AAn Empirical Analysis of Acting Whiten Empirical Analysis of Acting White

Journal of Public Economics xxx (2010) xxx–xxx

PUBEC-02935; No of Pages 17

Contents lists available at ScienceDirect

Journal of Public Economics

j ourna l homepage: www.e lsev ie r.com/ locate / jpube

ARTICLE IN PRESS

An empirical analysis of ‘acting white’

Roland G. Fryer Jr. a,b,⁎, Paul Torelli c

a Robert M. Beren Professor of Economics, Harvard University, United Statesb National Bureau of Economic Research, United Statesc Edgeworth Economics 1050 Connecticut Avenue, NW 10th Floor Washington, DC 20036, United States

⁎ Corresponding author. Harvard University, 1875 Cam02138, United States. Tel.: +1 617 495 9592.

E-mail addresses: [email protected] (R.G. Fryer(P. Torelli).

1 This fact was first established by Coleman et al. (19see Campbell et al. (2000), Carneiro and Heckman (22005), Neal (2004), or Phillips et al. (1998).

0047-2727/$ – see front matter © 2009 Published by Edoi:10.1016/j.jpubeco.2009.10.011

Please cite this article as: Fryer, R.G., Toreljpubeco.2009.10.011

a b s t r a c t
a r t i c l e i n f o
Article history:Received 22 November 2008Received in revised form 11 September 2009Accepted 23 October 2009Available online xxxx

Keywords:Social interactionsRacial achievement gap

Using a newly available data set, which allows one to construct a novel measure of a student's social status,we demonstrate that there are potentially important racial differences in the relationship between socialstatus and academic achievement. The effect is concentrated among students with a grade point average(GPA) of 3.5 or higher and more pronounced in schools with more interracial contact. Earlier studiesshowing a positive relationship between popularity and academic achievement for blacks are sensitive to theinclusion of more continuous achievement measures. We argue that the data are most consistent with amodel of ‘acting white’ in which investments in education are taken as a signal of one's opportunity costs ofpeer-group loyalty, though imprecise estimates make definitive conclusions difficult.

bridge Street, Cambridge, MA,

), [email protected]

66). For more recent analysis,002), Fryer and Levitt (2004,

2 It is imperativeregard. A version ofBuraku Outcastes ofWest End (Gans, 19working class in Brreceive a derogatorsociety documentedwere labeled “mobithese groups. Furthethis paper.

3 There are many w

lsevier B.V.

li, P., An empirical analysis of ‘acting white’,

© 2009 Published by Elsevier B.V.

“Go into any inner city neighborhood, and folks will tell you thatgovernment alone can't teach kids to learn. They know thatparents have to parent, that children can't achieve unless we raisetheir expectations and turn off the television sets and eradicate theslander that says a black youth with a book is acting white.”Then-State Senator Barack Obama, 2004 Democratic NationalConvention Keynote Address

to note that there is nothing unique about black culture in this‘acting white’ is also prevalent in ethnographies involving the

1. Introduction

The racial achievement gap in education is a vexing reality. Blackchildren enter kindergarten lagging whites, and these differences growthroughout their school years.1 On every subject, at every grade level,there are large achievement differences between blacks and whites(Campbell et al., 2000; Neal, 2004). The typical black seventeen-year-old reads at the proficiency level of the typical white thirteen-year-old(Campbell et al., 2000). On the Scholastic Aptitude Test, black college-goers score, on average,more thanone-standard-deviationbelowwhitecollege-goers (Card and Rothstein, 2004). Even in affluent neighbor-hoods, the racial divide is daunting (Ferguson, 2001, 2002; Ogbu andDavis, 2003). Accounting for a host of background factors, theachievement gap remains essentially unchanged (Jencks and Phillips,1998).

A wide variety of possible explanations for the test-score gap havebeen put forth. These explanations include differences in genetics(Herrnstein and Murray, 1994; Jensen, 1998), differences in familystructure and poverty (Armor, 1992; Brooks-Gunn and Duncan,1997), differences in school quality (Cook and Evans, 2000), racialbias in testing or teachers' perceptions (Delpit, 1995; Ferguson,1998a; Rodgers and Spriggs, 1996), and differences in peer culture,socialization, or behavior (Austen-Smith and Fryer, 2005; Cook andLudwig, 1997; Fordham and Ogbu, 1986; Steele and Aronson, 1998).The appropriate public policy choice to address the achievement gapmay depend critically on its underlying source.

In this paper, we focus on a highly controversial and well-publicized aspect of black peer interaction — the existence of a peerexternality commonly referred to as ‘acting white.’2 ‘Acting white’describes a set of social interactions in which some minorities incurcosts for investing in behaviors characteristic of whites (e.g., raisingtheir hand in class, making good grades, or having an interest inballet).3 A primary obstacle to the study of ‘acting white’ has been thelack of quantitative measures of the phenomenon. We focus on racialdifferences in the relationship between social status and academicachievement: our (albeit, narrow) definition of ‘acting white’.

Japan (DeVos and Hiroshi, 1966), Italian immigrants in Boston's62), the Maori of New Zealand (Chapple et al., 1997), and theitain (Willis, 1977), among others. In all cases high-achieversy label from their peer group. For example, in the peer-groupin Gans (1962), upwardly mobile youth interested in educationles” and “sissies.” See Fryer (2007) for a detailed discussion ofrmore, we examine Hispanics in addition to blacks throughout

orking definitions of ‘acting white’ which we discuss in Section 2.

Journal of Public Economics (2010), doi:10.1016/j.

mailto:[email protected]

mailto:[email protected]

http://dx.doi.org/10.1016/j.jpubeco.2009.10.011

http://www.sciencedirect.com/science/journal/00472727



2 R.G. Fryer Jr., P. Torelli / Journal of Public Economics xxx (2010) xxx–xxx

ARTICLE IN PRESS

To circumvent some of the problems inherent in self-reportedpopularity measures, we construct our own index of social status. Foreach student the index measures the number of same-race friendswithin her school, weighted by the social status of each friend (resultsare unaltered if we use all friends). We implement the index usingdetailed information on friendship networks within schools availablein the National Longitudinal Study of Adolescent Health (Add Health).Add Health is, in many respects, the ideal data for understanding theexistence of ‘acting white.’ To our knowledge, it is the only availabledata set that contains within-school friendship networks from whichit is possible to construct novel, and arguably more objective, measuresof social status, along with detailed data on parental characteristics,academic achievement, and so forth. The survey covers a sample ofmore than 90,000 junior-high and high-school students from 175schools in 80 communities around the US.

Our empirical analysis of ‘acting white’ uncovers a rich set of newfacts. In contrast to the previous literature (Ainsworth-Darnell andDowney, 1998; Cook and Ludwig, 1997), Fig. 1A demonstrates thatthere are racial differences in the relationship between social statusand academic achievement. Among whites, higher grades areassociated with higher social status. For blacks, higher achievementis associated with modestly higher social status until a grade pointaverage of 3.48, when the slope turns negative. A black student with a4.0 grade point average has, on average, 1.5 fewer same-race friendsthan a white student with a 4.0. Among Hispanics, there is littlechange in social status from a grade point average of 1 through 2.5.After 2.5, the gradient turns sharply negative. A Hispanic student witha 4.0 grade point average is the least popular of all Hispanic students,and has 3 fewer friends than a typical white student with a 4.0 gradepoint average. Put differently, evaluated at the sample mean, a one-standard-deviation increase in grades is associated with roughly a.103-standard-deviation decrease in social status for blacks and a.171-standard-deviation decrease for Hispanics. For students with a3.5 out of 4.0 grade point average or better, themagnitude of the racialdifference is three times as large. It is imperative to note at the outsetthat the majority of black and Hispanic students earn GPAs between2.0 and 3.0, where racial differences in the relationship between socialstatus and academic achievement exist only for Hispanics.

We caution against a causal interpretation of the coefficients on thecovariates, which we view as proxies for a broad set of environmentaland behavioral factors. Even our main parameter is not void of thepotential for reverse causality. We implicitly assume that highachievement causes lower social status; it is plausible that low socialstatus causes high grades. We failed to find a plausible instrument for astudent's academic achievement, butweremildly successful atfinding a

Fig. 1. A: Spectral popularity and grades by race, raw data. B

Please cite this article as: Fryer, R.G., Torelli, P., An empirical analysis ojpubeco.2009.10.011

way to vary popularity in ways that may not affect grades: interviewerratings of physical attractiveness for each student. If we think that aregression of popularity on grades gives a significant coefficient eitherbecause of simultaneity and/or omitted variables, we can write it as atwo-equation system. Then, if we have a valid instrument for the secondequation (where grades are the dependent variable), it gives a morevalid estimate of the causal effect of popularity on grades (Wooldridge,2002). Instrumenting for popularity using attractiveness has a strongfirst stage, but there are no racial differences in the effect on grades,suggesting that the causality arrow points in the assumed direction(grades cause popularity).

Racial differences in the relationship between social status andachievement are robust across many alternative empirical models,subsets of the data, and definitions of both social status and academicachievement. Estimating a fully interacted model does not alter theresults. The ‘actingwhite’ effect differs slightly as children age. For blackstudents in junior high school (grades 7–9) a one-standard-deviationincrease in grades is associated with a .13-standard-deviation decreasein social status; the decrease is .07 for high-school students. Accountingfor the number of students at each GPA level does little to temper the‘actingwhite’ effect, suggesting that the supply of potential friends doesnot explain the phenomenon. Self-selection of students into particularactivities (sports, band, debate, etc.) matters little, and substitutiontowards other-race friendships does not fully explain the starkdifference in the social status-achievement gradient. Using alternativemeasures of academic achievement confirms our results.

We argue that the empirical patterns are most consistent with asimplemodel of peer pressure. The principal idea is that individuals facea two-audience signaling quandary: signals that beget labor-marketsuccess are signals that induce peer rejection. The model's twodistinguishing predictions — racial differences in the relationshipbetween peer-group acceptance and academic achievement will existand these differences will be exacerbated in arenas that foster moreinterracial contact or increased mobility — are borne out in the data.Other models we consider, such as self-sabotage among black youth orthe presence of an oppositional culture identity,fitmuchof the data, butare inconsistent with the fact that racial differences in the relationshipbetween social status and academic achievement are larger inpredominantly white schools relative to predominantly black ones.

The remainder of the paper is structured as follows. Section 2provides a brief review of the literature on ‘acting white.’ Section 3describes and summarizes the data used in the analysis. Section 4provides a derivation of our measure of social status along with aconstructive example. Section 5 estimates the relationship betweensocial status and academic achievement among racial groups.

: Spectral popularity and grades by race, with controls.

f ‘acting white’, Journal of Public Economics (2010), doi:10.1016/j.



3R.G. Fryer Jr., P. Torelli / Journal of Public Economics xxx (2010) xxx–xxx

ARTICLE IN PRESS

Section 6 reconciles our results with the previous literature on ‘actingwhite.’ 7 uses economic and social theory to interpret the set of facts.Section Section 8 concludes. There are two appendices: Appendix Aprovides the details of our sample and definitions of relevant variablesused in our analysis; Appendix B provides a guide to programs used tocalculate our index of social status.

2. Background and previous literature

‘Acting white’ is a slippery and politically explosive phrase, withlittle consensus on a precise definition. Neal-Barnett (2001) assembledstudent focus groups in an attempt to deduce what specific behaviorsled to accusations of ‘acting white.’ The list included being enrolled inhonors or advanced placement classes, speaking standard English,wearing clothes from the Gap or Abercrombie & Fitch (instead ofTommy Hilfiger or FUBU), and wearing shorts in the winter. While weare cognizant of the complications and nuances in what is oftenmeantby ‘acting white,’ our data are not rich enough to test many of theplausible definitions.4 As such, for the purposes of this paper, we say‘actingwhite’ exists if there are statistically significant racial differences inthe relationship between social status and academic achievement.

For nearly two decades, there has been a divisive debate amongsociologists, cultural anthropologists, newspaper journalists, andpolicy makers on the existence of ‘acting white’. Fordham and Ogbu(1986) argue for the existence of an oppositional culture among blackyouth that eschew behaviors traditionally seen as a prerogative forwhites. Following this work, there has been a flurry of conflictinganalyses as to the nature, extent, and definition of ‘acting white.’Ethnographic evidence is hopelessly divided, and the only twonationally representative studies dismiss ‘acting white’ as nothingmore than an urban (or, more precisely, ethnographic) legend.

