The Winner-Take-All High School

Sociology of Education 2001, Vol. 74 (October): 267–295 267

The Winner-Take-All High School:Organizational Adaptations to

Educational Stratification

Paul AttewellGraduate Center, City University of New York

Across the United States, families seek schools with reputations for academic

excellence for their children, assuming that such schools improve a talented

child’s prospects for college admission. This article shows that students from

“star” public high schools experience a disadvantage in entering elite colleges

that stems from the attention played to class rank. In response, some high schools

have developed “winner-take-all” characteristics, enhancing the success of their

top students at the expense of other students. Students below the top of their

class do less well than would be expected in terms of grades and advanced place-

ment courses. They also take fewer advanced math and science subjects, given

their high academic skills, compared to equivalent students in less prestigious

schools. These patterns are linked to a system of educational stratification that

links colleges and high schools through several types of assessment games.

merican culture stresses theimportance of education forsocial mobility. Hence, many

families respond with strategies forenhancing their children’s educationalprospects. One strategy is to search fornew homes close to academically superiorpublic schools. Countrywide, this strategyresults in high concentrations of affluentfamilies clustered in school districts withreputations for academic excellence.

A second strategy is to send children toprivate “prep schools” that are known fortheir academic eminence (Cookson andPersell 1985, Powell 1996). A third strategyprevails in certain cities where elite publichigh schools recruit talented studentsthrough competitive examinations.1 Takentogether, these three patterns have gener-ated a system of roughly 200 “star” high

schools in the United States with nationalreputations for exceptional educationaloutcomes.2

Although parents choose star schools forthe rich educational experiences they pro-vide, most assume that a major advantageinheres in their preparing students forentry into the nation’s most selective uni-versities. Going to a star high school isbelieved to improve a talented child’sprospects for getting into the best colleges,a step on the way to the most sought-aftercareers.

This article examines the organizationaldynamics of these exceptional high schoolsand the colleges to which they send theirbest pupils. My central thesis is that star highschools are caught in an assessment systemin which top students vie for entry into selec-tive colleges. Many schools adapt to this sys-

A

268 Attewell

tem, seeking to maximize the chances of theirstrongest students. Some of these schools havedeveloped policies that resemble “winner-take-all” markets: Their strongest students benefit atthe expense of those below (Frank and Cook1995). I build a case that1. The reputation of a star school becomes

tied to the attainments of its academicallystrongest students, but may be bought atthe cost of reduced opportunities for thetalented graduates who are below the top.

2. Attempts to “game” or manipulate theselective college entrance system have ledto modifications in the grading systemsthat are used by many star high schoolsthat further distance their top studentsfrom all the other students.

3. In some star schools, this effort has led tochanges in curriculum that lower theinvolvement of highly talented students inrigorous courses, especially in math andscience and in advanced placement (AP)courses. In sum, there is an irony. In many cases,

the hopes that draw families to enroll theirchildren in star public schools tend to backfirefor all but the top-performing students. As Idemonstrate later, formulas used by elite col-leges in the admissions process, especially anemphasis on class rank in high school, createa higher hurdle for students who are educat-ed in public high schools where there is ahigh concentration of talented young peoplein one school. Students who have excellenttest scores and high grade point averages(GPAs) from rigorous courses but are not atthe top of their class are downgraded bythese formulas. For such students, entry intoelite colleges from star public schools requireshigher test scores than entry from elsewhere.

Faced with the difficulty of getting theirstrongest students into selective colleges, starpublic schools have adapted. Many haveimplemented internal grading systems thatgive greater prominence to the achievementsof their top students. Some star high schoolsalso “cream off” the best of their students,discouraging with tough grading policies allbut the strongest students from taking APexaminations, especially for more rigorouscourses in math and science. Such adapta-tions ensure that the best students in star

schools stand out from the rest, enhancingtheir chances in the stiff competition for themost selective colleges. The adaptationsunfortunately have negative consequencesfor the remaining ranks of students in thesehigh schools.

CONCEPTUAL BACKGROUND

Stratification viaBureaucratic Statistics

In our highly bureaucratic society, decisionsabout resource allocation are increasinglybased on quantitative measures of perfor-mance generated by organizations (Attewell1986). Corporations collect statistics on rev-enues from local branches to decide which toexpand and which to shrink. Colleges evalu-ate applicants using SAT scores, GPAs, andclass rank, among other factors. Prospectivestudents, in turn, read handbooks that rankcolleges on quality and selectivity.

Of course, quantitative measures or rank-ings are not the sole criteria on which deci-sions are made. Many decisions hinge onnonnumerical judgments. However, in aworld dominated by rationalistic or econom-ic ideas, quantitative assessments of perfor-mance, or deciding among alternativesaccording to their scores or rankings, have abroad appeal. A major advantage of suchmeasures is that they appear to provide anunbiased decision-making method.

Publicly available performance measurestake on a coercive aspect. Important outsiderscan hold institutions to account when measuresdecline over time, when one organization ranksbelow another, or when decisions seem unfair(DiMaggio and Powell 1983). Consequently,performance measures become powerfulincentives for action. Individuals and organiza-tions change their behavior to look as good aspossible on the numbers, in a process known asgaming (Blau 1955). I describe instances inwhich organizations provide certain informa-tion but conceal other data, create new mea-sures that will highlight certain performances atthe expense of others, and filter out data frompoorly performing individuals—all to improvethe educational organization’s profile.

The Winner-Take-All High School 269

The larger point, however, is that the edu-cational field is increasingly structured byassessment games, in which organizations andindividuals seek to maximize their standingon these measures. Assessment procedures,in this case those developed by Ivy Leaguecolleges, come to have implications not onlyfor the students who seek entry into thosecolleges, but for the reputations of collegesthemselves and for high schools. In gamingthe admissions process for their strongest stu-dents, star high schools have affected theeducational climate of the majority of theirstudents, even those who do not aim to go tohighly selective colleges. Assessment gamesthat originated in one institutional spherehave come to structure the internal workingsof another.

Winner-Take-All Markets

There are winners and losers in all competi-tive markets. Some sellers find that theirgoods are more desirable and command abetter price, while others cannot sell theirgoods at all or have to discount their pricesteeply to sell them. In economic theory, suchcompetition and unequal outcomes areviewed as efficient mechanisms for allocatingscarce resources and matching supply anddemand.

Frank and Cook (1995) coined the term win-ner-take-all markets for a subspecies of marketsin which competition takes on a different char-acter. In these special markets, sellers whosegoods are viewed as the best (or are thought ofas the top performers) reap far greater rewardsand capture a disproportionate share of themarket than do other sellers whose goods arealso of high quality. Sellers immediately belowthe top may perform almost as well as those atthe top, but gain far fewer rewards. In Frankand Cook’s example, a handful of top operatenors command high fees for performing andrecording, while artists who are technicallyclose to them remain obscure, are paid far less,and are rarely asked to record. Similar behavioris found among top athletes (where the com-petitor who came in fourth is barely acknowl-edged), as well as among law firms andauthors. In such cases, reputation is central tothe dynamics of the market, and reward shares

are far more highly skewed toward the top thanwould be expected, given the “objective” ormeasurable distribution of talent.

Frank and Cook (1995) view winner-take-all markets as inefficient. The disparities inrewards are far greater than would be neededto attract talented people to participate inthese markets. The high rewards at the toplure far more sellers than demand requires, sothat the efforts of many unsuccessful buthighly talented people are wasted. In addi-tion, the competition takes on a wastefulcharacter, as individuals feel pressure toexpend greater and greater efforts to com-pete, even though the number of winners isnot increased by these efforts. Frank andCook gave as an example that if a few newMBAs feel it advantageous to wear $1,000suits to job interviews, others feel obliged todo likewise, although this expenditure doesnot increase the overall success rate in termsof open job slots.

This article demonstrates that some acade-mically outstanding public high schools havecome to resemble winner-take-all markets,both in the disproportionate rewards earnedby the top few students, compared to similarstudents of a slightly lower rank, and in termsof increased stratification inside the schools.

The School as a Frog Pond

Davis (1959, 1966) applied the concept ofrelative deprivation to academic success andmotivation, demonstrating the truth in theaphorism: “It is better to be a big frog in asmall pond than a small frog in a big pond.”He examined the relationships between acad-emic achievement, GPA, and career choiceamong senior men from colleges with differ-ent degrees of academic selectivity. Withinany given college, Davis (1966) found that ahigher GPA increased the likelihood of choos-ing a high-prestige career. However, acrosscolleges, for a given level of scholastic aptitude,there was a negative association between col-lege selectivity and the choice of a high-pres-tige career. That is, of two equally ableseniors, the one who attended the less selec-tive college was more likely to try for a high-prestige professional career than the one whoattended the more selective college.

270 Attewell

The mechanism that accounted for thisfinding was that a student in a more selectivecollege received a lower GPA than a studentof equivalent scholastic aptitude in a lessselective college. A student who was consid-ered academically outstanding in a less selec-tive college might be viewed as merely ade-quate in a more selective institution. On aver-age, this lower GPA resulted in lower acade-mic self-esteem and in more humble occupa-tional choices for a given aptitude. Davis(1966:30) concluded: “At the level of theindividual [these findings] challenge thenotion that getting into the ‘best possible’school is the most efficient route to occupa-tional mobility.” This article documents a sim-ilar phenomenon in today’s high schools, butfocuses on the organizational, rather than thesocial psychological, aspects.

METHODS AND DATA

Information on the selectivity and rankings ofrecent college admissions were obtainedfrom handbooks published annually byKaplan/Newsweek, Peterson’s, PrincetonReview, and US News and World Report. Mydescription of college admissions draws frombooks about or by admissions staff (Duffy andGoldberg 1998; Fetter 1995; Hernandez1997; Paul 1995), from sociological works(Karabel 1984; Karen 1990a, 1990b;Klitgaard 1985), and from historical studies(Geiger 1986; Synnott 1979; Wechsler 1977).I collected data on test scores of high schoolgraduates and their college placements forseveral star high schools by way of schoolnewspapers, school home pages on the web,school catalogs, and handouts for parents.

Information on the relations between stu-dents’ SAT scores and patterns of honorscourses and AP exams taken were obtainedfrom the College Board, which provided amerged file containing data on all 1,196,213students in the graduating class of 1997 whohad taken any of the following: SAT I, SAT IIsubject tests, or AP exams (personal identifierswere removed). I aggregated these individ-ual-level data by high school code andcleaned the file to remove home-schooledand non-high school examinees.

Student-Reported Data

Drawn from the SAT Student DescriptiveQuestionnaire (SDQ), which is part of theCollege Board data, students’ self-reporteddata on race, religion, parental income andeducation, and high school GPA were used.College Board researchers have examined thevalidity of SDQ measures by comparing theself-reports to external administrative records(Freeberg 1988). Variables created from theseself-reports included the following:

Race-Ethnicity Dummy variables were creat-ed for black, Hispanic, Asian and PacificIslander, and other race (including NativeAmericans). White was the reference catego-ry used in the regressions.

Parental Education This variable was themean educational attainment of the respon-dent’s father and mother on a scale of from 1to 9, from elementary school to graduatedegree.

Parental Income This variable was measuredby an 11-category scale, ranging from less than$10,000 to more $100,000. Although CollegeBoard researchers found a correlation of .78between students’ reports of parental incomeand data on financial aid applications, patternsof nonresponse dictate caution (Freeberg1988). I used this variable solely to determinethe proportion of affluent parents in a school, inconjunction with parental education (as dis-cussed later), not as an individual-level predictor.

