Annual Review of Sociology, 25, 659–707.