An interesting feature of the ethnographic literature is that manystudies report some negative relationship between “white behaviors”and social status among blacks.5 The key differences lie in the fact thatmany of the authors do not report that academic achievement, per se, isidentified as a “white behavior.” Instead, black students are deemed“white” for behaviors that are often associated with good grades (e.g.,answering questions in class, being in advanced classes, or properdiction). Even when academic achievement is highly correlated with“white behaviors,”many authors make a distinction between resentingachievement versus resenting behaviors associated with achievement.6

The precise behaviors that lead to accusations of ‘actingwhite’maybe useful in designing effective policies, but the first-order problem iswhether racial differences in the relationship between social statusand achievement exist. These differences potentially feed into studentinvestment decisions regarding human capital, social affiliations, andother important economic decisions.7

There have been two previous studies on the existence of ‘actingwhite’ using nationally representative data (Cook and Ludwig, 1997;Ainsworth-Darnell and Downey, 1998). Both are based on theNational Educational Longitudinal Study (NELS). These studiesexamine the relationship between self-reported measures of popu-larity and dichotomousmeasures of academic achievement, providingevidence that peer-group norms are not significantly different

4 See Ferguson (1998b) for a very nice discussion.5 See, for example, Ferguson (2001), Horvat and Lewis (2003), Bergin and Cooks

(2002), Datnow and Cooper (1996), Farkas et al. (2002), and Tyson et al. (2004).6 This reticence could be due, in part, to the fear that some may equate ‘acting white’

with black cultural dysfunctionality. Yet, economic theory informs us that ‘actingwhite’ is an equilibrium phenomenon; it is the consequence of two-audience signaling(see Austen-Smith and Fryer, 2005). As such, any group presented with the same set ofpayoffs, strategies, and so on, would behave identically.

7 For example, some black students may be accused of ‘acting white’ because of theway they talk or dress, not because they get good grades. This may be because they arein high-ability classes (AP, Honors, etc.) where there are few black students, andconsequently their social circle may have few blacks. See Ferguson (2001).


between black and white tenth graders. Cook and Ludwig (1997)find that high-achievers are actually more popular than low-achievers, and that this positive achievement–popularity relationshipis not significantly different between whites and blacks — if anything,the relation is stronger among blacks, suggesting the ‘acting white’phenomenon is not empirically relevant.8

An important limitation of the previous work is that it relies onmeasures of popularity and academic achievement that have littlevariation. Section 6 investigates how popularity varies across a morecontinuous measure of achievement (GPA) using NELS data and ourbasic specification. We find that racial differences in the relationshipbetween social status and academic achievement exist and are robustto including a myriad of controls.

3. The data

The National Longitudinal Study of Adolescent Health (AddHealth) data set is a nationally representative sample of 90,118students entering grades 7 through 12 in the 1994–1995 academicyear. A stratified random sample of 20,745 students was given anadditional (and remarkably comprehensive) in-home interview;17,700 parents of these children were also interviewed. Thus far,information has been collected on these students at three separatepoints in time: 1995, 1996, and 2002. There are 175 schools in 80communities included in the sample, with an average of more than490 students per school, allowing within-school analysis. Studentswho are missing data on race or grade level, or who attend schoolsthat do not assign grades are dropped from the sample.

A wide range of data is gathered on the students, as described indetail on the Add Health website (http://www.cpc.unc.edu/projects/addhealth). For our purposes, the key feature of the Add Health data isthe detailed information regarding friendship associations in schools.All students contained in the in-school survey were asked, “List yourclosest male/female friends. List your best male/female friend first,then your next best friend, and so on.” Students were allowed to list asmany as five friends from either sex. Each friend can be linked in thedata and the full range of covariates in the in-school survey (race,gender, grade point average, etc.) can be gleaned from each friend.The next section details how we take this information on friendshiplinkages within schools and construct an index of social status.

Summary statistics for the variables we use in our core specifica-tions are displayed by race in Table 1, with white and black referringsolely to non-Hispanic whites and blacks, respectively.9 Hispanicsinclude any individual who checked yes to the question: “Are you ofHispanic or Spanish origin?” Our primary outcome variable is thesocial status index. The index is normalized such that it has a mean ofzero and a standard deviation of one. On average, blacks and Hispanicshave fewer same-race social connections than whites. Whites are .142above the mean in social status whereas blacks are .024 below themean, yielding a black–white difference in social status of .166.Hispanics are .141 below the mean social status, yielding a Hispanic–white gap of .283. The table also indicates the average number offriends students have from different racial backgrounds.10 Friendshipnetworks are remarkably segregated; the typical black and whitestudent each has exactly one friend of a different race.

The remainder of Table 1 presents summary statistics for the othervariables used in our analysis. The most important of these covariatesis a composite measure of grade point average (GPA) that weconstructed. Each student was asked, “At the most recent grading

8 Their results hold even when they control for nonacademic variables thatinfluence popularity, including family income and participation in varsity sports orband (Cook and Ludwig, 1997).

9 A detailed description of these variables and their construction can be found inAppendix A.10 The rank correlation between the social status index of an individual i and thenumber of students who list i as a friend is 0.74.


http://www.cpc.unc.edu/projects/addhealth

http://www.cpc.unc.edu/projects/addhealth



Table 1Summary statistics by race, add health.

Full sample Whites Blacks Hispanics

White .546 (.498) 1.000 (.000) .000 (.000) .000 (.000)Black .147 (.354) .000 (.000) 1.000 (.000) .000 (.000)Hispanic .160 (.367) .000 (.000) .000 (.000) 1.000 (.000)Asian .048 (.214) .000 (.000) .000 (.000) .000 (.000)Other .099 (.299) .000 (.000) .000 (.000) .000 (.000)Popularity .000 (1.000) .132 (1.162) −.022 (.697) −.149 (.879)Same-race friends 4.360 (3.989) 5.648 (4.129) 4.317 (3.616) 2.660 (2.992)Other-race friends 2.014 (2.696) 1.305 (1.577) 1.716 (2.167) 2.547 (3.459)Grade point average 2.800 (.808) 2.914 (.793) 2.561 (.754) 2.552 (.801)Age 15.029 (1.790) 15.057 (1.697) 14.954 (1.888) 15.185 (1.930)Male .488 (.500) .495 (.500) .452 (.498) .494 (.500)Effort 3.122 (.897) 3.149 (.775) 3.174 (1.001) 2.971 (1.113)Sports .522 (.500) .545 (.498) .516 (.500) .444 (.497)Student government .077 (.267) .078 (.269) .096 (.294) .055 (.228)Cheerleading .081 (.273) .076 (.265) .111 (.314) .072 (.259)Mother college educated .262 (.440) .298 (.457) .236 (.425) .135 (.342)Father college educated .254 (.435) .311 (.463) .149 (.356) .130 (.336)Mother professional .283 (.450) .317 (.465) .291 (.454) .172 (.377)Father professional .253 (.435) .321 (.467) .124 (.330) .135 (.342)Urban .330 (.470) .231 (.421) .418 (.493) .600 (.490)Suburban .571 (.495) .645 (.478) .470 (.499) .367 (.482)Private school .063 (.242) .074 (.261) .035 (.183) .032 (.177)Percent black in school 14.677 (20.136) 8.050 (11.503) 42.302 (26.997) 13.697 (16.371)Percent Hispanic in school 16.002 (19.562) 9.018 (10.138) 14.933 (15.985) 39.915 (29.295)

Frequency of missing valuesMissing age .002 (.041) .001 (.025) .004 (.060) .003 (.057)Missing mother's education .194 (.395) .138 (.345) .228 (.420) .290 (.454)Missing father's education .342 (.475) .247 (.431) .554 (.497) .450 (.498)Missing mother's occupation .224 (.417) .164 (.371) .311 (.463) .306 (.461)Missing father's occupation .368 (.482) .276 (.447) .599 (.490) .456 (.498)Missing effort .036 (.187) .017 (.129) .059 (.235) .079 (.270)Missing gender .005 (.067) .003 (.058) .004 (.061) .008 (.086)

Note: All data are drawn from the National Longitudinal Study of Adolescent Health. All summary statistics are unweighted. “Other” includesmixed students. The first column of eachsub-sample (including full sample) is the mean and the second column is the standard deviation.


ARTICLE IN PRESS

period, what was your grade in each of the following subjects?” Thesubjects included were English/Language Arts, Mathematics, History/Social Studies, and Science.11 Each of these grades was given on a 4.0scale (A=4, D or Lower=1). Relative to whites, blacks and Hispanicshave lower grades (3.0 grade point average relative to 2.5).12

Other variables used as controls include: parental education,parental occupation, various school activities and school character-istics. There are substantial differences across races on many of thesevariables. Black and Hispanic students come from lower socio-economic status families (mother and father less likely to be collegeeducated or working in professional jobs), are more likely to live inurban neighborhoods, and less likely to attend private schools, and soon. While this may seem like a stylized set of covariates to use, theycomprise all the social and demographic variables available in the in-school survey of 90,118 students.13

4. Measuring social status

We develop an index of social status using a measure of socialconnections in networks. The basic building block is a set ofindividuals V, who belong to the same race/ethnicity and information

11 In calculating the GPA of each student, we only used courses in which valid gradeswere received, as some students did not take particular subjects in every gradingperiod. Students who did not have a grading period that academic year were told toprovide grades from the last grading period of the previous academic year. Recall thatstudents in schools that do not assign grades were dropped from the sample.12 There is some evidence that blacks overstate their grades (Bauman 1996). Ourcalculations, using the National Educational Longitudinal Study which containstranscript data and self-reported grades, suggests that blacks exaggerate their gradesby 0.3 grade points, on average.13 A much more exhaustive set of covariates is available for individuals whocompleted the in-home survey (covering approximately one fourth of the sample).


on whether any two individuals in V are socially connected. Ourmeasure identifies social status of the members of a group with theintensity of the social connections among the members of that group.

Given two individuals, v and v′ in V, let rvv′∈ {0,1} denote whetheror not they interact socially. The data used in this paper asks studentsto name their ten closest friends. We say that rvv′=rv′v=1 if v′ namesv as one of her friends, or vice versa. In other words, if either student ior student j lists each other as a friend, both students i and j willreceive an increase in social status for the connection (rij=rji=1).14

Let B be amatrix with as many rows and columns as members of V,and typical entry rvv′. The matrix B summarizes all the information onsame-race connections among themembers of the race. A social statusindex is a function that assigns a real number S(B) to each matrix B ofsame-race connections, along with functions assigning a real numbersv(B) for each individual member v of V, such that S(B) is the averageof the individual sv(B).

There are three properties that jointly define the social statusindex. The first property requires that an increase in same-raceconnections implies an increase in social status. A matrix B′ has moreintense same-race connections than matrix B if each individual v hasmore friends of her race in B′ than in B. We say that a social status

14 What is essential for our framework is that rvv′N0 if and only if rv′vN0; seeEchenique and Fryer (2005). We have also calculated all empirical results using the“intersection” of the friendship list — i is a friend of j if and only if i and j both writeeach other down. The correlation between the intersection method and the “union”method is .41. We chose the “union” method because the friendship questionexplicitly states, “list your best friend first, your second best friend second,” and so on.Thus, the “union” method is most consistent with general notions of popularity orgeneral social status (where popular people may be listed frequently, but only list afew of their own close friends), whereas the “intersection” method is more closelyrelated to strong reciprocal friendships.





ARTICLE IN PRESS

index satisfies the property of monotonicity if, whenever B′ has moreintense interactions than B, S(B)≤S(B′).

The second property is a normalization of the index. A social statusindex is homogeneous if, when all individuals in B have exactly dsame-race friends, S(B)=d.

We introducesomeadditional concepts topresent the thirdproperty.LetNv be the set of v's friends of her same race— her connections in V. Ina similar vein, consider all friends (in V) of v's friends, and friends offriends of friends, and so on. The resulting set of individuals, with director indirect connections to v, is called the connected component of B thatv belongs to; denote this set of individuals by Cv.

The third property says that sv(B) should be the average sv′(B)among v's same-race friends, relative to the average social status of theindividuals in v's connected component. In symbols, if SCv is the averagesocial status of individuals in Cv, a social status index satisfies linearity if

svðBÞ =1SCv

∑v′∈Nv

rvv′ sv′ ðBÞ

There are two qualitative assumptions behind the linearity property.Thefirst is that v's social status depends on the social status of her friends.The second is that the dependence is modulated by the connectedcomponent's social status. A decrease in the social status of one of v'sfriends will affect v less if v is in a high social status component.

The social status index is the (unique) index that satisfies theproperties of monotonicity, homogeneity, and linearity (Echeniqueand Fryer, 2005).

