APPLY YOURSELF: RACIAL AND ETHNIC

DIFFERENCES IN COLLEGE APPLICATION

Sandra E. Noir

Department of Economics

Columbia University

New York, New York 10027

sblack@columbia.edu

Kalena E. Cortes

(corresponding author)

The Bush School of

Government and Public

Service

Texas A&M University

College Station, Tx 77843

kcortes@tamu.edu

Jane Arnold Lincove

School of Public Policy

University of Maryland

Baltimore County

Baltimore, MARYLAND 21250

jlincove@umbc.edu

Abstrait
Access to higher education begins with a student’s decision
whether and where to apply to college. This paper examines
racial and ethnic diﬀerences in college application behavior of
high school graduates, using two recent graduation cohorts from
Texas. We estimate racial and ethnic diﬀerences in the probability
of applying to college, controlling for a student’s college readi-
ness, high school quality, certainty of college admissions, et
high school ﬁxed eﬀects. We then investigate racial and ethnic
diﬀerences in the choice of where to apply. We enhance the typ-
ical model of college matching by considering the social setting
and high school feeder patterns of state universities. We ﬁnd that
racial and ethnic gaps in application rates, particularly for His-
panic students, are not explained by diﬀerential levels of college
readiness, high school quality, or information regarding college
admission processes. When applying to college, minorities are
inﬂuenced by more than just matching their academic ability to
the institution, and prefer institutions with a large proportion
of same-race students and campuses where same-race students
from their high school have been successful in the past.

https://doi.org/10.1162/edfp_a_00273

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
e
d
p
un
r
t
je
c
e
–
p
d

F
/

1
5
2
2
0
9
1
6
9
3
4
7
2
e
d
p
_
un
_
0
0
2
7
3
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

209

Apply Yourself

INTRODUCTION

1 .
Substantial disparities in college enrollment rates exist across racial and socioeconomic
groups in the United States. Nationally, only 62 percent of black and 60 percent of
Hispanic high school graduates enrolled in college the fall semester after high school
graduation, compared with 71 percent of white graduates (NCES 2016). A key goal of
education policy is to help remediate this inequity by providing equal opportunity and
access to all students. Cependant, to develop better policy, one needs to understand the
underlying causes of these disparities. There are many possible explanations, inclure-
ing lack of information and diﬀerent preferences. Selective admissions processes can
limit college access among minorities if high school quality and high school outcomes
are unequally distributed by race and ethnicity. Even before a student can enter an ad-
missions process, she must apply—a process that may be particularly daunting for a
potential ﬁrst-generation college student. Although there are a few studies that inves-
tigate the admissions and enrollment process for minority students, most begin with
a sample of students who express college interest by taking a college entrance exam
(SAT or ACT). Dans cette étude, we ﬁrst investigate racial and ethnic diﬀerences in the
initial step in college enrollment—the application decision—using two full cohorts of
recent Texas high school graduates. Because there is a common college application
in Texas (see www.ApplyTexas.org, hereafter referred to as ApplyTexas), we are able to
observe whether each student applied to each of the thirty-seven, four-year public post-
secondary institutions in the state. Surtout, because we observe the population of
students and not just those students who applied or took the SAT or ACT exam, nous
can learn more broadly about diﬀerences in college application behavior. We do so in
order to assess the diﬀerential roles of student demographics, high school quality, et
student preparedness in the college application decision. Surtout, we also allow for
eﬀects of race to diﬀer based on student and high school characteristics.

Once we have done this, we then examine the decision of where to apply, condi-
tional on applying. When deciding to apply to a college, students can incorporate a
number of diﬀerent elements, including their level of preparedness (which could be
a function of their high school quality and their performance in high school, famille
characteristics and ﬁnancial factors, as well as the probability of acceptance.) We focus
on understanding the importance of student–campus ﬁt by estimating the inﬂuence of
academic match, campus demographics, and high school feeder patterns on behavior
regarding where students apply to college.1

One possible explanation for the observed racial and ethnic disparities is that stu-
dents have incorrect assessments of the likelihood of gaining admission to a college,
and this varies by race and ethnicity. To address this possibility, we take advantage of a
unique institutional feature of the Texas public university system. Texas is one of several
states that uses both “percent plan” and holistic admissions policies. During the time
period under study, students who graduated from the top 10 percent of their senior class
at all Texas public high schools were eligible for automatic admissions to all Texas public
universities, including the highly selective ﬂagship campuses of the University of Texas

1. Note that the decision to apply to a particular college could depend on the characteristics of the other colleges
to which the student applies. Although this is likely to be the case, our estimation strategy does not allow for
this type of dependence.

210

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
e
d
p
un
r
t
je
c
e
–
p
d

F
/

1
5
2
2
0
9
1
6
9
3
4
7
2
e
d
p
_
un
_
0
0
2
7
3
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Sandra E. Noir, Kalena E. Cortes, and Jane Arnold Lincove

at Austin (UT Austin) and Texas A&M University at College Station (TAMU). Academic
slots not ﬁlled by automatically-admitted students were given to students who under-
went a more traditional holistic admissions process that included race as one of many
background factors. Top 10 percent students were informed during their junior year of
high school of their guaranteed admissions status and provided with information about
campus options (Top Ten Percent Plan), whereas students below the top 10 pour cent
faced the uncertainty and costs associated with a typical college search and application
processus. Ainsi, comparing the application choices of top 10 percent graduates, who were
guaranteed college admission, against all high school graduates is particularly useful
for understanding the interaction of race/ethnicity and admissions uncertainty. Enfin,
we examine whether black and Hispanic students respond diﬀerently than white and
Asian students to the Top Ten Percent Plan in terms of application choices to assess
whether perfect information about admissions might reduce inequalities in access. Nous
ﬁnd the largest eﬀects of automatic admissions on Hispanic students, suggesting that
admissions information and policies might be a pathway to improve college quality for
these students.

There are several key ﬁndings from our analysis. D'abord, we ﬁnd that lower Hispanic
college enrollment begins with a lower propensity to apply to college—that is, Hispanic
students are less likely to apply to college than all other races, despite having higher av-
erage college readiness than black students. These results are robust to controls for
student-level college readiness measures, high school characteristics, and high school
ﬁxed eﬀects. Deuxième, black students are in fact more likely to apply to college than stu-
dents of other races with similar college readiness and high school quality, but this eﬀect
is concentrated among black students who are less prepared for college. Enfin, we ﬁnd
racial and ethnic diﬀerences in the selection of college applied to, with black and Asian
students preferring campuses with students within their own racial group, and black
and Hispanic students responding to the prior experiences of their high school peers
on the campus. Ainsi, inequality in college access and quality begins with the decision
to apply to college, and is inﬂuenced by college factors including student demographics
and past high school feeder patterns.

The paper unfolds as follows. Section 2 provides a brief overview of related liter-
ature. Section 3 describes the data. Section 4 examines racial and ethnic diﬀerences
in college application decisions, controlling for individual preparation and high school
characteristics. Section 5 allows for heterogeneity in the decision to apply to college
by race and ethnicity based on high school characteristics and individual preparedness.
Section 6 then examines the decision of which college to apply to, conditional on apply-
ing, as a function of student, high school, and college characteristics, focusing primarily
on student racial and academic match. Enfin, we examine the role of information by
exploiting variation in student admission expectations induced by the Top Ten Percent
Plan. Section 7 includes a discussion of these results.

2 . R E L AT E D L I T E R AT U R E
When considering the application behavior of underrepresented minorities, it is im-
portant to consider the context in which these decisions are made. Most signiﬁcantly,
college decisions are made in light of existing aﬃrmative action policies; in Texas,

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
e
d
p
un
r
t
je
c
e
–
p
d

F
/

1
5
2
2
0
9
1
6
9
3
4
7
2
e
d
p
_
un
_
0
0
2
7
3
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

211

Apply Yourself

Par exemple, the admissions environment includes both holistic admissions, dans lequel
race and ethnicity can be a factor, and race-neutral automatic admissions through the
Top Ten Percent Plan (Dickson 2006; Daugherty, Martorell, and McFarlin 2014; Noir,
Cortes, and Lincove 2015). En outre, college application decisions are integrally re-
lated to issues of college mismatch; c'est, whether students choose colleges that are
the best ﬁt for their level of ability and interests (Walton and Cohen 2007; Arcidiacono
and Lovenheim 2016; Cortes and Lincove 2016, 2019).2 Par conséquent, our work relates to a
number of literatures on the application decisions of minority students. One strand of
literature examines how the presence or absence of aﬃrmative action changes college
application and enrollment decisions. Mark Long (2004un) uses data from the College
Board containing a 10 percent random sample of SAT takers, and considers how white
and minority applications changed with the elimination of aﬃrmative action in Cali-
fornia and Texas. He ﬁnds that both minority and nonminority students changed their
behaviors; white and Asian students increased the selectivity of schools to which they
appliqué, whereas underrepresented minority students applied to less selective schools.
Although he cannot observe the actual colleges to which the student applied, he can
observe SAT score reports. He ﬁnds that, relative to white and Asian Americans, et-
derrepresented minorities signiﬁcantly lowered the number of score reports sent to
in-state public colleges, regardless of quality; these results are slightly stronger in Cal-
ifornia than in Texas.3 Card and Krueger (2005) also use a sample of SAT takers to
examine how application decisions responded to the elimination of aﬃrmative action
and ﬁnd little evidence of a change in the propensity of high-ability minorities to apply
to at least one selective public institution in California and Texas.

Whereas the aforementioned papers analyze the implications of changes in aﬃr-
mative action policies on college application behaviors, we focus on a variety of other
factors, such as diﬀerences in student preparation and diﬀerences in high school qual-
ville, using data on the full population of high school graduates in Texas; we are thus
not limited to students who reveal their college preferences by taking entrance exams.
En outre, we consider black and Hispanic students separately while prior studies
consider both groups in an aggregate category of underrepresented minorities. A lim-
itation of our work is that it is purely descriptive in nature (and does not, Par exemple,
take advantage of a natural experiment for identiﬁcation.) Par conséquent, we view our study
as a complement to the existing work, not a substitute.

2. See Page and Scott-Clayton (2016) for an excellent review of the literature on additional barriers to college

access.

3. Considering the role of the Texas Top Ten Percent Plan on minority decision making, Niu, Tienda, and Cortes
(2006) analyze a representative survey of Texas high school seniors in the spring of 2002, who were rein-
terviewed one year later to evaluate diﬀerences in college preferences and enrollment decisions, according to
three criteria targeted by the Top Ten Percent Plan: high school type, class rank, and minority status. Ils
ﬁnd that Texas seniors, and top 10 percent graduates in particular, consider institutional selectivity in making
college choices. De plus, graduates from feeder and resource-aﬄuent high schools are more likely to prefer
selective institutions, whereas graduates from resource-poor high schools and high schools that rarely feed to
selective state universities are less likely to choose selective institutions as their ﬁrst preference. Aussi, both for
ﬁrst college preference and enrollment decisions, black and Hispanic students are less likely than white stu-
dents to opt for selective universities. Mark Long (2004b) uses data from the National Education Longitudinal
Étude (NELS) and documents that there would be substantial changes in the racial composition of universities
following the elimination of aﬃrmative action, even holding application behavior constant.

