The Frog Pond Revisited: High School Academic Context, Class Rank, and Elite College
Admission
Author(s): Thomas J. Espenshade, Lauren E. Hale, Chang Y. Chung
Source: Sociology of Education, Vol. 78, No. 4 (Oct., 2005), pp. 269-293
Published by: American Sociological Association
Stable URL: http://www.jstor.org/stable/4150499
Accessed: 12/08/2009 11:17
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=asa.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].
American Sociological Association is collaborating with JSTOR to digitize, preserve and extend access to
Sociology of Education.
http://www.jstor.org
The Frog Pond Revisited:
High School Academic Context, Class
Rank, and Elite College Admission
Thomas J. Espenshade
PrincetonUniversity
Lauren E. Hale
State Universityof New York,Stony Brook
Chang Y. Chung
PrincetonUniversity
In this article,the authorstest a "frog-pond"model of elite college admissionproposed by
Attewell,operationalizinghigh school academiccontext as the secondaryschool-averageSAT
score and numberof AdvancedPlacementtests per high school senior.Data on more than
45,000 applicationsto threeelite universitiesshow that a high school'sacademicenvironment
has a negativeeffect on college admission,controllingfor individualstudents'scholasticability.A given applicant'schancesof being accepted are reducedif he or she comes froma high
school with relativelymore highlytalentedstudents,that is, if the applicantis a smallfrog in
a big pond. Directevidence on high school class rankproducessimilarfindings.A school's
reputation or prestige has a counterbalancing positive effect on college admission.
Institutionalgatekeepersare susceptibleto context effects, but the influenceof school variables is smallrelativeto the characteristics
of individualstudents.The authorstie the findings
to priorworkon meritocracyin college admissionand to the role playedby elite educationin
promotingopportunityor reproducinginequality,and they speculateon the applicabilityof
frog-pondmodels in areasbeyondelite college admission.
rog-pondeffectsare typicallystudiedin elite careers. However, when students'
instances in which a social comparison
dynamic is presumedto influence individual behaviors. In a classic article, Davis
(1966) examinedthe careerdecisionsof college-age men. Hefound that on a given campus, students who had higher grade point
averages(GPAs)were more likelyto opt for
scholasticaptitudewas controlledand school
qualityvaried,a given student was less likely
to select a high-performancecareerfield the
more competitivehis college academicenvironment was. Davis cautioned parentswho
were sending theirsons to college: "Itis better to be a big frog in a small pond than a
269-293
2005, Vol.78 (October):
Sociologyof Education
269
270
smallfrog in a big pond"(p. 31). Marshand
his colleagues (Marsh 1991; Marsh et al.
1995; Marsh, K6ller,and Baumert 2001;
Marsh, Kong, and Hau 2000; Marsh and
Parker1984) found a similarpatternof results
and schoolwhen linkingindividual-student
averageachievementlevelsto academicselfconcept, educationaland occupationalaspirations,and the likelihoodof attending college.
Thefrog-pondmodel helpsto explainhow
the nation's3 millionhigh school graduates
get sorted among alternativecollege and
noncollege destinations. Class rank, along
with self-selectionby students;individualacademic performance;parents'education and
socioeconomic status; and the gatekeeping
roles of peers, schools, teachers, and guidance counselors,is an importantdeterminant
of whetherto applyto college, how manyto
apply to, and which ones (Hossler,Schmit,
and Vesper 1999; Manskiand Wise 1983;
McDonough1997).
Colleges and universities,especially the
most selectiveones, alsoexercisegatekeeping
functions. In deciding which students to
admit, admissionofficersevaluateapplicants
with an eye toward assembling a first-year
class that best matchesinstitutionalpriorities
and objectives(Bowen and Bok 1998; Duffy
and Goldberg 1998). Studiesof factorsthat
affect admission outcomes usually concentrate on the characteristicsof individualstudents, includingacademicperformance,sex,
race and ethnicity,athleticability,and legacy
status (Bowen and Bok 1998; Bowen and
Levin2003; Espenshade,Chung,and Walling
2004; Kane 1998; Karen 1990, 1991;
Shulmanand Bowen2001). Rarelyhas a secondary school's academic environment
receivedseriousattention.Some researchhas
includedcontrolsfor high school samplestratum (Kane 1998) or for the type of school
(public,parochial,or private).Privateschools,
especiallyelite boardingschools, have more
successthan do publichigh schoolsin placing
students at the most academically selective
institutions (Cookson and Persell 1985).
Lillardand Gerner (1999) included indicators
for the percentage of a high school's class
who go to college and whether the high
school is in an urban setting, but these con-
Espenshade, Hale, and Chung
textualmeasureswere treatedas controlvariables and were not featuredin the analysis.
Attewell (2001) introduced high school
context effects into studies of selective college admissions.He argued that admission
officersat IvyLeagueand other elite colleges
and universitiesuse an AcademicIndex(Al)to
summarizethe scholastic competence and
potentialof applicants.The Al is made up of
three equal parts: a student's SATI score,
averagescores on the SATIIor achievement
tests, and percentileclassrank.Becauseof the
importantrole that class rankplays,students
with very good gradesand strongtest scores
at talent-richhigh schools are at a competitive disadvantagerelativeto theireven-moregifted peers. Unlessthese studentsare at the
top of theirclass,they are more likelyto gain
acceptanceto selectivecolleges and universities if they attend a less prestigioushigh
school.
Attewell(2001) developed his conclusions
from simulationsthat were based on correlations between Al valuesand reportedprobabilitiesof admissionto DartmouthCollege.
One objectiveof this articleis to reexamine
Attewell'sconclusionswith the aid of actual
data on college applicationsand their outcomes. Our central question is this: Apart
from a student'sacademic abilityand other
personal characteristics,do the chances of
being admittedto an elite universitydepend
on where an applicantgoes to high school,
that is, on the high school's academicenvironment?
HYPOTHESES
We expect the followingfive outcomeswith
respectto the roleplayedby high schoolacademic context in decisions on admissionto
elite colleges:
Hypothesis1: If two students come from the
sameor comparable
highschoolsandif their
demographicand other social characteristics
are equivalent,there is a positiveassociation
between a student'sacademicmeritand the
probabilityof admission.
Academic ability is such a vital component
in admission to elite colleges and universities
TheFrogPond Revisited
271
We hypothesize that institutionalgatethat a student is unlikelyto be accepted if
admission officers have serious reservations keepersare susceptibleto the effects of high
about hisor herabilityto performin the class- school context. Thishypothesisparallelsfindroom. This does not mean that admission ings by Davis(1966) and by Marshand coldeans are searchingonly for applicantswho leagues, cited earlier,that there is a negative
score 1600 on the SATtest, but it does sug- effect of school-averageachievement levels
gest that there is a scholastic floor below on occupationalchoice and academic selfwhich applicantswill not be admitted(Karen concept, when individualstudents' achieve1991). Wethereforeexpect that, otherthings ment is controlled. But here, following
being equal, students who displaystronger Attewell's(2001) work, we conjecture that
academiccredentialswill have a higherprob- frog-pond effects also work at the receiving
end, where selective college admissionoffiabilityof being accepted.
cers
must admit a relativelysmall numberof
2: Withoutscreeningon informaHypothesis
from a largepool of highlyqualified
students
abilstudents'scholastic
tionaboutindividual
ities,applicantsfrom "better"high schools applicants.Testing the frog-pond model of
admission outcomes is the most important
standa greaterchanceof beingadmitted.
objectiveof this article.We hypothesizethat
Some of the more prestigioushigh schools with regardto acceptanceat an elite college,
go to great lengths to burnishtheir reputa- it is more advantageousto be the strongest
tions in the eyes of college admissiondeans. student in a less prestigioushigh school than
As Attewell(2001:279-80) noted, "Theydo an average student in an outstanding high
so in the belief that a student's success in school.
gainingentryto a selectivecollege is not sim4: The negativeeffectof schoolHypothesis
ply a matter of the candidate's personal
levelsin Hypothesis3
achievement
average
achievements,but also reflectsthe scholarly
more
talented
will
for
be
smaller
academically
reputationand rigorof the high school she or
students.
he is graduatingfrom."Collegesand universitiesrespondby deployingadmissionofficers
Attewell's (2001) simulationsimply that
to these elite secondaryschools, by cultivat- the size of the negativeeffect of high school
ing relationshipswith high school college academiccontext varieswith the abilitylevel
counselors,and by encouraginga feeder sys- of the student. As Attewellnoted, "Students
tem between particularsecondary schools at the top of starschoolsare not penalizedby
and elite colleges (Cooksonand Persell1985; the class-rankemphasis.Butstudentswho are
Yaqub 2002). The close relationshipsthat just below the top find that their SATscores
develop between elite colleges and similarly are 'worth'much less ... than if they received
elite secondary schools were illustratedby the same SATscores and GPAswhile attendCookson and Persell (1985:181), who ing a less selective high school" (p. 277).
described instances in which college coun- Marsh and Hau (2003) and Marsh et al.
selors from prestigiouspreparatoryschools (2001) examinedthe effect of an interaction
are invited to sit in on meetings of college between school-averageabilityand individual
admission committees. College admission students' ability on academic self-concept
officesare engaged, then, in a kindof "acad- and concludedthat it is small.We expect that
emic profiling"in which certainhigh schools an interactionbetween individualstudents'
are favoredover others on the basis of their achievement levels and school-average
reputationsfor producing highly qualified achievementlevelswill be positive.
students.
3: Iftwo studentsare equivalentin
Hypothesis
termsof individualacademicmeritand other
the one from the more
social characteristics,
competitive
high school has a
academically
smallerchance of being admittedto an elite
college or university.
5: Informationabout high schools'
Hypothesis
academic environmentmakes a statistically
significantcontributionto our understanding
of elite college admissions,net of other factors.