4.1. A constructive example

Below, we provide a precise example on how we calculate socialstatus using the theoretical apparatus explained above with data onfriendships within schools from the Add Health data set. Consider anelite middle school — Markovian — which consists of 14 students:Larry, Andrei, Franziska, Claudia, Phillip, Jeremy, Jens, Ronald, Gerhard,Katherine, Rachel, Drew, Jerry, and Matt. The top panel of Table 2shows the (hypothetical) data gleaned from these individuals whenthey were asked to list their five best friends. Larry includes Gerhard,Claudia, Andrei, and Jens as friends, while Jeremy only list Andrei andGerhard. Recall, our measure of social status for an individual i

Table 2Deriving the social status index from data on friendships, an example.

Larry Andrei Franziska Claudia Phillip Jeremy J

Hypothetical data on friendshipsGerhard Larry Gerhard Larry Larry Andrei LClaudia Jeremy Claudia Andrei Jens Gerhard GAndrei Franziska Larry Franziska Ronald PJens Gerhard Andrei Rachel Gerhard R

Katherine Gerhard

Matrix of social interactionsLarry 0 1 1 1 1 0 1Andrei 1 0 1 1 0 1 0Franziska 1 1 0 1 0 0 0Claudia 1 1 1 0 0 0 0Phillip 1 0 0 0 0 0 1Jeremy 0 1 0 0 0 0 0Jens 1 0 0 0 1 0 0Ronald 0 0 0 0 1 0 1Gerhard 1 1 1 1 1 1 1Katherine 0 1 0 0 0 0 0Rachel 0 0 0 1 0 0 0Drew 0 0 0 0 0 0 0Jerry 0 0 0 0 0 0 0Matt 0 0 0 0 0 0 0

Social status measures7.165 6.604 5.615 6.094 4.789 3.055 4


depends on the number of students, j, that list i as a friend, weightedby the social status of each j, coupled with i's original list.

The middle panel of Table 2 translates the data in the upper panelinto a matrix of social interactions (B from the formal model) thatdescribes who is socially connected to whom. Absent any informationon the strength of connections, we simply assume that individuals areeither connected or not and we denote a connection with a value of 1.For each individual in the school, the column and rows represent theirlist in the upper panel along with all individuals who listed them. Thematrix is symmetric.

The bottom panel of Table 2 presents the social status of eachstudent in Markovian Middle School. The most popular student isGerhard, with an index value of 9.22. Ten of the fourteen students listGerhard as a friend. Larry is the second most popular student, withsocial status of 7.17. Rachel is the least popular. Katherine and Rachelboth have two friends, but Rachel's friend (Claudia) is less popularthan Katherine's friend (Andrei), thus (by our linearity property)Katherine is more popular than Rachel. Drew, Matt, and Jerry are allfriends with one another and no one else at the school. By ourhomogeneity condition, each has a social status of 2.

There are several noteworthy caveats to our measure of socialstatus. First, it is possible that, given our data, we are not capturing themost popular kids in schools. It is possible that the captain of thefootball team or the most beautiful female students are quite“popular,” but few others report them as friends. Second, to theextent that students have friendship connections outside of school(through religious groups, boy scouts, and so on), these connectionswill not be captured in our index.

Despite these caveats, the index formalized in this paper isarguably more objective and appealing than measures that have beenpreviously used. The previous measure (Cook and Ludwig, 1997) tookon three values, with over 80% of the sample reporting two of thosevalues. Our index takes on values between 0 and 479 (mean=6.65,sd=14.89), which introduces more variation to understand therelationship between social status and academic achievement.

A more intuitive test of our social status measure is to correlate itwith activitieswhich one believes, ex ante, to containmore popular kids(e.g., comparing math club and non-math club students). A simplecomparison of means reveals that students in sports are .242-standard-deviations more popular than kids who are not in sports. Students in

ens Ronald Gerhard Katherine Rachel Drew Jerry Matt

arry Gerhard Larry Gerhard Gerhard Jerry Drew Jerryerhard Jens Andrei Claudia Matt Matt Drewhillip Phillip Franziskaonald Claudia

Ronald

0 1 0 0 0 0 00 1 1 0 0 0 00 1 0 0 0 0 00 1 0 1 0 0 01 1 0 0 0 0 00 1 0 0 0 0 01 1 0 0 0 0 00 1 0 0 0 0 01 0 1 1 0 0 00 1 0 0 0 0 00 1 0 0 0 0 00 0 0 0 0 1 10 0 0 0 1 0 10 0 0 0 1 1 0

.789 3.629 9.22 3.055 2.957 2 2 2




Fig. 2. A: Spectral popularity and grades by race, boys. B: Spectral popularity and grades by race, girls.


ARTICLE IN PRESS

government (.337) and cheerleading (.239) are also more popular,whereas students in the book club (−.101) and math club (−.02) areless popular.

16 In terms of capturing the ‘acting white’ phenomenon, it is not clear whether oneshould have grades and effort, grades alone, or effort alone as independent variables.

5. Racial differences in the relationship between social status andachievement

Fig. 1A presents the relationship between social status and gradesamong whites, blacks, and Hispanics in the raw data. At low gradepoint averages, there is little difference among racial groups in therelationship between social status and grades; blacks are morepopular than whites. At roughly a 2.5 GPA (an even mix of Bs andCs) racial differences start to emerge. Hispanic students lose socialstatus at an alarming rate after this cut-off — while blacks and whitescontinue to garner friends as their grades increase; the white slope issteeper. Black social status peaks at a grade point average of roughly3.48 (an even mix of As and Bs) and turns down afterward. Blackswith straight As are as popular as blacks with a 2.9 GPA. Whitescontinue to gain social status as their grades increase.

Racial differences in the social status-grades gradient may be dueto various background factors that are positively related to socialstatus (e.g., having high-income parents, attending a school that ismore socially connected). Fig. 1B estimates a non-parametricrelationship between social status and grades for blacks, Hispanics,and whites, controlling for background factors including school fixedeffects.15 Our core set of controls include gender, age, parentaleducation and occupation, a measure of (self-reported) effort, andvarious school activities such as cheerleading, athletics, and studentgovernment. Fig. 1B demonstrates that including these controlsshrinks the social status differential among low-achievers whilemaintaining the shape of the social status-grades gradient. Black andHispanic high-achievers continue to bemuch less popular than similarwhites.

Fig. 2A and B estimates a similar non-parametric relationshipbetween social status and grades for blacks, Hispanics, and whites, bygender. The relationship is remarkably similar between boys and girls.One caveat: black, high-achieving boys have fewer friends than black,

15 To understand the details, consider the following model: social status=H(grades)+Xγ+e.We cannot estimate this directly becausewe do not knowH(·). But, E(social status|grades)=H(grades)+E(X|grades)γ+E(e|grades). Taking the difference, social status−E(social status|grades)=[X−E(X|grades)]γ+ [e−E(e|grades)]. Thus, to obtain H(grades),one needs to: (1) obtain estimates of E(social status|grades) by running a non-parametricregression of social status on grades; (2) obtain estimates ofE(X|grades) by running a non-parametric regression of X on grades; (3) compute social status — E(social status|grades)and X−E(X|grades); (4) obtain γ̂ by estimating the linear regression: social status−E(social status|grades)=[X−E(X|grades)]γ+[e−E(e|grades)]; and (5) estimate socialstatus−Xγ=̂H(grades)+e (non-parametric regression of social status−Xγ̂ on grades).


high-achieving girls. Students in public and private schools face a verydifferent tradeoff between social status and grades, as demonstratedin Fig. 3A and B. Public schools mirror the aggregate data presented inFig. 1B. In private schools, whites with higher grades are not aspopular as their lower-achieving peers; the most popular studentshave a GPA of roughly 2.0 (C average). For blacks, the gradient isvirtually flat. In levels, blacks are substantially less popular thanwhites, which is an artifact of our same-race social status measure.

Table 3 presents a series of estimates designed to understand theexistence of ‘acting white’ among blacks and Hispanics, providingnumbers (and standard errors) to the figures we described.Econometrically, our analysis is of the form:

Social Statusi;s = β0 + β1⋅Achievementi + β2⋅Blacki+ β3⋅Blacki⋅Achievementi + β4⋅Hispanici+ β5⋅Hispanici⋅Achievementi + γ⋅Xi + ηs + εi;s

ð1Þ

where Xi denotes an array of individual-level variables for agent andηs denotes an exhaustive set of school dummies.16 The model isestimated on a sample of only black, white, and Hispanic students;thus β2 and β3 are interpreted as black coefficients relative to whitesand β4 and β5 are Hispanic coefficients relative to whites. We refer toβ1+β3 and β1+β5 as the ‘acting white’ coefficients for blacks andHispanics, respectively. The dependent variable is our social statusindex. The independent variables are constant across columns andinclude: (self-reported) grades and effort, socioeconomic background(parental occupation and education), and school activities (cheer-leading, sports, and student government). In all cases the estimation isdone using weighted least squares, with weights corresponding tosample weights provided in the data, and including school fixedeffects. By comparing students who attend the same school (and likelylive in similar neighborhoods), one controls for different gradingstandards, social norms, and mean social status levels acrossschools.17 As noted above, this may seem like a stylized set ofcovariates to include. Our approach is to control for these variables

We chose to concentrate on grades for three reasons. First, our specification is moreconsistent with the previous literature. Second, the effort variable in Add Health is a bitsuspicious. Respondents are asked, “In general, how hard do you try to do your schoolwork well.” Blacks report more effort than whites and cheerleaders report more effortthan students in math or science clubs. Third, we have transcripts to validate self-reported GPA, but no such measure to validate effort. If we substitute grades for effort,we find that there are no racial differences in the relationship between social statusand effort.17 For instance, if high-achieving black students attend schools where gradingstandards are lax and social status levels, inflated, then this specification will purge theupward bias.




Fig. 3. A: Spectral popularity and grades by race, public schools. B: Spectral popularity and grades by race, private schools.


ARTICLE IN PRESS

and see whether or not there are any racial differences in therelationship between social status and grades left to be explained.18

Fortunately, whether or not we include controls in addition to schoolfixed effects has little impact on the results.

Column 1 of Table 3 estimates our specification on the full sample ofstudents. These results confirm the results illustrated in Fig. 1; the ‘actingwhite’ coefficient is large and statistically significant for blacks andHispanics. To account for the non-linearity inherent in the relationshipbetween social status and achievement, we group students into fourcategories based solely onGPA: 1 to 2.0, 2.0 to 3.0, 3.0 to 3.5, and 3.5 andhigher.19 These categories are roughly alignedwith the changes in slopeapparent in Fig. 1B. The bottom two rows of Table 3 indicate the fractionof black and Hispanic students in each GPA category.

The salience of ‘acting white’ increases monotonically withachievement. For low-achievers (students with less than a 2.0 GPA)there is no ‘acting white’ effect, which is consistent with Fig. 1B. Forstudents with a 3.5 GPA or better, the ‘acting white’ coefficient is−.305, over three times the magnitude of the effect in the overallsample. This captures the divergence in social status among black andwhite high-achievers depicted in Fig. 1B. For Hispanics, the results aremore mixed. The smallest effect exists among low-achievers, but‘acting white’ is most salient among Hispanics with grade pointaverages between 3 and 3.5. Note: the majority of blacks andHispanics are in the 2.0 to 3.0 GPA category. For these students, racialdifferences between social status and academic achievement are smallfor blacks and relatively large for Hispanics.

5.1. Robustness checks and alternative empirical models

One of the biggest worries with our results thus far is that weimplicitly assume that higher grades cause lower popularity. It iscertainly plausible that high popularity causes lower grades through asimple time constraint or other mechanism. If causality runs in theopposite direction than we assume, linking racial differences in therelationship between social status and academic achievement to the‘acting white’ phenomenon is misleading.

The simplest and most direct way of addressing the potential forreverse causality would be to find a plausible instrument for gradesand re-estimate Eq. (1). We were unsuccessful at finding plausiblyexogenous variation in grades or any form of academic achievement.We were, however, mildly successful at finding a way to vary

18 We have also run all regressions with effort⁎race interactions. The results are verysimilar with slightly higher standard errors.19 The results to come are similar if one includes higher-order polynomials on gradesas regressors, though the interpretation of the coefficients is more convoluted.


popularity in ways that may not affect grades: a student's physicalattractiveness.

Recall, if we think that a regression of popularity on grades gives asignificant coefficient either because of simultaneity or omittedvariables, we can write it as a two-equation system. Then, if wehave a robust instrument for the second equation (where grades arethe dependent variable), it provides a valid estimate of the causaleffect of social status on grades (Wooldridge, 2002). If racialdifferences in that estimate are insignificant, and the sample is bigenough to have power, then it is a good piece of evidence that socialstatus does not cause grades.