212

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
e
d
p
un
r
t
je
c
e
–
p
d

F
/

1
5
2
2
0
9
1
6
9
3
4
7
2
e
d
p
_
un
_
0
0
2
7
3
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Sandra E. Noir, Kalena E. Cortes, and Jane Arnold Lincove

There is also a recent and inﬂuential literature examining underserved popula-
tions and their decisions to apply to selective institutions (Griﬃth and Rothstein 2009;
Hoxby and Avery 2013; Hoxby et Turner 2013; Forgeron, Pender, and Howell 2013; Pal-
lais 2015; Dillon and Smith 2017b). En général, this literature examines a sample of
students who are applying to college and tries to understand the college choices they
make. Work by Hoxby and Avery (2013) focuses on high-ability high school students
to examine the phenomenon of “undermatching.” They show that low-income, haut-
achieving students who are likely to undermatch are those who are more isolated from
other achieving students; those who do not undermatch tend to be highly concentrated
in a small number of high schools. Pallais (2015) documents that low-income students
may be more sensitive to application costs; she shows that a decline in the cost of ap-
plying to an additional college, generated through a decline in the cost of sending one’s
ACT scores to an additional university, leads students to apply to more selective uni-
versities. Enfin, Dillon and Smith (2017b) use the National Longitudinal Survey of
Youth 1997 to document the degree of “mismatch” among students in the survey. Im-
portantly, these data include a variety of demographic characteristics, and the authors
use a unique deﬁnition of mismatch—the diﬀerence between a student’s percentile in
the ability distribution and the percentile of her college in the student-weighted distri-
bution of a college quality index. They ﬁnd that a large fraction of students is either
undermatched or overmatched, and this is due to student decisions. They also ﬁnd that
features of the state university system the student faces aﬀect the probability of mis-
match. Our research focuses on racial and ethnic diﬀerences, considers the population
of high school graduates, and examines a much broader set of institutional character-
istics on which students may base their college application decisions.4

Enfin, there is also a recent literature examining racial and ethnic diﬀerences in
college enrollment. Reardon, Boulanger, and Klasik (2012) document diﬀerences in enroll-
ment patterns in highly selective colleges by race/ethnicity and income, and ﬁnd that
both black and Hispanic students are underrepresented in the most selective colleges,
even after controlling for family income. Clotfelter, Ladd, and Vigdor (2015) use data
from North Carolina to examine the role of the public university system in explaining
the racial disparities in college enrollment and completion. They document signiﬁcant
racial and socioeconomic disparities in the likelihood of obtaining a four-year college
degree at a University of North Carolina campus. Cependant, these disparities can be
fully explained by diﬀerential student readiness; once they control for eighth-grade test
scores, black students are more likely to enroll and succeed than non-Hispanic white
students, largely due to the presence of historically black universities.

Our study builds on this literature using the population of students from Texas,
with a focus on the racial and ethnic diﬀerences in college application behavior. En addition-
dition, because we have such a large sample, we are able to estimate diﬀerences in
application behavior across students within the same high school. Enfin, Texas’s auto-
matic admissions policy enables us to examine the issue of application behavior for a

4. More recent work by Dillon and Smith (2017un) also shows that student ability and college quality strongly
improve degree completion and earnings; this work suggests, on average, students beneﬁt from “overmatch”
of the sort generated by aﬃrmative action in admissions.

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
e
d
p
un
r
t
je
c
e
–
p
d

F
/

1
5
2
2
0
9
1
6
9
3
4
7
2
e
d
p
_
un
_
0
0
2
7
3
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

213

Apply Yourself

subsample of high school graduates who are guaranteed college admission, thereby
ruling out mistaken beliefs in the probability of acceptance as an explanation for diﬀer-
ential college application behavior.

3 . DATA
The data sources for this study were collected by the Texas Workforce Data Quality Ini-
tiative (WDQI) at UT Austin, funded by the United States Department of Labor. Le
dataset includes high school enrollment and performance measures for all Texas pub-
lic school students who graduated in 2008 et 2009. High school measures of college
readiness (such as type and number of courses completed, and performance on high
school exit exams) and basic demographics (race and ethnicity, eligibility for free lunch,
English proﬁciency) were obtained from high school academic records. The WDQI
database improves on datasets from prior studies of race/ethnicity and admissions by
including all high school graduates rather than just those who expressed interest in col-
lege and by including a suﬃcient minority population to disaggregate eﬀects for blacks
and Hispanics. En outre, links between high school and university administrative
datasets allow for estimation of within high school diﬀerences by race and ethnicity, comme
well as measurement of a high school’s historical feeder relationship with a university
campus. Enfin, the Texas context allows us to directly test the role of admissions un-
certainty by comparing automatically admitted students to students who must undergo
holistic admissions.

The two high school graduation cohorts include over 490,000 individuals. Reﬂect-
ing the diversity of Texas, 44 percent of high school graduates in our analysis are white,
39 percent Hispanic, 14 percent black, et 4 percent Asian American.5 We are able
to include approximately 430,000 graduates in our regressions after excluding those
with incomplete data and high schools that graduated fewer than ﬁve students within
a year.6

Tableau 1 displays post–high school college enrollment choices by race and ethnicity.
Both black and Hispanic students are underrepresented in enrollment at elite state
ﬂagship universities, other selective state universities, Texas private universities, et
out-of-state universities.7 Black students make up some ground in total enrollment at
open-enrollment state universities, and Hispanic students are most likely to not enroll
in any type of higher education. Dans cette étude, we examine whether these inequalities
begin with college application behavior.8

5. Appendix table A.1 provides additional summary statistics by various college application behaviors.
6. Typiquement, missing data are college readiness measures for students who transferred into Texas public schools
in their junior or senior year. These students have incomplete high school coursework records and were not
required to take the Texas high school exit exam.
In table 1, and throughout our analysis, we identify three types of Texas public universities based on the se-
lectivity of their holistic (non–top 10 pour cent) admissions processes. Texas ﬂagship universities (UT Austin
and TAMU) are the state’s elite campuses and considered the most selective, “other universities with selective
admissions” are non-ﬂagship campuses where non–top 10 percent applicants undergo a holistic admissions
revoir, “public universities with open enrollment” admit all applicants who meet minimum criteria and no
qualiﬁed applicants are rejected.

8. The cost of postsecondary education in Texas is about the same across public universities, and it is signiﬁ-
cantly more expensive for a Texas resident to enroll out-of-state. Par exemple, dans 2006, the total cost of atten-
dance (tuition, fees, plus room and board) ranged from $11,919–$12,845 at Texas ﬂagship universities, et de

214

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
e
d
p
un
r
t
je
c
e
–
p
d

F
/

1
5
2
2
0
9
1
6
9
3
4
7
2
e
d
p
_
un
_
0
0
2
7
3
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

Sandra E. Noir, Kalena E. Cortes, and Jane Arnold Lincove

Tableau 1. Summary Statistics—–Racial and Ethnic Composition of High School Graduates by College Enrollment

Other Texas
Public
Universities
with Selective
Admissions

(2)

0.474

0.115

0.337

0.070

Texas
Flagship

(1)

0.638

0.049

0.188

0.123

Texas Public
Universities
with Open
Enrollment

Texas Private
4-Year
College or
University

Out-of-State
4-Year
College or
University

Texas Public
2-Year
Collège

(3)

0.438

0.403

0.136

0.020

(4)

0.584

0.126

0.234

0.053

(5)

0.622

0.180

0.139

0.053

(6)

0.468

0.125

0.369

0.033

Not
Enrolled

(7)

0.351

0.140

0.482

0.023

Blanc (non-Hispanic)

Noir

Hispanic

Asian

Number of observations

23,836

66,744

14,233

21,155

18,660

128,341

217,738

Note: Based on enrollment in fall semester following high school graduation.

Sources: Authors’ calculations from Texas Workforce Data Quality Initiative Database, graduating student cohorts from spring 2008 et
2009. National Student Clearinghouse data from the 2008—09 and 2009—10 academic years. Universities with selective admissions have
an applicant screening process and do not admit all applicants. Open enrollment universities admit all applicants who meet minimum
requirements with no selective admissions processes.

Tableau 2 displays summary statistics for all high school graduates and then disaggre-
gated statistics by race and ethnicity. The graduates were 50 percent female, 44 pour cent
eligible for free or reduced-price lunch (FRPL) and were, on average, 17.1 years old the
September of their senior year in high school. In terms of college readiness, graduates
average 2.5 semesters of Advanced Placement (AP) or International Baccalaureate (IB)
courses. Twenty-four percent took AP English, 15 percent took AP mathematics, et
12 percent took AP science.9 The average graduate attended a high school that is 44
percent FRPL-eligible, 14 percent black, et 41 percent Hispanic. In terms of the high
school’s emphasis on college readiness, graduates on average attended schools where
11 percent of students earn college credit by passing an AP exam (by achieving a score
de 3 or higher), 67 percent took SATs (or ACTs), and SAT (or converted ACT equivalent)
scores averaged 976.

En moyenne, black and Hispanic students have both lower individual college readi-
ness and lower average high school college readiness than white and Asian students.
Par exemple, black and Hispanic students attended high schools where, on average,
fewer than 9 percent of students earned credit by AP exam, and average SAT scores
were below 950. Black students had lower average individual college readiness than
Hispanic students across AP coursework and exit exam scores. Black and Hispanic
students attended high schools with similar college readiness, but Hispanic students
attended schools with a larger percent of minority and FRPL-eligible students, sur
average. En outre, black and Hispanic students were more likely to attend high
schools whose students, on average, had received a scholarship to a ﬂagship campus.
Speciﬁcally, the Century Scholars (CS) program at TAMU, which began in 1999, is a

$7,445–$13,027 at other Texas public universities. Ainsi, cost of attendance at state universities, relative to other
college options, is not the driving mechanism behind the low enrollment rates, in particular for Hispanic stu-
bosses, observed in table 1.

9. AP mathematics include the following subjects: Calculus, Computer Science, and Statistics. AP science in-

cludes the following subjects: Biology, Chemistry, and Physics.