Student’s High School GPA on the SDQThis variable had 12 response categories,from A+, A, A-, B+, in half grade steps until afinal category of “E or F.” In the SDQ, eachletter grade is accompanied by its corre-sponding numerical grade (e.g., an A grade isalso 93-96). For regression analyses, this vari-able was reversed, so that a score of 0 indi-cated E/F and a score of 12 indicated A plus.External verification of self-reported GPAsagainst high school transcripts revealed goodreliability for this measure (Freeberg 1988).

Class Rank Decile This self-report was recod-ed into a dummy variable with a value of 1 for


students who reported that they ranked inthe top decile of their high school class, zerootherwise.

Honors Math The SDQ asked for years ofcourse work in mathematics, followed by ayes-no question as to whether the studenthad taken honors courses in mathematics.This variable was recoded as zero or 1.

Trigonometry Course Work The SDQ askedfor the number of years of trigonometry coursework, from none to one semester through oneto four years. This variable was recoded, with avalue of 1 if any amount of trigonometry coursework was reported, zero otherwise.

Calculus Course Work Similarly, the numberof years of course work in calculus was recod-ed, with a value of 1 if the student reportedone or more semesters of calculus.

Variables from AdministrativeRecords

The College Board file reported the followingfor each student who took any SAT or APexamination:

Student’s Gender Recoded into a dummywith a value of 1 for male, zero for female.

SAT I Verbal Score and SAT I Math ScoreEach score was measured on a scale from 200to 800.

Combined SAT This variable was the sum ofa student’s verbal and math SAT I scores. (Thecorrelation between these two components isabout 0.7.)

Sat II Subject Tests Students who had takenthree or more examinations out of the 24possible SAT II subject examinations werecoded 1, and zero otherwise. (Three SAT IItests are requested by the admissions officesof many selective colleges.)

Any AP Examination A dummy variable wascreated with a value of 1 if a student hadtaken one or more of 33 possible AP subjectexaminations, and zero otherwise.

AP Math A dummy variable had a value of 1if a student had taken one or more of the APmathematics and statistics subject examina-tions, and zero otherwise.

AP Science A dummy variable was createdfor students who had taken one or more of 6AP subject examinations, including biology,chemistry, computer science, and physics.(Psychology and environmental science werenot included in this variable.)

School-level Variables

Each senior in the College Board database isidentified by a high school code. For school-level analyses, only students who wereenrolled in U.S. high schools were included.This process entailed dropping examineesfrom foreign schools, junior colleges, and“other” codes, as listed in a College Board(1998b) manual.

Certain information, such as the type ofschool (public, private, or religious), is pro-vided by the College Board database. In addi-tion, I constructed the following three school-level variables by aggregating test scores forstudents within a school. Only one school-level variable (school percent affluent)entailed aggregating students’ self-reportdata by school.

School Size The database did not provideinformation on high school enrollment.Instead, a count of the number of SAT-takingseniors in each school was used as a proxy.

School SAT Standard Deviation For eachschool, I calculated the standard deviation forall 1997 SAT-seniors of their combined SAT Imath and verbal scores.

School Percent Affluent For each student, Iconstructed a dummy variable “affluent,”with a value of 1 if either parent had a post-BA education or if the family income was over$80,000. I then aggregated nonmissingresponses into a percent affluent variable foreach school.

School Percent High Scorers. This variablewas constructed by calculating the percent-

272 Attewell

age of SAT seniors in each school whose SATI math score was 750 or higher. (The mathSAT I was chosen because this measure wasused primarily to predict who took AP scienceand math exams.)

School Type I classified schools into fivemutually exclusive categories: star private,nonstar private, exam star public, nonexamstar public, and nonstar public. Star schoolswere defined as schools having 11 or moreseniors with SAT I verbal scores of 750 orgreater.

Modeling Issues

In regression models contrasting schooltypes, I limit the analyses to schools with 40or more SAT seniors. Thus, all the schools areones in which a substantial number of col-lege-bound seniors take the SAT. In regres-sions predicting the probability of taking APcourses from individual or school characteris-tics, I omit schools in which no SAT studenttook an AP. Hence, those models predict theproportion of students taking APs, acrossschools where students do take APs. In thelatter part of the article, I present logisticalregressions that predict whether a studenttook, say, an AP exam, using individual stu-dents’ characteristics and (in some cases)school-level variables as predictors.3

My analyses use population data—all stu-dents who took the SAT in 1997—not a sam-ple. For this reason, the tables report regres-sion coefficients but not tests of statistical sig-nificance. When analyzing population data,one should focus on the sign and magnitudeof these coefficients.

FINDINGS

The Concentration ofTalent in Star High Schools

Skills of the kind measured by SAT I examina-tions develop from the interplay of familybackground, material and cultural resources,opportunities for learning, motivation, andeffort. Given the importance of resourcesamong these factors, it is not surprising that a

disproportionate number of the highest-scor-ing students on SATs (or on AP or SAT IIexaminations) are found in relatively fewschools. Nevertheless, the degree of concen-tration of high scorers is striking.

Nationwide in 1997, the top 0.7 percentof test takers in the graduating class scored780 or higher on the SAT I verbal test. Thatyear, 81 percent of high schools had no onegraduating with SAT I scores this high, and 12percent of schools had only one person. Atthe other extreme, 57 schools (0.3 percent ofall high schools) each had 10 or more of thesehigh-scoring students, accounting for 11 per-cent of all high scorers. Broadening theboundary, 217 schools (under 1 percent ofschools) graduated 24 percent of these highscorers.

Choosing a score of 780 is arbitrary, but alower cutoff does not change the pattern:Students at the top of the test distributionremain highly concentrated in a few highschools. Of the schools, 217 enrolled 23 per-cent of all seniors with verbal SAT scores over750 and 18 percent of those over 700, forexample.

In this article, I refer to these 217 schoolsas “star high schools,” but the term requirescareful qualification. As the College Board(1988) noted, SAT and other test scores donot measure the performance of any school.The existence of many high-scoring studentsin a school may reflect the quality of teachingto some extent, but to an even greaterextent, it reflects the social composition ofthe student body and the selection processesthat bring students into that high school.However, schools with a high concentrationof high-scoring students typically offeradvanced courses and send many of their stu-dents to the most selective colleges. As aresult, parents, staff, and especially collegeadmissions officers view them as extraordi-nary schools. It is in this sense that I use thelabel “star.”

Of the 217 star schools just mentioned, 67are private “prep schools” that select studentson their ability to pay and on applicants’scores on entrance examinations. Theremaining star schools are public. Thirteen ofthese public schools select students throughentrance examinations (“exam schools”). I


refer to the 137 others as “star publicschools.” Most of this last group are in afflu-ent communities and are open to all local res-idents.

Selective Colleges and Universities

The United States supports about 2,200 four-year colleges and universities, most of whichaccept almost all applicants who meet certainbasic educational requirements. Selectiveinstitutions—those with many qualified appli-cants for each opening—are a small minority(Bowen and Bok 1998). There is a spectrumof selectivity: The most sought-after IvyLeague colleges report about 8 applicants forevery undergraduate opening, while a fewdozen selective state universities typicallyreport from 1.5 to 3 applicants per under-graduate admitted. The rest have close to onequalified applicant for every place in thefreshman class (see Table 1).

Scholars have documented that the “mar-ket” for colleges has become increasinglypolarized in recent decades. Already highlyselective colleges have experienced greaterincreases in applicants than have less-selec-tive colleges (Duffy and Goldberg 1998). Thisincreased polarization reflects a widespreadbelief that a student gains a lifetime advan-

tage if she or he gets into a top college(Karabel and Astin 1975). As economists putit, there has been a “flight to quality.” Thistendency is especially marked among thehighest performing students (Cook and Frank1993; Hoxby 1997; McPherson and Shapiro1998).

Bowen and Bok (1998) found that gradua-tion from the most selective colleges was agreat advantage in terms of entry to elite pro-fessional schools and long-run posteducation-al incomes, net of a student’s initial academicperformance or skills (SAT and GPA). Thus,the economic payoff to a given academicability is greatly enhanced by entry to a high-ly selective college (see also Kingston andSmart 1990; Loury and Garman 1995;Zweigenhaft 1993).4

Most researchers focus on the periodbefore and during World War II as central tounderstanding current private college orien-tations. After World War II, the SAT becamewidespread as a yardstick for assessing col-lege entrants. By taking this one examination,students could apply to multiple colleges. In1959, the College Board published its firstapplication and admission figures for col-leges, and by the early 1980s, commercialpublishers provided comparative ratings forprospective students as part of guides for

Table 1. Selectivity of Some Colleges and Universities

Institution % of Applicants Accepted in 25th and 75th Percentile1980 1998 Scores for SAT Verbal in 1998

Harvard 16 13 700 790Princeton 20 12 670 770Stanford 19 15 670 770Yale 20 18 670 770Dartmouth 23 22 660 760

Chicago 66 58 640 740Duke 37 30 640 730Northwestern 55 29 620 720Vanderbilt 67 58 590 680

Ann Arbor, MI 72 69 590 660Austin, TX 74 61 540 650Berkeley, CA 70 36 570 700Chapel Hill, NC 46 37 560 670Madison, WI 83 77 520 650

Sources: Peterson’s Annual Guide to Undergraduate Study (1982) and Kaplan’s College Catalogue (1999). Note: The admissions data refer to undergraduate applicants, and the SAT verbal scores refer to those enter-

ing the freshman class.

274 Attewell

choosing a college. These publications cameto list the SAT test-score profile of each col-lege’s entering class, the ratio of applicants toadmissions, and the proportion of studentsdrawn from the top 5 percent or 10 percentof their high school classes, as well as morequalitative information about each school.Today, these annual publications are best-sell-ers. (The best-known ones are the annual USNews and World Report issue on colleges andsimilar guides by Kaplan, Peterson’s, andPrinceton Review.)

These measures have developed into aniron cage, constraining colleges and creatingincentives to look good on the measures.Students typically apply to colleges whosemean SAT scores approximate, or are some-what higher than, their own scores(Chapman and Jackson 1987). Colleges thathave refused to report statistics for publica-tion have experienced disastrous declines inthe number of applicants, leading them toprovide data in subsequent years (Duffy andGoldberg 1998). Performance statistics, nowmade public, have increased stratificationamong colleges. The top ones, all private, areswamped with applicants, but neverthelessremain sensitive about their standing relativeto one another. Many other private collegesstruggle to find enough paying customers,despite reputations for excellence in teach-ing.

To thrive, today’s private colleges need topresent evidence of their selectivity and per-formance to potential students and their fam-ilies. Admissions staff therefore strive for (1) astudent body that is drawn from all parts ofthe country, (2) a large pool of applicantsfrom which the college accepts a small pro-portion (selectivity), (3) maintaining highacademic standards for the incoming class interms of SAT scores and high school classrank, and (4) maintaining the college’s stand-ing vis-à-vis competing elite colleges. This lastcriterion is represented bureaucratically bythe “yield”—the proportion of accepted stu-dents who actually come to the collegeinstead of going to competing colleges.

These measures may be gamed or manip-ulated, given that colleges provide data aboutthemselves to the publishers of collegeguides. The reported SAT scores may not

include those of athletes and other lower-scoring persons, and applicant-to-admitratios may be inflated by overcounting thosewho applied (Duffy and Goldberg 1998;Fetter 1995). Selectivity or admissions ratiosdepend on the effort expended on mailingsand visits by college representatives to highschools. There is a strong incentive for col-leges to drive up the number of students whoapply without changing the number admit-ted, in order to appear more selective.Colleges whose SAT profiles are not strongmay also attract talent by offering scholar-ships to high performers (merit scholarships.)The income from high tuition may be recy-cled to offer scholarships to applicants whosehigh scores will make the school seem moredesirable (Duffy and Goldberg 1998,McPherson and Shapiro 1998).