Table 4 estimates a series of models, including Eq. (1), andadditional models of the form:

Gradesi;s;R = β0 + β1⋅SocialStatusi + γ⋅Xi + ηs + εi;s;R; ð2Þ

in column 1 and using interviewer ratings for attractiveness for eachstudent as an instrument for popularity in column 2. Each interviewerwas asked to rate the physical attractiveness of each student in one offive categories: very attractive (unattractive), attractive (unattrac-tive), and average. Columns 3 through 5 estimate a similar equationwhere social status is the dependent variable. Note: the subscript Rdenotes that each regression is run within each racial groupseparately.

Column 1 of Table 4 estimates Eq. (2) using weighted least squaresand our standard set of controls. The relationship between grades andsocial status is significantly more muted when grades are thedependent variable, and there are no racial differences betweenblack andwhites and small differences between Hispanics andwhites.Column 2 estimates the same equation instrumenting for social statuswith a student's physical attractiveness. The instrument has a strongfirst stage (F-statistic of roughly 40), and the second stage has similarcoefficients for blacks (.774 (.426)), Hispanics (1.480 (1.219)), andwhites (.692 (.165)). One clearly cannot distinguish between racialgroups in the magnitudes of the coefficients, which provides someevidence that reverse causality cannot explain our results.

Another important check is to ensure our results are robust toestimating a fully interacted model — allowing coefficients to vary byrace for every independent variable, not just academic achievement.Columns 3 through 7 estimate equations identical to Eq. (1), withinrace. Column 3 reports the results for all grade levels and columns 4through 7 provide estimates within GPA categories. In our simpleinteraction framework, we estimated racial differences in therelationship between social status and achievement to be −.098 onthe full sample and−.305 among students with a 3.5 GPA or higher. Inthe fully interacted model, the estimates are strikingly similar —




Table 3The prevalence of acting white, by grade point average.

Full sample b2.0 2.0–3.0 3.0–3.5 N3.5(1) (2) (3) (4)

Black .148 (.031)⁎⁎ .018 (.094) .122 (.113) .467 (.473) .921 (.423)⁎

Hispanic .182 (.030)⁎⁎ −.099 (.091) .180 (.119) .761 (.404) .271 (.371)Grades .127 (.007)⁎⁎ .109 (.041)⁎⁎ .148 (.033)⁎⁎ .211 (.086)⁎ .088 (.054)Black⁎grades −.098 (.011)⁎⁎ −.036 (.064) −.089 (.048) −.187 (.153) −.305 (.113)⁎⁎

Hispanic⁎grades −.165 (.011)⁎⁎ −.034 (.064) −.164 (.050)⁎⁎ −.335 (.130)⁎⁎ −.184 (.099)Effort −.039 (.006)⁎⁎ −.002 (.010) −.044 (.010)⁎⁎ −.065 (.013)⁎⁎ −.039 (.016)⁎

Male .027 (.009)⁎⁎ .022 (.016) .020 (.012) .055 (.017)⁎⁎ .020 (.020)Age −.012 (.003)⁎⁎ −.009 (.006) −.012 (.005)⁎ −.026 (.007)⁎⁎ .000 (.008)Cheerleading .232 (.018)⁎⁎ .173 (.049)⁎⁎ .178 (.027)⁎⁎ .243 (.034)⁎⁎ .299 (.036)⁎⁎

Athlete .234 (.008)⁎⁎ .123 (.017)⁎⁎ .207 (.012)⁎⁎ .239 (.017)⁎⁎ .318 (.019)⁎⁎

Student government .211 (.019)⁎⁎ −.032 (.057) .210 (.036)⁎⁎ .253 (.039)⁎⁎ .205 (.030)⁎⁎

Father househusband −.158 (.045)⁎⁎ −.039 (.058) −.120 (.087) −.149 (.085) −.267 (.098)⁎⁎

Mother housewife −.026 (.012)⁎ −.021 (.025) .008 (.020) −.039 (.025) −.041 (.027)Father education .042 (.014)⁎⁎ .017 (.030) .038 (.021) .037 (.027) .043 (.030)Father professional .065 (.013)⁎⁎ .064 (.030)⁎ .054 (.020)⁎⁎ .074 (.027)⁎⁎ .063 (.027)⁎

Mother education .031 (.013)⁎ .014 (.027) .041 (.019)⁎ .042 (.026) .001 (.027)Mother professional .045 (.012)⁎⁎ .007 (.024) .040 (.017)⁎ .015 (.025) .088 (.026)⁎⁎

Observations 58,305 8249 20,874 13,605 15,577R-squared .10 .10 .08 .09 .12Fraction black – 18.60 44.84 21.50 15.06Fraction Hispanic – 20.26 41.85 21.35 16.55

Note: All data are drawn from the National Longitudinal Study of Adolescent Health. The dependent variable is Social Status. Robust standard errors used. Standard errors aredenoted parenthetically under coefficients. Dummies for missing values for all variables except race and grades are included. All regressions contain school fixed effects. Weightsused in all regressions.⁎ Denotes significant at 5% level.⁎⁎ Denotes significant at 1% level.

Table 4Robustness checks and alternative empirical models.

Full-sample-OLS Full sample-IV Full-sample-OLS b2.0 2.0–3.0 3.0–3.5 N3.5

Dependent variable

Grades Grades Popularity Social status

Black .053 (.013)⁎⁎ .774 (.426)⁎ .040 (.100)⁎⁎ .091 (.050) .061 (.035) .148 (.132) −.217 (.097)⁎

White .053 (.004)⁎⁎ .692 (.165)⁎⁎ .113 (.007)⁎⁎ .091 (.041)⁎ .133 (.033)⁎⁎ .190 (.087)⁎ .067 (.055)Hispanic −.024 (.011)⁎ 1.480 (1.219) −.017 (.008)⁎ .056 (.048) .009 (.035) −.056 (.093) .026 (.061)

Note: All data are drawn from the National Longitudinal Study of Adolescent Health. Only Black, White and Hispanic respondents included. The dependent variable is Social Status.Each regression is limited only to the students with the GPA interval and race specified. Other covariates include effort, male, age, cheerleading, athlete, student government, fatherhousehusband, mother housewife, father education, father profession, mother education, and mother profession. See Appendix A for full description. Robust standard errors andschool fixed effects are included. Standard errors are denoted parenthetically under coefficients. Dummies for missing values for all variables except grades. Weights used in allregressions.⁎ Denotes significance at the 5% level.⁎⁎ Denotes significance at the 1% level.

20 Plotting race-specific social status and grades by race and gender reveals thatsocial status decreases for black males earlier (3.25 GPA relative to a 3.5 GPA) and thegradient for high-achievers is steeper.21 We have also investigated whether ‘acting outlier’ is the underlying problem andnot ‘acting white’ as such, by estimating the relationship between social status andgrades among whites in schools with low average test scores. Whites in these schoolscontinue to have a positive relationship between social status and grades.


ARTICLE IN PRESS

−.109 and −.272. The results are also consistent with the patternspresented in Fig. 1B.

5.2. Sensitivity analysis

Tables 5 and 6 explore the sensitivity of our results across a widevariety of sub-samples of the data and alternative specifications. Weestimate separate regressions for each indicated sub-sample of thedata. Table 5 estimates regressions on the full sample of students;Table 6 restricts the sample to students with a 3.5 GPA or better(hereafter referred to as “high-achievers”). In most cases thequalitative results are identical, though the magnitude of the ‘actingwhite’ coefficient is three to four times as large among high-achievers.We report only coefficients on grades and the race-grades interaction(and their associated standard errors) for blacks and Hispanics. Thetop row of the tables present the baseline results from columns 1 and5 in Table 3, respectively.

As one peruses the tables, it becomes evident that the ‘actingwhite’ phenomenon is robust across most subsets of the data, thoughthere are several notable exceptions. ‘Acting white’ is modestly moreprevalent among black males relative to black females, and this is


especially true among high-achievers.20 Whether or not one uses thesample weights provided in the Add Health data matters little. ‘Actingwhite’ is mainly in public schools and non-existent among blacks inprivate schools. This latter finding may partially explain why blackstudents in private schools appear to do especially well (Akerlof andKranton, 2002; Neal, 1997; and Grogger and Neal, 2000).21

5.2.1. Alternative measures of social statusThe ‘acting white’ coefficient is also robust to different measures of

social status. We include the following three alternative measures ofsocial status: (1) a non-normalized version of our index; (2) same-race friends — measured as how many individuals j put i down as afriend; (3) all-race social status; and (4) other-race social status.Unless otherwise noted, these measures have been normalized such




Table 5Sensitivity analysis and extensions of the basic model, full sample.

Grades Grades⁎black Grades⁎Hispanic Grades Grades⁎black Grades⁎Hispanic Grades Grades⁎black Grades⁎Hispanic Grades Grades⁎black Grades⁎Hispanic

Baseline Weighted Unweighted.131 (.007)⁎⁎ −.100 (.011)⁎⁎ −.165 (.011)⁎⁎ .137 (.007)⁎⁎ −.091 (.010)⁎⁎ −.172 (.011)⁎⁎

Otherpopularitymeasures

Non-normalized spectral popularity Same-race friends All-race popularity Other-race popularity2.010 (.107)⁎⁎ −1.533 (.175)⁎⁎ −2.541 (.164)⁎⁎ .587(.035)⁎⁎ −.439 (.066)⁎⁎ −.730 (.058)⁎⁎ .516(.038)⁎⁎ −.273 (.068)⁎⁎ −.449 (.064)⁎⁎ .006 (.006) .037 (.015)⁎⁎ .060 (.026)⁎

Alternativemeasures ofachievement

Math and Science grades Academic clubs Peabody vocabulary test−.086 (.006)⁎⁎ .114 (.010)⁎⁎ .065 (.010)⁎⁎ −.064 (.022)⁎ −.132 (.021)⁎⁎ .010 (.015) −.003 (.023) −.039 (.021)

Grade level Seventh grade Ninth grade Twelfth grade.082 (.013)⁎⁎ −.105 (.026)⁎⁎ −.083 (.024)⁎⁎ .196 (.017)⁎⁎ −.140 (.026)⁎⁎ −.244 (.023)⁎⁎ .101 (.016)⁎⁎ −.065 (.032)⁎ −.150 (.027)⁎⁎

School type Private Public.014 (.033) .051 (.048) −.038 (.053) .138 (.007)⁎⁎ −.108 (.012)⁎⁎ −.175 (.011)⁎⁎

Gender Male Female.129 (.010)⁎⁎ −.099 (.017)⁎⁎ −.176 (.015)⁎⁎ .137 (.010)⁎⁎ −.111 (.016)⁎⁎ −.164 (.015)⁎⁎

Urbanicity Suburban Urban/rural.139 (.009)⁎⁎ −.102 (.016)⁎⁎ −.164 (.015)⁎⁎ .115 (.010)⁎⁎ −.093 (.016)⁎⁎ −.152 (.016)⁎⁎

Parental college Both-college educated Other.055 (.025)⁎ −.029 (.039) −.114 (.039)⁎⁎ .132 (.007)⁎⁎ −.098 (.012)⁎⁎ −.158 (.011)⁎⁎

Extracurricularactivity

Athletes Cheerleading Music/band.162 (.011)⁎⁎ −.131 (.017)⁎⁎ −.192 (.017)⁎⁎ .229 (.033)⁎⁎ −.233 (.045)⁎⁎ −.245 (.043)⁎⁎ .160 (.016)⁎⁎ −.086 (.025)⁎⁎ −.189 (.024)⁎⁎

Co-curricularactivity

Student government Academic clubs National Honor Society.114 (.034)⁎⁎ −.070 (.050) −.184 (.048)⁎⁎ .120 (.016)⁎⁎ −.093 (.024)⁎⁎ −.148 (.022)⁎⁎ .075 (.041) −.019 (.069) −.048 (.058)

Level ofinterracialcontact

High segregation Low segregation b20% Black N80% Black.092 (.008)⁎⁎ −.055 (.013)⁎⁎ −.114 (.013)⁎⁎ .153 (.010)⁎⁎ −.136 (.023)⁎⁎ −.208 (.017)⁎⁎ .134 (.008)⁎⁎ −.150 (.015)⁎⁎ −.174 (.013)⁎⁎ −.241 (.091)⁎⁎ .352 (.097)⁎⁎ .271 (.098)⁎⁎

N50% Black N50% Black b50% White N50% White.132 (.007)⁎⁎ −.130 (.012)⁎⁎ −.172 (.011)⁎⁎ −.012 (.033) .064 (.036) .039 (.035) .055 (.012)⁎⁎ −.011 (.017) −.079 (.016)⁎⁎ .136 (.008)⁎⁎ −.155 (.015)⁎⁎ −.174 (.012)⁎⁎

By gender High segregation — males High segregation — females Low segregation — males Low segregation — females.077 (.012)⁎⁎ −.038 (.019) −.106 (.018)⁎⁎ .109 (.011)⁎⁎ −.079 (.018)⁎⁎ −.132 (.018)⁎⁎ .163 (.014)⁎⁎ −.157 (.033)⁎⁎ −.233 (.025)⁎⁎ .149 (.015)⁎⁎ −.130 (.032)⁎⁎ −.188 (.022)⁎⁎

Note: All data are drawn from the National Longitudinal Study of Adolescent Health. Separate regressions are estimated for each indicated subsample. The dependent variable is Social Status, except for the second panel. Regressions run onblacks, whites, and Hispanics as in previous tables. Fixed-effects specification from previous table is used. Robust standard errors used. Standard errors under coefficients. All regressions weighted except for unweighted line. In schools withgreater than 50% black population, 10.4% of the students are Hispanic; in schools that are more than 80% black, 6.7% of students are Hispanic.⁎ Denotes significant at 5% level.