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

:
/
/

d
je
r
e
c
t
.

je
t
.

e
d
toi
e
d
p
un
r
t
je
c
e
–
p
d

F
/

1
5
2
2
0
9
1
6
9
3
4
7
2
e
d
p
_
un
_
0
0
2
7
3
p
d

b
oui
g
toi
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

215

Apply Yourself

Tableau 2. Summary Statistics—–Student and High School Characteristics by Race and Ethnicity

Noir

(1)

Hispanic

(2)

Asian

(3)

Blanc

(4)

All Graduates

(5)

Demographics

Femelle

Age, années

Free/reduced-price lunch eligible

Limited English proficient

High school coursework:

Non. of AP/IB courses completed (semesters)un

Took AP English language arts

Took AP mathematics

Took AP science

Non. of courses failed (semesters)

High school exam exit (z-scores)

English language arts

Mathematics

Science

Social studies

High school characteristics:

% Free or reduced-price lunch eligible

% Noir

% Hispanic

% Earning AP credit

% SAT tested

Average SAT score

Flagship scholarship programb

Number of graduatesc

Percent of graduates

0.518

17.12
(0.500)

0.503
<0.01 1.56 (3.31) 0.182 0.074 0.064 3.54 (4.67) −0.224 (0.947) −0.366 (0.860) −0.34 (0.872) −0.24 (0.936) 0.490 (0.213) 0.338 (0.237) 0.327 (0.193) 0.087 (0.110) 0.659 (0.195) 941.51 (100.02) 0.159 67,215 0.14 0.507 17.16 (0.526) 0.626 0.059 1.94 (3.66) 0.200 0.096 0.081 4.01 (5.20) −0.187 (0.958) −0.212 (0.928) −0.268 (0.907) −0.243 (0.949) 0.587 (0.247) 0.105 (0.136) 0.636 (0.291) 0.086 (0.114) 0.636 (0.216) 934.13 (92.98) 0.161 188,835 0.38 0.488 17.03 (0.539) 0.251 0.055 6.72 (7.08) 0.449 0.437 0.387 1.47 (3.22) 0.313 (1.003) 0.660 (1.094) 0.477 (1.049) 0.379 (0.983) 0.300 (0.200) 0.168 (0.131) 0.285 (0.186) 0.188 (0.141) 0.767 (0.182) 0.494 17.11 (0.434) 0.134 <0.01 2.90 (4.59) 0.277 0.184 0.141 2.01 (3.66) 0.213 (0.975) 0.243 (0.982) 0.301 (0.983) 0.261 (0.962) 0.305 (0.175) 0.108 (0.105) 0.256 (0.183) 0.130 (0.132) 0.688 (0.209) 0.502 17.12 (0.486) 0.435 0.036 2.50 (4.34) 0.241 0.145 0.117 2.97 (4.54) 0.008 (0.989) 0.008 (0.992) 0.008 (0.991) 0.009 (0.989) 0.439 (0.249) 0.141 (0.163) 0.414 (0.292) 0.110 (0.125) 0.667 (0.211) 1031.51 (88.93) 0.023 19,405 0.04 1018.32 (75.44) <0.01 213,486 0.44 976.00 (96.19) 0.087 490,707 1.00 Notes: Summary statistics for graduates of Texas public high schools from 2008 and 2009. Standard errors are in parentheses for continuous variables. aTotal number of Advanced Placement (AP) and International Baccalaureate (IB) courses completed (semesters). AP mathematics include the following subjects: Calculus, Computer Science, and Statistics. AP science includes the following subjects: Biology, Chemistry, and Physics. bThe Century Scholars program at Texas A&M University began in 1999. cThe sum of black, Hispanic, Asian, and white students does not add up to the total number of all graduates because there were 1,766 students who did not report their race/ethnicity or reported multiple races/ethnicities. Sources: Authors’ calculations from Texas Workforce Data Quality Initiative Database, graduating student cohorts from spring 2008 and 2009. scholarship program that targets high-ability students, namely, top 10 percent students, at low-income high schools in Texas. The CS program provides both ﬁnancial aid and academic support once students enroll at TAMU (see, for example, Domina 2007; 216 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d / f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Sandra E. Black, Kalena E. Cortes, and Jane Arnold Lincove Andrews, Ranchhod, and Sathy 2010; and Andrews, Imberman, and Lovenheim 2016, for further details on Texas ﬂagship scholarship programs).10 K–12 public school data were merged with college application data for all those who applied to Texas public colleges and universities. Because all Texas public universities use ApplyTexas, we are able to observe application behavior to any Texas public uni- versity. We used these records to identify students who applied to any public four-year university within one year of high school graduation. The application data also provide additional information on family income, parent education, and college readiness, in- cluding SAT scores and eligibility for top 10 percent automatic admissions, which is not provided in the high school dataset.11 We also can observe enrollment, ﬁnancial aid, col- lege grades, and college graduation data for all those who enrolled in any Texas public university. Unfortunately, we do not have access to college application information for students who applied only to out-of-state schools. However, we were able to use the National Student Clearinghouse (NSC) data from the 2008–09 and 2009–10 academic years to identify Texas high school graduates who did not apply to any four-year Texas public university but who did enroll at a four-year university in the United States in the fall following graduation.12 Individuals who applied to any Texas public university or en- rolled in any other university are considered to be college applicants in our analysis.13 We are thus missing individuals who applied only to private or out-of-state colleges and ultimately did not enroll anywhere; as discussed below, this is likely to be a very small fraction and is unlikely to aﬀect our results. When we consider applications to 11. 10. The goal of the CS program is to oﬀer ﬁnancial assistance and academic support services to students from schools that did not historically place many students at TAMU. To be eligible a student must attend a high school identiﬁed by TAMU as historically under- or nonrepresented at this ﬂagship campus. These students received a scholarship of $5,000 per year for four years and access to a four-year learning community (the learning community aims to help students develop during their time at TAMU). UT Austin had a similar scholarship program called the Longhorn Opportunity Scholarship (LOS), but in 2003 the LOS program was discontinued and replaced with the Discovery Scholars Program (DSP). DSP has many of the same features as the LOS program, but DSP eligibility is individual based rather than high school based (Andrews, Ranchhod, and Sathy 2010; Andrews, Imberman, and Lovenheim 2016). In the data, we only observe a top 10 percent indicator for students who applied to college. It is important to note that class rank may be an important omitted determinant of the college application behavior; to the extent that it is correlated with other variables of interest, our results may be biased. However, we do include a standardized test score as a proxy for class rank, which should limit the bias induced by this omission. A concern, however, is that the measurement error is correlated with other unobserved characteristics also correlated with race. As an example, it may be the case that, in high-performing high schools, there is little correlation between test scores and class rank (as a larger fraction of the students perform well on the standardized tests), while in lower-performing schools there is a stronger correlation between test scores and class rank. To the extent that minorities are disproportionately at lower-performing schools and are lower-ranked, that might bias the coeﬃcients on the race dummies to be more negative. However, when we break our sample by high school characteristics, including the percentage of Hispanics in the school, we ﬁnd very similar results, suggesting that this type of bias is unlikely to be driving our conclusions. (Results available on request.) 12. National Student Clearinghouse, a nonproﬁt organization, collects data on student enrollment and comple- tion and covers over 90 percent of U.S. universities. See Dynarski, Hemelt, and Hyman (2015) for more details. In 2008 and 2009, Texas had a total of thirty-seven public four-year universities, all of which participated in ApplyTexas. This includes campuses of the UT system, TAMU system, Texas State system, University of Houston system, Texas Tech University, and several other campuses. A full list of included public universities is available from the authors. 13. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d . f / f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 217 Apply Yourself Table 3. Summary Statistics by Four-Year College Application Rates for Texas High School Graduates Applied to any college Applied to any college or enrolled Applied to flagship Applied to flagship or enrolled at higher-ranked campus Applied to an open enrollment (nonselective) campus Applied only to flagship Applied only to open enrollment campus Applied to multiple campuses Applied to a high-SAT campus Applied to a low-SAT campus Number of graduates Percent of graduates Black (1) 0.351 0.393 0.050 0.053 0.154 0.016 0.077 0.183 0.094 0.159 Hispanic (2) 0.267 0.295 0.054 0.056 0.024 0.017 0.010 0.110 0.079 0.077 Asian (3) 0.524 0.613 0.295 0.326 0.032 0.144 0.011 0.304 0.365 0.047 White (4) 0.364 0.434 0.141 0.147 0.052 0.076 0.029 0.174 0.204 0.046 67,215 188,835 19,405 213,486 0.14 0.38 0.04 0.44 Notes: Includes all Texas high school graduates from 2008 and 2009. Application data are available for stu- dents who applied to Texas public universities. Additional enrollment data are available for all U.S. universities through the National Student Clearinghouse (NSC). A high-SAT campus is defined as one in the top quartile in the state in terms of median SAT score, and a low-SAT campus is in the bottom quartile in the state in terms of median SAT score. Sources: Authors’ calculations from Texas Workforce Data Quality Initiative Database, graduating student co- horts from spring 2008 and 2009. NSC data from the 2008—09 and 2009—10 academic years. Schools ranked higher than Texas flagships universities were identified from the 2008 U.S. News & World Reports U.S. college rankings. the most selective institutions, we include individuals who applied to the state ﬂagship campuses of UT Austin and TAMU or who were observed to have enrolled at an out-of- state college that was ranked higher by U.S. News & World Report (2007) than either Texas ﬂagship university for 2008 (UT Austin was ranked 44th, and TAMU was ranked 62nd). Table 3 presents the application decisions by race and ethnicity for our graduation cohorts. Row 1 presents statistics for the sample observed in the Texas public university application sample, and row 2 then presents similar statistics when we augment the sample with the NSC enrollment data. For four-year public universities in Texas, black and white application rates were similar, at 35 percent for black and 36 percent for white students, and the Asian student application rate was substantially higher at 52 percent. Hispanic students, who make up the second largest group of graduates, have a much lower application rate at only 27 percent. The adjustment resulting from incorporating the NSC data is relatively small; the gaps are largest for white and Asian students, with 9 percent of Asian students and 7 percent of white students appearing as enrolled in the NSC without applying to any four-year public university in Texas. The gaps are smaller for black and Hispanic students at only 4 and 3 percent, respectively. When we look at applications to elite state ﬂagship universities in row 3, black and Hispanic student application rates fall to only 5 percent, compared with 14 percent for white and nearly 30 percent for Asian students. Row 4 then uses the NSC data to in- corporate those who enrolled in four-year universities ranked higher than the ﬂagship universities (according to U.S. News & World Report) and who did not apply to either ﬂagship in Texas. We see that approximately 3 percent of Asian students appear as en- rolled at elite universities without applying to either ﬂagship university in Texas. The 218 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d / . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Sandra E. Black, Kalena E. Cortes, and Jane Arnold Lincove gaps for other groups are quite small. Only 0.6 percent of white, 0.3 percent of black, and 0.2 percent of Hispanic students appear as enrolled in higher-ranked universities without also applying to at least one Texas ﬂagship university. Because the gaps for black and Hispanic students are always lower than for white students, omitted information for students who apply only to non-Texan universities and do not enroll is likely to re- sult in understating racial and ethnic diﬀerences, as it is likely that more white than black or Hispanic students also apply and do not enroll. Finally, we can also look in more detail at a student’s portfolio of applications and how he diﬀers by race and ethnicity. Black students are substantially more likely to apply to an open enrollment campus, with 15 percent of them doing so, whereas only 5 percent of white, 3 percent of Asian, and 2 percent of Hispanic students do. In contrast, Asian students are much more likely to apply to multiple campuses, with 30 percent doing so, whereas only 11 percent of Hispanic, 17 percent of white, and 18 percent of black students are doing so. A total of 36.5 percent of Asian students apply to a high- SAT campus (deﬁned as the top quartile in the state in terms of median SAT score), and only 5 percent of Asian students apply to a low-SAT campus (deﬁned as the bottom quartile in the state in terms of median SAT score). This pattern reverses among black students, with 16 percent applying to a low-SAT campus and only 9 percent applying to a high-SAT campus. Hispanic students split evenly, with about 8 percent applying to a high-SAT campus and 8 percent applying to a low-SAT campus. 