Several authors have reported that high
272
Espenshade,Hale,and Chung
school contextualeffectsare typically"small"
or "modest"in relationto the importanceof
individualstudents'attributes(Alexanderand
Eckland1975, 1977; Hauser, Sewell, and
Alwin 1976). Theory alone cannot tell us
whether to expect that the effects of a secondary school's academic environmentwill
be largeor small.Butour resultswould stand
in contrast to a large literatureif school
effects, as perceived by college admission
deans, outweighed the importanceof appliWe
cants'socialand academiccharacteristics.
to
find
that
school
academic
conhigh
expect
text is statisticallysignificantwhen added to
models that include individual students'
attributes.We also expect that the impactof
such institutionalvariableswill be small in
relationto students'academicand nonacademic merit.
Information on Students
ward from elite colleges to applicants. Even
though the analysis sample is a highly unrepresentative sample of all high school students
who applied to college, the data are appropriate for analyzing the nature of elite college
admissions.
significant about the academic strength of
high schools and that our findings would be
more robust if both indicators were related to
admission outcomes in the same way. The
first school-level variable was high school
average SATscore. The College Boardprovid-
Forthe purposesof thisanalysis,we useddata
from fall 1997 for three highly selectiveprivate researchuniversitiesthat representthe
top tier of Americanhighereducation.SATI
scores average almost 1400 for entering
freshmen. Data from these institutional
recordsincludedthe applicant'ssex, race,citizenship, athletic and legacy status, SAT I
scores,secondaryschool attended,and place
of permanentresidence.Academicabilityis
usually considered sufficientlymultidimensionalthat no singlemeasureadequatelycapturesit. Initially,we consideredtwo measures
of individual students' academic achievement:the combinedscore on the SATI math
and verbaltests and the numberof Advanced
Placement(AP)tests an applicanthad taken.
These data were obtained from the
EducationalTestingService(ETS).
SAT score is a much-used measure of
scholasticability.Studentscan alsosignaltheir
DATA AND METHODS
academicaptitudeto elitecollegesby takinga
We used data from the National Study of large number of AP examinations(Attewell
College Experience(NSCE),a collaborative 2001). Studentswho take AP courses earn
study involving10 academicallyselectivecol- highergradesin college, majorin more diffileges and universitiesfor the purposeof gain- cult subject areas, drop out of college at
ing a better understandingof the paths that lowerrates,and are more likelyto seek gradstudents follow through higher education. uate and professionaldegrees than are stuThe 10 institutionsare a subset of the 34 dents who do not take AP classes (Curry,
schools in the College and Beyonddatabase MacDonald, and Morgan 1999; Santoli
in colis
an
(Bowen and Bok 1998). They contain geo- 2002). Not only there advantage
AP
on
courses
from
lege acceptance
having
graphicspreadand representpublicuniversi- the
the
numbut
it
is
school
transcript,
high
ties, private researchuniversities,and small
liberalarts colleges.' Eachschool was asked ber and subject areas of the courses,rather
to supply informationon all individualstu- than the gradesreceivedon the APexaminadents who applied for admission in 1983, tions, that matter most to admissiondeans
1993, and 1997. This informationincluded (Santoli2002).
personalidentifiers,whetherthe studentwas
Contextual Variables
accepted, other data from the application
form,and supplementaldatafromthe offices We included two indicatorsof high school
of financialaid and the registrarif the appli- academic environmentin our analyses. By
cant subsequentlyenrolledat that institution. using multiplemeasures,we had greaterconInconstructingthe data set, we workedback- fidence that we would capturewhat is most
TheFrogPond Revisited
273
ed the 1997 SATI mean score for each sec- that these three institutionsare among the
ondary school, calculated from the latest most selective in the United States.
scoresof college-boundseniors.
Acceptance rates by category are shown in
Thesecond measureof high school acade- the second column of Table 1. Among the
mic contextwas the numberof APtests taken major racial/ethnicgroups, admission rates
per high school senior(Mathews1998). This were the highest for blacks, followed by
ratiois high when a large proportionof stu- Hispanics,and the lowestfor Asians.Athletes
dents takesAPtests and when the numberof and legacieswere accepted at ratesthat were
tests taken per student is large. Experienced more than double those for the typicalstuadmissionofficerssuggest that elite colleges dent.
The SATscore distributionfor these applioften relyon the numberof APcoursesa high
schooloffersas an indicatorof the rigorof the cants is shown in the second panelof Table1.
curriculum(Toor 2001). The numeratorof Two percent of the students had combined
this ratio is based on ETSdata on the total SATscores that were below 1000. Most stunumberof AP tests that were taken at each dents had test scores that were well above
high school in the 1996-97 academicyear. 1300, and SATscores overallaveraged1344.
The denominatoris the sizeof the 12th-grade Becausethe averagecombinedSATI scorefor
class in 1996-97. Data on the number of all SATtakers in 1997 was 1016, one sees
seniors in public schools came from the how gifted the applicantpool was (College
CommonCoreof Data,whichwere obtained Board2003). The chances of being admitted
from the National Center for Education rose monotonically with each 100-point
Statistics (NCES). NCES's Private School increasein SATscores. Buteven in the highUniverseSurveycontainscomparabledatafor est SATcategory (1500 and above), more
than half the applicantswere denied admisprivateschools.
We used two additionalhigh school con- sion. Table 1 also shows the number of AP
textualvariables.One is type of school (pub- tests that these applicantstook. Nearlyone
lic, private,or religious),which is containedin thirdof the sample,and almost40 percentof
data from the Student Descriptive the cases in which the number of tests is
Questionnaire(SDQ) that we acquiredfrom known,took fewer than three APtests. Most
ETS.A measureof a high school's academic studentswho took relativelyfew APexaminareputationis also important.We asked sea- tions in all likelihoodattended high schools
soned admissionofficersat an IvyLeagueuni- that did not offer many AP classes. Selective
versityto identifywhat they consideredto be colleges and universitieslook unfavorablyon
the "top U.S. secondaryschools."Theirrec- applicantswho could have taken AP courses
ommendationsincludedboth largeand small and chose not to (Santoli2002). Nearly7 perprivateand publicschools.We groupedtheir cent of the applicantstook eight or more AP
suggestions into a list of 72 elite secondary tests. Thesestudentswere twice as likelyto be
schools-ones that would be consideredby admittedto the three universitiesin 1997 as
knowledgeableobserversto be among the were those with fewer than threeAPtests.
Table2 containsdata on the characteristics
nation'sbest.
of
the
information
on
the
social
and
Descriptive
applicants'high schools.Nearly60 perof the 45,549 appli- cent of the applicationsreceivedby the three
academiccharacteristics
cants in our analysisfile is presentedin Table universitiesin our study came from students
1.2 These three highly selective universities who attended public high schools. Another
received an average of more than 15,000 sixthwere from applicantsat private,nonreliapplicationsfor fall 1997. Slightlymore than gious institutions.Students who attend prihalf the applicants were male; whites made
up the largest applicant pool, followed by
Asian students; most applicants were U.S. citizens; and only small proportions were athletes or legacies. The average acceptance rate
in 1997 was just over 20 percent, meaning
vate schools have a somewhat better chance
of being admitted to an elite university than
have applicants from other kinds of secondary schools. Roughly 1 out of every 12
applications came from a candidate at one of
the 72 elite secondary schools in our sample.
Espenshade,Hale,and Chung
274
Table 1. Students' Personal Characteristics and Academic Performance and Percentage of
Applicants Admitted: Fall 1997
Category
TotalSample(N = 45,549)
Students'PersonalCharacteristics
Sex
Male
Female
Percentageof Applicants
100.0
PercentageAdmitted
21.9
52.1
47.9
21.2
22.8
Race
White
Black
Hispanic
Asian
Other
Unknown
49.0
5.9
6.5
29.8
4.1
4.7
23.5
33.6
26.7
17.6
24.6
9.6
Citizenship
U.S.
Non-U.S.
83.9
16.1
23.2
15.2
Athlete
No
Yes
95.5
4.6
20.6
49.3
Legacy
No
Yes
96.9
3.1
21.2
46.5
2.0
3.7
8.6
17.2
26.3
26.5
13.3
2.5
1.8
5.1
10.6
16.1
20.0
26.9
40.8
10.0
11.3
9.8
10.3
11.1
10.8
9.5
7.4
5.0
3.4
3.5
18.0
15.0
17.8
18.7
20.9
23.0
25.8
27.3
29.3
32.5
36.2
19.4
Students'AcademicPerformance
SATscore
< 1000
1000-1099
1100-1199
1200-1299
1300-1 399
1400-1499
1500-1600
Unknown
Sample mean = 1344
National mean = 1016a
Numberof APtests taken
0
1
2
3
4
5
6
7
8
9+
Unknown
Sample mean = 4.2b
Nationalmean = 2.5c
Note: Percentagesmay add up to 100.0 + 0.1 because of rounding.
a College Board(2003).
b Both the sample and nationalmean for the numberof AP tests taken are based on students who took at
least one test.
c The data are for 1999 (Curry,MacDonald,and Morgan1999).
Source:NationalStudy of College Experience.
275
TheFroqPond Revisited
Table2. High School AcademicEnvironmentand Percentage of ApplicantsAdmitted: Fall
1997
Category
Percentageof ApplicantsPercentageAdmitted
TotalSample(N = 45,549)
100.0
21.9
Typeof high school
Public
Private
Religiousa
Unknown
58.9
16.2
13.6
11.4
21.2
26.4
21.6
19.8
Elite72
No
Yes
91.6
8.5
21.5
27.1
Numberof APtests taken
per high school senior(n)
0
0 < n < 0.4
0.4 < n < 0.8
0.8 < n < 1.5
1.5+
Unknown
13.6
17.9
17.0
16.9
18.0
16.5
19.9
20.7
21.4
21.1
25.4
22.5
Averagehigh school SATscore
< 1000
1000-1099
1100-1199
1200-1299
1300-1 600
Unknown
11.9
28.5
28.4
13.9
8.0
9.4
19.6
20.1
21.3
25.2
29.4
21.4
HighSchoolAcademicEnvironment
Note: Percentagesmay total 100.0 + 0.1 because of rounding.
a IncludesCatholicparochialschools and other independentschoolswith a
ation.