⁎⁎ Denotes significant at 1% level.

9R.G

.FryerJr.,P.Torelli

/Journal

ofPublic

Economics

xxx(2010)

xxx–xxx

ARTICLE

INPRESS

Pleasecite

thisarticle

as:Fryer,R.G

.,Torelli,P.,Anem

piricalanalysis

of‘acting

white

’,Journalof

PublicEconom

ics(2010),doi:10.1016/j.

jpubeco.2009.10.011



Table 6Sensitivity analysis and extensions of the basic model, high achievers (3.5 grade point average or better).

Grades Grades⁎black Grades⁎Hispanic Grades Grades⁎black Grades⁎Hispanic Grades Grades⁎black Grades⁎Hispanic Grades Grades⁎black Grades⁎Hispanic

Baseline Weighted Unweighted.095 (.054) −.329 (.112)⁎⁎ −.191 (.096)* .061 (.055) −.36 5 (.105)⁎⁎ −.23 5 (.092)⁎

Other popularitymeasure

Non-normalized spectral popularity Same-race friends All-race popularity Other-race popularity1.452 (.830) −5.059 (1.716)⁎⁎ −2.940 (1.476)⁎ .275 (.211) −2.165 (.582)⁎⁎ −1.568 (.521)⁎⁎ .307 (.228) −1.790 (.599)⁎⁎ −2.556 (.545)⁎⁎ −.011 (.037) −.045 (.134) −.680 (.228)⁎⁎

Alternativemeasuresof achievemen

Math and Science grades Academic clubs Peabody vocabulary test.026 (.035) .121 (.069) −.056 (.055) −.030 (.025) −.058 (.047) −.116 (.041)⁎⁎ −.056 (.033) .136 (.052)⁎⁎ −.061 (.042)

Grade level Seventh grade Ninth grade Twelfth grade−.047 (.079) .203 (.230) −.042 (.162) .320 (.137)⁎ −.543 (.280) −.177 (.225) .076 (.118) −.131 (.274) −.563 (.204)⁎⁎

School type Private Public.082 (.183) .465 (.320) −.292 (.378) .095 (.057) −.375 (.117)⁎⁎ −.165 (.101)

Gender Male Female.138 (.084) −.513 (.170)⁎⁎ −.245 (.137) .071 (.071) −.178 (.146) −.182 (.130)

Urbanicity Suburban Urban/rural.121 (.074) −.404 (.164)⁎ −.268 (.156) .049 (.074) −.212 (.142) −.075 (.124)

Parental college Both-college educated Other.109 (.109) −.549 (.250)⁎ −.142 (.223) .075 (.062) −.269 (.124)⁎ −.143 (.102)

Extracurricularactivity

Athletes Cheerleading Music/band.109 (.076) −.525 (.157)⁎⁎ −.353 (.132)⁎⁎ .008 (.204) .178 (.345) −.107 (.334) .189 (.098) −.371 (.222) −.352 (.169)⁎

Co-curricularactivity

Student government Academic clubs National Honor Society−.157 (.148) .016 (.360) .237 (.272) .034 (.091) −.223 (.172) −.271 (.154) −.099 (.111) .032 (.279) −.118 (.239)

Level ofinterracialcontact

High segregation Low segregation b20% Black N80% Black−.011 (.066) −.144 (.135) −.083 (.122) .158 (.078)⁎ −.536 (.195)⁎⁎ −.260 (.122)⁎ .123 (.060)⁎ −.112 (.139) −.202 (.114) −.510 (.497) .310 (.644) .349 (.734)b50% Black N50% Black b50% White N50% White.093 (.054) −.260 (.129)⁎ −.164 (.100) −.297 (.320) .009 (.370) .001 (.356) .074 (.093) −.403 (.154)⁎⁎ −.150 (.126) .089 (.060) −.157 (.140) −.256 (.122)⁎

By gender High segregation — males High segregation — females Low segregation — males Low segregation — females.008 (.096) −.236 (.212) −.093 (.189) −.033 (.091) −.032 (.179) −.112 (.156) .218 (.122) −.845 (.267) −.313 (.184) .135 (.102) −.353 (.261) −.311 (.162)

Note: All data are drawn from the National Longitudinal Study of Adolescent Health. Separate regressions are estimated for each indicated subsample. The dependent variable is Social Status, except for the second panel. Regressions run onblacks, whites, and Hispanics as in previous tables. Fixed-effects specification from previous table is used. Robust standard errors used. Standard errors under coefficients. All regressions weighted except for unweighted line.⁎ Denotes significant at 5% level.

⁎⁎ Denotes significant at 1% level.

10R.G

.FryerJr.,P.Torelli

/Journal

ofPublic

Economics

xxx(2010)

xxx–xxx

ARTICLE

INPRESS

Pleasecite

thisarticle

as:Fryer,R.G

.,Torelli,P.,Anem

piricalanalysis

of‘acting

white

’,Journalof

PublicEconom

ics(2010),doi:10.1016/j.

jpubeco.2009.10.011



25 Thus for each school, there are 12 categories of students (three race categoriestimes four GPA categories) and we count how many students there are in eachcategory.


ARTICLE IN PRESS

that they have a mean of zero and a standard deviation of one. Thesign and magnitude of the ‘acting white’ coefficient for blacks andHispanics is robust across the first three measures. The positivecoefficient on other-race social status suggests that blacks andHispanics substitute towards other-race friends in the presence of‘acting white’. The magnitude of the coefficient, however, issignificantly smaller so that, on net, ‘acting white’ (race effect minusthe substitution effect) still dominates.22 Blacks and Hispanics with aGPA above 3.5 are actually losing other-race friends, exacerbating theeffect of ‘acting white’ among this select group. This result issurprising, as one might expect high-achieving minority students tofind refuge among high-achieving whites.23

We also investigated the number of fights that a student had beeninvolved in as a measure of social status (not shown in tabular form).There were marked differences in the relationship between gradesand fights, with black students much more likely to be involved infights as their grades increased. Hispanics followed a similar pattern.

5.2.2. Alternative measures of achievementOur results are also robust to different notions of academic

achievement. We include three additional measures of achievement:math and science grades, participation in academic clubs such asmathclub, book club, foreign language clubs and so on, and Peabody PictureVocabulary Test scores. In all cases, themagnitude of the ‘actingwhite’coefficient is less than our baseline specifications, though math andscience grades and academic clubs are both negative and statisticallysignificant.

Test scores, in lieu of grades, provide different results. Thecoefficient on the race-achievement interaction is −.003 (.023) forblacks and −.04 (.021) for Hispanics. This is surprising, but quiteconsistent with the intuition behind ‘acting white,’ assuming that testscores are less observable by one's peer group than grades. Inparticular, grades are likely more observable to peers than test scores.In an environment where ‘acting white’ exists, one would expect tosee a steeper (negative) relationship between grades and socialstatus. The correlation between test scores and grades is .27.

5.2.3. By grade levelTheoretically, one could imagine that the prevalence of ‘acting

white’ could change as students are maturing and developing theiridentities (Cross and Fhagen-Smith, 2001). We estimate our mainfixed-effects equation separately for each grade level from 7 to 12,though we only report ‘acting white’ coefficients for seventh, ninth,and twelfth grades. There are no discernible patterns by grade level.

5.2.4. Accounting for the supply of high-achieving black studentsOne possible reason that blacks face a steeper social status-grades

gradient is a supply shortage of high-achieving black students. Ifstudents tend to associate with other students with similar back-grounds (in terms of race, grades, neighborhood location, etc), thenone might observe a steep trajectory for blacks at higher GPA levelssimply due to the fact that there are few blacks in their classes.Sociologists have argued for some time that this is a likely reason forthe alienation of high-achieving black students. If so, this potentiallyprovides a different qualitative understanding of the ‘acting white’phenomenon.24

We account for the number of same-race students with similargrades in each school by dividing the GPA distribution into four

22 The bulk of the substitution effect is occurring among students with a GPAbetween 3.0 and 3.5, especially among blacks. Including controls for number of“potential friends” (i.e. number of same-race students in GPA category by school) doeslittle to change this.23 We are by no means implying that high-achieving whites are in any way refusingto befriend minorities; only reporting a correlation in the data.24 The supply of students with particular grades is itself an endogenous outcome of‘acting white,’ thus these results should be interpreted with some care.


categories: less than 2.0, 2.0–3.0, 3.0–3.5, and greater than 3.5. Withineach grade category, we count the number of students by raceseparately for each school, and we include this variable as anindependent regressor in our basic model, linearly.25 Accounting forthe supply of black high-achievers has a modest effect on the ‘actingwhite’ coefficient, reducing it from −.103 to −.081, and virtually noeffect on other covariates (not shown in tabular form).26

5.2.5. Student activitiesThere are many opportunities in schools for students to self-select

into particular activities, ensuring that most of their peer interactionsare with other students with similar interests. These includeorganized sports and cheerleading, various clubs (languages, mathor science, drama, or debate), band and music, or the National HonorSociety. These activities might differ in their affirmation of highachievement; the honor society likely affirms achievement in a waythat organized sports do not. Yet, across this diverse set of potentialstudent activities, only one eliminates the racial difference in therelationship between social status and achievement: the NationalHonor Society. In all other activities, there remains a substantial gap,suggesting that self-selection into extra-curricular activities does noteliminate the effects of high grades among blacks and Hispanics. Thus,the hypothesis by some that high-achieving blacks take up extracur-ricular activities to deflect hostility (e.g., Farkas et al., 2002) is notborne out in the data.

5.2.6. Interracial contactThe most striking aspect of Tables 5 and 6 is how the coefficient on

‘acting white’ varies in arenas with more or less interracial contact.We partition the sample of schools in several ways: high versus lowsegregation, schools where blacks and whites are the simple majorityand schools that are 80% black versus those that are 20% black.Remarkably, schools that are less than 20% black have the largest‘acting white’ effect for blacks and Hispanics.27

It is difficult to discern the true effect in schools that are greater than80% black. In schools that are 80% black or greater, the coefficient ongrades is large and negative and the coefficient on black interactedwithgrades is large, positive, and imprecisely measured. A joint significancetest confirms that one cannot distinguish between the coefficients ongrades and black⁎grades in the full sample and schools that are greaterthan 80% black. We can be confident, however, that the ‘acting white’coefficients are statistically different between schools that are less than20% black and those that are greater than 80%.

Results in Tables 5 and 6 are also contrasted between high- andlow-segregated schools. The amount of segregation in any school A ismeasured as:

SegregationA =ExpectedðCrossRaceFriendsÞ−ObservedðCrossRaceFriendsÞ

ExpectedðCrossRaceFriendsÞ ;

ð3Þ

where cross-race friends refers to the total number of friends fromeach race with students from different races.28 The expected numberof cross-race friends is the sum of the expected value for each racecombination, omitting friends of the same-race. The expected number

26 We have also partitioned the number of potential friends into twenty equally sizedbins and included them as dummy variables. Using this approach modestly decreasesthe ‘acting white’ effect.27 This is true even after controlling for the number of blacks within a school orcontrolling for the number of blacks with similar grades.28 This index is a modification of Freeman (1978) and related to Echenique and Fryer(2005). For a complete derivation, see the Add Health's School Network VariablesCodebook available at: http://www.cpc.unc.edu/projects/addhealth/codebooks/wave1.


http://www.cpc.unc.edu/projects/addhealth/codebooks/wave1

http://www.cpc.unc.edu/projects/addhealth/codebooks/wave1



Fig. 4. A: Spectral popularity and grades by race, low segregation. B: Spectral popularity and grades by race, high segregation.


ARTICLE IN PRESS

of friends between race i and race j is equal to the total number offriends of race imultiplied by the probability that an individual of racei is friends with a student of race j.

The segregation index ranges from−1 (pure out-race preference)to 1 (total segregation). A value of 0 indicates that there is no grouppreference: friends are random with respect to race. The measure ofsegregation was calculated for each school and divided the set intohigh- and low-segregation school; cutting at the school median.