4 . D E C I S I O N TO A P P LY These summary statistics suggest that racial and ethnic gaps in college access can be partially explained by diﬀerential application behavior; however, these statistics do not account for students’ preparation and ability, which vary by race/ethnicity and inﬂuence the likelihood of college admission. In this setting, it is possible that high school sorting and college readiness explain much of the diﬀerence in college application behavior for Hispanic students, and it is unclear why black students have similar application rates to white students. To better understand the relationship between student characteristics and their ap- plication decisions in the context of high school sorting, we next examine the decision to apply to college, controlling for individual and high school characteristics. Impor- tantly, we want to identify whether racial and ethnic diﬀerences in the likelihood of applying to college (or the likelihood of applying to a ﬂagship university) are a function of the level of preparation, high school characteristics, and geographic proximity. To do so, we estimate the following simple model: yist = α · Racei + Xi · β + γ · Zst + ηt + πs + εist, (1) where yist is an indicator of whether student i, who attended high school s and graduated at time t, applied to any four-year college. Racei represents several indicator variables for race or ethnicity (black, Hispanic, Asian, and white [omitted comparison group] are included in all regressions), Xi is a vector of individual student characteristics, such as college readiness and indicators for FRPL-eligibility and limited English proﬁciency (LEP). Zst is a time-varying vector of high school characteristics related to both the de- mographics, college geographic proximity, and the college preparedness of graduates, l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d / . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 219 Apply Yourself and ηt represents graduation year indicators.14 We estimate this equation as a linear probability model both with and without high school ﬁxed eﬀects, πs. Table 4 presents results for regression speciﬁcations for the full sample of high school graduates for our two main outcomes of interest. As noted earlier, our ﬁrst out- come is an indicator of whether an individual applies to a four-year university—this is equal to one if the student either applied to any four-year Texas public university or appeared in the NSC as enrolled at any four-year university (panel A). Our second out- come is an indicator of whether an individual applied to a selective university—this is equal to one if the individual applied to either Texas ﬂagship university or appeared in the NSC as enrolled in a four-year university ranked higher by U.S. News & World Report than the Texas ﬂagships (panel B). We ﬁrst consider the probability of applying to a four-year college. Column 1 presents the results from a simple model that controls for race and ethnicity along with a par- simonious set of demographic controls that includes graduation year, age, gender, and indicators for LEP and FRPL-eligibility in high school. This speciﬁcation conﬁrms sub- stantial diﬀerences across races, controlling for this limited set of demographics, in the probability of applying to a public university, with black and Asian students signiﬁ- cantly more likely to apply and Hispanic students signiﬁcantly less likely to apply than white students. However, these application diﬀerences could be due to diﬀerences in college prepa- ration or ability. Column 2 adds controls for observable college readiness including the total number of AP course semesters taken, indicators for whether the student took AP English, AP mathematics, and AP science courses, the total number of fail- ing grades received on semester report cards, and performance on the state exit exam (standardized composite of English and math scores). All of these additional variables are statistically signiﬁcant predictors of college application behavior in the expected di- rection, and the estimated racial and ethnic diﬀerences change in distinct ways. The positive black coeﬃcient becomes even more positive, the positive Asian coeﬃcient becomes smaller in magnitude, and the negative signiﬁcant Hispanic coeﬃcient be- comes insigniﬁcant and smaller in magnitude. Thus, diﬀerences in individual college readiness variables appear to explain much of the diﬀerence between white and His- panic and white and Asian application rates, but not the diﬀerences between black and white students. Even controlling for student characteristics, diﬀerences in application behavior may still be due to diﬀerences in high school quality or proximity to the nearest four-year institution. School quality can aﬀect students through a variety of channels, including college preparedness, opportunities to take AP coursework, and expectations for higher education. School quality will vary by race and ethnicity due to historic segregation of neighborhoods and school districts, as well as selective admissions at magnet high schools. It is likely that race and ethnicity also inﬂuence a student’s experiences within a racially integrated high school in ways that aﬀect college choices. To examine this, we add observable measures of high school quality, including demographic measures (logged total enrollment, percent FRPL eligible, percent black, and percent Hispanic), 14. Throughout this study, high school characteristics are lagged one year prior to the student’s senior year in high school. 220 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . f / / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d / . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Sandra E. Black, Kalena E. Cortes, and Jane Arnold Lincove Table 4. Ordinary Least Squares Regression Results for Applying to College Panel A: Any Four-Year University Panel B: Flagship University or Better (1) (2) (3) (4) (5) (6) (7) (8) Race and ethnicity Black Hispanic Asian Other demographics 0.138* −0.074* −0.016* −0.009* −0.008* 0.146* 0.127* 0.023* (0.005) (0.002) (0.003) (0.004) (0.008) (0.006) −0.065* −0.061* −0.060* −0.025* −0.022* −0.021* −0.077* −0.002 (0.002) (0.007) (0.009) (0.003) (0.003) 0.018* −0.036* −0.019* 0.026* 0.177* (0.005) (0.006) (0.007) (0.009) (0.009) (0.002) 0.048* (0.008) (0.004) 0.185* (0.011) (0.002) 0.031* (0.007) (0.002) (0.004) eligible Free or reduced-price lunch (FRPL) −0.114* −0.047* −0.052* −0.054* −0.070* −0.030* −0.020* −0.020* (0.001) (0.002) (0.003) (0.002) 0.011* −0.097* −0.060* 0.011* (0.002) (0.003) (0.005) (0.002) (0.006) (0.005) −0.204* −0.096* (0.006) (0.006) (0.002) (0.001) 0.009* −0.002* (0.001) (0.002) Limited English proficient (LEP) High school coursework (college readiness) No. of AP and IB semesters Took AP English language arts Took AP mathematics Took AP science Total semesters failed High school exit exam (z-score)a Geographic proximityb Distance to nearest university (100 miles) Distance squared 0.013* 0.010* 0.014* (0.001) (0.001) (0.001) 0.138* 0.129* 0.132* (0.004) (0.005) (0.007) 0.116* 0.121* 0.121* (0.004) (0.004) (0.005) 0.048* 0.044* 0.051* (0.004) (0.005) (0.006) −0.020* −0.018* −0.020* (0.000) (0.000) (0.001) 0.067* 0.069* 0.079* (0.001) (0.001) (0.002) −0.107* (0.038) 0.074* (0.036) 0.018* 0.015* 0.017* (0.001) (0.001) (0.001) 0.019* 0.020* 0.013* (0.004) (0.005) (0.005) 0.140* 0.146* 0.151* (0.004) (0.005) (0.006) 0.056* 0.054* 0.058* (0.004) (0.005) (0.005) −0.003* −0.003* −0.003* (0.000) (0.000) (0.000) 0.031* 0.034* 0.038* (0.001) (0.001) (0.001) −0.063* (0.005) 0.009* (0.001) Number of observations 427,300 427,300 427,300 427,300 427,300 427,300 427,300 427,300 Other controls High school characteristicsc High school fixed effects Yes Yes Yes Yes Notes: Robust standard errors (shown in parentheses) are clustered at the high school level. Overall college application rate to a four-year public university is 0.382 (panel A) and flagship university application rate is 0.106 (panel B). Regressions also control for graduation year, gender, and age (not reported). Linear probability models for high school graduates from 2008 and 2009. Panel A dependent variable is equal to one if the student applied to any Texas four-year university or enrolled at any four-year university within one year of graduation. Panel B dependent variable is equal to one if the student applied to a Texas flagship public university (UT Austin or Texas A&M), enrolled in a Texas flagship public university, or enrolled within one year of graduation at an elite four-year university that is ranked higher than UT Austin by Barron’s. Enrollment data are available for all U.S. universities through the National Student Clearinghouse. aHigh school exit exam scores are a composite z-score of both English language arts and mathematics. Advanced Placement (AP) mathematics include the following subjects: Calculus, Computer Science, and Statistics. AP science includes the following subjects: Biology, Chemistry, and Physics. bThe distance variables are generated using longitude and latitude to compute the distance between all high schools and public universities in Texas. Panel A includes the distance to the nearest four-year public university. Panel B includes the distance to the nearest flagship university. cCharacteristics include logged enrollment, percent FRPL, percent black, percent Hispanic, lagged measures of percent of graduates entering four-year colleges, percent earning AP credit, percent SAT tested, and Flagship scholarship program. *Statistical significance at p < 0.05 level. indicators of typical college expectations (percent of graduates who enroll in four-year colleges, percent of students earning AP credit, and percent of students taking SATs, all lagged one year to reﬂect prior graduating classes), and geographical proximity to l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . f / / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d . / f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 221 Apply Yourself the nearest public university.15 Minority students may live farther from universities than white students; thus, the costs of attending college may be higher.16 In addition, we also include an indicator variable for high schools whose students were eligible for the CS program. Column 3 presents the results when we include these controls. Controlling for college readiness, observable high school characteristics, and distance to college, we ﬁnd once again that a black student is signiﬁcantly more likely than a white student to apply to college, ceteris paribus. In contrast, Hispanic students are again signiﬁcantly less likely to apply to college than similar white students, but Asian students are now signiﬁcantly less likely to apply than similar white students. In our ﬁnal speciﬁcation, we include high school ﬁxed eﬀects, thereby controlling for any school characteristics that are constant across this two-year time period (col- umn 4). Here, we are comparing students with equal college preparedness who grad- uated from the same high school to see if there are racial and ethnic disparities in application behavior within high schools. With high school ﬁxed eﬀects, the positive coeﬃcient for black students relative to white students and the negative coeﬃcient for Hispanic students relative to white students (from column 3) are conﬁrmed and have similar magnitudes. We estimate that within high schools, a black student is 13.8 per- centage points more likely to apply to college than a similar white student, whereas a Hispanic student is 6.1 percentage points less likely to apply than a similar white student. For Asian students, the within-high-school diﬀerence in application behavior is statistically signiﬁcant and negative, suggesting that Asian students are less likely to apply than white peers from the same high school by approximately 2 percentage points.17 Our results so far indicate that the large Hispanic minority in Texas is least likely to apply to college, while black students are more likely to apply than equally qualiﬁed white students with similar high school characteristics. Although racial and ethnic dif- ferences in application to public universities in general is important, work by Hoxby and Avery (2013) suggests that another important diﬀerence for long-run economic out- comes may be racial and ethnic diﬀerences in application to elite universities. Texas’s automatic admissions policy is designed to increase minority admissions to these elite 15. The distance variables are generated using longitude and latitude to compute the distance between each high school and each public university campus. The program used in the computation of the distance variables is called “Distance and Bearing between Matched Features” (distbyid.avx) by Jenness (2004), which is an applica- tion for ArcView. The extension distbyid.avx calculates the distance and bearing between features with identical attribute values, allowing one to generate connecting lines and calculate data for speciﬁc sets of features. The output options in this extension include a results table containing various user-selected ﬁelds such as: distance and bearing between features, x/y coordinates, centroids versus closest edges, etc. Because we had all school addresses, we ﬁrst generated x/y coordinates based on longitude and latitude of all of the Texas high schools. Then, using the option x/y coordinates, we compute a 2,412 distance matrix. Lastly, the function option in Stata Statistics/Data Analysis called min(x1, x2, x3, …, xk) is used to generate miles to the nearest public uni- versity and public ﬂagship university. In the case of missing distance data, we used the average distance for non-missing observations within the same school district or county. 