The Selection Process inElite Universities

The college admission process has been sub-jected to considerable scrutiny. Few scholarsdoubt that there was ethnic and racial dis-crimination in the past (Duffy and Goldberg1998; Geiger 1986). In reaction, admissionsoffices nowadays have explicit policies thatare designed to prevent bias. Several admis-sions officers from selective colleges haveallowed studies of their procedures or havewritten books about the process themselves(Fetter 1995; Hernandez 1997; Paul 1995; cf.Karen 1990a, 1990b; Klitgaard 1985).

At an early stage, candidates’ applicationmaterials are typically coded into two numer-ical scores, one representing academicachievements and the other representingextracurricular activities (Fetter 1995; Paul1995). These numerical scores are modifiedduring a series of readings by admissions offi-cers, using their judgment of the toughnessof a student’s choice of courses and otherinformation. Review committees choose per-haps one in six applicants out of a pool thathas already been filtered to contain only goodprospects.5

In their writings, admissions officers takegreat pains to explain that their decisions areas fair as possible, stating that they are con-strained by standardized procedures and


shielded from outside influences (Fetter1995). I accept that officers are unbiased, butargue that the initial scoring scheme never-theless disadvantages certain kinds of appli-cants. The calculation of academic standingfrom SATs and class rank, which is often per-formed by computer according to a standardalgorithm, is the point at which students whoattend star high schools initially become dis-advantaged. This calculation requires no atti-tudinal bias on any staff person’s part.

An Admissions Algorithm andSimulation

Hernandez (1997), the assistant director ofadmissions at Dartmouth College, detailedthe “Blue Book” algorithm used there and inother Ivy League colleges to calculate an aca-demic index (AI) for comparing students’applications. She reported that there is a highdegree of agreement between admissionsdecisions using this method and decisionsmade by other highly selective colleges thatuse the same basic inputs but in a slightly dif-ferent way.

The formula calculates an AI by combiningthree components: SAT I scores, SAT II scores,and the student’s class rank in his or her par-ticular high school.6 The last is called a “con-verted rank score” or CRS. It drops off steeplyas one moves down from the head of theclass.

Table 2 presents a simulation using the IvyLeague admissions algorithm. In each of threepanels, I grouped hypothetical students fromdifferent high schools who have identical SATscores. The students differ only on their rank

in their school’s graduating class. I utilized thealgorithm to calculate the AI for each hypo-thetical applicant and the likelihood of beingaccepted into the freshman class atDartmouth College, using a translation fromAI values to admissions probabilities reportedby Hernandez.7

The simulations show that using class rankin combination with SAT scores creates a dra-matic valedictorian effect. Being at the top ofone’s class is worth about 70 points on theSAT I plus 60 points on the SAT II in terms ofoutcome (compare valedictorian G to studentB in Table 2). However, this class-rank benefitis extended only to the top handful in a highschool class (compare G with H and D withE). Conversely, the simulation shows thatthere is a substantial disadvantage of beingbelow the top of one’s class. Even having veryhigh SAT scores and an excellent GPA will nothelp if a substantial number of students rankhigher in one’s high school class (comparestudent C to A, D, or G).

In effect, the Ivy League algorithm treatsevery school in the United States as equiva-lent, so that a student at the 95th percentilein his or her high school class rank wouldreceive the same CRS score, no matter whatschool he or she attended. This equivalence isespecially questionable for star high schools,where the distribution of students’ talent dif-fers dramatically from the U.S. distribution.Using the SAT I verbal score as a yardstick forpurposes of illustration, one can see howextraordinarily aberrant star schools are com-pared to typical high schools. In Table 3, thefirst column presents the published nationalSAT I verbal score distribution for all students

Table 2. Simulated Effect of the Admissions Formula and the Likelihood of Acceptance to an Ivy LeagueSchool

Student SAT I SAT II Class Rank AI % Accepted

A 770 760 1 233 94 B 770 760 10 224 76 C 770 760 25 220 52

D 750 740 1 229 94 E 750 740 5 224 76 F 750 740 20 217 52

G 700 700 1 221 76 H 700 700 5 215 25

276 Attewell

in the class of 1997 (College Board 1999).This distribution could be imagined as theSAT profile of a quintessentially average highschool. The second column reports the actu-al SAT distribution for one star public schoolthat accepts all children in an affluent suburb.The third column shows the distribution foran urban exam star public school.

Table 3 indicates that nationwide, 1 per-cent of the takers of the SAT I verbal scored760 or over, compared to 7.4 percent at thesuburban star school and 28.9 percent at theexam-based public school. In terms of theSAT I, the 29th-ranked student in the examschool performed as well as the 9th-rankedstudent in the suburban star school and aswell as the top-ranked student in the averagehigh school. Yet because of the class-rank partof the Ivy League algorithm, the valedictorianat the average school would receive 170

more SAT-equivalent points than the 29th-ranked student at the exam star high school,despite their both having identical SAT scoresand (potentially) similar grades.

A multivariate analysis of students in theCollege Board database confirms this case-study finding (see Table 4). The model pre-dicts whether a student is in his or herschool’s top decile in class rank from the stu-dent’s SAT I scores and type of high schoolattended. The odds of being in the top decilefor a student in an exam star public schoolwas only 24 percent of the odds of a studentwith the same SAT scores from a nonstar pub-lic school (the reference category). The oddsof a student from a nonexam star publicschool being in the top decile was 30 percentof the odds of an equivalent-scoring studentin a nonstar public school. In sum, academi-cally strong students attending star schools

Table 3. SAT I Verbal Distributions for the Graduating Class at Three Schools

“Average High” “Affluent High” “Exam High”

SAT I Verbal N % N % N %

790–800 2 0.7 14 5.2 34 20.0760–780 1 0.3 6 2.2 15 8.9740–750 1 0.3 8 3.0 18 10.4720–730 3 1.1 10 3.6 27 15.6700–710 4 1.5 8 3.0 15 8.4

Under 700 256 95.8 221 82.7 63 36.6

Total 267 100.0 267 100.0 172 100.0

Note: “Average High” is a fictional school whose SAT distribution is set to the national SAT distribution. Theother two schools are real.

Table 4. Logistic Regression Predicting Rank in Top Decile

Predictor B Exp(B)

SAT I verbal 0.0051 1.0051SAT I math 0.0091 1.0092Star exam -1.43 .238Affluent star -1.17 .309Prep star -1.59 .202Nonstar private -.429 .651Constant -9.03

N of cases 837,845Nagelkerke

R-squared 0.39

Source: College Board 1997 college-bound seniors file.


have a substantially lower class rank than dotheir counterparts with identical SAT scores innonstar schools (cf. Davis 1966).

By constituting one-third of the overall AIfrom class rank, the Ivy League admissionsprocedure imposes a large penalty on manystudents in star schools. Students at the topof star schools are not penalized by the class-rank emphasis. But students who are justbelow the top find that their SAT scores are“worth” much less, in terms of the AI, than ifthey received the same SAT scores and GPAswhile attending a less selective high school.This is the “Big frog in a small pond or smallfrog in a big pond” phenomenon (Davis1966, Frank 1985).

Other things being equal, the disadvan-tage of being in a star high school is worse,the more intellectually exceptional its studentbody. For example, in 1998, Stuyvesant HighSchool in New York City had 11,397 eighth-grade applicants for its 850 high schoolplaces (Bumiller 1998). Admission was strictlyon the basis of a special entrance examina-tion, which is considered tougher than theSAT. The ratio of admissions to applicants tothis public high school—a selectivity of 7.5percent—was considerably higher than forthe most selective universities (cf. Harvard at13 percent).

It is not surprising, then, that the graduat-ing class of Stuyvesant seniors (like those fromother exam and star schools) contains anastonishing number of gifted students. Thisconcentration of talent makes class rank high-ly misleading compared to a nonselectiveschool. As the principal put it: “[Class rank]doesn’t make any sense when you havedozens and dozens of kids with a 97.2 aver-age” (Bumiller 1998). All but the top handfulof these “dozens and dozens” would receivea low converted rank score from the IvyLeague algorithm.

The claim that these high schools are dis-advantaged in the college admission processmay puzzle readers. Star exam high schoolsare renowned for getting students into IvyLeague colleges: Stuyvesant High Schoolnoted in the press that from its 700 graduatesin 1997, 21 went on to Harvard, 20 went toYale, 43 went to Columbia, and 103 went toCornell (Bumiller 1998). Similarly, Hernandez

(1997:200) insisted: “[I]t is not harder to getaccepted from a strong high school likeStuyvesant because even though Dartmouthreceives over one hundred applications ayear, it typically accepts 30 to 35 percent,since many are extremely qualified academi-cally.”

Nevertheless it is harder. The test scores ofstudents who are admitted to Dartmouth(and to other Ivy League schools) from examschools are substantially higher than the col-lege’s norm, as I document later. Conversely,many of the applicants from exam schoolswho are rejected by elite colleges have sub-stantially higher scores than do students whoare admitted to those same institutions fromless selective high schools. In essence, the starstudents who do gain admission do sobecause their various SAT scores are so highthat they more than compensate for the dele-terious effects of the class-rank procedure.

This disadvantage can be illustrated bydata on admissions and rejections collectedfrom the school newspaper of one exam highschool, which I call “Exam High.” Year afteryear, the SAT scores of students who wereadmitted to each Ivy League college fromExam High were far higher than the medianSAT score for the incoming class of that col-lege. In 1998, for example, the SAT scores ofaccepted students from Exam High exceededthe median score of admitted students to thatIvy League college 96 percent of the time andexceeded the score for the top 25 percent ofthe college class 58 percent of the time. Inother words, Exam High students who wereadmitted to Ivy League schools typicallyscored far higher than the average admission.Conversely, many Exam High students whowere rejected from Ivy League schools hadGPAs and SATs that were well above themedian score of students who were admittedfrom other high schools.

These case-study data complement thesimulation model in suggesting that studentsfrom high schools with a high concentrationof talented students face a higher hurdle toenter elite colleges than do students from lesstalent-rich schools, although a substantialnumber of students do leap that hurdle. Toreiterate: This disadvantage is not due to quo-tas or bias, but to the fact that class rank is used

278 Attewell

in admissions decisions. An emphasis on classrank makes it harder for a large number ofstudents from any high school to gain admis-sion to a particular elite college. Emphasizingclass rank in admissions prevents studentsfrom exceptional high schools from dominat-ing any college’s admissions pool.8 However,as I detail later, the class-rank procedure is notapplied equally to public and private highschools.

Adaptations by Star High Schools

The Ivy League colleges and universitiesadmit roughly 13,000 freshmen each year,while star high schools alone graduate56,000 seniors, and a million more college-bound seniors are close on their heels. Theresult is a recipe for frustration, an arena offierce struggle among the academically privi-leged. Faced with this dilemma, star schoolsexhibit a number of responses:1. Some high schools try to improve the

competitive standing of all their graduatesvis-à-vis students from other schools.Among other strategies, they encouragetheir students to take many AP courses andSAT II tests to demonstrate their academicprowess.

2. Many high schools carefully manage thepresentation of their institution—theschools’ collective statistical “presentationof self”—to improve their students’chances.