Source:NationalStudyof College Experience.
Theseindividualswere admittedat somewhat
higher rates than were students at nonelite
high schools.
Datain the lowerhalfof the table indicate
that 14 percentof all the applicationsto the
three universitiescame from students at
schools where no AP examinations were
taken, and slightly more (18 percent) came
from studentswhose high schools averaged
morethan 1.5 APtests per high schoolsenior.
In addition, 12 percent of the applications
religiousaffili-
came from students in high schools whose
averageSATscorewas lowerthan 1000. Most
applicants, however, were from "better"
schools. When attentionis restrictedto cases
in which the averagehigh school SATscore is
known,nearlyone quarterof the applications
came fromschoolswith an averageSATscore
of 1200 or higher.The probabilityof being
accepted by an elite universityis positively
associated with both measures of a high
school'sacademicstrength.3
Espenshade,Hale,and Chung
276
andfor high
students'personalcharacteristics
school academic environment.A student's
The resultsfrom the empiricalanalysisare score on the SATI examinationis positively
shown in Table3. We used logisticregression and significantlyrelatedto the probabilityof
because the responsevariableis whether a admission.So steep is the SATgradientthat
was accepted(= 1) or not (= the odds of being admitted for a student
givenapplication
coefficientsareexponentiatedto whose score is in the 1500-1600 range are
0). Regression
reflectodds ratios.Model1 includesapplicants' more than 200 times as high as those for a
Model studentwhose score is lowerthan 1000. The
demographicand socialcharacteristics.
2 adds individualstudents'academicperfor- numberof APtests displaysa similarpattern.
mance.In Model3, students'scholasticability Studentswith one AP test in their portfolio
is removed,and two indicatorsof high school have 31 percenthigherodds of being acceptacademic reputation are added. Model 4 ed than do students with no AP tests. The
variables. admissionadvantageis statisticallysignificant
includesallstudentand institutional
We begin by consideringthe influenceof and cumulativethe more APexaminationsa
candidates'demographicand socialcharacter- student has taken. As expected, both meaistics.The effectsof sex, race,citizenship,ath- suresof individualstudents'academicperforletic ability,and legacy status in Model 1 are mance have a positive effect on admission
similarto those in Model 3. Female,black, outcomes (see Figures1 and 2).
Hispanic,recruitedathlete, and legacy applicants receivestatisticallysignificantadmission Hypothesis 2
preferences,whereasAsiansand non-U.S.citizens are at a disadvantagerelativeto those in The second hypothesis expects that applithe reference groups. Preferencesare the cantsfrom high schoolswith a reputationfor
largestfor athletes,followedby those for chil- academicexcellencewillhavea betterchance
dren of alumni. Many of these effects are of admissionto an elite college, apartfrom
Private
strongerin Models2 and 4, in whichmeasures informationabout individualstudents.
admission
are
schools
generallyperceivedby
of students'academicperformanceare included. Comparedto Models1 and 3, the odds of deans as more venerablethan are publicor
admission for applicants who are black, parochialschools(Cooksonand Persell1985),
to
Hispanic,or recruitedathletesincreaseby a fac- and the elite72 variableis designed captwo
We
used
these
ture
institutional
prestige.
tor of 2 or more. The preferencefor athletes
to
test
for
measures
effects,
reputational
Smaller
outweighsthat for any other group.
increasesare registeredby women and lega- ratherthan school-averageSATscoresor the
cies, and the Asian disadvantageincreases. number of AP tests per senior. These latter
Asianapplicantsare morethan 30 percentless indicatorsare directlyinfluencedby students'
likelyto be admittedthanarewhitecandidates academic performance,which we already
The know is positivelyrelatedto admissionoutwiththe sameacademicaccomplishments.
most striking of these changes can be comes. Model 3, which does not controlfor
explainedby the factthat blacksand Hispanics individual students' academic credentials,
have,on average,weakeracademiccredentials providessupportfor this hypothesis.Students
than havewhite applicants,and recruitedath- who attend privatesecondaryschoolsare 26
letes are less well preparedthan are nonath- percentsignificantlymore likelyto be admitletes. Asianstudentstypicallyscore higheron ted than are public school applicants.
the SATtest and take more AP tests than do Candidatesfrom schoolswith a religiousaffilWe turn next to an iationare somewhatless likelyto be accepted
theirwhite counterparts.4
RESULTS
examination of the five hypotheses.
Hypothesis
1
The results to test Hypothesis 1 are presented
in Model 4 of Table 3. Here, we control for
than are public school students. The odds of
being admitted if one attends an elite secondary school-regardless of the type of
school-are 34 percent significantly higher
than are the chances of admission for students at nonelite high schools.
TheFrogPond Revisited
277
Table 3. Odds Ratios from Logistic Regression Estimates of the Effect of Students' and
High Schools' Characteristics on the Probability of Admission
PredictorVariables
Model 1
Model 2
Model 3
Students' PersonalCharacteristics
Sex
(Male)
Female
1.112***
1.435***
1.114***
Model 4
1.452***
Race
(White)
Black
Hispanic
Asian
Other
Unknown
Citizenship
(U.S.)
Non-U.S.
-
-
-
-
1.784***
1.540***
0.892***
1.014
0.875
5.791 ***
3.379***
0.682***
1.143*
0.728***
1.799***
1.573***
0.894***
0.989
0.854
5.685***
3.294***
0.690***
1.114
0.771**
0.637***
0.799***
0.630***
0.791***
Athlete
(No)
Yes
Legacy
(No)
Yes
Students'AcademicPerformance
SATscore
< 1000
1000-1099
1100-1199
(1 200-1 299)
-
3.163***
2.560***
-
6.102***
2.894***
0.049***
0.1 61***
0.459***
-
3.181**
2.470***
-
6.486***
2.995***
0.041***
0.147***
0.442***
1300-1 399
1.682***
1.701***
1400-1499
1500-1600
Unknown
3.371***
8.524***
0.398***
3.433***
8.742***
0.366***
1.192**
1.258***
1.312***
1.516"***
1.746***
1.823***
2.036***
2.307***
2.751 ***
2.41 3***
1.312***
1.409***
1.519***
1.795***
2.145"***
2.261***
2.571"***
2.964***
3.524***
2.240***
Continued
Number of AP tests taken
(0)
1
2
3
4
5
6
7
8
9+
Unknown
278
Esoenshade,Hale, and Chuna
Table 3. Continued
PredictorVariables
Model 1
High School AcademicEnvironment
Type of high school
(Public)
Private
Religious
Unknown
Elite 72
(No)
Yes
Model 2
Model 3
Model 4
1.263***
0.923*
1.267***
1.395***
1.105*
0.899
1.344***
1.109*
Number of AP tests taken
per high school senior (n)
(0)
0 < n < 0.4
0.4 5 n < 0.8
0.8 5 n < 1.5
1.5+
Unknown
0.813*
0.636***
0.550***
0.469***
0.587***
Average high school SATscore
(<1000)
1000-1 099
1100-1199
1200-1 299
1300-1 600
Unknown
Numberof Cases
LikelihoodRatioChi-squared(do)
PseudoR2
0.789***
0.774***
0.81 7**
0.968
1.080
45,549
45,549
45,549
45,549
3245.97(65) 8817.54(82) 3386.86(69) 9105.25(96)
0.0677
0.1840
0.1900
0.0707
Note: Referencecategoriesare shown in parentheses.Geographicarea (51 U.S. state dummyvariables, includingPuertoRicoand Washington,DC, and "other,""foreign,"and "unknown"areas)and
two institutiondummyvariables(totalof 56 variables)are includedin allthe models presentedhereas
controlvariables,but the estimatedcoefficientsare not reportedin the table.
* p < .05, ** p
< .01, ***p < .001.
Source:NationalStudyof College Experience.
Hypothesis 3
We expect that school-average achievement
levels will have a negative effect on admission
probabilities, net of an individual student's
own academic qualifications and other personal attributes. Model 4 contains the data
for this hypothesis. Note first that both school
type and elite72 behave as expected.
Students at private schools have a 40 percent
higher admission advantage over their public
school counterparts. Admission officers also
give a slight edge (roughly 10 percent) to students at religiously affiliated schools.
Attending one of the six dozen elite secondary schools is equivalent to having an 11
percent admission advantage.s
With regard to one of the two measures of
high school academic context, there is a signi-
279
TheFrogPondRevisited
8
6
5
4 -
2
-
< 1000
1000-1099
1100-1199
1200-1299
1300-1399
1400-1499
1500-1600
SATScore
Figure1. EstimatedOdds Ratiosfor Admission,by Students'AcademicPerformance:SAT
Score
Note: Basedon the estimatesfor Model4, Table3. The lightercoloredbarrepresentsthe reference category. * p < .05, ** p < .01, ***p < .001.
5-
4.
3.
0
1
2
3
4
5
6
7
8
9+
Numberof APTests Taken
Figure 2. Estimated Odds Ratios for Admission, by Students' Academic Performance:
Number of AP Tests Taken
Note: Basedon the estimatesfor Model4, Table3. The lightercoloredbarrepresentsthe reference category. * p < .05, ** p < .01, *** p < .001.
ficantlynegativerelation,as expected,between
the numberof APtests per high schoolsenior
and admissionoutcomes. A student has the
bestodds of beingacceptedby an elitecollege
if he orshe comesfroma highschoolwhereno
APtests aretaken(and, presumably,
whereno
AP coursesare offered).These odds steadily
deteriorateas a highschool'sacademicclimate
improves.Ifa studentwiththe sameacademic
credentialsappliesfrom a high school where
the averagenumberof APtests per senioris
between0.4 and0.8, the oddsof admissionare
36 percentlower.Andat the mostcompetitive
high schools-those with more than 1.5 AP
tests per senior-the same applicanthas 53
percentlowerodds of admission.These contextual effects strongly support our central
hypothesis.