16. Although many Texas cities have large Hispanic populations, Hispanics are more likely than other groups to live in rural areas near the Texas–Mexico border. 17. To further support these results, we reestimated all regression speciﬁcations on a subset of colleges that ex- cludes Texas’s two Historically Black Colleges and Universities (HBCUs) to determine whether our results on black application rates could be driven by these colleges. These regression results are shown in appendix table A.2 of this paper. Without HBCUs included, we predict a smaller, but still statistically signiﬁcant, gap with black students more likely to apply than white students. 222 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d / f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Sandra E. Black, Kalena E. Cortes, and Jane Arnold Lincove college campuses, but the eﬀects on minority application are unclear (M. Long 2004a; Card and Krueger 2005; Niu, Tienda, and Cortes 2006). Panel B of table 4 replicates the four speciﬁcations above for the outcome of apply- ing to a ﬂagship university or another campus ranked higher than the Texas ﬂagships by U.S. News & World Report.18 Across all four speciﬁcations, black and Hispanic students are signiﬁcantly less likely than white students to apply to a ﬂagship university. Asian students are signiﬁcantly more likely to apply than white students, but a large portion of this eﬀect is explained by diﬀerences in college readiness and high school quality. The point estimates for all races and ethnicities are larger in the parsimonious speciﬁ- cation in column 5, suggesting that some, but not all, racial and ethnic diﬀerences are explained by diﬀerential levels of college preparation and high school quality. 5 . H E T E RO G E N E I T Y So far, our analysis has assumed that our point estimates are estimating constant eﬀects of race and ethnicity across both student and high school characteristics. It is possible, however, that racial and ethnic diﬀerences in the decision to apply vary depending on the characteristics of the high school a student attended. Minorities attending a high school where more graduates take AP exams or enroll in college might be more likely to consider college options than those attending a high school where graduates rarely attend college.19 Application patterns could also vary depending on the characteristics of the student, and in particular on the student’s level of college readiness. For example, less college-ready minorities may behave diﬀerently than minorities who are more pre- pared for college. We next examine whether we observe diﬀerential college application patterns based on high school and student characteristics. High School Characteristics Minority students may have diﬀerent college application behaviors when they attend a high school with greater postsecondary expectations or with diﬀerential treatment of minorities relative to college expectations. Table 5 presents results when we examine whether gaps in application behavior are diﬀerent at schools with a high fraction of Hispanic students versus a low fraction of Hispanic students, and schools with high college enrollment rates for graduates versus low rates. In each case, a high school is characterized as “high” for a given characteristic if it is in the top 20 percent of all high schools statewide and “low” if it is in the bottom 20 percent.20 Each speciﬁcation includes controls for gender, age, race and ethnicity, demographics, college readiness, graduation year, and high school ﬁxed eﬀects. The ﬁndings that black students are more likely to apply to college than white stu- dents, and that Hispanic students are less likely to apply to college than white students, 18. For estimation of the probability of applying to a ﬂagship university, the “distance to college” variable is replaced with the distance to the nearest of the two ﬂagship campuses. 19. For example, Hoxby and Avery (2013) ﬁnd that low-income, high-achieving students who are in schools with other high-achieving students are more likely to apply to elite institutions, whereas those who are more isolated are not. 20. These cutoﬀ values are 0.12, 0.30, and 0.67 for percentage Hispanic (bottom 20 percent, median, and top 20 percent, respectively), and 0.09, 0.22, and 0.37 for percentage of graduates from the high school who enrolled in a four-year college (bottom 20 percent, median, and top 20 percent, respectively). l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d / . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 223 Apply Yourself Table 5. Ordinary Least Squares Regression Results for Applying to College by High School Characteristics Panel A: Any Four-Year University Panel B: Flagship University or Better % Hispanic College Entrance Rate % Hispanic College Entrance Rate Top 20% (1) Bottom 20% (2) Top 20% (3) Bottom 20% (4) Top 20% (5) Bottom 20% (6) Top 20% (7) Bottom 20% (8) Race and ethnicity Black Hispanic Asian Other demographics Free or reduced-price lunch (FRPL) eligible Limited English proficient (LEP) High school coursework (college readiness) No. of AP and IB semesters Took AP English language arts Took AP mathematics Took AP science Total semesters failed High school exit exam (z-score)a 0.114* (0.012) 0.113* (0.008) 0.049* (0.009) 0.038* 0.157* (0.016) (0.008) −0.026* −0.049* −0.072* −0.017* −0.022* −0.049* −0.068* −0.015* (0.009) (0.005) (0.006) (0.009) 0.017 −0.040* −0.022* (0.006) (0.010) (0.023) (0.009) (0.006) (0.009) 0.017 −0.038* −0.022* (0.007) (0.011) (0.022) 0.075* (0.007) 0.084* (0.012) 0.125* (0.015) 0.007 (0.018) 0.007 (0.018) (0.005) −0.040* −0.062* −0.072* −0.008* −0.039* −0.053* −0.066* −0.007* (0.004) (0.004) −0.117* −0.159* −0.170* −0.021* −0.116* −0.164* −0.162* −0.022* (0.009) (0.009) (0.005) (0.005) (0.004) (0.006) (0.021) (0.004) (0.011) (0.009) (0.022) (0.005) (0.011) (0.009) 0.028* (0.005) 0.027* (0.005) 0.015 (0.018) 0.121* (0.032) 0.005* (0.002) 0.144* (0.013) 0.115* (0.010) 0.056* (0.013) 0.007* (0.001) 0.123* (0.007) 0.095* (0.006) 0.037* (0.006) 0.021* (0.002) 0.115* (0.009) 0.062* (0.010) 0.040* 0.019 (0.010) (0.031) −0.015* −0.028* −0.032* −0.002* −0.015* −0.024* −0.029* −0.002* (0.000) (0.001) 0.014* 0.056* (0.002) (0.002) 0.007* (0.001) 0.123* (0.007) 0.102* (0.006) 0.038* (0.006) 0.006* (0.002) 0.135* (0.013) 0.126* (0.011) 0.053* (0.012) 0.022* (0.002) 0.113* (0.009) 0.062* (0.010) 0.041* (0.011) 0.010 (0.018) 0.124* (0.033) (0.001) 0.066* (0.002) (0.002) 0.073* (0.003) (0.001) 0.055* (0.002) (0.000) 0.015* (0.002) (0.001) 0.067* (0.002) (0.001) 0.078* (0.004) 0.024 (0.031) Number of observations 94,289 55,795 132,050 19,938 94,289 55,795 132,050 19,938 Other controls High school fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Notes: Robust standard errors (shown in parentheses) are clustered at the high school level. Regressions also control for graduation year, gender, and age. Linear probability model regressions for high school graduates from 2008 and 2009. Panel A dependent variable is equal to one if the student applied to any Texas four-year university or enrolled at any four-year university within one year of graduation. Panel B dependent variable is equal to one if the student applied to a Texas flagship public university (UT Austin or Texas A&M), enrolled in a Texas flagship public university, or enrolled within one year of graduation at an elite four-year university that is ranked higher than UT Austin by Barron’s. Enrollment data are available for all U.S. universities through the National Student Clearinghouse. Regression subsamples are based on characteristics of high school attended compared to statewide quintiles. Quintile cutoff values are 0.115 and 0.669 for percentage Hispanic (bottom 20% and top 20%, respectively); 0.086 and 0.366 for percentage of graduates from the high school who enrolled in a four-year college (bottom 20% and top 20%, respectively). aHigh school exit exam scores are a composite z-score of both English language arts and mathematics. Advanced Placement (AP) mathematics include the following subjects: Calculus, Computer Science, and Statistics. AP science includes the following subjects: Biology, Chemistry, and Physics. *Statistical significance at p < 0.05 level. are remarkably consistent across high school types. The estimated racial and ethnic diﬀerences are smallest in high schools with low college entrance rates (column 4 of panel A). Interestingly, the estimated coeﬃcient for Hispanic students relative to white students was negative even in high schools with the highest college entrance rates (col- umn 3) and percent Hispanic students (column 1). Other high school characteristics 224 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d / . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Sandra E. Black, Kalena E. Cortes, and Jane Arnold Lincove were also tested (results not displayed) with consistent results for black and Hispanic students.21 In panel B of table 5, we examine whether racial diﬀerences in application to se- lective universities vary by high school type. Some minorities should be more likely to apply from high schools where minorities are more likely to be represented in the top 10 percent—for example, a high school with more than 90 percent minority students. The results in panel B also control for gender, age, demographics, college readiness, graduation year, and high school ﬁxed eﬀects. Again, we observe that Hispanic stu- dents are less likely to apply to elite ﬂagship universities compared with their white student counterparts, and their lower application rate is remarkably consistent across all high school types. The results for black and Asian students vary by high school type. Most notably, black students are signiﬁcantly more likely than white students to apply at schools with high college enrollment. College Readiness We next examine whether racial and ethnic diﬀerences are constant across the distri- bution of student preparation for college. Students should be more likely to apply to college when their high school outcomes signal the potential for college success, and racial/ethnic gaps might diminish for minority students who are highly qualiﬁed for postsecondary education. To examine this, we divide high school graduates by their ob- served high school preparation in comparison to what is typical for Texas public univer- sities. This comparison proxies for information students might know about their own college readiness through observation of peers. First, we measured average levels of col- lege readiness based on AP courses completed and high school exit exams for freshman entering all non-open enrollment public universities in Texas in the year prior to the student’s graduation. We then selected three institutions from across the distribution of average freshman college readiness—the top-ranked campus (UT Austin), the me- dian campus (UT San Antonio), and the bottom-ranked campus (UT El Paso)—among campuses that are not open enrollment. Next, from among the sample of high school graduates, we identiﬁed students whose exit exams and AP courses exceeded that of the average entering freshman at those three institutions during the student’s senior year in high school. Thus, a “highly qualiﬁed student” is above average compared to an entering freshman at the top-ranked public university, a “somewhat-qualiﬁed stu- dent” is above average for a median public university but not the top-ranked, and a “less-qualiﬁed student” is above average for a bottom-ranked public university but not the median university. Students who are below average for the bottom-ranked campus are excluded from this analysis. High school students who are similar to students who successfully gain admissions and enroll should view themselves as college-ready, and should receive encouragement to apply from teachers and counselors. In addition, a student’s college readiness should partially overcome the information problem regard- ing the probability of admissions. 