3. Some high schools try to lower students’expectations and reduce competitionwithin the schools over admission to first-rank colleges, in part by enforcing anextremely tough grading curve from thefreshman year on.

4. Some schools perform a kind of education-al triage, focusing on the prospects of theirmost talented students, knowing that theremainder of their students will still findplaces at less-prestigious colleges. Theykeep all but the smartest students out ofcertain AP and honors courses.

These strategies—documented next—are notmutually exclusive, but I suggest that certaintypes of star schools tend to emphasize onestrategy rather than another.

Raising the Competitive Stakes The CollegeBoard’s AP examinations provide an opportu-nity for students to demonstrate that theyhave learned advanced material during highschool and to gain college credit for thesecourses. They test a student’s knowledge in32 subjects, from art to statistics. Having APcourses on one’s transcript fulfills a “signal-ing” function to college admissions staff andallows students to skip certain college coursesand graduate college faster. Adelman (1999)showed that taking AP examinations is agood predictor of the completion of collegeand that taking a rigorous curriculum in highschool is by far the best predictor of successin college. College admissions officers knowthis.

There has been a marked expansion in thenumber of high schools offering AP coursesand the number of students who take thesecourses. The College Board (1997) reportedthat 581,554 students took AP courses in1997, up from 400,000 in 1992. About11,500 high schools currently offer AP cours-es, and roughly 90 percent of U.S. collegesand universities grant academic credit forqualifying AP grades.

It is not only star schools that offer honorsand AP courses. Riehl, Pallas, and Natriello(1999) found that urban high schools thatserve poor communities place considerableeffort and resources into honors courses evenwhen these courses are not heavily enrolled.Honors and AP courses have become impor-tant for the academic prestige and legitimacyof all kinds of high schools.

Table 5 indicates that star schools encour-age many of their students to take multipleAP courses. However, this effect is by far thestrongest in the exam-based public highschools and the prep schools. These schoolshave the greatest concentrations of high SATscorers, and their students signal their acade-mic prowess to colleges by taking a largenumber of AP examinations.

A similar pattern is evidenced by SAT IIsubject examinations. Most elite private col-leges, and some elite public universities,require three SAT II examinations for admis-sion. Hence students who take three SAT IItests may be presumed to have ambitions to


attend elite colleges. For example, 77 percentof SAT seniors in star prep schools had takenthree SAT II examinations, compared to 63percent in star exam schools, 34 percent instar public schools, and 13 percent in nonstarpublic schools.

A Statistical Presentation of Self High schoolsexpend considerable effort to build a positiveimage of themselves in the eyes of collegeadmissions officers. They do so in the belief thata student’s success in gaining entry to a selec-tive college is not simply a matter of the candi-

Table 5. Academic and Demographic Profiles of SAT Seniors in 1997, by Type of School

Nonstar Star Exam Star Affluent Star PrepPublic Public Public Private

Mean SAT verbal 497 635 560 631σ SAT verbal 108 96 113 90Mean SAT math 503 657 577 636σ SAT math 109 92 113 85

% Combined SATScore > 1400 3 31 11 23Score > 1500 0.6 10 4 7

% of Students withAny SAT II 16 79 42 813 or more SAT IIs 13 63 34 77

% of Students withAny AP 27 69 42 673 or more APs 2 10 5 10

% of Students TakingHonors math 27 46 36 28Calculus 21 51 29 32AP math 5 16 8 14AP science 6 28 11 19

Race% black 12 10 6 6% Hispanic 8 7 5 3% Asian 8 38 16 11

Family Income% > $80,000 25 26 34 65% < $25,000 21 18 10 6

% of Students with CUM GPAA- to A + 37 55 40 42B- or worse 26 15 22 15

% of Schools with a Tough Grade Distributiona 41 15 31 9

N of SAT seniors 823,274 4,360 42,296 10,095N of schools 8,484 13 137 67

A “tough grade distribution” is defined as 25 percent or more of SAT takers receiving a cum GPA of B minusor worse.

Note: The universe consists of those seniors in 1997 who had taken the SAT I examination. Seniors who didnot take the SAT are not represented in the data. Private nonstar schools are in the data set but are not shown.

280 Attewell

date’s personal achievements, but also reflectsthe scholarly reputation and rigor of the highschool she or he is graduating from (Powell1996). The school’s impression management iscultivated partly by visits with college admis-sions staff. In addition, many schools prepareannual “fact sheets” about the school that theysend to colleges, often attached to each appli-cant’s high school transcript.

A typical fact sheet reports the proportionof the school’s seniors who are usually accept-ed to four-year colleges, the SAT score distri-bution of the graduating class, the number ofstudents taking AP courses and the AP gradesthey receive, the number of National MeritScholarship finalists, and other awards. Theseitems demonstrate the exceptional academicquality of the school as a whole. In the past,many fact sheets also reported the mean GPAfor the graduating cohort and the grade dis-tribution. The latter information indicateshow “tough” a school’s grading curve is. Italso allows any applicant’s GPA to be translat-ed into a class rank. (Some schools directlyreport class rank on the individual student’srecord.)

Reporting class rank has been viewed asproblematic by staff in star schools for someyears. Class rankings make it clear to studentsin a school that they are engaged in a zero-sum game. They also encourage students’complaints about particular grades and makeinvidious distinctions between students whoare similar in ability. Moreover, some highschool principals believed that providing col-leges with seniors’ ranks disadvantaged them,though they lacked explicit evidence as tohow ranking formulas worked. (The use ofclass ranks in a decision algorithm was a“trade secret” until recently, though the gen-eral attention to class rank was not; seeHernandez 1997.)

One adaptation that appears to havebecome widespread by 1990 is that manyhigh schools abandoned the use of class rank;the schools neither calculated it for internaluse nor informed colleges of a student’s rankvia the transcript. In place of the class rank,some high schools reported class deciles sothat a student (or a college) knew whethershe or he was in the top or second or thirdtenth of the high school class. However,

removing class ranks from transcripts andsubstituting deciles did not prevent collegesfrom calculating a pseudo-class rank for anapplicant. Hernandez (1997) reported thatadmissions staff assign any applicant whosetranscript indicates that he or she was in thefirst decile to a rank at the midpoint for thatdecile. So, in a class of 400, any college appli-cants in the top decile would be assigned apseudo-rank of 20. Given the way that CRSscores are sloped, this procedure creates asevere disadvantage for anyone in the firstdecile whose actual class rank (were it known)is above 20th place.

Some high schools stopped reporting evendeciles, but college admissions officers werenot stymied. They then calculated pseudo-class ranks using an individual applicant’s GPAwith the overall distribution of grades report-ed for the senior class in the fact sheet.9However, in the past few years, more andmore high schools have also stopped provid-ing that grade-distribution curve to colleges.Here is the explanation on the fact sheet ofone star public high school:

Although [this school] has long had a systemof class rank by decile as a way in which tominimize negative competition between andamong students, increasingly this systemcaused problems. The difference betweendeciles became extremely small, especially inthe middle ranges. The School Council feelsthat the new system will result in a muchhealthier school climate. The Council furtherdecided not to include histograms or othercomparative charts that could easily result instudents determining a more absolute classrank.

A school’s refusal to provide grade distrib-utions prevents colleges from calculating classranks, causing one college admissions officerto complain: “GPAs by themselves are nothelpful if the college cannot calculate rough-ly where that GPA would put you in relationto your classmates. . . . When high schools tryto withhold information in this way, collegestend to assume that the high school is hidingsomething” (Hernandez 1997:61, 52). Inthese situations of incomplete information,colleges reluctantly calculate CRS pointsdirectly from a student’s GPA, irrespective ofclass rank.


Here we see an assessment game evolvingbetween colleges and high schools. Collegeswant to rank students against their own highschool peers, while high schools want to havetheir students ranked against the nationalpool of applicants, irrespective of the achieve-ments of other students in their own highschool. The back and forth over reporting orwithholding class ranks, deciles, and gradedistributions is an informational struggle overhow students and schools are to be assessed.

It is important to note that a few highschools have successfully avoided the class-rank issue for years. They have done so byusing unconventional grading systems (scalesother than 1 to 4 or 1 to 100) and/or “meth-ods that are not specific enough for deter-mining rank.” Hernandez (1997:68-69) listedthe schools in this special category. They con-stitute a “Who’s Who” of America’s leadingprivate preparatory schools, plus a few of themost prestigious star public schools.Evidently, these schools have long avoidedthe problem of having their many strong stu-dents handicapped by the CRS class rankprocess.

Weighted GPAs Another important strategyadopted by many star schools is the use of a“weighted” GPA. The College Board (1998a)reported that 7 percent of U.S. high schoolsuse this kind of system. Under this system,grades from an honors or AP course areweighed differently in calculating a student’scumulative GPA than are grades from non-honors courses. For example, the rare studentwho received all A grades while taking solelyhonors classes would graduate with a 4.5average (instead of a 4.0), and a student whograduated with Bs in all honors classes wouldhave a 3.5 average (not a 3.0). The overalleffect is to make the academically most success-ful students in a school stand out even furtherfrom their peers.

It is noteworthy that weighted GPA sys-tems are widespread among nonexam starpublic schools, but are not found in the exam-based high schools with which I am familiaror in the star prep schools. In both privateprep and public exam schools, the variance inacademic ability is lower (see Table 5), andthere is an ethos that all classes in the school

are highly demanding and therefore roughlyequivalent (Powell 1996). By contrast, innonexam star public schools, with a greaterdegree of academic heterogeneity amongstudents, the rationale for weighted GPAs isthat students who take honors classes aremastering more difficult material than theirpeers are. Schools argue that to avoid strongstudents selecting less-challenging courses inwhich they could get better grades, thegrades in honors courses should be weightedto reflect the extra effort required.

Weighted GPA systems do not differentiatecourses intended for college-bound studentsfrom courses intended for those who will notgo on to college. The distinction betweenhonors and nonhonors is within collegepreparatory courses, between the “high fly-ers” and the rest of the college-bound stu-dent body.

Weighting substantially improves thechances of individuals’ admission to selectivecolleges, since it pushes the best studentsabove the 4.0 ceiling that once defined a topperformance. It also affects competitionbetween different high schools over their stu-dents’ entry to selective colleges. For examplethe average GPA for entering freshmen atUCLA is now 4.16, and in 1998, theUniversity of California at Berkeley rejectedhundreds of Hispanic and African Americanstudents who had 4.0 averages in favor of stu-dents with higher GPAs. This practice has ledethnic minorities to cry foul and sue, sincepoorer schools rarely provide sufficient APcourses to allow even their best students toattain 4.5 averages (Berthelsen 1999; Takaki1999). This linkage between taking AP cours-es and the possibility of obtaining higherGPAs has also contributed to a new intensityof tracking within high schools, as I discusslater.

Grading Curve Policies High schools face abalancing act in terms of their course-gradingpolicies. On the one hand, an easier gradecurve may help individual students in compe-tition for college places with students fromelsewhere. On the other hand, a school thatis perceived as too generous in its grade poli-cy (“grade inflation”) is thought to lower itsreputation in the eyes of college admissions

282 Attewell

staff, possibly disadvantaging its students. Table 5 shows that the grade distributions

of SAT-taking students vary systematically byschool type. Considering only SAT takers, starschools, on average, have more generousgrade distributions than do nonstar publicschools. Exam star schools have the highestgrade distributions, followed by star prepschools. Nonexam star public schools arecloser to regular public schools in having atougher grade distribution. The contrastsbetween types of schools become moremarked when subsets of schools that aretough graders are identified. I defined atough grading distribution to be over 25 per-cent of SAT takers with cumulative GPAs of Bminus or worse. In Table 5, only 9 percent ofstar prep schools are tough graders, in thissense, compared to 15 percent of examschools and 31 percent of star public schools.