Surprisingly, blacks in more segregated schools incur less of atradeoff between social status and achievement. The coefficient on the‘acting white’ term is twice as large in schools that are above themedian in terms of segregation (−.055 compared to−.136).29 Amonghigh-achievers, the differences are starker (−.144 and −.536). Inaddition, Fig. 4A and B which estimate (non-parametrically) the socialstatus-grades relationship in high- and low-segregation schools,accounting for our core set of controls, provides more evidenceattesting to the fact that ‘acting white’ is particularly salient amonghigh-achievers and those in schools with more interracial contact.

Using the Moving to Opportunity (MTO) experiment, whichassigned housing vouchers via random lottery to public housingresidents in five large cities, Kling and Liebman (2004) and Kling et al.(2005) provide evidence of gender differences in the effect of thetreatment on a variety of outcomes for youth. Females exhibit lowerarrests rates, improvements in education and mental health, and areless likely to engage in risky behaviors. Males, on the other hand, weremore likely to engage in risky behaviors, had no decrease in arrestsrates, and experiencedmore physical health problems (e.g., injuries oraccidents). These results suggest that males and females may respondto their environments in distinct ways; ‘acting white’may well be theunderlying cause.

One way to test this general hypothesis with the current data is toexamine gender/race differences in the relationship between socialstatus and grades in schools with more or less interracial contact.Results are presented in the bottom row of Tables 5 and 6. Consistentwith the evidence from MTO, ‘acting white’ is twice as large for high-achievingmales relative to high-achieving females in low-segregationschools. The coefficient on black⁎grades for males in low-segregationschools is seven times as large as the identical coefficient estimated onthe full sample. The fact that ‘acting white’ is more prevalent inschools withmore interracial contact is surprising, but consistent withgrowing evidence that there can be significant pressure in raciallyheterogeneous schools to toe the racial line (Tatum, 1997).

29 This result is robust to accounting for the potential supply shortage of blackstudents in very integrated schools.


6. Reconciling our results with previous work on ‘acting white’

In this section, we will attempt to reconcile our results with Cookand Ludwig (1997) and Ainsworth-Darnell and Downey (1998)— theonly two empirical papers on ‘acting white’ which use nationallyrepresentative data — both of which report very different findingsthan those presented here.

An important potential difference between our analysis and theprevious work is that it relies on measures of popularity and academicachievement that have less variation. The NELS contains a questionthat asks if the student “thinks others see him/her as popular.” Theanswer choices are: ‘not at all,’ ‘somewhat,’ or ‘very’; over 80% of therespondents categorized themselves as ‘somewhat’ or ‘very’ popular.Cook and Ludwig (1997) also use two dichotomous achievementmeasures: (1) whether the student earns “mostly As in math”; andwhether the student is in the honor society. Another possibility is thatAdd Health and NELS are just different data sets with differentsampling protocols and yield different conclusions.

To understand whether the differences in our results are due todifferent data or better measures, we construct measures of socialstatus and academic achievement in the Add Health that are as similarto the NELS variables as we could make them and estimate Eq. (1)with our standard set of controls in each dataset. If this approachyields different results for each data set, we will conclude that there issomething different about the individuals in the AddHealth relative tothe NELS and we will further investigate these differences. If we findsimilar results across the datasets, we will know that differentmeasures are the culprit.

Table 7 estimates the prevalence of ‘acting white’ among eighth-and tenth-grade students in the NELS. The equation estimated isidentical to that implemented in the Add Health data, described byEq. (1). Social status is measured as a dichotomous variable; equal toone if students in class see the respondent as very popular and equalto zero if not.30 The independent variables vary by column and aregenerally increasing from left to right. Grades are grade point averagescalculated from each student's transcript.

Columns 1 and 5 of Table 7 show that there is a positiverelationship between grades and popularity, though the relationshipis smaller for blacks, which is consistent with ‘acting white.’ The‘acting white’ coefficient is large and statistically significant for blacksin both eighth and tenth grades, but never statistically significant for

30 In the raw data, the measure takes on three values, whether or not students seethe respondent as: (1) very popular; (2) somewhat popular; or (3) not popular at all.Only 15% of the tenth-grade sample (and 18% of the eighth-grade sample) reportedbeing not popular at all — thus we merged responses (2) and (3). See Appendix A fordetails.




Table 7The relationship between social status and achievement, NELS.

8th grade 10th grade

(1) (2) (3) (4) (5) (6) (7) (8)

Black .261 (.040)⁎⁎ .214 (.040)⁎⁎ .206 (.040)⁎⁎ .210 (.042)⁎⁎ .144 (.058)⁎ .143 (.064)⁎ .158 (.064)⁎ .189 (.047)⁎⁎

Hispanic .086 (.035)⁎ .075 (.035)⁎ .084 (.036)⁎ .094 (.037)⁎ .061 (.054) .073 (.054) .090 (.055) .112 (.046)⁎

Grades .034 (.004)⁎⁎ .037 (.006)⁎⁎ .037 (.006)⁎⁎ .037 (.006)⁎⁎ .013 (.006)⁎ .030 (.008)⁎⁎ .030 (.008)⁎⁎ .037 (.008)⁎⁎

Black⁎grades −.054 (.014)⁎⁎ −.047 (.014)⁎⁎ −.047 (.014)⁎⁎ −.042 (.015)⁎⁎ −.037 (.021) −.046 (.023)⁎ −.047 (.023)⁎ −.058 (.017)⁎⁎

Hispanic⁎grades −.023 (.012) −.022 (.012) −.021 (.012) −.018 (.013) −.022 (.018) −.028 (.018) −.025 (.017) −.033 (.016)⁎

Current test score −.004 (.000)⁎⁎ −.004 (.000)⁎⁎ −.004 (.000)⁎⁎ −.004 (.001)⁎⁎ −.004 (.001)⁎⁎ −.004 (.001)⁎⁎

Ability group .061 (.008)⁎⁎ .062 (.008)⁎⁎ .061 (.008)⁎⁎ .036 (.010)⁎⁎ .038 (.010)⁎⁎ .032 (.010)⁎⁎

Effort measure −.078 (.012)⁎⁎ −.080 (.012)⁎⁎ −.086 (.012)⁎⁎ −.120 (.023)⁎⁎ −.123 (.023)⁎⁎ −.125 (.021)⁎⁎

SESb25th percentile −.030 (.007)⁎⁎ −.028 (.007)⁎⁎ −.038 (.008)⁎⁎ −.024 (.013) −.023 (.013) −.031 (.009)⁎⁎

SESN75th percentile .023 (.007)⁎⁎ .020 (.008)⁎⁎ .028 (.008)⁎⁎ .012 (.011) .010 (.011) .017 (.010)Male .027 (.006)⁎⁎ .027 (.006)⁎⁎ .027 (.006)⁎⁎ .006 (.009) .007 (.009) .012 (.007)Athlete .080 (.006)⁎⁎ .079 (.006)⁎⁎ .081 (.006)⁎⁎ .058 (.009)⁎⁎ .056 (.009)⁎⁎ .063 (.008)⁎⁎

Student government .139 (.012)⁎⁎ .139 (.012)⁎⁎ .145 (.012)⁎⁎ .154 (.019)⁎⁎ .152 (.019)⁎⁎ .166 (.017)⁎⁎

Cheerleading .084 (.012)⁎⁎ .082 (.012)⁎⁎ .086 (.012)⁎⁎ .080 (.024)⁎⁎ .079 (.024)⁎⁎ .091 (.018)⁎⁎

Urban school .003 (.009) −.016 (.015)Suburban school −.005 (.007) −.021 (.009)⁎

% Black in school (0–100) .000 (.000) .000 (.000)% Asian in school (0–100) .001 (.001) −.001 (.001)⁎

% Hispanic in school (0–100) .000 (.000) −.001 (.000)⁎

%American Indian in school (0–100) .000 (.000) .000 (.001)Private school .034 (.009)⁎⁎ .036 (.021)School fixed effects? N N N Y N N N YObservations 20,766 20,766 20,766 20,766 13,706 13,706 13,706 13,706R-squared .01 .06 .07 .13 .00 .05 .05 .18

Note: All data are drawn from the National Educational Longitudinal Study. The dependent variable is dichotomous version of self-reported popularity (see Appendix A for details).Robust standard errors used. Standard errors under coefficients. Dummies for missing values for all variables except race and grades. Regression only on Blacks, Whites, andHispanics. Weights used in all regressions. The fraction of 8th and 10th graders who “see themselves as very popular,” is 17.5% and 13.5% respectively.⁎ Denotes significant at 5% level.⁎⁎ Denotes significant at 1% level.


ARTICLE IN PRESS

Hispanics. Columns 2 and 6 include controls for test score, effort, SES,gender, and extracurricular activities (athletics, student government,and cheerleading). The coefficient on grades and the ‘acting white’coefficient change little, suggesting that black–white differences incovariates are not driving the negative ‘acting white’ coefficient(similar to the results in the Add Health).31

Columns 3 and 7 include controls for school characteristics. Privateschool attendance is associated with greater reported social status,but the inclusion of school characteristics does not alter the ‘actingwhite’ coefficient. Columns 4 and 8 present results with the inclusionof school fixed-effects. These results are consistent with our previousanalysis using the Add Health, though the coefficients are smaller.There is no statistical difference between Hispanics and whites in therelationship between popularity and grades in the NELS. Othercoefficients stay essentially the same.

These results differ substantially from those published in Cook andLudwig (1997) and Ainsworth-Darnell and Downey (1998). Theestimated racial differences in the relationship between social statusand academic achievement are smaller than Add Health, but quitedifferent (i.e., the opposite sign!) from the original analysis in the NELS.

Wealso runa similar analysis usingAddHealth data. Tobe consistentwith thepopularitymeasures used in theNELS,we construct two similarmeasures in Add Health: (a) whether the respondent feels accepted bypeople at the school and (b) whether the respondent feels close topeople at this school. These measures are ranked 1 to 5. A value of 1implies the respondent strongly disagrees and 5 implies they stronglyagree. Note: we reversed the coding from the survey so that thedirection is consistent with the NELS analysis — higher numbers implymore popular. To further match the NELS analysis, we created twodichotomous achievement metrics: (a) a student earned an A in math(as opposed to a B or C) and (b) the student is in the nationally honorsociety.We also limit the sample to tenth graders tomatch their sample.

31 Including the interactions Black⁎Test Score, Black⁎Effort, or Black⁎SES does littleto alter the results.


Table 8 demonstrates that there are no racial differences in therelationship between social status and academic achievement in theAdd Health using measures that closely resemble data available in theNELS, though large standard errors make definitive conclusionsdifficult. Columns 1 through 4 estimate our standard specificationwith the index of social status as the dependent variable and “As inMath” as the achievement metric in columns 1 and 2 and NationalHonor Society in columns 3 and 4. Odd-numbered columns containidentical controls to Table 3, including an exhaustive set of racedummies, and even-numbered columns add school fixed effects tothis set of controls. We only show race coefficients for blacks andHispanics. White is the omitted category.

Making A’s in math is associated with lower social status on ourindex of social status, but is not appreciably different between blacksand whites. For Hispanics, the differences are statistically significantand substantively large. Across the three measures of popularity inTable 8, there are no significant differences between blacks andwhites. On two out of the four popularity measures, Hispanics havesignificant differences.

These results, coupled with Table 7, suggest that an importantreason for the differences between the previous literature and thecurrent analysis is how one defines academic achievement. Dichot-omous measures mask the variation needed to estimate racialdifferences in the relationship between social status and academicachievement — at least for Blacks. We were unsuccessful atreconciling the results in both data sets for Hispanics.

7. Interpreting the data through the lens of social theory

A number of stylized facts emerge from the analysis of thepreceding sections. There are racial differences in the relationshipbetween social status and grades. These differences are roughly threetimes as large for black students with a 3.5 grade point average orhigher. ‘Acting white’ is more salient in public schools and amongchildren from low-education families. Accounting for self-selectioninto extra-curricular activities or the supply of black high-achievers




Table 8Understanding the differences between NELS and Add Health.