Espenshade,Hale,and Chung
280
admissionwill weakenas students'academic
performanceimproves.We examined these
interactionsand found that the negative
effect of school academic context is indeed
less for better students. Model 4 was first
reestimatedusing continuousforms of SAT
scoreand numberof APtests at both the individual and school levels. Logisticregression
coefficientson the two measuresof students'
academic performanceare positiveand significant(at the .001 level). Bothschool context effects are negative and significant(at
the .001 level).Whentwo interactionterms,
one for individualand school SATscore and
anotherfor individualand school numberof
AP tests, are added to the model, the four
main effects have the same signs as before
and are significantat the .001 level. Both
interactionterms have significantlypositive
coefficients(p-valuesare 0.021 for SATscore
and 0.002 for numberof AP tests), suggesting that the negativeeffect of school context
fades as students'SATscoresand the number
of APtests increase.The negativeinfluenceof
school-averageSATscores disappearscompletely when students' SAT scores reach
1730. Thereforethere is stilla smallnegative
Hypothesis 4
effect of school-averageSATeven for top stuWe expect that the negativeeffect of school- dents with SATscores of 1600. Likewise,the
wide achievementlevelson the probabilityof adverseinfluenceof school-averageAPtests is
Moreover,with regardto the other measure of high school academic context, an
applicantfrom a high school whose average
SATscore is in the 1000-1099 or 1100-1199
range has more than 20 percentsignificantly
lower odds of being accepted than does a
studentwho attendsa school whose average
SATscore is lower than 1000. In the latter
case, the studenthasa muchbetterchanceof
risingto the top of his or her class.Attending
a high school with a schoolwideSATscore of
between 1200 and 1300 is associatedwith 18
percentsignificantlylowerodds of admission.
Beyonda high school average SATscore of
1300, the odds of admissionare not statistically differentfrom what they would be for
studentsfrom schools with an averagescore
that is lowerthan 1000. However,morethan
ninetenthsof applicationsin oursampleoriginated from schools whose average SAT
scores were below 1300. The frog-pond
model receivesclearsupportover the broad
range of school SATscores in which most of
the students in our sample are concentrated
(see Figures3 and 4).6
2-
15
...... ....... .. ...........
1..
. ............... . .. .. . ....... .
0
.............
0< n <0.4
.....
............. .... ......
.
.
0.4<= n <0.8
.
.......... .................... .... ..
0.8<= n <1.5
1.5+
Numberof APTests TakenperHighSchoolSenior(n)
Figure 3. Estimated Odds Ratios for Admission, by High School Academic Environment:
Number of AP Tests Taken per High School Senior
Note: Basedon the estimatesfor Model4, Table3. The lightercoloredbarrepresentsthe reference category. * p < .05, ** p < .01, *** p < .001.
281
The Frog Pond Revisited
15
-
<1000
1000-1099
1100-1199
1200-1299
1300-1600
AverageHighSchoolSATScore
Figure4. EstimatedOdds Ratiosfor Admission, by High School Academic Environment:
Average High School SATScore
Note: Basedon the estimatesfor Model4, Table3. The lightercoloredbarrepresentsthe reference category. * p < .05, ** p < .01, ***p < .001.
exhausted for students who take 13.3 AP
tests. The data in Table 1 suggest that few
students reach such rarifiedlevels. We conclude that there is a negativeschoolAPeffect
for most students in our sample, but the
effect is the weakestfor studentswho take a
largenumberof APtests.
TESTS FOR ROBUSTNESS
The evidenceprovidesstrong supportfor our
principal hypothesis. Admission officers at
elite colleges care how studentsperformacademicallyrelativeto their peers at the same
high school and rewardapplicantswho rank
high not only absolutely,but in comparison
to
their classmates.The purposeof this sec5
Hypothesis
tion is to probedeeperand to reexaminethis
To test whether information about high conclusionfromthree differentangles.
schoolacademiccontext improvesourunderstandingof elite college admissions,we need Clustered Observations
to comparethe fullModel4 againstModel2.
There is a modest gain in R2from 0.184 to A potential methodologicalconcern arises
0.190. A chi-squaretest of the null hypothe- fromthe fact that our45,549 applicantswere
sis that the newlyincludedvariablesin Model grouped into roughly 8,800 secondary
4 add nothing to our understanding of schools. Clustered observations within
admissionoutcomes has a test statisticvalue schools are not statisticallyindependent,and
of 287.7 on 14 degrees of freedom,which is while estimationproceduresthat ignoreclussufficientto rejectthe hypothesisat the .001 tering yield unbiased coefficient estimates,
level. There is ample evidence in favor of they also producestandarderrorsthat aretoo
Hypothesis5. Atthe same time, it is clearthat small.Ourdata appearto be influencedby a
these schoolvariablesare not as importantto mild degree of clustering.Eachof morethan
the analysis,and presumablyto admission 40 percentof the high schools in the sample
deans, as are students'academiccredentials. accounted for just a single application,and
The increasein the explainedvariancewhen another16 percentproducedonly two applimeasuresof students'academicperformance cationsperschool.At the otherextreme,one
are added is consistentlymuch greaterthan secondary school sent 287 applicationsto
when high schoolacademiccontext is includ- these three elite universities.We reestimated
ed.
282
Espenshade,Hale,and Chung
Model 4 in Table 3 (hereafter called the alike with respect to individualand high
"baseline" model) using the Huber-White school averageSATscoresand all othercharcorrectionin Stata for correlated,clustered acteristicsin the baselinemodel, a negative
errors. This correction produces standard SATschool effect could lead to the student
errorsthat are robustto clustering.The only from the ACThigh school having a higher
difference that the Huber-Whitecorrection probabilityof college admission.Admission
makes is that the coefficienton high school officersmay believethat the ACThighschool
SATscores in the 1200-1300 range is now is somewhat less competitivethan the nonsignificantat the 5-percent level instead of ACT high school if its school-averageSAT
the 1-percent level. Anotherapproachis to score is based on the performanceof a relaestimate a random-effectsmodel with one tivelysmallgroup of the best students.
To examinethis possibility,we reestimated
randomeffect per high school. Doingso produces odds ratios that are qualitativelythe the baseline model by includinga dummy
same as are those in baseline Model 4 and variable (ACTSchool)for applicationsfrom
significancelevelsthat are identical.The neg- secondaryschools in ACTstates. A total of
ative effect of school-average AP tests is 6,095 applications,or 13.4 percent of all
strongerusing randomeffects. In short, two applications,came from ACT high schools.
standardadjustmentsfor clusteredobserva- The resultsin column 1 in Table4 show that
the estimated school effects are similarto
tions corroborateour originalconclusions.
those in the baselinemodel.As expected,the
coefficienton ACTSchoolis positive;students
Applications from ACTStates
from high schools in ACTstates have 31 perMorethan 2 millionhigh schoolstudentstake cent higher odds of college admission,but
the SATtest each year, and 1.7 milliontake the effect is not significant.
the ACTtest administeredby the American
The negativeSATschool effect in the baseCollege Testing Program(Zwick2002). The line model of Table 3 is an average effect
SATis the preferredtest in the Northeastand between high schools in ACTand non-ACT
the West,while the ACTpredominatesin the states. Does the influenceof school-average
Midwest and the South. Students in ACT SATscore depend on where a high school is
states who take the SATtypicallyhave the located?To answerthis question,we added
strongest academic backgroundsand apply an interactiontermto the model in column1,
out of state to the nation'smost selectivecol- withthe resultsshownin column2 of Table4.
leges and universities(College Board2003). The SATschool effect in non-ACTstates is
Usingschool-averageSATscoresas a measure shown by the maineffectsof high schoolSAT
of high school academic context may not score. These effects are still negative, albeit
therefore have the same meaning in ACT somewhat weaker when compared to colstates as in non-ACTstates. Specifically,not umn 1, especiallyfor average scores in the
accountingfor the proportionof studentsin a 1200-1 300 range.Thenegativeeffectof high
high schoolwho takethe SATcouldaffectour school SATscore is more pronouncedin secinferencesconcerningschool-leveleffects.