While Table 5 indicates that star schoolshave higher overall grade distributions thando nonstar schools, the picture changes onceone controls for their students’ SAT scores.Table 6 shows that for a given level on the SAT,seniors in star public schools have lower GPAsthan do their counterparts in nonstar publicschools (cf. Davis 1966).

In sum, a student in a star public school isless likely to be in her or his school’s top decileand is more likely to have a considerablylower GPA than a student with the same SATscores who attends a nonstar high school.The implications of this finding, within a win-ner-take-all framework, are elaborated in thenext section.

Academic Tracking Among StrongStudents Research on academic tracking inhigh schools initially focused on the separa-tion of students into academic, general, andvocational tracks. Once assigned to a track, astudent took most, if not all, her or his cours-es with other students in the same track(Kerckhoff 1976; Rosenbaum 1976).Rosenbaum (1979: 223) applied the conceptof tournament mobility to high schools: “Theresearch revealed a tournament patternwhere students who were moved out of thecollege track any time between 7th and 12thgrades had no chance of getting back in thattrack and very little chance of getting intocollege, regardless of how hard they strived.”

Tracking policies were loudly criticizedfrom the late 1960s on, both for reinforcingsocial inequality (Oakes 1985) and becauseresearch indicated that homogeneous group-ings of students according to skill-ability didnot improve students’ educational perfor-mance (Slavin 1990). Many schools appar-ently responded to such criticism because by1981, most high schools had abandoned theold system of tracks (Carey, Farris, andCarpenter 1994; Lucas 1999; Moore andDavenport 1988). In its place has emerged anew flexible form of course assignment.Instead of a high school student beingassigned to one track for all her or his cours-es, a student now typically receives individualassignments to courses of different levels ofdifficulty. Thus, one student may be in honorsFrench, general-level English, and remedialmath. Courses have become stratified, and“the vast majority of students” take math and

Table 6. OLS Regression of Grade Point Average, Predicted by SAT and Type of Public School

Variable b Beta

SAT math .008 .456SAT verbal .003 .168Star exam -.958 -.034Affluent star -.586 -.586Black -.002 -.020Asian .284 .040Hispanic .285 .039Male -.776 -.195

Adjusted R-squared .339N of cases 806,786

Source: College Board SAT database for 1997 college-bound seniors.


English courses at different levels (Lucas1999; see also Garet and DeLany 1988;Stevenson, Schiller, and Schneider 1994).

Lucas (1999) argued that this is inequality inanother guise because class, race, and genderremain predictive of course-taking patterns.Adelman (1999) reported that one-third of stu-dents who believe they are in a college-preptrack are not taking an academic curriculum.Lucas and Good (2001) criticized the tourna-ment concept because they found that nowa-days a substantial number of students do moveback up into more difficult classes after drop-ping down; however, they also reported thatdownward mobility predominates.

This article considers tracking in its newerform, especially the role of AP and honorsmath and science. While confirming that stu-dents’ characteristics play a role, I also exam-ine school-level predictors of the relative like-lihood of a student taking honors courses.

Useem (1990, 1991, 1992) studied 27school districts in Massachusetts and docu-mented that districts varied widely in the pro-portions of students they enrolled inadvanced math courses. She noted a steadyattrition in the proportions of students in theadvanced math track, from 17 percent in the8th grade to 6 percent in the 12th grade. Thispattern resembled tournament mobility with-in a college-bound student body. Useem dis-covered two contrasting ideologies aboutadvanced math education and noted thateach school district tended to hold one or theother ideology. Teachers in some districtsviewed advanced math as requiring excep-tional ability and restricted their advancedcourses to a small percentage of students.They made it difficult for students to move upinto advanced math courses, and evenamong those who initially entered theadvanced track, the number of students whostayed enrolled in advanced math and sci-ence classes declined steadily over timebetween the 8th and 12th grade. By contrast,teachers in the other type of school districtviewed mathematics as accessible to manyand encouraged more students to attemptadvanced courses. They allowed a larger ini-tial enrollment, which persisted over time,resulting in a higher number of students tak-ing calculus in the 12th grade.

Useem reported that affluent communitieswere more likely to have the second, moreinclusive approach, while blue-collar commu-nities tended to be more restrictive. However,some affluent communities shared the restric-tive approach with its more rigid tracking orability grouping. Socioeconomic status (SES)was not the whole story.

The College Board data allowed me toreexamine Useem’s findings on a broaderscale, looking at school differences in whotakes advanced math or science courses, cal-culus, AP math, AP science, and any APcourse. Table 7 presents logistic regressionmodels predicting whether a 1997 senior hadtaken, for example, AP mathematics, control-ling for individual-level characteristics (SAT Iverbal and math scores, gender, race, andparental education) and for school type.Table 8 presents an equivalent analysis but forstudent-reported course work in honorsmath, calculus, and trigonometry.

Two models, in adjacent columns, are pro-vided for each dependent variable in Table 7.The first column, Model 1, predicts one out-come (e.g., taking one or more AP examina-tions) from the type of school attended andfrom the student’s demographic characteristics.The second model adds the student’s SAT ver-bal and math scores to the prior predictors.Thus, it adds a control for certain academicskills in addition to school type and demo-graphics. The reader may wish to focus on thecolumns headed “exp (B).” This measure, e tothe power B, is known as the odds ratio. Itreports the change in the odds of the outcome(taking the specific AP or honors course) that isassociated with a one-unit increase in the par-ticular predictor, after holding all the other pre-dictors constant. For example, in the firstmodel, the odds of a student from a nonexampublic star school taking one or more AP exam-inations is 1.375 times as high as a student of asimilar gender, race, and parental educationwho is enrolled in a nonstar public school (thereference category or yardstick). In the secondmodel, after controlling for SAT scores, theodds of a student in an affluent star school tak-ing an AP examination is .823, or only 82 per-cent as high as the odds of a demographicallysimilar student with an equal SAT score who isenrolled in a nonstar public school.

284 AttewellTa

ble

7.

Log

isti

c R

egre

ssio

ns:

Eff

ect

of

Sch

oo

l Typ

e o

n T

akin

g A

P Ex

amin

atio

ns

Took

Any

AP

Took

AP

Mat

hTo

ok A

P Sc

ienc

e

Mod

el 1

Mod

el 2

Mod

el 1

Mod

el 2

Mod

el 1

Mod

el 2

Pred

icto

rsB

exp

(B)

Bex

p(B

)B

exp

(B)

Bex

p(B

)B

exp

(B)

Bex

p(B

)

Con

stan

t-2

.285

-9.6

76-4

.093

-10.

05-4

.021

-9.9

11

Exam

sta

r1.

472

4.35

6-.

110

.896

.798

2.22

2-.

344

.709

1.22

63.

408

.013

Pr

ep s

tar

1.14

93.

155

.093

1.

098

.689

1.99

1-.

033

.675

1.

964

-.16

2.8

51Pu

blic

sta

r.3

181.

375

-.19

5.8

23.1

831.

201

-.33

5.7

15.2

351.

265

-.27

5.7

60Pr

ivat

e .1

011.

106

-.02

9.9

72-.

001

-.01

7 -.

039

.962

-.11

6.8

91

Mal

e-.

164

.849

-.49

2.6

11.3

361.

400

-.14

3.8

67.2

521.

286

-.05

7 .9

45Bl

ack

-.75

4.4

70.3

251.

384

-.66

6.5

14.4

011.

493

-.69

3.5

00.3

091.

362

His

pan

ic.4

011.

494

1.16

53.

205

-.27

5.7

60.1

701.

185

-.29

0.7

48.1

571.

170

Asi

an.6

931.

999

.821

2.27

2.8

962.

450

.328

1.38

9.9

312.

536

.700

2.01

3O

ther

rac

e.0

391.

040

.274

1.31

6-.

093

.911

.026

.0

861.

090

.222

1.24

9Pa

rent

aled

ucat

ion

.243

1.27

5.0

581.

059

.167

1.18

1-.

008

.992

.211

1.23

5.0

251.

025

SAT

mat

h—

—.0

091.

009

——

.015

1.01

5—

—.0

101.

010

SAT

verb

al—

—.0

071.

007

——

-.00

3.9

97—

—.0

031.

003

Nof

cas

es85

7,79

885

7,79

885

7,79

885

7,79

885

7,79

885

7,79

8R-

squa

red

.108

.398

.051

.241

.069

.232

Sour

ce:

Col

lege

Boa

rd d

atab

ase

of a

ll 19

97 c

olle

ge-b

ound

sen

iors

.N

ote:

Pop

ulat

ion:

all

SAT-

taki

ng s

enio

rs a

tten

ding

sch

ools

with

40

or m

ore

SAT

seni

ors

in 1

997.

Ref

eren

ce c

ateg

ory

for

scho

ol d

umm

ies:

non

star

pub

lic s

choo

ls.

“Priv

ate”

mea

ns n

onst

ar p

rivat

e sc

hool

s. R

efer

ence

cat

egor

y fo

r ra

ce:

whi

tes.

The Winner-Take-All High School 285Tab

le 8. Log

istic Reg

ression

s: Effect of Sch

oo

l Type o

n A

dvan

ced M

ath C

ourse W

ork

Some C

alculus Course W

ork Som

e Trigonometry C

ourse Work

Honors M

ath Course W

ork

Model 1

Model 2

Model 1

Model 2

Model 1

Model 2

PredictorsB

exp(B)

Bexp

(B)B

exp(B)

Bexp

(B)B

exp(B)

Bexp

(B)

Constant

-2.033-9.705

-.194-3.487

-1.956-9.237

Exam star

1.2723.570

-.289.749

1.3483.849

.4861.626

.6191.857

-1.146 .318

Prep star

.8492.337

-.185.831

.8572.357

.2671.306

.1451.158

-1.131.323

Public star.225

1.252-.408

.665.105

1.111-.186

.830.186

1.204-.414

.661Private

-.002-.087

.917.440

1.553.461

1.586-.028

.973-.140

.869

Male

.3331.393

-.127.881

.2181.243

.0131.013

-.025.975

-.471.624

Black-.599

.594.655

1.926-.313

.732.295

1.343-.566

.568.625

1.868H

ispanic

-.156.855

.4461.563

-.170.843

.1161.123

.000 .634

1.885A

sian.848

2.335.705

2.024.633

1.883.532

1.703.589

1.802.392

1.479O

ther race-.016

.941.152

1.164.022

.1741.190

-.112.894

.0831.087

Parental education

.2131.237

.0131.014

.1341.144

.0301.030

.1891.209

-.005 .995

SAT math

——

.0151.015

——

.0071.007

——

.0131.014

SAT verbal—

—.001

1.001—

—.001

1.001—

—.002

1.002

Nof cases

567,838567,838

669,853669,853

857,798857,798

R-squared

.110.445

.056.172

.066.389

Source: College Board database of all 1997 college-bound seniors.

Note: Pop

ulation: All SAT-taking seniors attending schools w

ith 40 or more SAT seniors in 1997. Reference category for school d

umm

ies: nonstar public schools. “Private”

means nonstar p

rivate schools. Reference category for race: whites.