Variables Social status index Feels accepted at school Feel close to school

As in Math −.3220(.124)⁎⁎⁎

−.3640(.121)⁎⁎⁎

.0250(.033)

.0162(.034)

.0102(.035)

.0308(.036)

Honor society .6960(.172)⁎⁎⁎

.6760(.165)⁎⁎⁎

.0194(.034)

.0187(.036)

.0388(.039)

.0409(.041)

Hispanic .5730(.126)⁎⁎⁎

.9830(.145)⁎⁎⁎

−.1420(.084)⁎

.5310(.099)⁎⁎⁎

.0991(.029)⁎⁎⁎

.0924(.043)⁎⁎

.0676(.021)⁎⁎⁎

.0341(.029)

−.0190(.048)

−.1920(.056)⁎⁎⁎

−.1030(.036)⁎⁎⁎

−.2220(.040)⁎⁎⁎

Hispanic⁎Achievement −.6220(.203)⁎⁎⁎

−.5180(.215)⁎⁎

.2020(.668)

−.0729(.688)

−.0179(.050)

−.0583(.059)

.1640(.185)

.1700(.183)

−.2370(.098)⁎⁎

−.1770(.086)⁎⁎

.4270(.195)⁎⁎

.3690(.203)⁎

Black −.4320(.118)⁎⁎⁎

−.5730(.133)⁎⁎⁎

−.5950(.084)⁎⁎⁎

−.5030(.097)⁎⁎⁎

.1780(.031)⁎⁎⁎

.0978(.038)⁎⁎⁎

.1610(.024)⁎⁎⁎

.1060(.029)⁎⁎⁎

−.1780(.036)⁎⁎⁎

−.1430(.043)⁎⁎⁎

−.2450(.027)⁎⁎⁎

−.1970(.033)⁎⁎⁎

Black⁎Achievement −.0507(.212)

.0825(.208)

.4620(.347)

.1090(.338)

−.1170(.065)⁎

−.0975(.067)

.0442(.078)

−.0031(.081)

−.0744(.073)

−.1210(.074)

−.0307(.090)

−.0451(.091)

School fixed effects N Y N Y N Y N Y N Y N YObservations 10,375 10,375 15,660 15,660 9468 9468 14,287 14,287 9752 9752 14,540 14,540R-squared .125 .242 .098 .222 .036 .077 .031 .057 .057 .132 .034 .08

Note: All data are drawn from the National Longitudinal Study of Adolescent Health. Each column represents a separate regression. All regressions are weighted and include the fullset of controls in Table 3. An exhaustive set of Race dummies are included, white is the omitted category. bRaceN⁎Achievement uses the appropriate achievement measure (mostlyAs or honor society). Robust standard errors are below each coefficient.

⁎ Denotes significance at the 10% level.⁎⁎ Denotes significance at the 5% level.⁎⁎⁎ Denotes significance at the 1% level.


ARTICLE IN PRESS

does little to undermine the effect. Environments with moreinterracial contact are associated with a steeper social status-gradesgradient among blacks; seven times as large for high-achieving blackmales. The results are not an artifact of our particular measure ofsocial status and they hold in the two nationally representativesamples in which one can construct measures of popularity whenmore continuous measures of academic achievement are used.

In this section, we consider the extent to which existing stories cansuccessfully account for this set of facts. It is important to note that thefirst two theories are our attempt to put more structure on what wereessentially verbal accounts of the phenomenon.

7.1. An oppositional culture identity model

The most prominent theory to explain the stylized facts put forth isthe oppositional culture hypothesis, developed in Fordham and Ogbu(1986) and Fordham (1996).32 The hypothesis states that the observeddisparity between blacks and whites stems from the following factors:(1) white people provide them with inferior schooling and treat themdifferently in school; (2) by imposing a job ceiling, white people fail toreward them adequately for their academic achievement in adult life;and (3) black Americans develop coping devices which, in turn, furtherlimit their striving for academic success. Fordham and Ogbu (1986)suggest the problem arose partly becausewhite Americans traditionallyrefused to acknowledge that black Americans were capable ofintellectual achievement, and partly because black Americans subse-quently began to doubt their own intellectual ability, began to defineacademic success as white people's prerogative, and began todiscourage their peers, perhaps unconsciously, from emulating whitepeople in striving for academic success.

The Fordham and Ogbu hypothesis can be directly imputed into asimple economic model of identity and human capital acquisition,according to Akerlof and Kranton (2000, 2002). In their language,identities are accompanied by certain “prescriptions” that defineappropriate behaviors for a person of that type. When an individualmakes decisions in line with these prescriptions, there is a utilitybenefit. In standard models, a student invests in human capital untilthe marginal cost of investment equals the marginal benefit of thatinvestment. Fordham and Ogbu (1986) argue that institutionaldiscrimination lowers the marginal benefit of investment for certain

32 Since then, efforts have been focused on refuting Fordham and Ogbu's hypothesis.


minority groups. As a rational response, these minorities began toequate educational achievement with whiteness, thereby altering theprescriptions (e.g., what it means to “be black”).

The predictions of the oppositional culture identity model facemixed success in terms of the patterns observed in the data. Consistentwith the theory, there are no racial differences in the relationshipbetween social status and achievement among blacks in private schools,as these environments likely change identity prescriptions.

The theory, however, does quite poorly in explaining why ‘actingwhite’ is more salient in schools that are less than 20% black relative tothose that are greater than 80% black. In fact, Fordham and Ogbu(1986) base the oppositional culture hypothesis on behavior observedin an all-black high school in Washington, DC.

7.2. A sabotage model

Another explanation for racial differences in the relationshipbetween social status and achievement is that blacks simply sabotagetheir high-achieving peers.33 McWhorter (2000) is a prominentadvocate of this view.

Consider the following skeletal outline of a sabotagemodel. Imagineaworldwith two neighborhoods: amajority community and aminoritycommunity, each containingmany individuals. Individuals come in twoflavors: high-ability and low-ability, which are determined by Natureand publicly observed within a community. There are two states of theworld: discriminatory and fair, which are unknown. Individuals observetheir ability type and make a dichotomous human capital decision.Assume that human capital is less costly to obtain for high-abilityworkers. Individual utility is partly determined bywhat others (outsidethe community) perceive about their ability.

Firms observe each individual's level of human capital and decidewhether or not to hire them. When the state of the world isdiscriminatory, firms refuse to hire minority workers. In a fair state,workers are hired if and only if they have invested in human capital.Assume that payoffs are such that low-ability types never haveincentive to acquire human capital.

In this framework, low-ability agents have incentive to “hold back”individualswith high abilitywhen the cost of doing so is less than the netbenefit, which will hold when the state of the world has a high enough

33 There are several mechanisms that could lead to such behavior. The mostsimplistic is that blacks and Hispanics punish, rather than reward, successful membersof their group.





ARTICLE IN PRESS

chance of being fair. As more high-ability individuals invest in humancapital and garner success, the world may be revealed to be fair, puttingthe onus of non-achievement on low-ability individuals. Thus, the netbenefit of sabotage is increasing in the fraction of individualswho escape.

While we do not have definitive proof, there is evidence that blackstudents with college-educated parents in schools that are less than20% black know that success is possible. They are more likely tograduate from high school, have higher Scholastic Aptitude Testscores, and are more likely to be admitted into Ivy League universities(Massey et al., 2002). The sabotage theory would predict that therewould be no racial differences in the relationship between academicachievement and social status among this set of students. Yet, Table 6suggests the opposite: the coefficient on grades is .109 and thecoefficient on black⁎grades is −.549 and statistically significant.

7.3. A two-audience signaling model

Suppose there aremany individuals, a set offirms, and a peer group.There are two discrete stages of an individual's life: “school years” and“employment years.” Nature moves first and distributes an innateability to each student. Abilities, once disseminated, are fixed. Anindividual is endowed with one unit of non-storable time in eachperiod; the allocation of which is common knowledge. At the start ofeach period, an individual's time allocation problem is influenced bywhether or not she is an accepted member of her peer group. Peergroups are valued because, other things equal, leisure time spent in thegroup ismore enjoyable than leisure time spent outside the group. If anindividual is not an accepted member, then she makes decisionswithout reference to the group. If she is an accepted group member,then she may be called upon to make some observable timecontribution to the group.

In the “school years,” individuals allocate effort between leisure,group commitments, and a once-and-for-all investment in education. Atthe end of the school years, an individual's education level is fixed andfirms choosewage offers tomaximize expected profit. If an individual isrejected by the group during the school years, she cannot be accepted inthe post-school years; however, an individual accepted by the group inthe school years may be rejected in the post-school years.

At the beginning of the employment years, the group makesanother acceptance decision, and Nature reveals an individual's post-school-year time commitment to the group. Then, each individualdecides whether or not to contribute to the group, and makes a work-force effort decision. Any student who is accepted by the group andcontributes in the post-school years receives a lifetime utility benefit.The key feature of the group's payoffs is that the group is strictlyworse off having accepted an individual who chooses not to makeher required contribution than it would be was such an individualrejected.

The two-audience signaling model has two clear predictions —

racial differences in the relationship between social status andacademic achievement will exist and these differences will tend tobe exacerbated in environments with more interracial contact andincreased mobility — both of which are consistent with the empiricalevidence presented in Section 4. In addition, black high-achievers alsoreport more risky behaviors (smoking, drinking, lying to their parents,fighting, etc.) and less happiness than white high-achievers, which isconsistent with the signaling aspects of the model.

8. Conclusion

For nearly twenty years, there has been a debate among socialscientists on the extent and potential impact of racial differences inpeer sanctions often referred to as ‘acting white.’ The consensusgentium is: (1) ‘acting white’ does not exist in nationally represen-tative samples, and (2) to the extent that it is discernible in data, it isconcentrated in low-income minority schools. This paper demon-


strates that an additional set of facts points in a very differentdirection due to an adjusted definition of academic achievement.There is, however, a crucial point of agreement. We, like Ferguson(2001), Cook and Ludwig (1997), and Ainsworth-Darnell and Downey(1998), find scant empirical support for the oppositional culturehypothesis described in Fordham and Ogbu (1986).

Acknowledgements

We are grateful to George Akerlof, Pol Antras, David Card, PrudenceCarter, Kenneth Chay, William Darity, Federico Echenique, RonaldFerguson, Edward Glaeser, Michael Greenstone, Jennifer Hochschild,Christopher Jencks, Adriana Lleras-Muney, Lawrence Kahn, LawrenceKatz, Rachel Kranton, Steven Levitt, Glenn Loury, Linda Loury, JensLudwig, Franziska Michor, TedMiguel, Emily Oster, Paul Peterson, MicaPollock, Gavin Samms, Jesse Shapiro, and Andrei Shleifer for helpfulcomments and suggestions. The editor and two anonymous refereesprovided valuable feedback that greatly helped the paper take shape.Patricia Foo, Alexander Kaufman, Sue Lin, Katherine Penner, and DavidToniatti provided exceptional research assistance. Fryer thanks theEducation Innovation Laboratory at Harvard University for invaluablesupport.

Appendix A. Data description

The Longitudinal Study of Adolescent Health (Add Health)

GradesStudents were asked “At the most recent grading period, what was

your grade in each of the following subjects?” where the subjectswere English/Language Arts, Mathematics, History/Social Studies, andScience, and possible answers were A, B, C, or “D or lower”. Assumingstudent answered that their school grades on a letter basis, weaveraged the grades in these four subjects according to a 4.0 scale (i.e.A=4.0, B=3.0, C=2.0, and D or lower=1.0).

RaceStudents were asked “What race are you?” and “Are you of

Hispanic or Spanish origin?” Non-Hispanic white, black, or Asianswere coded as separate values. Students answering yes to the latterquestion were coded as Hispanic. Answers to the former questioncould include multiple races; non-Hispanic mixed race students (i.e.,students selecting multiple races) were also coded separately.

GenderStudents were asked “What sex are you?” Male or female.

AgeStudentswere asked “How old are you?” Answers range from 10 to

19, where 19 indicates age is 19 or older.

EffortStudents were asked “In general, how hard do you try to do your

school work well?”Possible answers were “I try very hard to do my best”; “I try hard

enough, but not as hard as I could”; “I don't try very hard at all”; and Inever try at all”. We coded these on a 1–4 scale where 4 represents thehighest level of effort (i.e., the first response) and 1 represents thelowest.

AthleticsStudents were asked whether they were participating or planned

to participate in a number of clubs or teams. This variable was codedas one if any of the following teams were indicated: baseball/softball,basketball, field hockey, football, ice hockey, soccer, swimming,tennis, track, volleyball, wrestling, or other sport.





ARTICLE IN PRESS

Student governmentThis variable was coded as one if student indicated he/she

participated in the student council.

CheerleadingThis variable was coded as one if student indicated he/she

participated in cheerleading or the dance team.

Parental educationStudent was asked how far in school their mother and father went.

This variable is coded as one if the parent graduated from a college oruniversity, or if they had professional training beyond a four-yearcollege. If the student does not know the exact level of education, it iscoded asmissing. Otherwise, other educational levels are coded as zero.