ondaryschools in ACTstates.Applicantsfrom
ACTschoolswhose school-averageSATscores
Indirect Adjustment We tested the seri- are in the 1000-1100 range have one third
ousness of this omission in two ways. We lower odds of admissionto elite universities
began by defining ACT states as states in than do studentsin ACTschoolswithaverage
which a largerfractionof high school seniors SATscores that are lower than 1000. And
takes the ACTthan the SATtest. Thisdefini- compared to the same students, applicants
tion produced 26 ACT states (American from ACT high schools where the average
College Testing Program 1997; Snyder and SATscore is 1100 or higher have just half the
Hoffman 2000). We define secondary schools chance of admission to an elite college. When
within ACT states as ACT high schools. If school-average SAT scores are lower than
there is an applicant from an ACThigh school 1000, applicants from ACT schools have 97
and one from a non-ACThigh school who are percent significantly higher odds of admission
TheFrogPond Revisited
283
Table4. AdjustingHigh School Average SATScores for the Proportionof Students Taking
the SATTest (logistic regression coefficients shown as odds ratios)
IndirectAdjustment
Additive
Predictor
Variables
DirectAdjustment
Interactive Slope = -2.53 Slope = -3.79
(1)
(2)
(3)
(4)
0.815*
0.638***
0.552***
0.470***
0.592***
0.820*
0.644***
0.556***
0.471 ***
0.915
0.743**
0.648***
0.546***
0.665***
0.920
0.750**
0.659***
0.565***
0.673***
Number of AP TestsTaken
per High School Senior(n)
(0)
0 < n < 0.4
0.4 5 n < 0.8
0.8 5 n < 1.5
1.5+
Unknown
0.586***
AverageHigh School SATScore
-
(< 1000)
1000-1099
1100-1199
1200-1299
1300-1 600
Unknown
-
0.787***
0.770***
0.814**
0.966
1.078
0.782***
0.791 ***
0.856*
1.004
1.066
-
-
0.768***
0.668***
0.680***
0.802**
1.449**
0.844**
0.682***
0.678***
0.749***
1.494***
ACTSecondarySchool
(No)
-
Yes
1.311
InteractionTerms
ACTSchoolx
ACTSchoolx
ACTSchoolx
ACTSchoolx
ACTSchoolx
-
1.971*
HSSAT10-11
HSSAT11-12
2-1 3
HSSAT1
HSSAT13-16
HSSATunk
Numberof Cases
LikelihoodRatioChi-squared(df)
PseudoR2
0.833
0.615
0.563*
0.553
1.475
45,549
45,549
45,549
45,549
9107.82(97) 9133.69(102) 9163.24(96) 9158.29(96)
0.1901
0.1912
0.1911
0.1906
Note: All four models in Table 4 include as additional predictor variables all the variables in
Model 4 of Table 3. Coefficients on other predictor variables not shown in Table 4 are similar
to those in the baseline model in Table 3.
* p < .05, **
p < .01, *** p < .001.
Source: National Study of College Experience.
than have students from non-ACT schools.
The ACT school admission advantage diminishes as school-average SATscores rise. These
indirect adjustments to account for high
schools in ACT states leave intact our principal conclusion that there is a significantly
negative high school SAT effect. This effect
appears to be somewhat stronger in ACT
states. The negative AP school effect is essentially unchanged between columns 1 and 2.
Direct Adjustment
The College Board
(2003:1) observed that, "The most significant
factor to consider in interpreting SATI scores
Espenshade,Hale,and Chung
284
for any group, or subgroup,of test-takersis
the proportionof students taking the test."
Our second reassessmentrelies on a direct
adjustmentof each high school'saverageSAT
score for the proportionof seniorstakingthe
SAT.
Consider the following "parallel lines"
model,
mSATs,t= at + 0s + yPcts,t + Es,t,
(1)
where S individualstates are the units of
observation,mSATs,tis the mean SATscore in
state s in year t, at is an interceptterm in year
t, Ps representsS-1 state-leveldummy variables or "fixed"effects,y is the change in the
average state SATscore correspondingto a
one-unit increasein the percentageof high
school graduateswho take the SAT,Pcts,tis
the percentageof graduatestakingthe SATin
state s in year t, and Es,t is a random error
term.7 We expect that y < 0. Now assume
that the same model also holds k yearsearlier, except that the entire system of parallel
lines could have shifted up or down during
the period.Thenwe have
mSATs,t-k = "t-k +
s + YPCts,t-k+ Es,t-k.
(2)
Subtracting Equation 2 from Equation 1
yields
(mSATs,t- mSATs,t-k)= (ot at-k) + y (PCts,tPcts,t-k)
-
+
(•s,t
(3)
•s,t.k).
Equation3 was estimatedover two alternative time intervalsusing data on all 50
states from various issues of the National
Centerfor EducationStatistics'annualDigest
of EducationStatistics. For the period begin-
ning 1987-88 and ending 2001-02, the estimated intercept term is 28.4, slope (y) is
-2.53, and R2is 0.32. Both estimated parametersare significantat the .001 level.These
results imply that each percentage-point
increase in the proportionof a state's high
school graduateswho take the SATis associated with an expected declineof 2.53 points
in the state's average SAT score. When
Equation3 is estimatedfor the period from
1987-88 to 1996-97, the estimatedintercept
is 19.7, the slope is -3.79, and R2is 0.47. Both
the estimatedinterceptand the slope coefficient are significantat the .001 level.
We assume that the slope coefficients
derivedfromaggregatestate datatranslateto
the secondaryschoolleveland can be used to
maketwo independentestimatesof what each
secondaryschool'saverage SATscore would
have been if 42 percentof the studentstook
the SAT--the national average in 1996-97
(Snyderand Hoffman2000:151). These estimates requiredata on the percentageof high
school graduateswho took the SATat each
high school in 1996-97. Numeratordata on
the number of SATtakers came from the
datawerederived
CollegeBoard.Denominator
from the SDQ variableon the size of a
high school's senior class. Suppose that a
high school'sunadjustedaverageSATscore is
1000 and 76 percentof seniorstakethe examination.Then,witha slope coefficientof -2.53,
the adjustedschool-averageSATscoreis 1086
(= 1000 - 2.53(42-76)). The baseline model
sets
was then reestimatedusingtwo alternative
SATscore.
of adjustedvaluesforschool-average
The resultsare shown in columns3 and 4 of
Table4.
The negativeeffect of school-averageSAT
score is now much stronger,more persistent
acrossSATranges,and more statisticallysignificantthan in the baselinecase. Attendinga
high school whose average SATscore is
between 1200 and 1300 versusone with an
averagescore below 1000 lowersthe odds of
admissionby 32 percentin columns3 and 4,
compared with 18 percent in the baseline
case. High schools with average SATscores
above 1300 are associatedwith an insignificant 3 percentlowerodds of admissionin the
baseline model, but with 20 percentsignificantly lower odds using a 2.53 slope adjustment and with 25 percentsignificantlylower
odds with a 3.79 adjustment.The effects of
school AP in columns 3 and 4 are negative
and significant,althoughnot quiteas steep as
in the baseline model. To summarize,the
resultsusing high school SATscoresthat are
adjusted for the proportion of graduates who
took the SATtest lend added support to our
central conclusion that in making decisions,
admission officers at elite universitiesconsider
how students perform relative to their peers
in the same high school.8
TheFrogPond Revisited
High School Class Rank
We have inferreda student'sacademicstanding among peers at the same high school by
comparingthe student'sSATscore and number of AP tests with the school-averageSAT
score and number of AP tests. A more customary way of evaluating applicants with
respectto classmatesis to relyon high school
GPAand class rank.In this section, we incorporate these measuresinto the analysis.We
use SDQ data that students supplied when
they registeredfor the SAT.Students were
instructed,"Pleaseindicateyour cumulative
grade pointaveragefor all academicsubjects
in high school."Theywere also asked,"What
is your most recent high school class rank?"
We have SDQ responsesfor 75.6 percentof
the applicants.In the remainingcases, the
studentschose not to completethe voluntary
SDQ questionnaire,or it was not possibleto
match their answersto institutionalrecords.
The GPAdistributionfor students with SDQ
data is A+ (29.2 percent),A (35.7 percent),
and A- (20.9 percent). Just 14.2 percent
respondedwith a lower GPAor refused to
answerthe question.Amongthe same group
of applicants,62.3 percentreportedbeing in
the top 10 percentof theirhigh school class,
another18.7 percentwere in the top 11-20
percent, 8.4 percent gave a lower ranking,
and a tenth refusedto answer.
Table5 incorporatesthese new variables.
Highschool GPAis positivelyand significantly relatedto elite college admission.In column 1, students with an A+ average have
odds of admissionthat are more than four
timesas highas studentswith a B+averageor
less. In column 2, controllingfor GPAand
other indicatorsof academicmerit,an applicant'sclass rankis positivelyand significantly
related to the probability of admission.
Studentsin the top 10th of their high school
class have more than twice the chance of
being admitted to an elite universitythan
have otherwise comparable students who
rank in the bottom 80 percent. Applicants in
the second decile have 40 percent higher
odds of admission than have lower ranked
candidates. Class rank matters, and it matters
in the direction that Attewell's (2001) simulations predict.
285
The effects of GPAare somewhatstronger
in Model 3 than in Model 2, and the effects
of class rankare about the same. Compared
with the baselinemodel in Table3, attending
a private or religiousschool gives a larger
boost to admissionchances. School-average
AP effects are significantlynegative, but the
relationto admissionoutcomes is somewhat
less steep. Admissionoutcomes are also significantlynegativelyrelatedto school-aggregate SAT scores when average scores are
below 1200, but the overallrelationis marginallyshallowerthan in the baseline.9Tests
that substitute data on class rankand GPA
from high school transcriptsfor theircounterpartsfromthe SDQin Model3 of Table5 produce the same patternof resultsfor GPA,class
rank,and the four high school context variables. The fact that the negativeschool-level
SATand AP effects are preservedwhen class
rank and GPA are included suggests that
these high school contextualmeasuresconstitute resilientand independentbarometers
of students'relativeclassstandingwhen they
are juxtaposedwith individualstudents'performance on the SATand AP tests. At the
same time, these resultsexplicitlyshow the
importanceof high school classrankoverand
above other indicatorsof comparativeacademic performance.
The effect of class rankcould depend on
it
high school academiccontext. Inparticular,
for
less
matter
most
from
acaapplicants
may
demically competitive high schools, where
there are relativelyfew excellentstudentsand
academicdistinctionsamong them are clear.