286 Attewell

A regular pattern repeats across the depen-dent variables in Tables 7 and 8. In Model 1,in which school type, race, gender, andparental education are predictors, students instar schools are more likely than students innonstar public schools (the reference group)to take an AP examination, specifically a mathor science AP examination, honors math,trigonometry, or calculus. Boys tend to bemore likely than girls to take these subjects orexaminations (with two exceptions). Whites(the reference racial category) are more likelyto take them than blacks or Hispanics. Asiansare more likely to take these subjects thanwhites, and children of highly educated par-ents are more likely to take them than stu-dents with less educated parents.

These findings fit our common-senseimpression that star schools are high-pow-ered academically and that gender and race-ethnicity shape students’ specialization inadvanced science and math. The effect sizesare substantial: Students in star public schoolshave odds of taking AP ranging from 20 per-cent higher to 37 percent higher (odds ratiosof 1.201 to 1.375).

The impression shifts dramatically, howev-er, in the Model 2 columns. Here, across thesame dependent variables, the same predic-tors are examined, but now also controlling foran individual student’s SAT verbal and mathscores. The signs of many predictors arereversed. Boys are less likely than girls to takehonors courses and AP examinations in sci-ence and math, relative to their SAT scores.Blacks and Hispanics are now more likely totake these advanced subjects than are whitesrelative to their SAT scores. The racial findingsare consistent with studies by Gamoran andMare (1989), Garet and DeLany (1988),Oakes (1985), and Lucas (1999), using differ-ent data sets and measures.10

Most important for my argument, theseregression models indicate that students whoattend star schools are considerably less likelyto take advanced math and science subjectsor AP examinations than are students in non-star public schools, once their SAT scores arecontrolled. Relative to their SAT scores, stu-dents in nonstar schools are more likely totake AP math and science than are students instar schools.11 The effect sizes are substantial:

The odds of a student in an affluent star pub-lic school taking AP math or AP science are71-76 percent of the odds of a student of thesame demographics and SAT who attends anonstar public school.

This finding suggests that nonstar schoolshave a less-talented student mix, but never-theless tend to push their better students intoadvanced work. Similarly, minority studentswhose SAT scores are high for their school butare not on national benchmarks are oftenencouraged to take honors and AP courses(Riehl et al. 1999). Conversely, star schoolshave such a rich mix of SAT students thateven though they have many students whoare enrolled in advanced work, these studentstend to have high SAT scores. Students withlower SAT scores, ones that are still high bynational benchmarks, are likely to be pusheddown to the nonhonors or non-AP level instar schools. Elsewhere, students of this cal-iber would be near the top of the schools andwould likely be tracked into the mostadvanced courses, but not in star schools.

Useem (1990) captured this tracking phe-nomenon in a vignette. An adviser in a starschool, explaining to a student why he wasnot selected for the honors math trackalthough he had high scores on a nationalmath assessment, stated: “You’re good, butyou’re not good enough.” Mathews(1998:140) noted that one star school“turned down several dozen students whowished to take AP courses. Principal [X] saidthe policy grew from the school’s fear that iftoo many marginal students took the test,they would lower the school’s pass rate andtarnish its reputation with colleges.”

Model 2 does not contradict the finding inModel 1 that in sheer numbers and in per-centage terms, star schools have substantiallymore students taking advanced math and sci-ence than do nonstar schools. But viewingthe data while controlling for SAT providesinsight into the dynamics of how markets (orschools) that are rich in talent work: Relativeto their ability, many students in talent-richschools are underchallenged. Here, one againsees the “large frog in small pond, small frogin large pond” phenomenon in operation.

Tracking students away from AP coursesmay be partly a matter of advising students


not to take them, but another mechanism forattaining the same end is adopting a toughgrading curve, especially for advanced cours-es. Students who receive low grades inadvanced math and science courses subse-quently drop down into less stringent coursesand/or shift their focus to the humanities.

Some star schools keep not only potential-ly failing students out of AP courses, but alsostudents who would probably pass the test.The grade profile of AP grades, especially inscience and math, becomes truncated in suchschools: The only students who get to takethe courses or exams are those who are likelyto get top grades. Relatively few studentswho take AP courses at star schools get mod-est but passing grades. This tendency to haveall AP takers pass with flying colors makes theschools look strong in the materials they sendto colleges and is an aspect of the “statisticalpresentation of self” discussed earlier. But ifthe point is to avoid AP failures, star schoolsovershoot and prevent AP passes.

To illustrate this truncation, one affluentstar school noted in its information sheet sentto colleges that 93 percent of its students’ APexaminations were scored 3 or above andthat 50 percent of the AP examinations takenby its students earned the top score of 5. Bycontrast, at the national level, 64 percent ofAP examinations were graded at 3 or betterand about 10 percent received the top gradeof 5 (College Board 1997:12). Using theCollege Board database, Table 9 presentsgrade distributions for several AP examina-tions, contrasting the grade distribution of AP

students in star public schools with those ofstudents in nonstar public schools. The latterare clearly far more likely to let students tryand fail or get low passing scores.

Predicting the Winner-Take-AllPattern

The ideal-type of a winner-take-all highschool is one that distinguishes between hon-ors and nonhonors courses in its grading pol-icy (by using a weighted GPA system), thusadvantaging students who take mainly hon-ors. It limits access to AP and honors coursesamong strong students, partly through coun-seling and partly through a tough gradingcurve. The next task is to identify the charac-teristics of schools that manifest this patternand the possible causal mechanisms behindthe phenomenon. The explanation I advancehas two parts: The first invokes institutionalideologies for certain types of schools, andthe second tests competing sociologicalhypotheses in predicting dimensions of theideal type.

Two kinds of star schools largely avoid thewinner-take-all pattern. Private prepschools—even those that enroll academicallyoutstanding students—seem resistant to thetendency. As Powell (1996:172) noted:

Challenge is . . . sustained at the [elite Prep]high school level by a serious but limited cur-riculum and by requirements that substantial-ly restrict student choice. Fewer courses andfewer options create a relatively egalitarianstudent experience. The curriculum does not

Table 9. AP Grade Distributions in Two Types of Schools (%)

Affluent Star Nonstar Public

AP U.S. History% grade 5 18 9 % grade 1–2 28 66

AP Math AB% grade 5 19 10% grade 1–2 23 47

AP Chemistry% grade 5 27 10 % grade 1–2 23 50

Source: College Board data for all SAT seniors in 1997.Note: AP grades go from 1 to 5, with 5 indicating the best performance.

288 Attewell

function to widen the aptitudinal and motiva-tional differences students inevitably bring totheir studies: its function is the reverse.Schools can show parents who are paying thesame price that their different children arereceiving the same service. One benefit ofeconomic privilege is greater equity in accessto a challenging curriculum.

Despite their poorer and more ethnicallydiverse student bodies, urban exam publicschools share this approach with their prepschool brethren. Exam public schools are“purposeful educational communities.” Theirstudents chose to apply and then were select-ed on the basis of their performance ondemanding examinations. Upon this base ofshared ability and commitment, an institu-tional ideology of uniqueness and superioritydeveloped. Having defined their students asan elite, exam schools treat all their scholarsas capable of taking rigorous programs.

“Constrained curriculum” is anotherphrase used to describe schools with a limit-ed curriculum of demanding academic cours-es that almost all students take (Lee,Croninger, and Smith 1997). Such a patterncharacterizes prep, exam, and Catholicschools. It has been shown to be associatedwith students’ higher performance in math(Lee et al. 1997; see also Gamoran 1992;Jones, Vanfossen, and Ensminger 1995).

By contrast, among academically hetero-geneous public high schools, the winner-take-all pattern tends to emerge, but even inthese schools, it is not ubiquitous. I advancefour alternative hypotheses for where andwhy this pattern may emerge among nonex-am public high schools:Hypothesis 1: Winner-take-all traits increase

with school size. As organizations growlarger, they increase their internal special-ization and differentiation. Hence, ceterisparibus, the separation of college-boundstudents into multiple tracks (“top flight”versus “others”) should increase withschool size.

Hypothesis 2: Winner-take-all traits increasewith greater heterogeneity in students’ abili-ty, as measured by the variance in test scoresin a school. Tracking may emerge as a func-tional response to widely varying students’abilities, from the belief that it is more

effective to teach students who aregrouped according to similar ability thanto teach mixed-ability classes (a belief dis-puted by Slavin 1990). So winner-take-allmay be a response to students’ intellectualheterogeneity.

Hypothesis 3: Winner-take-all traits increasewith the school’s proportion of affluent pro-fessional parents. Parents who hold profes-sional jobs may be highly concerned withaccess to elite colleges because they rec-ognize the mobility advantages providedby Ivy League credentials. As babyboomers, they have experienced the high-ly competitive labor markets that accom-pany large birth cohorts, and they maytransfer that insecurity into concern abouttheir children’s prospects (cf. Easterlin1987). In public high schools, some edu-cated affluent parents exert influence toincrease their children’s opportunitiesbecause they view their own children as“special” (Lareau 1989, Useem 1991). Oneinstitutional response to “specialness” is toprovide different classes that are reservedfor the most gifted students. Thus, winner-take-all tracking may grow out of this class-linked parental pressure (cf. Kohn 1998).Alternatively, professional parents may fearthat their children’s education will suffer ifthey are in classes with lower-class childrenand may push for tracking to obtain SEShomogeneity in the classroom (Oakes1994a, 1994b).

Hypothesis 4: Winner-take-all traits increasewith increasing proportions of very high-scor-ing students. Frank and Cook’s (1995)model suggests that winner-take-all strati-fication is associated with a glut of talentseeking high rewards for success. Applyingthis framework to education, the moreacademic talent in a school, the greaterthe likelihood that a school will skew itsrewards toward its strongest students.Merely smart students will be left on theeducational sidelines.In Table 10, these four hypotheses are tested

simultaneously through student-level logisticregressions predicting taking any AP examina-tion, taking an AP math examination, or takingan AP science examination. The population isall students in nonselective public schools with


40 or more SAT takers. To facilitate comparisonsacross the four hypotheses, I standardized allvariables (except dummies) with a mean ofzero and a standard deviation of one. Thus, theodds ratios—exp(B)—represent the change inthe odds of, say, taking an AP examinationassociated with a one-standard deviationincrease in each predictor.

The models in Table 10 include student-level controls (sex, race, parental education,and SAT scores), but the focus is on theschool-level characteristics. Hypothesis 1 pro-posed that increased school size should beassociated with decreases in the likelihood oftaking an AP, net of other factors. Thishypothesis finds little support: In the case ofany AP, school size has no discernable effect.In the case of AP science, it has a positive asso-ciation. Only with regard to AP math is thehypothesis supported: Students in largerschools are less likely to take AP math, net oftheir general scholarly ability.

Hypothesis 2 posited that academic het-erogeneity would be associated with a lowerprobability of taking an AP, holding a stu-

dent’s ability constant. Here, too, the effectsare mixed. For AP math, the hypothesis issupported, but for any AP and AP science, theeffect is reversed: The likelihood of takingthese subjects is increased in a more hetero-geneous school.

Hypothesis 3 suggested that students inschools with higher percentages of affluentparents (here defined as the percentage ofstudents whose parents have a post-BA edu-cation or a family income of over $80,000)would be less likely to take AP subjects, net ofpersonal attributes and other school charac-teristics. One mechanism for this outcomewould be if professional parents pressed forschool policies in which higher-SES or moregifted children tend to take different classesfrom their lower-SES or less-gifted school-mates. The hypothesis is supported: In eachcase, the greater presence of affluent parentsin a school is significantly associated with thelower likelihood of taking AP, after individualstudents’ SAT scores, sex, race, parental edu-cation, and for other school-level variables arecontrolled.