Parental occupationWe coded two variables each for the mother and father. The first is

based on the student's description of their mother or father's job —

whether they are “white collar professionals”. It was coded as one if,according to the student, theparent's occupation is “Professional 1” (suchas doctor, lawyer or scientist); “Professional 2” (such as teacher, librarian,nurse); “Manager” (such as executive or director); or “Technical” (suchas computer specialist or radiologist). Other professions are coded aszero. If the parent doesn't work or is disabled or retired, it is coded asmissing. The second variable is coded as one if the parent is a housewifeor househusband, according to the student, and zero otherwise.

Urban/suburban/rural and public/private schoolWe created dummies for whether the school is public or private

and whether it is located in an urban, suburban, or rural setting,according to the school administrator questionnaire.

Percent teachers black/Asian/HispanicThese variables are taken from the school administrator's answer

to the question “Approximately what percentage of your full-timeclassroom teachers are of each of the following races?” where raceslisted include black, Asian, and Hispanic.

Physical attractivenessThese variables are taken from section 40 of the In-Home Survey

Instrument. Each interviewer was asked, “How Physically Attractive isthe Respondent?” The answer choices were: very attractive, attrac-tive, average, unattractive, very unattractive.

Appendix B. Guide to programs

The programs used to calculate the index of social status are writtenin Matlab. There are three program files: cs_actingwhite.m, bs_acting-white.m, andn_actingwhite.m. They are adapted fromprogramsused tocalculate the Spectral Segregation Index (Echenique and Fryer, 2005).

cs_actingwhite.m is executed and calls the other two files. It takesas its input an identification string corresponding to a list of schools.Each list is contained in a text file called list#.txt, where # is theidentification string. This string need not be numeric. For example, alist of three schools might look like:

“001”“002”“003”

This list would tell cs_actingwhite.m to compute indices for theschools identified by 001, 002, and 003. Identification numbers shouldbe in double quotes, with each number on a separate line. The list#.txtfiles should be in the same folder as the three m-files.

cs_actingwhite.m calls bs_actingwhite.m for each of the schools inlist#.txt, then bs_actingwhite.m calls n_actingwhite.m. Each school


has a corresponding file, school_#.txt, which contains data on within-school friendships. The school_#.txt files should also be in the samefolder as the other files.

In the school_#.txt files, each row corresponds to an individualstudent. The first column of each row is an identifier, which should bein double quotes. The second is the school identification number foran individual student. The next 10 columns contain the ID numbers ofthe friends listed by that student. If the student lists fewer than 10friends, some of these columns will be empty. The final column liststhe group identifier for that student. All of the columns should becomma-separated. A line from a school_#.txt file might look like:

“106685”,94579955,92579959,96579956,97579953,98579959,95579950,98579957,,,,,4

In this case, “1006685” is a unique identifier for the studentwith ID94579955. The student belongs to group 4 and lists six friends.

n_actingwhite.m takes the school text files and counts bothfriendships and same-group friendships. A friendship is countedbetween students i and j if i lists j as a friend or if j lists i as a friend. Theinformation on friendships is passed to bs_actingwhite.m, whichperforms the calculations to find indices of social status.

The programs generate two kinds of output files. Information oneach school is outputted in files named si_#.txt, where # is the schoolidentifier. Sipartial.mat and spectralindex.dat contain information onthe entire set of schools in list#.txt.

sipartial.mat and spectralindex.dat each contain 12 columns:

Column 1: school identifierColumn 2: group identifierColumn 3: index for schoolColumn 4: number of connected components for groupColumn 5: number of singletons for groupColumn 6: median connected component size for groupColumn 7: largest connected component size for groupColumn 8: smallest connected component size for groupColumn 9: total number of students in groupColumn 10: percent of students belonging to groupColumn 11: average number of same-group friends

This information is stored continuously in sipartial.mat. The sameinformation is stored in spectralindex.dat after calculations for allschools have been completed.

The si_#.txt files each have six columns:

Column 1: school identifierColumn 2: group identifierColumn 3: connected component identifierColumn 4: student identifierColumn 5: individual indexColumn 6: connected component index

A separate file is created for each school in list#.txt.Like the files used to calculate the Spectral Segregation Index,

these m-files can be adapted to other applications.

References

Akerlof, G., Kranton, R., 2000. Economics and identity. Quarterly Journal of Economics115, 715–753.

Akerlof, G., Kranton, R., 2002. Identity and schooling: some lessons for the economics ofeducation. Journal of Economic Literature 40, 1167–1201.

Ainsworth-Darnell, J., Downey, D., 1998. Assessing the oppositional culture explanationfor racial/ethnic differences in school performance. American Sociological Review63, 536–553.

Armor, G., 1992. Why is black educational achievement rising? Public Interest 108,65–80.

Austen-Smith, D., Fryer, R., 2005. An economic analysis of ‘acting white’. QuarterlyJournal of Economics 551–583.





ARTICLE IN PRESS

Bauman, K., 1996. Trying to Get Ahead: School Work and Grades in the EducationalAdvance of African-Americans. Institute for Research on Poverty, Washington, DC.

Bergin, D., Cooks, H., 2002. High school students of color talk about accusations of‘acting white’. Urban Review 34, 113–134.

Brooks-Gunn, J., Duncan, G. (Eds.), 1997. The Consequences of Growing Up Poor. RussellSage, New York.

Campbell, J., Hombo, C., Mazzeo, J., 2000. NAEP 1999 Trends in Academic Progress:Three Decades of Student Performance. U.S. Department of Education's Office ofEducational Research and Improvement, Washington, DC. NCES 2000-469.

Card, D., Rothstein, J., 2004. Racial segregation and the black–white test score gap.Journal of Public Economics 91 (11–12), 2158–2184.

Carneiro, P., Heckman, J., 2002. Human capital policy. Working Paper, University ofChicago.

Chapple, S., Jefferies, R., Walker, R., 1997. Maori education participation and performance:a literature review and research programme. Paper for the New Zealand Ministry ofEducation.

Coleman, J.S., Campbell, E.Q., Hobson, C.J., McPartland, J., Mood, A.M., Weinfeld, F.D.,York, R.L., 1966. Equality of Educational Opportunity. U.S. Government PrintingOffice, Washington, DC.

Cook, M., Evans, W., 2000. Families or schools? Explaining the convergence in white andblack academic performance. Journal of Labor Economics 18 (4), 729–754.

Cook, P., Ludwig, J., 1997. Weighing the “burden of ‘acting white’”: are there racedifferences in attitudes towards education? Journal of Public Policy and Analysis 16,256–278.

Cross, W., Fhagen-Smith, P., 2001. Patterns of African American identity development: alife span perspective. In: Wijeyesinghe, C.L., Jackson, B.W. (Eds.), New Perspectiveson Racial Identity Development: A Theoretical and Practical Anthology. New YorkUniversity Press, New York.

Datnow, A., Cooper, R., 1996. Peer networks of African American students inindependent schools: affirming academic success and racial identity. Journal ofNegro Education 65, 56–72.

Delpit, L., 1995. Other People's Children: Cultural Conflict in the Classroom. New Press,New York.

DeVos, G., Hiroshi, W., 1966. Japan's Invisible Race: Caste in Culture and Personality.University of California Press, Berkeley.

Echenique, F., Fryer, R., 2005. On the measurement of segregation. Working Paper11258, National Bureau of Economic Research.

Farkas, G., Lleras, C., Maczuga, S., 2002. Comment: does oppositional culture exist inminority and poverty peer groups? American Sociological Review 67 (1), 148–155.

Ferguson, R., 1998a. Teachers' perceptions and expectations and the black–white testscore gap. In: Jencks, C., Phillips, M. (Eds.), The Black–White Test Score Gap.Brookings Institution Press, Washington, DC, pp. 273–317.

Ferguson, R., 1998b. Can schools narrow the black–white test score gap? In: Jencks, C.,Phillips, M. (Eds.), The Black–White Test Score Gap. Brookings Institution Press,Washington, DC, pp. 318–374.

Ferguson, R., 2001. A diagnostic analysis of black–white GPA disparities in ShakerHeights, Ohio. In: Ravitch, D. (Ed.), Brookings Papers on Education Policy 2001.Brookings Institution Press, Washington, DC, pp. 347–414.

Ferguson, R., 2002. What doesn't meet the eye: understanding and addressing racialdisparities in high-achieving suburban schools. Working Paper, Wiener Center forSocial Policy, Harvard University.

Fordham, S., 1996. Blacked Out: Dilemmas of Race, Identity, and Success at Capital High.University of Chicago Press, Chicago.

Fordham, S., Ogbu, J., 1986. Black students' school successes: coping with the burden of‘acting white’. The Urban Review 18, 176–206.

Freeman, L.C., 1978. Segregation in social networks. Sociological Methods and Research6 (4), 411–430.


Fryer, R., 2007. Amodel of social interactions and endogenous poverty traps. Rationalityand Society 19 (3), 335–366.

Fryer, R., Levitt, S., 2004. Understanding the black–white test score gap in the first twoyears of school. Review of Economics and Statistics 86, 447–464.

Fryer, R., Levitt, S., 2005. The black–white test score gap through third grade. WorkingPaper 11049, National Bureau of Economic Research.

Gans, H., 1962. The Urban Villagers: Group and Class in the Life of Italian-Americans.The Free Press, New York.

Grogger, J., Neal, D., 2000. Further Evidence on the Benefits of Catholic SecondarySchooling, Brookings–Wharton Papers on Urban Affairs 2000. Brookings InstitutionPress, Washington, DC, pp. 151–193.

Herrnstein, R., Murray, C., 1994. The Bell Curve: Intelligence and Class Structure inAmerican Life. The Free Press, New York.

Horvat, E.M., Lewis, K.S., 2003. Reassessing the ‘burden of ‘actingwhite’: the importanceof peer groups in managing academic success. Sociology of Education 76, 265–280.

Jencks, C., Phillips, M., 1998. The black–white test score gap: an introduction. In: Jencks,C., Phillips, M. (Eds.), The Black–White Test Score Gap. Brookings Institution Press,Washington, DC, pp. 1–51.

Jensen, A., 1998. The G Factor: The Science of Mental Ability. Greenwood PublishingGroup, Westport, CT.

Kling, J., Liebman, J., 2004. Experimental analysis of neighborhood effects on youth.Working Paper, Princeton University.

Kling, J., Ludwig, J., Katz, L., 2005. Neighborhood effects on crime for female and maleyouth: evidence from a randomized housing voucher experiment. QuarterlyJournal of Economics 120 (1), 87–130.

Massey, D., Charles, C.Z., Lundy, G., Fischer, M.J., 2002. The Source of the River: TheSocial Origins of Freshmen at America's Selective Colleges and Universities.Princeton University Press, Princeton, NJ.

McWhorter, J., 2000. Losing the Race: Self-Sabotage in Black America. Free Press, NewYork.

Neal, D., 1997. The effects of catholic secondary schooling on educational achievement.Journal of Labor Economics 15, 98–123.

Neal, D., 2004. Has black–white skill convergence stopped? Mimeo, University ofChicago.

Neal-Barnett, A.M., 2001. Being black: new thoughts on the old phenomenon of actingwhite. In: Neal-Barnett, A.M., Contreras, J., Kerns, K. (Eds.), Forging Links: AfricanAmerican Children Clinical Developmental Perspectives. Greenwood PublishingGroup, Westport, CT, pp. 75–88.

Ogbu, J., Davis, A., 2003. Black American Students in an Affluent Suburb: A Study ofAcademic Disengagement. Lawrence Erlbaum Associates, Inc., Mahwah, NJ.

Phillips, M., Crouse, J., Ralph, J., 1998. Does the black–white test score gap widen afterchildren enter school? In: Jencks, C., Phillips, M. (Eds.), The Black–White Test ScoreGap. Brookings Institution Press, Washington, DC, pp. 229–272.

Rodgers, W., Spriggs, W., 1996. What does AFQT really measure: race, wages, schooling,and the AFQT score. The Review of Black Political Economy 24 (4), 13–46.

Steele, S., Aronson, J., 1998. Stereotype threat and the test performance of academicallysuccessful African Americans. In: Jencks, C., Phillips, M. (Eds.), The Black–WhiteTest Score Gap. Brookings Institution Press, Washington, DC, pp. 401–430.

Tatum, B., 1997. Why are all the black kids sitting together in the cafeteria? And otherconversations about race. Basic Books, New York.

Tyson, K., Darity Jr., W., Castellino, D., 2004. Breeding Animosity: The ‘Burden of ActingWhite’ and Other Problems of Status Group Hierarchies in Schools. Working paper,Department of Sociology University of North Carolina Chapel Hill.

Willis, P., 1977. Learning to Labor: How Working Class Kids Get Working Class Jobs.Columbia University Press, New York.

Wooldridge, J., 2002. Econometric Analysis of Cross Section and Panel Data. MIT Press,Cambridge, MA.




Documents

AAn Empirical Analysis of Acting Whiten Empirical Analysis of Acting White