We hypothesizethat admissionofficersplace
less weight on class rankif applicantscome
fromschoolswherethere is a high concentration of talented students and where differences among the top studentsare smalland
difficultto discern.To examinethis hypothesis, we interactedSDQ class rankwith high
school type, elite72, school-averagenumber
of APtests, and school-averageSATscoreand
added each group of interactiontermsone at
a time to Model 3 in Table 5. Eachset of interactions typically is statistically significant at
the .001 level, and the class-rank gradient is
the steepest for applicants from weaker
schools. Admission officers normally give
more weight to class rank in public high
286
Espenshade,Hale,and Chung
Table 5. Effect of High School GPA and Class Rank on Admission Outcomes: Evidence
from the Student Descriptive Questionnaire (Coefficients shown as odds ratios)
Adding High School GPAand Class Rank
PredictorVariables
(1)
High School GPA
A+
A
A-
(2)
(3)
4.229***
2.837***
2.859***
2.054***
3.308***
2.333***
1.687***
1.424***
1.506***
1.995***
1.573***
1.641***
2.196***
1.399***
2.221***
1.41 6***
1.811***
1.699***
(B + or less)
No response
High School Class Rank
Top 10 percent
Next 10 percent
(Bottom 80 percent)
No response
Typeof High School
(Public)
Private
Religious
Unknown
1.625***
1.152**
1.118
Elite72
(No)
Yes
1.144*
Numberof AP TestsTaken
per High School Senior(n)
(0)
0 < n < 0.4
0.4 5 n < 0.8
0.8 5 n < 1.5
1.5+
Unknown
0.860
0.742**
0.672***
0.630***
0.723***
AverageHigh School SATScore
(< 1000)
1000-1099
1100-1199
1200-1299
1300-1600
Unknown
Number of Cases
LikelihoodRatio Chi-squared (do)
Pseudo R2
0.788***
0.815***
0.920
1.099
1.018
45,549
10239.93(94)
0.2137
45,549
10386.54(97)
0.2168
45,549
10690.24(111)
0.2231
Note: Reference categories are shown in parentheses. All three models also include dummy variables
for 54 geographic areas and two institutions, all other student-level variables in the baseline model, plus
the number of SATIIAchievement (or Subject) tests that a student has taken and the student's average
score on these tests. SATIIdata were provided by ETS.
* p < .05, ** p < .01, ***p < .001.
Source: National Study of College Experience.
The Frog Pond Revisited
schools, in schools where no seniors have
taken any APtest, and where school-average
SATscoresare lowerthan 1000. Forexample,
when SDQclass rankis interactedwith high
school SATscore, the ratio of the odds of
being admittedfrom the top 10 percent of
one's classto the odds of admissionfor applicants in the 80th-90th percentileis 2.2 when
school-averageSATscoresare less than 1000,
1.6 when school-averagescores are in the
1000-1100 range, and about 1.3 when
school-averagescores are higherthan 1100.
DISCUSSION
Most previousresearchon access to postsecondary education has been incomplete
because it has focused on schools that students actuallyattend and has overlookedthe
intervening processes of application and
admissionto college. It is not that gatekeeping functionsof colleges and universitiesare
believedto be unimportantor uninteresting
but, rather,that data on who applies and
who is acceptedaredifficultto obtain(Bowen
and Levin2003; Karen1991). In this article,
we enteredthe "littleunderstoodblackbox"
(Hearn1991:168) in which high school graduates are matchedto particularcollege destinations and investigatedthe role of college
admission officers in the overall sorting
process. Studiesof factorsthat affect admission outcomes have identifiedthe attributes
of individualstudents, includingtest scores,
sex, race, athletic ability,and legacy status.
of secondaryschools have
The characteristics
rarelybeen considered.An importantaim of
this article was to assess the simultaneous
of individualapplirolesof the characteristics
cants and the academic context of high
schools. We examined the selection mechanismsof elite universitiesat the highestechelon of Americanhighereducation.Eliteeducation has been neglected, partlybecause its
students constitute a small fraction of all
undergraduates (Carnevale and Rose 2004).
Moreover, it is in these schools where one
expects the salience of selection criteria to
emerge (Kane 1998). What is most important, prestigious educational institutions at all
levels play critical roles in the transmission of
287
power and privilegeto those in high-status
educationaland careertracks(Cookson and
Persell1985; Kingstonand Lewis1990a).
The origins of our analysisare in relative
deprivationand social comparisontheories.
Variouslyknown as the frog-pond model
(Davis1966) or the big-fish-little-pondeffect
(Marsh1987), the theory highlightsa social
comparisondynamicthat is presumedto animate individualbehavior.In particular,when
individualstudents' ability is controlled, an
increasein schoolwide achievementlevels is
expected to depress academic self-concepts
and loweroccupationaland educationalaspirations. This frameworkhas been used to
evaluatethe effectof affirmativeactionon the
supplyof minorityfacultyat leadingcolleges
and universities(Cole and Barber2003) and
the numberof AfricanAmericansin the legal
profession(Sander2004).
Attewell(2001) adaptedthese concepts to
college admissions. He hypothesized that
institutionalgatekeepers are susceptible to
context effects and that a given student
would have a betterchance of being accepted by an elite college or universityif he or she
attended a high school with a smallerconcentrationof high-achievingpeers.Admission
deans care about high school class rankfor
several reasons. Colleges compete for students with proved scholastic ability, and
being rankedat or nearthe top of one's class
is a way of demonstrating this ability.
Admissionofficersalso know that their institutions' reputationsdepend on enrolling a
largeproportionof high-rankingenteringstudents because the publishedannual standings of U.S. colleges and universitiesfactor
class rank into the formula (Hossler et al.
1999). Theremay also be a "diversityrationale" for emphasizingclass rankin admission
decisions. Its use preventstalented students
at a relativelysmall number of exceptional
secondary schools from dominating a college's first-yearclass(Attewell2001).
Using data on more than 45,000 applications to three highly selective universities, we
found broad support for Attewell's (2001)
hypothesis. Whether a student's relative class
standing is inferred by comparing individual
students' achievement to school-average
achievement levels or measured directly,
288
Espenshade,Hale,and Chung
there is a significantnegative effect of aver- 1975). But high school contextualvariables,
age-studentachievementlevels on the likeli- especiallythose that are relatedto the acadehood of admissionto an elite college. This mic qualityof schools and theirstudentbodconclusion is robust to tests for clustered ies, have not been examined in studies of
observationsand to adjustmentsto school- admission to college. Our four contextual
wide SATscores for the proportionof stu- indicatorsare individuallyand jointlystatistidents who took the SAT.However,class rank callysignificant,but they do not add as much
does not apply with equal strength every- to the explainedvarianceas do the attributes
where. The negativeeffectsof higherschool- of individualstudents.Thesefindingssupport
averageachievementsarethe weakestfor the the school context hypothesisbut also reinmost outstandingstudents.In addition,class force the secondaryimportanceof context in
rankis a less importantcriterionfor students relationto individualstudenteffects.
To what extent does admissionto an elite
who applyfrom schoolswith a high concentrationof academicallyable studentsbecause, college reflectmeritocraticideals? Ourfindin these cases, fine distinctionsamong stu- ings show that academic merit mattersfirst
dents in the uppertail of the distributionare and foremost at highly selective institutions.
difficultto makeand interpret.
Significantlypositiveand monotonicrelations
AlthoughAttewell(2001) did not consider between the chances of admissionand such
them, not all the effects that are associated academic credentials as high school GPA,
with a strongerhigh school academic envi- standardizedtest scores, and the numberof
ronmentare necessarilynegative. Forexam- AP and SATIItests that are taken, together
ple, Marshet al. (2000) showed that per- with the fact that the greatest incrementin
ceived school status has counterbalancing the explainedvarianceoccurswhen measures
effectson acade- of students' academic performance are
positiveor "reflected-glory"
mic self-concept. In addition, despite their included, provide convincing evidence that
negative effects on grades and class rank, meritocraticstandards are being applied.
institutionalselectivityand prestigehave pos- These findings are consistent with prior
itive outcomes for rates of passing the bar researchon access to colleges at particular
(Sander2004), college graduationand later levels of selectivity (Hearn 1984, 1991;
income (Kane 1998; Kingston and Smart Karabeland Astin1975; Karen2002). At the
1990), and the likelihoodof earninga profes- same time, nonacademiccriteriaare also in
sional or doctoral degree (Bowen and Bok play. Admissiondoors are open wider for
1998). We, too, found counterbalancing members of traditionallyexcluded groups,
effects on college admission.Studentswho including women and especially African
attend privateor religioussecondaryschools Americanand Hispanicapplicants.And suband those whose high schools are viewed by stantialpreferencesare extendedto recruited
veteranadmissionofficersas elite are signifi- athletesand childrenof alumni.Some scholcantly more likelyto be admittedto an elite arshave referredto the use of ascriptivecharuniversity.In addition,it is plausiblethat stu- acteristicsas reflectingnonmeritocratictenit maybe
dents learnmore in a moreintellectuallychal- dencies(Hearn1991). Alternatively,
our
to
broaden
of
The
existence
these
environment.
conceptionof merit
necessary
lenging
of studentsthat
all
to
characteristics
include
that
admission
deans
effects
means
positive
must weigh trade-offs between a school's contributeto the mission of an institution
academic reputationand an individualstu- (Bowenand Bok1998).
dent's relativeacademicperformance.A posDespitesome evidencethat eliteeducation
siblesolution,as we noted earlier,is to down- createsthe possibilityof upwardmobilityfor
grade the importance of a candidate's class
rank at the most competitive high schools.
Our results also provide another test of the
"school context hypothesis." The received
wisdom suggests that high school effects are
generally modest (Alexander and Eckland
a few high-achieving students from modest
socioeconomic backgrounds (Levine and
Nidiffer 1996; Persell and Cookson 1985),
research has typically shown that higher education is characterized by persistent inequality (Hearn 1984, 1991; Karabel and Astin
TheFrogPond Revisited
289
1975; Karen2002). Students from higher
socioeconomic (SES)backgroundsare more
likelyto attend more selective colleges and
universities(Kingstonand Lewis1990b). Our
findings generally support this conclusion.