Table 10. Logistic Regressions: Testing Hypotheses Regarding School-level Correlates of AP Exam Taking

Any AP AP Math AP Science

Predictors B exp(B) B exp(B) B exp(B)

Constant -1.382 -3.844 -3.425

Male -.494 .610 -.153 .859 -.067 .935Black .329 1.389 .428 1.535 .297 1.345Hispanic 1.210 3.355 .261 1.298 .192 1.212Asian .897 2.451 .443 1.557 .753 2.123Other race .293 1.341 .050 .233 1.262Parental

education .158 1.172 .034 1.035 .073 1.076Verbal SAT .789 2.202 -.296 .744 .287 1.332Math SAT .972 2.644 1.721 5.588 1.125 3.079

School size .002 -.098 .907 .059 1.061School SAT SD .053 1.055 -.065 .937 .056 1.057School %

affluent -.099 .906 -.087 .917 -.092 .912School %

high scorers -.073 .930 -.056 .946 -.061 .940

N of cases 736,670 736,670 736,670R-squared .390 .243 .232

Source: College Board 1997 college-bound seniors.Note: Population: All SAT-taking seniors in schools with 40 or more SAT takers. Variables: except for race and

gender dummies, all variables are standardized with a mean of zero and an SD of 1.

290 Attewell

Hypothesis 4, the winner-take-all hypothe-sis, suggests that the greater the proportionof very high-scoring students, the greater theinternal stratification and exclusion in aschool, leaving a substantial number ofstrong (but not as outstanding) students outof the running. This hypothesis is also sup-ported. In each case, the presence of higherproportions of top-scoring students is associ-ated with a lower likelihood of taking AP, afterSAT, personal, and school-level characteristicsare controlled.

In sum, a pattern was described earlier forstar public schools in which their studentbodies were less likely to take AP examina-tions and advanced math and science, con-trolling for their general academic skill level asmeasured via SAT scores. Table 10 suggeststhat this observed pattern is more prevalentin schools with higher proportions of giftedstudents and in schools with higher propor-tions of affluent families. Both these factorsare independently, and roughly equally, asso-

ciated with a lower-likelihood of an AP pat-tern.

Table 11 translates the logistic models intoprobabilities of taking various AP subjects.The probabilities reported are for white malestudents of average parental education. Thetable contrasts probabilities for schools thatdiffer on the percentage of high-scoring stu-dents and the percentage of affluent parents.It also contrasts strong, very strong, and out-standing students, as measured by their SATverbal and math scores in standard deviationunits. The school effects on AP are substantialfor all types of students and all three subjects,but the effects are the largest for exceptionalstudents in AP math. Such a student has a .42probability of taking AP math if he attends aschool below the mean in terms of the per-centage of high scorers and affluent families,compared with a probability of .30 in aschool with many very high scorers and afflu-ent families.

Table 11. Influence of High-Scoring Students and School Affluence on the Probability of Taking Any APExam, an AP Math Examination, and an AP Science Examination

School Characteristics: Probability of AP for a Probability of AP for a Probability of AP for an% High Scores and % Affluence Strong Student Very Strong Student Exceptional Student

Probability of Taking Any AP Examination1 SD below the mean .51 .86 .94At the mean .47 .84 .931 SD above the mean .43 .81 .912 SD above the mean .39 .79 .902.5 SD above mean .37 .77 .89

Probability of Taking the AP Math Examination1 SD below the mean .08 .26 .42At the mean .07 .23 .381 SD above the mean .06 .21 .352 SD above the mean .05 .19 .322.5 SD above the mean .05 .18 .30

Probability of Taking the AP Science Examination 1 SD below the mean .13 .37 .55At the mean .11 .34 .511 SD above the mean .10 .31 .472 SD above the mean .08 .27 .432.5 SD above mean .08 .26 .41

Note: A strong student means SAT I scores 1 SD above the mean, a very strong student means SAT I scores 2SD above the mean, and an exceptional student means SAT I scores 2.5 SD above the mean.

Source: Calculated from Table 10 logistic regressions.


CONCLUSION

The larger purpose of this article is to inte-grate a series of disparate phenomena—admissions formulas for elite universities,grade and tracking policies in top highschools, the surge in AP examinations, andthe attrition of able students from advancedmath and science courses. Scholars have pre-viously studied these phenomena separately. Ihave sought to show that they may be fruit-fully considered as interlocking aspects of oneassessment system.

That system is driven by competition forscarce resources. At the college level, theresource is reputation, which, in turn, affectsthe number of applicants and hence thefinancial viability of private colleges. At thehigh school level, reputation also looms large:School staff are convinced that their school’sreputation affects the ability of their strongeststudents to gain admission to the mostsought-after colleges. So some schools limitaccess to AP and other advanced courses to“sure bets” and adopt grading policies thatdistance top students from their fellows,while maintaining a “tough” grading curve.At the student level, there is a sequence ofscarce goods: The opportunity to obtain ahigh GPA, which is related to access to hon-ors courses, struggles over class rank (a zero-sum resource) and ultimately access to elitecolleges en route to a career.

All the actors in this system, individualsand organizations alike, are caught in a sys-tem of assessment games. That some of theinstitutions originated the assessment proce-dures does not matter. They are now caughtin a system in which a college’s reputationwill dip if its selectivity or test-score profilechanges and a high school will lose face if itceases to send enough of its best students totop colleges.

Faced with the existence of this system,actors try to do as well within its boundariesas possible. They game the system. Collegesencourage a larger number of applicants, andhigh schools keep their grade curves toughand their AP pass rates high to emphasizetheir seriousness of purpose, but refuse toreport class ranks. High school students crafta balance between taking demanding cours-

es and participating in extracurricular activi-ties to impress college admissions staff. Theyavoid courses that may lower their GPAs.

Weighted GPAs, which at first appear to beesoteric accounting devices, turn out to allowexceptional students to leap to extraordinaryheights of GPA, outdoing others who merelyhave 4.0 averages, in the intense competitionfor selective colleges. Grade systems withweighted GPAs make AP examinations takeon a whole new aspect. No longer just a wayof jumping over college requirements, insome schools AP courses have become agateway to exceptional grades and thus topcolleges. This situation disadvantages poorerschools that do not offer lots of AP coursesand places their students at a disadvantageGPA-wise.

In some star public schools, access to thehonors track has become limited to the creamof the cream. Accomplished through advisingand tough grading policies, this new form oftracking leads to a steady attrition out ofadvanced math and science courses, causingexperts to wonder why talented youngAmericans avoid these subjects.

The intense competition among collegessustains a form of stratification that is trans-ferred into high schools with many talentedstudents. The result resembles a winner-take-all market. Each year, from a large pool ofable students, a handful goes on to top col-leges. But both the lucky ones and the “alsorans” experience a kind of academic speedupin their high schools, an unanticipated conse-quence of stratification among colleges andthe educational assessment games that fol-low. The human costs at the high school levelof this interlocking system have yet to be doc-umented, but are likely to be severe.

Stratification in America is not solely a mat-ter of distancing the affluent from the poor.As this study demonstrated, stratificationprocesses also affect advantaged strata. In aneducational equivalent of the “tragedy of thecommons,” affluent parents, in seekingexceptional high schools to advantage theirchildren, fuel organizational practices thatrebound negatively on many of their own off-spring.

292 Attewell

NOTES

1. Some examples are the Stuyvesant andBronx Science public high schools in NewYork City, the Boston Latin School in Boston,and Lowell High School in San Francisco.

2. A preliminary definition of the term starhigh school is a school from which many grad-uates are admitted to selective colleges. I usethe term star, rather than elite, because someof these schools serve nonaffluent urban stu-dent bodies. An operationalization of starschool is detailed in this article. Mathew’s(1998) book Class Struggle presents a list ofstar schools.

3. Multilevel data of this type are some-times analyzed using hierarchical linear mod-eling (HLM). The logic for using HLM is thatthe standard errors estimated by traditional“level 1” (logistic or OLS) models are inaccu-rate owing to clustering. The slope estimatesfor predictors are almost identical in HLM andin conventional regression models (Bryk andRaudenbush 1992). The data used in this arti-cle are a census covering an entire populationof over 1 million cases. There is no samplingof students within schools, and all studentshave an equal probability of entering thesample (a probability of 1). Hence the issue ofavoiding deflated standard errors—the prima-ry rationale for HLM—is moot for this dataset. Standard errors and tests of significancefor estimates are appropriate for analyses ofsamples, not for populations. Given this con-text, I have opted to use logistic regressionmodels (not HLM) so that the effects of indi-vidual and school-level characteristics can bestraightforwardly compared in one model.

4. This more recent research differs from ear-lier studies summarized by Pascarella andTerenzini (1991) and Astin (1993) that foundonly minor differences in economic attainmentassociated with the quality or exclusiveness ofthe undergraduate college or university attend-ed after students’ precollege socioeconomiccharacteristics were controlled. Dale andKrueger (1999) also reported finding noincome payoff to attending a highly selectivecollege. However, their analyses included twopredictors: SAT selectivity and the average realcost of attending (with financial aid subtract-ed). The former variable was not a significant

predictor of postgraduation income, net of thesecond. However the second predictor wasboth significant and substantively important.One may interpret Dale and Krueger’s findingsas follows: The degree of academic/SAT selec-tivity of a college is not what predicts high post-graduation incomes, but attending an expen-sive college, where a high proportion of stu-dents pay full tuition, pays off handsomely. Onesociological interpretation would be that it isnot the intellectual but the SES exclusivity of acollege that is associated with higher income.

5. There are separate considerations forminority applicants and for children of alum-ni. Athletes are also treated in a special way.Applicants in these categories have a clearadvantage in terms of their chances foradmission. These preferences clearly offendsome critics’ sense of equity, but are defend-ed by admissions officers and college presi-dents, including Bowen and Bok (1998).However “preferences” are not a concern ofthis article. I focus on the dynamics applyingto “ordinary” applicants, especially thosefrom star high schools.

6. High school GPA does not contributedirectly to the AI. It has an effect only when itis used to calculate the rank of each studentrelative to his or her high school classmates.

7. The simulation does not deal with theextracurricular activity score that admissionsofficers use, in conjunction with the AI, inadmissions decisions. In effect, that score isheld equal for all applicants in this table.

8. The origins of the emphasis on classrank are to be found in the 1920s, when ananti-Semitic panic swept the Ivy League uni-versities, resulting in the adoption of variousexclusionary devices. After World War II,when formal quotas against Jews weredropped, the use of class rank became linkedto pressures for geographic “diversity” thatprevented urban public high schools withconcentrations of academically gifted Jewishstudents from dominating Ivy League admis-sions. Nowadays, urban Asian students aresimilarly affected by class rank. See Wechsler(1977), Synnott (1979), Karabel (1984), andKlitgaard (1985) for this history.

9. Hernandez (1997) included charts forcalculating rank from these pieces of informa-tion.


10. It is possible that taking AP coursesmay raise a student’s academic skills andthereby improve her or his SAT scores.However, this kind of reverse causation fromAP to SAT would not obviate the resultsreported here regarding the influence ofschool type on AP enrollment.

11. The results of the regressions wereconfirmed by cross-tabulations, which indi-cated the percentage of students at a givenSAT level who took a given AP or honorscourse across high school types. These tablesare available from the author on request.They confirm that the multivariate findingswere not due to modeling artifacts.