21. Speciﬁcally, we tested speciﬁcations by the high school’s percent FRPL-eligible students, percent SAT-tested, and percent earning AP credit, along with speciﬁcations by the high school’s rate of Hispanic SAT testing, Hispanic AP credit, and Hispanic college entry, with similar results. Even in schools with high rates of Hispanic college readiness indicators and high rates of Hispanic college entry, Hispanic students were signiﬁcantly less likely to apply to college than white students with similar college readiness. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d f / . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 225 Apply Yourself Table 6 displays regression results by college-readiness level for speciﬁcations that include both student college readiness measures and high school ﬁxed eﬀects. Thus, we are estimating within-school racial and ethnic diﬀerences for each qualiﬁcation group. We again present our results for our two main outcomes of interest: application to any four-year university (panel A) and application to a ﬂagship university (panel B), as de- ﬁned earlier. For either outcome, the results for Hispanic students are remarkably sim- ilar to prior point estimates. At all college-readiness levels, a Hispanic student is less likely to apply to college than a similarly prepared white student within the same high school. These eﬀects are statistically signiﬁcant for all groups except highly qualiﬁed applicants to a ﬂagship university. A highly qualiﬁed Hispanic student is 3.2 percent- age points less likely to apply to any college than a white student, and a less-qualiﬁed Hispanic student is 6.1 percentage points less likely to apply to any college than a white student. The results for black students in table 6 provide new insight into the positive ef- fects previously reported. Black students from the medium- and less-qualiﬁed groups are signiﬁcantly more likely to apply to college than similarly qualiﬁed white students, but black students from the highly qualiﬁed group are less likely to apply to college than highly qualiﬁed white students, although this diﬀerence is not statistically signif- icant. Thus, the college application rate for black students is high because of a higher propensity to apply among students who are more marginal for college admissions. In addition, a highly qualiﬁed black student is 5 percentage points more likely to apply to a ﬂagship university than a similar white student from the same high school. Interestingly, these results suggest that Hispanic students are less likely to apply to any college at most levels of college readiness relative to white students, but that less- prepared black students are more likely to apply to college than white students, and better-prepared black students are more likely to apply to ﬂagships than white students. Asian students are more likely to apply to ﬂagship universities than white students. Put diﬀerently, among highly qualiﬁed graduates, Hispanic students are signiﬁcantly more likely to “undermatch” (i.e., apply to campuses for which they are overqualiﬁed) relative to white students, but Asian students are more likely to “overmatch” (i.e., apply to campuses for which they are underqualiﬁed) relative to white students.22 6 . C H O I C E A M O N G C O L L E G E S C O N D I T I O N A L O N A P P L I C AT I O N So far, we have examined the decision of whether to apply to college and have shown that Hispanic students are less likely to apply to four-year postsecondary education relative to white students, black students are more likely to apply than white students, and Asian students are more likely to apply to elite universities in comparison to white students. Next, we examine how college characteristics inﬂuence the decision of where to apply by race and ethnicity. Why do students choose to apply to one university over another? 22. In other work (Black, Cortes, and Lincove 2015), we estimated interactions between college qualiﬁcations, as deﬁned above, and the propensity to apply to elite ﬂagship public universities by race and ethnicity. Here, we ﬁnd that Hispanic students in the top 11–25 percent who were highly qualiﬁed for ﬂagships were less likely to apply to ﬂagships than similar white students. In addition, we ﬁnd that black undermatch occurs among top 10 percent graduates who choose not to take advantage of automatic admissions to ﬂagships (but do apply to less competitive campuses), whereas the Hispanic undermatch occurs among highly qualiﬁed students who miss the cut for automatic admissions and must compete in a holistic process. 226 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d / f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Sandra E. Black, Kalena E. Cortes, and Jane Arnold Lincove Table 6. Ordinary Least Squares Regression by College-Readiness Levels for Applying to College College Readiness Based on High School Exit Exam and Advanced Placement Courseworka Panel A: Any Four-Year University Panel B: Flagship University or Better Highly Qualified Somewhat Qualified Less Qualified Highly Qualified Somewhat Qualified Less Qualified (1) (2) (3) (4) (5) (6) −0.021 (0.017) −0.032* (0.010) −0.004 (0.008) −0.043* (0.010) −0.119 (0.105) −0.001 (0.001) −0.000 (0.017) 0.030* (0.012) 0.010 (0.007) −0.034* (0.005) −0.004 (0.006) 1.035* (0.155) 13,037 0.889 0.073* (0.008) −0.032* (0.006) −0.004 (0.007) −0.059* (0.006) −0.280* (0.038) 0.008* (0.001) 0.024* (0.006) 0.059* (0.004) 0.018* (0.004) −0.039* (0.002) 0.032* (0.004) 0.475* (0.095) 53,412 0.815 0.136* (0.011) −0.061* (0.009) 0.004 (0.013) −0.064* (0.007) −0.257* (0.040) 0.013* (0.001) 0.033* (0.008) 0.067* (0.006) 0.040* (0.007) −0.040* (0.002) 0.061* (0.006) 0.190 (0.145) 30,521 0.698 0.050* (0.023) −0.022 (0.015) 0.073* (0.012) −0.074* (0.016) −0.130 (0.111) 0.006* (0.002) −0.018 (0.026) 0.175* (0.017) 0.041* (0.012) −0.059* (0.006) 0.044* (0.009) 0.298 (0.217) 13,037 0.653 −0.000 (0.010) −0.027* (0.006) 0.042* (0.009) −0.068* (0.006) −0.166* (0.032) 0.016* (0.001) 0.038* (0.007) 0.110* (0.006) 0.054* (0.005) −0.041* (0.002) 0.088* (0.004) −0.091 (0.110) 53,412 0.404 −0.012 (0.010) −0.040* (0.007) 0.002 (0.013) −0.041* (0.006) −0.076* (0.027) 0.014* (0.001) 0.032* (0.007) 0.093* (0.007) 0.053* (0.007) −0.025* (0.001) 0.093* (0.007) −0.144 (0.127) 30,521 0.218 Race and ethnicity Black Hispanic Asian Other demographics Free or reduced-price lunch (FRPL) eligible Limited English proficient (LEP) High school coursework (college readiness) Number of AP and IB semesters Took AP English language arts Took AP mathematics Took AP science Total semesters failed High school exit exam (z-score)b Constant Number of observations Dependent variable mean Other controls High school fixed effects Yes Yes Yes Yes Yes Yes Notes: Robust standard errors (shown in parentheses) are clustered at the high school level. All regressions control for graduation year, gender, and age. Linear probability model regressions for high school graduates from 2008 and 2009. Panel A dependent variable is equal to one if the student applied to any Texas four-year university or enrolled at any four-year university within one year of graduation. Panel B dependent variable is equal to one if the student applied to a Texas flagship public university (UT Austin or TAMU), enrolled in a Texas flagship public university, or enrolled within one year of graduation at an elite four-year university that is ranked higher than UT Austin by Barron’s. Enrollment data are available for all U.S. universities through the National Student Clearinghouse. IB = International Baccalaureate. aCollege qualification levels are based on the average performance of entering freshmen on high school exit exams and Advanced Placement (AP) coursework completion at Texas state universities of varying levels of selectivity. “Highly qualified” is greater than the average entering freshman at the UT Austin (most selective state university). “Somewhat qualified” is less than average for UT Austin, but greater than average for the University of Texas at San Antonio (mid-range state university). “Less qualified” is less than average for UT San Antonio, but greater than average for UT El Paso (less selective state university). bHigh school exit exam scores are a composite z-score of both English language arts and mathematics. AP mathematics include the following subjects: Calculus, Computer Science, and Statistics. AP science includes the following subjects: Biology, Chemistry, and Physics. *Statistical significance at p < 0.05 level. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . f / / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d / f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 227 Apply Yourself Are there particular institutional characteristics more appealing to students of diﬀerent races and ethnicities? A number of papers have estimated college choice behavior. We build on work by Bridget Long (2004), which ﬁrst examined college enrollment choices using McFad- den’s (1973) choice model. This strategy exploits the variation across college character- istics in a student’s choice set to estimate the inﬂuence on enrollment of observable characteristics (such as tuition, distance from home, and college inputs). Hoxby and Avery (2013) applied the same approach to college application decisions to examine the application behavior of high-achieving, low-income high school students to explain the phenomenon of “undermatch,” where highly qualiﬁed students opt for less-competitive universities, despite a high probability of admissions and ﬁnancial support at elite uni- versities.23 Here, we expand on Hoxby and Avery (2013) to better understand racial and ethnic diﬀerences in application choices. This strategy provides insight into how the characteristics of public universities may contribute to racial gaps in applications. We also expand on prior models of college choice by examining academic, social, and informational inﬂuences on student choices. Prior studies primarily test how stu- dents are inﬂuenced by campus academics by estimating the eﬀects of similarities or diﬀerences between the student’s SAT scores and campus average, with controls for geographic and ﬁnancial accessibility. To better understand racial and ethnic diﬀer- ences in application, we also consider characteristics describing the social setting of the college campus. Minority students may be reluctant to enroll in campuses with few students from the same racial or ethnic group. We also include measures of the feeder patterns between the student’s high school and the college campus. Students may have better information about campuses attended by prior graduates from their own high school and may be more willing to attend a campus if their older peers are currently attending that campus. A key advantage of applying this methodology is, by looking at the pool of applicants and examining how individual and school characteristics aﬀect these decisions, we are able to incorporate information on eligibility for automatic admissions through class rank. As noted earlier, we only observe information on eligibility for top 10 percent admissions for students who choose to apply to college, and, as a result, we cannot use this information when analyzing the decision to apply to college at all. Following B. Long (2004) and Hoxby and Avery (2013), we model the college appli- cation decision as a conditional logit: (cid:3) (cid:2) yi = j Pr = eZi j·β (cid:4) j eZi j·β , Zi j · β = β1 · C j + β2 · Distancei j + β3 · Academici j + β4 · Demog j + β5 · Feederi j + εi j, (2) (3) where Pr(yi = j) is an indicator if student i applied to university j, C j represents in-state tuition cost at institution j, Distancei j is the distance from student i’s high school address 23. Bettinger and Long (2004) use the same empirical strategy to investigate the eﬀects of a speciﬁc college input and academic remediation on college outcomes. 