We haveshown that SATscoresand the number of APtests taken are positivelyrelatedto
admissionoutcomes, but academic credentials are also positivelycorrelatedwith SES
(Karabeland Astin 1975). The admission
advantage conferred on alumni children is
anotherindicationthat privilegeis tied across
generations.Parentswho can affordto send
theirsons and daughtersto privatesecondary
schoolsor who livein affluentneighborhoods
whose schools have provedacademicreputations give theirchildrena leg up in the competition for spots at prestigiousuniversities.
We should remarkin passing that evidence
that being in an elite track in secondary
school enhancesadmissionto an elite college
begins to establishthe "socialarithmetic"for
how high-status tracks are linked across
school settings (Kingston and Lewis
1990a:xxv).Finally,in an expandedversionof
this article,we showed that although some
elite universitiespractice economic affirmative actionfor studentsin the lowest SEScategories-first-generation college students
and those from families whose annual
incomes are lower than $30,000-this small
basis
advantageon an all-other-things-equal
is not sufficientto overcome the fact that
higher-SESstudents have been consistently
favored over ones from lower-SESbackgrounds in simple tabulationsof admission
rates by social class (Espenshade,Hale, and
Chung2004).
We have shown that high school academic context and a student'srelativeclassstanding matter to admissiondeans at elite colleges and universities. To flesh out the
remainingfeaturesof the blackbox that connects high school students to particularcollege destinations,researchersshould investigate the relativeimportanceof individualstu-
dents' undergraduate institutions affect
admission to graduate and professional
schools.
Contextualeffectsmay be relevantto institutional gatekeepersoutside education.One
fruitfularea to explore is the job
particularly
market.In his review of this article,Herbert
Marshsuggested that all but the top students
in Davis's(1966) originalstudy perhapshad
good reasonto aim for lower academic-performance careersto the extent that evaluators of job applicationsuse selection criteria
that aresimilarto those of evaluatorsof applications to colleges and universities.In addition, afterexaminingthe job marketfor new
lawyers, Sander (2004) concluded that
employersof lawyersin both privatelawfirms
and governmental agencies place more
weight on grades than on the eliteness of
schools unless the lawyers graduated from
Yaleor Harvard.On the other hand, Kingston
and Smart(1990:169) emphasizedthe generally greater effects of prestige on early
careerearningsfrom attending a more elite
institution:"Tostate these findings in terms
of financialadvice,we are inclinedto say:go
to an elite college (and don't worry about
your grades unlessyou want to go to a professionalschool), but if you can'tget into one
of these schools, don'tworrytoo muchabout
where else to go because the name of your
alma materwon't be important."These disparate conclusions suggest the following
hypothesis:The relativeimportanceof frogpond effects to institutionalgatekeepersmay
depend on the particularorganizationalsetting, on who is doing the evaluating,and on
how essentialhigh intellectualabilityis to the
mix of skillsthat are being sought.
dents' attributes versus high school characteristics at other levels of college selectivity.
Furtherwork is needed, too, on students and
their decisions about where to apply and
where to enroll once they are accepted, as
well as on whether the characteristics of stu-
dentiality of all information on students.
2. More accurately, 45,549 represents the
number of applications, not the number of
applicants. A small number of students
applied to more than one of these institutions.
NOTES
1. Inexchangefor these schools' participation, we promisedto protect the identityof
each institutionand to safeguardthe confi-
290
Espenshade,Hale, and Chung
3. When the variablesare expressed in the proportionof applicantsthey accept,and
continuous form, the correlation between we controlledfor it with institutionaldummy
individualstudents'SATscoresand the num- variablesin the main model. In addition,we
ber of AP examinationsis 0.46. The correla- estimatedsome variantsof Model 4 to test
tion between their school-averagecounter- whetherthe inclusionof informationon SAT
partsis 0.62. Neitheris largeenough to sug- (or AP)adds anythingto the explainedvariance beyond AP (or SAT)variables.The 12
gest problemswith collinearity.
4. Amongapplicantsto these three univer- SATvariables(7 at the student leveland 5 at
sities, the average SATI score and average the school level) are jointlysignificantwhen
numberof AP tests, respectively,were 1208 they are added to a versionof Model4 that
and 2.1 for blacks, 1245 and 3.1 for excludes them, whether or not AP variables
Hispanics,1359 and 3.5 for whites,and 1371 are included.The 15 APvariables(10 at the
and 4.3 for Asians.The numbersfor athletes student level and 5 at the school level) are
were 1294 and 2.5 versus 1346 and 3.7 for jointlysignificantwhen they are added to a
version of Model 4 that excludes them,
nonathletes.
5. Baltzell(1958) identified16 of the most whetheror not SATvariablesare included.All
socially prestigious American boarding the resultsare statisticallysignificantat the
schools-what Cookson and Persell(1985) .0001 level.Therefore,the SATvariablesmeacalledthe "select16." Nearly3 percentof the suresomethingdifferentabout individual-stuapplicantsin our sampleattendedone of the dent and school-averageachievementfrom
select 16 schools, and 26.6 percentof them theirAPcounterparts.Put differently,the set
were accepted by an elite university,com- of SATvariablesadds explanatorypower on
pared with 21.8 percent of all other candi- top of the AP variables,and vice versa.As a
dates. We created a dummy variablefor the consequence,it is preferableto includeboth
select 16 schoolsand substitutedit for elite72 constructsin models of the effect of high
in Model 4. Its regressioncoefficientimplies school academiccontext.
7. We are gratefulto GermanRodriguez
that students from one of these schools are
for
are
4.3 percentlesslikelyto be acceptedthan
suggestingthis model.
It is not clearwhy studentswho attend
the
8.
but
16
from
nonselect
schools,
applicants
effect is not statisticallysignificant(p-valueof high schools whose average SATscores are
0.609). Nor does the select 16 variableper- unknownhave such a largeadmissionadvanform as well when it is substitutedfor elite72 tage over other students. One possibilityis
in Model 3. Studentswho attend select 16 that the unadjusted school-average SAT
schools are 9 percent more likelythan are scores were inflated (as suggested by the
other applicants to be admitted, but the example in the previous paragraph)by an
effect is not significant(p-valueof 0.251).
averageof 52 pointsusing the 2.53 slope cri6. Model4 was fit separatelyto data from terion and by 79 points using 3.79. After
each institution.All three universitiesexhibit adjustment,schoolswhose averagescoresare
qualitativelysimilarbehaviors,as one would unknownand thereforecannot be adjusted
expect, given the common aimsand valuesof appearto be relativelyless competitivethan
the institutionsand the wide varietyof stu- before. A negative school SATeffect makes
dents from which to choose. Underrepre- students at these schools now seem more
sented minoritystudents, women, athletes, attractiveto admissionofficers.The fact that
legacies, and applicantswith high academic the boost to odds of admissionwhen school
achievement have an admissionadvantage. SAT scores are unknown is higher when
In addition, preferenceis normallyextended adjustmentsto school-averageSATscoresare
to students from private and elite high
schools and to candidates whose high school
academic environment is such that it permits
them to distinguish themselves academically
from their classmates. The most important
way in which the three universities differ is in
larger (see column 4) is consistent with this
interpretation.
9. In results that are not reported in Table
5, estimated odds ratios for individual-level
variables are similar across all three models.
But there are important differences with prior
291
TheFrogPond Revisited
results from the baseline model. Stronger
admission preferences are estimated for applicants who are black, Hispanic, athletes, or
legacies. Lessemphasis is placed on individual
SAT scores and the number of AP tests, as
indicated by the flatter slopes to the pattern
of odds ratios. This outcome is at least partially explained by the inclusion of additional
measures of academic merit, principally GPA
and SAT II (Achievement) tests, with which
SATscore and AP tests are positively correlated. Taking more than three SATII tests and
scoring well on them are positively and significantly related to admission outcomes.
REFERENCES
Alexander, Karl L., and Bruce K. Eckland. 1975.
"Contextual Effects in the High School
Attainment Process." American Sociological
Review40:402-16.
1977. "High School Context and College
Selectivity: Institutional Constraints in
Educational Stratification." Social Forces
56:166-88.
American College Testing Program. 1997. "ACT
Average Composite Scores by State, 1997 ACTTested Graduates."Availableon-line at http://
www.act.org/news/data/97/97states.html
Attewell, Paul. 2001. "The Winner-Take-AllHigh
School: Organizational Adaptations to
Educational Stratification." Sociology of
Education74:267-95.
Baltzell, E. Digby. 1958. PhiladelphiaGentlemenTheMakingof a National UpperClass.Chicago:
Quadrangle Books.
Bowen, William G., and Derek Bok. 1998. The
Shape of the River:Long-TermConsequencesof
Considering Race in College and University
Admissions.Princeton, NJ:Princeton University
Press.
Bowen, William G., and Sarah A. Levin. 2003.
Reclaiming the Game: College Sports and
Educational Values. Princeton, NJ: Princeton
UniversityPress.
Carnevale,Anthony P., and Stephen J. Rose. 2004.
"Socioeconomic Status, Race/Ethnicity, and
Selective College Admissions." Pp. 101-56 in
America's Untapped Resource: Low-Income
Studentsin HigherEducation,edited by Richard
D. Kahlenberg.New York:Century Foundation
Press.
Cole, Stephen, and ElinorBarber.2003. Increasing
Faculty Diversity:The Occupational Choices of
High-AchievingMinorityStudents. Cambridge,
MA: HarvardUniversityPress.
College Board. 2003, August 26. SAT Verbaland
Math Scores Up Significantlyas a Record-breaking Number of Students Take the Test (Report
N0218). New York:Author.
Cookson, Peter W., Jr.,and Caroline Hodges Persell.
1985. Preparing for Power: America's Elite
BoardingSchools. New York:Basic Books.
Curry,Wade, Walt MacDonald, and Rick Morgan.