REFERENCES

Adelman, Clifford. 1999. Answers in the Tool Box:Academic Intensity, Attendance Patterns, andBachelor’s Degree Attainment. Washington,DC: Office of Educational Research andImprovement, U.S. Department of Education.

Astin, Alexander. 1993. What Matters in College:Four Critical Years Revisited. San Francisco:Jossey Bass.

Attewell, Paul. 1986. “Imperialism in ComplexOrganizations” Sociological Theory: 4:115–25.

Berthelsen, Christian. 1999, July 28. “Suit SaysAdvanced-Placement Tests Show Bias.” NewYork Times:A18

Blau, Peter. 1955. The Dynamics of Bureaucracy.Chicago: University of Chicago Press.

Bowen, William, and Derek Bok. 1998. The Shape ofthe River: Long-Term Consequences of ConsideringRace in College and University Admissions.Princeton, NJ: Princeton University Press.

Brewer, Dominic, Eric Eide, and Ronald Ehrenberg.1996. Does It Pay to Attend an Elite PrivateCollege? Santa Monica, CA: RAND Corp.

Bryk, Anthony, and Stephen Raudenbush. 1992.Hierarchical Linear Models: Applications andData Analysis Methods. Newbury Park, CA:Sage.

Bumiller, Elisabeth. 1998, April 2. “Putting Dreamsto the Test.” New York Times:A1.

Carey, Nancy Laine, Elizabeth Farris, and JudiCarpenter. 1994. Curricular Differentiation inPublic High Schools. Rockville, MD: Westat.

Chapman, Randall, and Rex Jackson. 1987. CollegeChoices of Academically Able Students: TheInfluence of No-Need Financial Aid and OtherFactors. New York: College EntranceExamination Board.

College Board. 1988. Guidelines on the Uses ofCollege Board Test Scores and Related Data.New York: Author.

—-. 1997. The 1997 Advanced Placement Yearbook.New York: Author.

—-. 1998a, May. “High School Grading Policies”(Research Note RN-04). New York: Office ofResearch and Development, College Board.

—-. 1998b. Secondary School and Junior CollegeCode List: SAT Program. New York: Author.

—-. 1999. 1997 College-Bound Seniors SAT I TestScore Distributions and Percentiles [On-line].Available: http://www.collegeboard.org/sat/-counselors/ cbs/cbs97.html

Cook, Philip, and Robert Frank. 1993. “TheGrowing Concentration of Top Students atElite Schools.” In Supply and Demand in HigherEducation, edited by Charles Clotfelter andMichael Rothschild. Chicago: University ofChicago Press.

Cookson, Peter, and Carolyn Persell. 1985.Preparing for Power: America’s Elite BoardingSchools. New York: Basic Books.

Dale, Stacey Berg, and Alan Krueger. 1999.“Estimating the Payoff to Attending a MoreSelective College” (Working Paper 409,Industrial Relations Section). Princeton, NJ:Department of Economics, PrincetonUniversity.

Davis, James A. 1959. “A Formal Interpretation ofthe Theory of Relative Deprivation” Sociometry23:280–96.

—-. 1966. “The Campus as a Frog Pond: AnApplication of the Theory of RelativeDeprivation to Career Decisions of CollegeMen.” American Journal of Sociology 72:17–31.

DiMaggio, Paul, and Walter Powell. 1983 “The IronCage Revisited: Institutional Isomorphism andCollective Rationality in Organizations.”American Sociological Review 48:147–60.

Duffy, Elizabeth, and Idana Goldberg. 1998.Crafting a Class: College Admissions andFinancial Aid 1955–1994. Princeton, NJ:Princeton University Press.

Easterlin, Richard. 1987. Birth and Fortune: TheImpact of Numbers on Personal Welfare.Chicago: University of Chicago Press.

Fetter, Jean H. 1995. Questions and Admissions:Reflections on 10,000 Admissions Decisions atStanford. Stanford, CA: Stanford UniversityPress.

Frank, Robert. 1985. Choosing the Right Pond:Human Behavior and the Quest for Status. NewYork: Oxford University Press.

Frank, Robert, and Philip J. Cook. 1995. TheWinner-Take-All Society. New York: Penguin.

Freeberg, Norman E. 1988. Analysis of the Revised

294 Attewell

Student Descriptive Questionnaire, Phase I:Accuracy of Student-Reported Information. NewYork: College Entrance Examination Board.

Gamoran, Adam. 1992. “The Variable Effects ofHigh School Tracking.” American SociologicalReview 57:812–28.

Gamoran, Adam, and Robert D. Mare. 1989.“Secondary School Tracking and EducationalEquality.” American Journal of Sociology94:1146–83.

Garet, Michael S. and Brian DeLaney. 1988.“Students, Courses, and Stratification”Sociology of Education 61:61–77.

Geiger, Roger. 1986. To Advance Knowledge: TheGrowth of American Research Universities1900–1940. New York: Oxford UniversityPress.

Hernandez, Michele. 1997. A is for Admission. NewYork: Warner Books.

Hoxby, Caroline. 1997, December.. How theChanging Market Structure of US HigherEducation Explains College Tuition (NBERWorking Paper W6323). Cambridge, MA:National Bureau of Economic Research.

Jones, James D., Beth E. Vanfossen, and MargaretEnsminger. 1995. “Individual and Organi-zational Predictors of High School TrackPlacement.” Sociology of Education 68:287–300.

Kaplan/Newsweek. (annual). College Catalogue.New York: Simon & Schuster.

Karabel, Jerome. 1984. “Status-Group Struggle,Organizational Interests, and the Limits ofInstitutional Autonomy.” Theory and Society13:1–40.

Karabel, Jerome, and Alexander Astin. 1975.“Social Class, Academic Ability, and College‘Quality.’” Social Forces 53:381–98.

Karen, David. 1990a. “Access to Higher Educationin the United States, 1990 to the Present.”Pp.264–79 in Education and Society: A Reader,edited by Kevin Dougherty and FloydHammock. San Diego: Harcourt BraceJovanovich.

—-. 1990b. “Toward a Political-OrganizationalModel of Gatekeeping: The Case of EliteColleges.” Sociology of Education 63:227–40.

Kerkhoff, Alan. 1976. “The Status AttainmentProcess: Socialization or Allocation?” SocialForces 55:368–81.

Kingston, Paul, and John Smart. 1990. “TheEconomic Payoff of Prestigious Colleges.”Pp.147–174 in The High Status Track: Studies ofElite School and Stratification, edited by PaulKingston and Lionel Lewis. Albany, NY: StateUniversity of New York Press.

Klitgaard, Robert. 1985. Choosing Elites. New York:Basic Books.

Kohn, Alfie. 1998, April. “Only for My Kid: HowPrivileged Parents Undermine SchoolReform.” Phi Delta Kappan:568.

Lareau, Annette. 1989. Home Advantage: SocialClass and Parental Intervention in ElementaryEducation. New York: Falmer Press.

Lee, Valerie, Robert Croninger, and Julia R. Smith.1997. “Course-Taking, Equity, and MathematicsLearning: Testing the Constrained CurriculumHypothesis in U.S. Secondary Schools.”Educational Evaluation and Policy Analysis19:99–121.

Loury, Linda, and David Garman. 1995. “CollegeSelectivity and Earnings.” Journal of LaborEconomics 13:289–308.

Lucas, Samuel R. 1999. Tracking Inequality:Stratification and Mobility in American HighSchools. New York: Teacher’s College Press.

Lucas, Samuel R., and Aaron D. Good. 2001.“Race, Class, and Tournament Track Mobility.”Sociology of Education 74:139–56.

Mathews, Jay. 1998. Class Struggle: What’s Wrong(and Right) with America’s Best Public HighSchools. New York: Random House.

McPherson, Michael, and Morton Shapiro. 1998.The Student Aid Game. Princeton, NJ:Princeton University Press.

Moore, Donald R., and Suzanne Davenport. 1988.The New Improved Sorting Machine. Madison:National Center on Effective Secondary Schools,School of Education, University of WisconsinMadison. (ERIC document No. ED 316942)

Oakes, Jeanne. 1985. Keeping Track: How SchoolsStructure Inequality. New Haven, CT: YaleUniversity Press.

—-. 1994a. “More than Misapplied Technology: ANormative and Political Response to Hallinanon Tracking.” Sociology of Education 67:84–89.

—-. 1994b. “One More Thought.” Sociology ofEducation 67:91.

Pascarella, Ernest, and Patrick Terenzini. 1991. HowCollege Affects Students. San Francisco: Jossey-Bass.

Paul, Bill. 1995. Getting In: Inside the CollegeAdmissions Process. Reading MA: Addison-Wesley.

Peterson’s. (annual). Four Year Colleges. Princeton,NJ: Author.

Powell, Arthur. 1996. Lessons from Privilege: TheAmerican Prep School Tradition. Cambridge,MA: Harvard University Press.

Princeton Review. (annual). The 311 Best Colleges.New York: Random House.

Riehl, Carolyn, Aaron Pallas, and Gary Natriello.1999. “Rites and Wrongs: InstitutionalExplanations for the Student’s Course-Scheduling Process in Urban High Schools”


American Journal of Education 107:116–54.Rosenbaum, James. 1976. Making Inequality: The

Hidden Curriculum of High School Tracking.New York: John Wiley & Sons.

—-. 1979. “Tournament Mobility: Career Patternsin a Corporation.” Administrative ScienceQuarterly 14:220–41.

—-. 1980. “Track Mis-perceptions and FrustratedCollege Plans: An Analysis of the Effects ofTracks and Track Perceptions in the NationalLongitudinal Survey.” Sociology of Education53:74–88.

Slavin, Robert E. 1990. “Ability Grouping andStudent Achievement in Secondary Schools: ABest Evidence Synthesis.” Review of EducationalResearch 60:471–99.

Stevenson, David L., Katherine S. Schiller, andBarbara Schneider. 1994. “Sequences ofOpportunity in Learning.” Sociology ofEducation 67:184–98.

Synnott, Marcia. 1979. The Half-Opened Door.Westport, CT: Greenwood Press.

Takaki, Ronald. 1999, August 10. “On Admissions,More Inequality” [Letter to the Editor]. NewYork Times:A18.

US News and World Report. (annual). America’sBest Colleges. Washington, DC: Author.

Useem, Elizabeth. 1990. “You’re Good But You’reNot Good Enough: Tracking Students Out ofAdvanced Mathematics.” American Educator14(3):24–46.

—-. 1991. “Student Selection into CourseSequences In Mathematics: The Impact ofParental Involvement and School Policies.”Journal of Research on Adolescence 1:231–50.

—-. 1992, May. “Getting on the Fast Track inMathematics: School OrganizationalInfluences on Math Track Assignment.”American Journal of Education:325–53.

Wechsler, Harold. 1977. The Qualified Student. NewYork: John Wiley & Sons.

Zweigenhaft, Richard. 1993. “Prep School andPublic School Graduates of Harvard.” Journalof Higher Education 64:211–25.

Paul Attewell, Ph.D., is Professor of Sociology, Doctoral Program in Sociology, Graduate Center, CityUniversity of New York. His main fields of interest are sociology of organizations, sociology of tech-nology, and the impacts of computer technology. He is currently studying the educational conse-quences of the digital divide, funded by the National Science Foundation.

Address all correspondence to Dr. Paul Attewell, Doctoral Program in Sociology, Graduate Schooland University Center, City University of New York, 365 Fifth Avenue, New York, NY 10016; e-mail:[email protected].

Documents

The Winner-Take-All High School