1999. "The Advanced Placement Program:
Access to Excellence," journal of Secondary
GiftedEducation11:17-22.
Davis, James A. 1966. "The Campus as a Frog
Pond: An Application of the Theory of Relative
Deprivation to Career Decisions of College
Men." Americanjournal of Sociology72:17-31.
Duffy, Elizabeth A., and Idana Goldberg. 1998.
Crafting a Class: College Admissions and
Financial Aid, 1955-1994. Princeton, NJ:
Princeton UniversityPress.
Espenshade, Thomas J., Chang Y. Chung, and Joan
L. Walling. 2004. "Admission Preferences for
Minority Students, Athletes, and Legacies at
Elite Universities." Social Science Quarterly
85:1422-46.
Espenshade,ThomasJ.,LaurenE.Hale,and Chang
Y. Chung. 2004, September. "High School
Academic Environment, Class Rank, and Elite
College Admission." Unpublished manuscript,
Office of Population Research, Princeton
University.
Hauser,Robert M., WilliamH. Sewell, and Duane F.
Alwin. 1976. "High School Effects on
Achievement." Pp. 309-41 in Schooling and
Achievement in American Society, edited by
WilliamH. Sewell, RobertM. Hauser,and David
L. Featherman.New York:Academic Press.
Hearn, James C. 1984. "The Relative Roles of
Academic, Ascribed, and Socioeconomic
Characteristics in College Destinations."
Sociologyof Education57:22-30.
1.1991. "Academicand Nonacademic Influences
on the College Destinations of 1980 High
School Graduates." Sociology of Education
64:158-71.
Hossler, Don, Jack Schmit, and Nick Vesper. 1999.
Going to College: How Social, Economic, and
Educational Factors Influence the Decisions
Students Make. Baltimore: Johns Hopkins
UniversityPress.
Kane,Thomas J. 1998. "Racialand EthnicPreferences
in College Admissions." Pp. 431-56 in The
Black-White Test Score Gap, edited by
Christopher Jencks and Meredith Phillips.
Washington, DC: BrookingsInstitutionPress.
Karabel, Jerome, and Alexander W. Astin. 1975.
292
Espenshade,Hale,and Chung
Cross-Cultural(26-Country) Test of the
"SocialClass,AcademicAbility,and College
Negative Effects of AcademicallySelective
'Quality.'"SocialForces53:381-98.
58:364-76.
Schools."American
Karen, David. 1990. "Toward a PoliticalPsychologist
Herbert
Olaf
and
The
of
Model
Kller,
W.,
Marsh,
Gatekeeping:
Organizational
JirgenBaumert.
of Eastand WestGerman
2001. "Reunification
Case of EliteColleges."Sociologyof Education
School Systems: Longitudinal Multilevel
63:227-40.
-.
1991. "'Achievement' and 'Ascription' in
Modeling Study of the Big-Fish-Little-Pond
Effecton AcademicSelf-Concept."American
Admissionto an Elite College: A PoliticalEducational
Research
38:321-50.
Journal
OrganizationalAnalysis."SociologicalForum
6:349-80.
Marsh,HerbertW., Chit-KwongKong,and Kit-Tai
Multilevel
Modelsof
Hau.2000. "Longitudinal
-. 2002. "Changesin Accessto HigherEducation
Effecton AcademicSelfthe Big-Fish-Little-Pond
in the UnitedStates:1980-1992." Sociologyof
75:191-210.
Education
Concept: CounterbalancingContrast and
ReflectedGloryEffectsin HongKongSchools."
Kingston,PaulWilliam,and LionelS. Lewis.1990a.
"Introduction:Studying Elite Schools in
Journalof Personalityand Social Psychology
78:337-49.
Track:
America."
Pp.xi-xxxivin TheHigh-Status
edited Marsh,HerbertW., and John W. Parker.1984.
Studiesof EliteSchoolsand Stratification,
"Determinantsof Student Self-Concept:Is It
by PaulWilliamKingstonand LionelS. Lewis.
Betterto Be a RelativelyLargeFishin a Small
of New YorkPress.
Albany:StateUniversity
Pond Even If You Don't Learnto Swim as
at EliteInstitutions:
-. 1990b. "Undergraduates
Well?" Journal of Personalityand Social
The Best,the Brightest,and the Richest."Pp.
47:213-31.
105-20 in The High-StatusTrack:Studiesof
Psychology
edited by Paul Mathews,Jay.1998. ClassStruggle:What'sWrong
EliteSchoolsand Stratification,
WilliamKingstonand LionelS. Lewis.Albany:
(and Right)with America'sBest PublicHigh
Schools.New York:TimesBooks.
StateUniversityof New YorkPress.
Kingston,PaulWilliam,and JohnC. Smart.1990. McDonough,PatriciaM. 1997. ChoosingColleges:
How Social Class and Schools Structure
"The Economic Pay-Off of Prestigious
Opportunity.
Albany:State Universityof New
Colleges." Pp. 147-74 in The High-Status
YorkPress.
Track:
Studiesof EliteSchoolsand Stratification,
edited by PaulWilliamKingstonand LionelS. Persell,CarolineHodges,and PeterW.Cookson,Jr.
1985. "Chartering and Bartering: Elite
Lewis.Albany:State Universityof New York
Educationand Social Reproduction."Social
Press.
Problems33:114-29.
Levine,Arthur,and JanaNidiffer.1996. Beatingthe
Odds: How the Poor Get to College. San Sander,RichardH. 2004. "A SystemicAnalysisof
Affirmative
Actionin AmericanLawSchools."
Francisco:
Jossey-Bass.
StanfordLawReview57:367-483.
Lillard,Dean, and JenniferGerner.1999. "Getting
to the Ivy League:How FamilyComposition Santoli, Susan P. 2002. "Is There an Advance
PlacementAdvantage?"AmericanSecondary
Affects College Choice." Journalof Higher
Education
30:23-35.
Education
70:706-30.
Manski, Charles F., and David A. Wise. 1983. Shulman,JamesL., and WilliamG. Bowen.2001.
The Game of Life: College Sports and
CollegeChoicein America.Cambridge,MA:
EducationalValues.Princeton,NJ: Princeton
HarvardUniversityPress.
Marsh,HerbertW. 1987. "TheBig-Fish-Little-Pond UniversityPress.
Effecton AcademicSelf-Concept."Journalof Snyder,Thomas D., and CharleneM. Hoffman.
2000. Digest of EducationStatistics,1999.
79:280-95.
Educational
Psychology
to
Schools
1991.
of
"Failure
-.
Washington, DC: National Center for
High-Ability
High
Education Statistics, U.S. Department of
Deliver Academic Benefits Commensurate
Education.
with TheirStudents'AbilityLevels."American
An
Research
Educational
Toor, Rachel. 2001. AdmissionsConfidential:
Journal28:445-80.
Insider'sAccountof the EliteCollegeSelection
Marsh, Herbert W., Danuta Chessor, Rhonda
Process.New York:St. Martin'sPress.
Craven, and LawrenceRoche. 1995. "The
Effectsof Gifted and TalentedProgramson Yaqub, Reshma Memon. 2002, September.
AcademicSelf-Concept:The Big Fish Strikes
"Getting Inside the Ivy Gates." Worth,pp.
94-104.
Research
journal
Again."AmericanEducational
32:285-319.
Zwick, Rebecca. 2002. Fair Game? The Use of
Standardized Admissions Tests in Higher
Hau.2003. "Big-FishMarsh,HerbertW.,and Kit-Tai
New York:RoutledgeFalmer.
Education.
A
Little-Pond
Effecton AcademicSelf-Concept:
The Frog Pond Revisited
293
of Sociologyand FacultyAssociateat the Officeof Population
ThomasJ. Espenshadeis Professor
Hisresearchinterestsincludehighereducationin the UnitedStates,
Research,PrincetonUniversity.
the racialdimensionsof collegeadmissionand campuslife, intergrouprelationson collegecamand is a CoHeis directingthe NationalStudyof CollegeExperience
puses,andsocialdemography.
Life
in
Student
on
the
America
Survey.
Principal
Investigator
Campus
in 2003. She is curLaurenE. Halereceivedher Ph.D.in publicaffairsfromPrincetonUniversity
of New York,StonyBrook.
of Preventive
Medicineat the State University
rentlyAssistantProfessor
and relatedtopicsin socialdemography.
Herresearchincludeshighereducation,healthdisparities,
Princeton
at the Officeof PopulationResearch,
Chang Y.Chungis a SeniorStatisticalProgrammer
and
a
has
a
in
from
the
of
South
Carolina
He
Ph.D.
master's
University.
University
degree
sociology
in systemsengineeringfrom the Universityof Pennsylvania.He is currentlyinvolvedin major
researchprojectson migrationin Thailandand highereducationin the UnitedStates.
and CenterGrantP30
Supportforthisresearchwasprovidedby theAndrewW.MellonFoundation
HD32030fromthe NationalInstituteof ChildHealthand HumanDevelopment.Theauthorsthank
Don Betterton,EmilyEspenshade,Jean Grossman,Fred Hargadon,MichaelHout, Stephen
KeithLight,Janet Rapelye,lake Rosenfeld,ChrisSimmons,Christopher
Weiss,and
LeMenager,
and
CharlesWestofffor theirusefulcommentsand suggestions;Scott Lynch,GermdnRodriguez,
BruceWesternfortheirmethodological
advice;ElanaBroch,JesseRothstein,the CollegeBoard,the
fordata; Alexandria
Waltonforherresearch
Educational
TestingService,and the NSCEinstitutions
foradministrative
assistance;and MelanieAdams,JudithTilton,and KristenTurner
support.Direct
to Thomasi. Espenshade,Officeof PopulationResearch,249 WallaceHall,
all correspondence
PrincetonUniversity,
Princeton,NJ08544-2091; e-mail:[email protected].