228 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d / . f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Sandra E. Black, Kalena E. Cortes, and Jane Arnold Lincove to institution j, Academici j is a vector of indicator variables representing student i’s aca- demic readiness compared to mean readiness on campus j (i.e., the distance between the student’s SAT scores and the institution mean), Demog j is a vector of campus racial and ethnic demographics based on prior year’s freshman class at institution j (i.e., % Black, % Hispanic, and % Asian enrolled on campus), Feederi j is a vector of variables indicating feeder patterns of student i’s high school to each institution j (i.e., number enrolled from the same high school in prior year, number of college graduates from the same high school in prior year, number enrolled of the same race/ethnicity from the same high school, and number of college graduates of the same race/ethnicity from the same high school—all lagged one year prior to the student’s high school gradua- tion), and lastly, εi j is independent and identically distributed with the extreme value distribution.24 In the estimation, a student appears in one observation for each four-year public university in Texas, reﬂecting the full choice set of public universities.25 The outcome variable is coded as one if the student applied to that campus and zero otherwise, and multiple positive outcomes occur for students who applied to more than one university. The conditional logit requires that individuals must have at least one positive outcome across the choice set, so this analysis is limited to students who applied to at least one university. Six universities that received fewer than ﬁve total applications are also omit- ted, creating a choice set of thirty-one public universities.26 The conditional logit is estimated with student ﬁxed eﬀects, controlling for all ﬁxed student characteristics. All independent variables must vary across college campuses within a student’s choice set. Our estimations include robust standard errors for clustering within high schools. Table 7 presents our regression results on what speciﬁc institutional attributes are appealing to students. We estimate the conditional logit model for all college appli- cants (panel A) and top 10 percent college applicants (panel B) across racial and ethnic groups.27 The conditional logit results are displayed as odds ratios of the change in the 24. A key underlying assumption of the conditional logit speciﬁcation is the Independence of Irrelevant Alterna- tives (IIA) assumption. In this model, the ratio of the probabilities of any two choices is unrelated to any of the other choices or their characteristics. A change in the characteristics of any of the other choices should not aﬀect this ratio. (See Hoﬀman and Duncan 1988 for more discussion.) This is a particularly strong assumption (despite being commonly made in the literature), and so results should be interpreted with caution. We have also estimated the models as Linear Probability Models, where the dependent variable is whether or not an in- dividual applied to a college with a speciﬁc attribute (based on characteristics such as school quality, geographic proximity, and racial composition). The conclusions are invariant to this speciﬁcation choice. 25. Niu and Tienda (2008) highlight the sensitivity of conditional logit estimations to the selection of the choice set of the individual, with no attention to diﬀerences in application behavior by race and ethnicity. They ex- amine how college choice sets diﬀer if left unconstrained to include all postsecondary institutions relative to choice sets that are constrained by students’ academic achievement and the high school attended. They ﬁnd that constraining the criteria used to deﬁne choice sets produces quite diﬀerent institutional proﬁles. In addi- tion, when one uses data from Texas to estimate a conditional logit of the inﬂuence of college characteristics on students’ top institutional preference, the choice set constrained by type of high school attended gives very diﬀerent estimates from the unconstrained choice set (while the choice set delimited only by class rank pro- duces essentially similar results.) We are using all public colleges in our choice set and not constraining choices based on individual characteristics or institutional quality. 26. The omitted campuses oﬀer highly speciﬁed programs, primarily for graduate students. 27. Based on numbers of high school graduates and college applicants in our data with top 10 percent status, ap- proximately 72 percent of the population of top 10 percent graduates applied to a Texas public university. The remaining 28 percent might have applied only to out-of-state or private institutions, or they might have not ap- plied to college all. Because we cannot assess why top 10 percent students do not appear in our application data, l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . f / / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d f . / f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 229 Apply Yourself Table 7. Factors Associated with Applying to College by Race and Ethnicity—–Results of a Conditional Logit Estimation (Expressed in Odds Ratios) Panel A: All College Applicants Panel B: Top 10% College Applicants Black Hispanic (1) (2) Asian (3) White Black Hispanic Asian White (4) (5) (6) (7) (8) Geographic proximitya Can commute (distance (cid:2) 60 miles) Can visit home (distance between 61 and 250 miles) College preparedness: Measures of SAT Matchb Own SAT 100—149 pts above campus mean Own SAT 150—199 pts above campus mean Own SAT greater than 200 pts above campus mean Own SAT 1—99 pts below campus mean Own SAT 100—149 pts below campus mean Own SAT 150—199 pts below campus mean Own SAT 200 or more pts below campus mean 4.407* (0.278) 2.929* (0.112) 0.439* (0.015) 0.246* (0.010) 0.053* (0.004) 2.249* (0.059) 3.259* (0.121) 3.804* (0.158) 4.319* (0.252) % Black enrolled on campus Campus racial demographics (prior year’s freshman class) 6.566* (0.558) 0.442* (0.037) % Hispanic enrolled on campus % Asian enrolled on campus 0.676 (0.140) Feeder patterns of student’s high school to college campus 1.029* (0.004) Number enrolled from same high school in prior year Number graduated from same high school in prior year Number enrolled of same race from same high school Number graduated of same race from same high school 0.996 (0.004) 1.004 (0.009) 1.109* (0.017) 15.335* (0.953) 4.171* (0.176) 0.506* (0.014) 0.302* (0.012) 0.077* (0.006) 1.913* (0.047) 2.569* (0.097) 2.768* (0.135) 2.501* (0.177) 0.306* (0.023) 1.269* (0.092) 0.458* (0.081) 1.032* (0.006) 1.002 (0.005) 0.989 (0.007) 1.048* (0.012) 5.485* (0.436) 2.905* (0.146) 0.739* (0.040) 0.579* (0.048) 0.203* (0.026) 0.965 (0.061) 0.794* (0.082) 0.659* (0.090) 0.287* (0.057) 0.208* (0.022) 0.799 (0.105) 294.081* (102.809) 1.031* (0.006) 1.026* (0.006) 1.016 (0.015) 0.981 (0.012) 5.323* (0.338) 2.612* (0.072) 0.653* (0.014) 0.462* (0.015) 0.141* (0.007) 1.271* (0.035) 1.200* (0.059) 0.965 (0.060) 0.507* (0.045) 0.067* (0.003) 0.129* (0.007) 0.013* (0.003) 1.017* (0.004) 1.019* (0.007) 1.017* (0.005) 0.996 (0.009) 3.698* 13.207* (1.138) (0.412) 4.301* 3.121* (0.220) (0.244) 4.100* (0.580) 3.909* (0.379) 4.710* (0.314) 3.309* (0.158) 0.511* (0.041) 0.311* (0.031) 0.082* (0.012) 1.957* (0.127) 2.788* (0.284) 3.073* (0.353) 4.548* (0.651) 4.280* (0.760) 0.371* (0.057) 1.865 (0.664) 1.016* (0.005) 1.047* (0.008) 1.020 (0.011) 1.081* (0.028) 0.526* (0.028) 0.341* (0.023) 0.092* (0.010) 1.792* (0.086) 2.562* (0.181) 3.230* (0.282) 3.570* (0.429) 0.831* (0.068) 0.667* (0.066) 0.340* (0.048) 0.833 (0.085) 0.793 (0.102) 0.685* (0.111) 0.429* (0.084) 0.416* (0.041) 0.302* (0.053) 1.187 0.943 (0.183) (0.131) 0.354* 103.599* (49.354) (0.088) 1.008 (0.007) 1.071* (0.009) 1.008 (0.009) 0.987 (0.016) 1.011* (0.005) 1.060* (0.009) 1.030 (0.024) 1.025 (0.020) 0.772* (0.024) 0.653* (0.027) 0.290* (0.018) 1.075* (0.038) 1.146* (0.068) 0.905 (0.070) 0.731* (0.073) 0.108* (0.007) 0.129* (0.010) 0.030* (0.008) 1.007 (0.004) 1.062* (0.009) 1.000 (0.007) 1.010 (0.013) Number of observations 616,559 1,374,664 305,660 2,250,259 70,215 286,998 115,010 576,321 Notes: Conditional logit estimates are reported in odds ratios (with student fixed effects) of the probability of application to thirty-one state universities. Robust standard errors for clustering within high schools. Sample includes students who graduated from a Texas public high school in 2008 and 2009 and applied to at least one Texas public university for admission in the fall following high school graduation. A small number of students (<1%) are omitted who were missing distance to college or SAT scores. aDistance is measured from the student’s high school to each college campus. bDichotomous measures of SAT match (omitted group has SATs 0—99 points [pts] above campus mean). Specifications also control for in- state tuition. All college campus variables (i.e., enrollment, graduation, SAT, and tuition) are lagged one year to information the student would have as she entered her senior year of high school. *Statistical significance at p < 0.05 level. odds of applying to a campus based on marginal changes in a campus characteristic, ceteris paribus. Similar to prior work (B. Long 2004; Hoxby and Avery 2013), we control this likely introduces some bias. Our results are only generalizable to similar populations of high-performing students who might apply to public institutions. 230 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . f / / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d f / . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Sandra E. Black, Kalena E. Cortes, and Jane Arnold Lincove for tuition, college proximity, and academic match. Building on this model, our second set of covariates controls for social match, measured as the proportion of each minority race and ethnic group on the campus. Our third and ﬁnal set of covariates controls for high school feeder patterns both for the whole high school and for students of the same race and ethnicity as the student. As shown in panel A, students of all races and ethnicities are more likely to apply to universities that are closer to home (either within commuting or visiting distance), but Hispanic students are three times more likely than any other racial group to apply to a university that is easily commutable to their home (i.e., distance less than sixty miles). Speciﬁcally, controlling for academic match, Hispanic students are the most sensitive to distance, followed by Asian and white students, and black students are the least in- ﬂuenced by distance. In terms of academic campus match, all races and ethnicities are more likely to apply to campuses where their SATs are below the campus mean, and less likely to apply to campuses where their SATs are above the campus mean. However, Hispanic and black students are more likely to apply to campuses that are a far reach academically, with average SATs 200 or more points above their own score, whereas white students are signiﬁcantly less likely to apply to a campus that is a far reach. With higher average SAT scores, white and Asian students have fewer “reach” cam- puses than black and Hispanic students, but the evidence does not suggest that a lack of close academic match campuses explains racial and ethnic diﬀerences in application behavior. Regarding social match, measured through the racial demographic composition of the college campus, we observe that black, Hispanic, and Asian students are all more likely to apply to campuses with a higher concentration within their own racial and eth- nic group. The eﬀect is particularly large for Asian and black students. Interestingly, white students’ application behavior is negatively associated with minority concentra- tions of all other racial and ethnic groups on campus.28 Because diﬀerences in applica- tion behavior by race and ethnicity could mask associated diﬀerences in income levels across racial and ethnic groups, we further disaggregate students into two brackets of family income: below $40,000 and above $80,000. Those additional analyses are re- ported in Appendix table A.3. Our ﬁnal set of covariates, high school feeder patterns, controls for the information and familiarity students would gain about a campus by attending a high school that frequently sends graduates from their own school to that particular college campus. Our measures include the number of students from one’s own high school who enrolled at the college in the previous year, as well as the number of students from one’s own high school who successfully graduated from that college in the previous year. We calculate 28. We also ran models where we added in separately the three sets of covariates (college preparedness, campus racial demographics, and feeder patterns of student’s high school) to see how the coeﬃcients might change with the inclusion of each set of covariates. Those results are very similar to those reported in panel A of table 7 and are available from the authors upon request. A notable diﬀerence, however, is for Asian students, where the coeﬃcients for academic match variables are inﬂuenced by the addition of campus racial demographic variables. Speciﬁcally, controlling for the percent Asian on campus, Asian students are most likely to enroll on a campus where their SAT scores are within 100 points of the campus mean. This suggests that Asian students are most likely to apply to campuses with a concentrated population of other Asian students who have similar academic readiness. Hispanic and black students, in contrast, remain more likely to apply to reach schools, holding racial demographic composition of the university constant. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d . / f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 231 Apply Yourself both of these measures for the high school overall, as well as for the student’s own racial group. Hispanic and black students are more likely to apply to a campus that has recently enrolled any students from the same high school and recently graduated students from the same high school of the same race. Asian students are positively inﬂuenced by both recent enrollment and recent graduation from their high school, but the race of the enrollers and completers is not relevant. White students are inﬂuenced by enrollment and completion of any race, as well as by enrollment speciﬁcally by white students. It appears that all students learn something about a college campus through the feeder relationship of their high school peers, but for black and Hispanic students, information about successful completion by same race and ethnicity of older peers is also important.29 We also consider the role of information in college application behavior. Hoxby and Avery (2013) hypothesize that college undermatch is related to a lack of information about admissions and ﬁnancial aid among low-income, high-achieving high school students. We exploit Texas’s unique college admissions policy to test this hypothesis in relation to racial and ethnic application gaps in the state. Speciﬁcally, we examine whether racial and ethnic diﬀerences in college application behavior are similar be- tween students in the top 10 percent, who have full information about their admission to the state’s top universities, and students who face typical college admissions uncer- tainty.30,31 The results, reported in panel B of table 7, show very little eﬀect of automatic ad- missions on the application choices of black, Asian, and white students. In comparison, Hispanic students in the top 10 percent are the only minority group not signiﬁcantly inﬂuenced by same-race enrollment on campus. Similarly, Hispanics in the top 10 per- cent are not signiﬁcantly inﬂuenced by the feeder history of same-race students from their high school to the campus, while Hispanic students subject to admissions uncer- tainty are more likely to apply to campuses where Hispanic students from their high school have recently completed a college degree. Also salient in this analysis is the ef- fect of college proximity on college application, in particular for Hispanic students—top 10 percent Hispanic students are about three times more likely compared with other students to apply to a university if that institution is near their home. College proxim- ity is not only an important predictor for college application for Hispanics students in general, but also for high-achieving top 10 percent Hispanic students. Finally, in table 8 we analyze the portfolio of college choices among top 10 per- cent students relative to all other students who apply to colleges. Here, we estimate linear probability models (LPMs) predicting diﬀerent types of application behavior (application ﬂagships, application to open-enrollment campuses, etc.), conditioned on 29. Although we interpret this eﬀect as students following the behavior of similar students from earlier cohorts, this could also reﬂect teacher or counselor behavior, such as encouraging students of a particular race to attend a speciﬁc institution. We are not able to distinguish this with our data. 30. We also ran regression models for students who were ranked in the top 11–25 percent and the bottom 75 percent. Those regression results are similar to those found in panel A of table 7. These results are available upon request from the authors. 31. Each student in the top 10 percent receives a letter informing her of her status and of the associated beneﬁts, suggesting that individuals do in fact have information about their admission status. Note that we assume the only value of being classiﬁed as top 10 percent is the associated admissions certainty. However, the classiﬁcation itself could change individual preferences; see work by Papay, Murnane, and Willett (2011) for more discussion. 232 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d / f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Sandra E. Black, Kalena E. Cortes, and Jane Arnold Lincove Table 8. Ordinary Least Squares Regression Results of the Effects of Automatic Admis- sions Status on Application Behavior by Race and Ethnicity Applied to flagship Applied to open enrollment Applied only to flagship Applied only to open enrollment Applied to HBCU Applied to multiple campuses Applied to a high SAT campus Applied to a low SAT campus Black (1) 0.200* (0.012) −0.075* (0.011) 0.159* (0.011) −0.065* (0.010) −0.055* (0.010) −0.040* (0.014) 0.150* (0.013) −0.093* (0.013) Hispanic (2) 0.214* (0.009) −0.018* (0.004) 0.149* (0.009) −0.012* (0.002) −0.003* (0.001) 0.015 (0.009) 0.187* (0.009) −0.033* (0.007) Asian (3) 0.137* (0.014) −0.023* (0.006) 0.346* (0.015) −0.012* (0.003) −0.001 (0.002) −0.209* (0.023) 0.091* (0.014) −0.031* (0.008) Number of observations 22,405 48,384 9,945 Other controls White (4) 0.193* (0.007) −0.039* (0.004) 0.283* (0.007) −0.027* (0.003) −0.001* (0.000) −0.111* (0.006) 0.158* (0.006) −0.040* (0.003) 75,934 Student characteristics High school fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Notes: Each cell from table 8 is from a separate regression and only reports on the esti- mated regression coefficients on the indicator of top 10 percent student by race. Samples include only students who applied to college. Students in the top 10 percent are automat- ically admitted to all types of campuses contingent on application. All regression specifi- cations control for student demographics (e.g., female; age; free or reduced-price lunch eligible; and limited English proficient status), college readiness measures (e.g., number of Advanced Placement [AP] and International Baccalaureate semesters; took AP English language arts, mathematics, and/or science; total semesters failed; and high school exit exam), and high school fixed effects. HBCU = historically black colleges and universities. *Statistical significance at p < 0.05 level. application to at least one Texas public university. Each cell reported in table 8 comes from a separate LPM regression model estimated by race/ethnicity, and we report the estimated regression coeﬃcients on the indicator of a top 10 percent student relative to a same-race/same-ethnicity student who is not in the top 10 percent but is otherwise similar in observable college readiness (and including high school ﬁxed eﬀects).32 As previously mentioned, students in the top 10 percent of their graduating high school class are automatically admitted to all types of campuses contingent on application. So, in theory, top 10 percent students may choose from a wide array of in-state four-year public colleges. We observed interesting application patterns among minority and nonminority top 10 percent students. We ﬁnd that across all racial groups, top 10 percent students are more likely to apply to only a ﬂagship campus, but the magnitudes of those coeﬃcients diﬀer substantially by race and ethnicity. Black and Hispanic top 10 percent students 32. Speciﬁcally, all regression speciﬁcations control for student demographics (e.g., female; age; FRPL eligible; and LEP status), college readiness measures (e.g., number of AP and IB semesters; took AP English language arts, mathematics, and/or science; total semesters failed; and high school exit exam), and high school ﬁxed eﬀects. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d f / . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 233 Apply Yourself are 15.9 and 14.9 percentage points, respectively, more likely to apply only to a ﬂagship campus. In contrast, both Asian and white top 10 percent students are substantially more likely to apply only to a ﬂagship campus by 34.6 and 28.3 percentage points, re- spectively. We also observed that black top 10 percent students are 5.5 percentage points less likely to apply to HBCUs, perhaps opting to apply to more selective institutions. In fact, the inﬂuence of top 10 percent status on application rates to high-SAT campuses are about the same for both black and white students (roughly 15 percentage points). Lastly, across all groups, top 10 percent students are less likely to apply to an open en- rollment campus. 7 . C O N C L U S I O N Minority access to higher education is a growing concern across the nation. Obstacles to higher education for minorities will limit growth in human capital and competitive- ness for the United States in the very near future. In this study, we identify the college application decision as a pivotal ﬁrst step in college access that precedes the processes of admissions and enrollment. Our study beneﬁts from the use of a statewide student dataset that includes the full range of student ability and college readiness existing within two high school graduation cohorts, and a single college application procedure that is common to most four-year universities in the state. We are also able to provide further insight into the decision process by analyzing campus preferences of applicants in a state where highly qualiﬁed students are automatically admitted to the top public universities. From a standpoint of population size, the growing Hispanic minority (soon to be a majority in Texas) is of great concern in Texas and across the nation. We consistently ﬁnd that Hispanic students are least likely of all ethnic groups to apply to college over- all and to elite ﬂagship universities in particular. This ﬁnding is robust to controls for college readiness, high school quality, and high school ﬁxed eﬀects. The gap between Hispanic and white students in college application is consistent across levels of observ- able college readiness and high school quality. Even when Hispanic students attend high schools where a majority of students move on to college or where Hispanic stu- dents are statistically highly likely to achieve automatic admissions, Hispanic students are signiﬁcantly less likely to apply to college than white students. We ﬁnd a more nuanced type of inequality in the college application behavior of black high school graduates, an issue not previously identiﬁed in studies that aggre- gate black and Hispanic students as a single group. On average, black students are actually more likely to apply to college than white and Hispanic students with similar levels of college readiness and high school quality. However, this is driven mostly by high application rates among less-prepared black students. Among the most qualiﬁed students, black students have similar overall application rates to white students but are more likely to apply to ﬂagship universities. We also ﬁnd that college application decisions for minorities are responsive to more than just the average academic performance of students on a campus. Black, Hispanic, and Asian students are more sensitive than white students to distance to college; and black, Hispanic, and Asian students are all inﬂuenced by the presence of same-race students on campus. Black and Hispanic students are also inﬂuenced by the historical 234 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d / f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Sandra E. Black, Kalena E. Cortes, and Jane Arnold Lincove feeder pattern of their high school to a campus, including past successful degree com- pletion of same-race students from their high school. These social and informational eﬀects are mitigated only among high-income Hispanics and Hispanic students who are guaranteed admission. Thus, minority application rates respond not only to stu- dent college readiness, but also the enrollment and outcomes of minority students on a campus. Automatic admissions might expand the application choices of Hispanic stu- dents to include campuses with fewer Hispanic students, but black students are highly responsive to the racial composition of a campus even when they are guaranteed ad- missions to any campus. Finally, we ﬁnd that admissions policy interacts with race and ethnicity, potentially altering the application behavior of minority students. Andrews, Imberman, and Loven- heim (2016) ﬁnd evidence that intensive recruitment and support programs increased minority enrollment at Texas ﬂagship universities. Adding to this ﬁnding, we show that state application policy also shapes how minority students select to which colleges they apply. Most notably, black and Hispanic students who are eligible for automatic admissions are 20 percentage points more likely to apply to a ﬂagship campus than same-race/same-ethnicity peers who are not eligible, and also less likely to apply only to lower-tiered, open-enrollment campuses. Combined with prior evidence on racial dif- ferences in responses to the Top Ten Percent Plan (Black, Cortes, and Lincove 2015), our results suggest that statewide, race-neutral admissions policy can also improve access and quality for qualiﬁed minority students. Although our results provide signiﬁcant insight into the racial and ethnic dispari- ties in college application behavior, we can oﬀer little insight into the source of these inequalities. Our ﬁndings of race and ethnicity eﬀects are robust to the inclusion of high school ﬁxed eﬀects, and the signiﬁcant eﬀect of high school characteristics on minority students suggests that high schools are an important source of information regarding college application. Overall, these potential frictions are not oﬀset by the prospect of guaranteed admissions policies in higher education. ACKNOWLEDGMENTS The authors are grateful to the Texas Workforce Data Quality Initiative at the University of Texas at Austin’s Ray Marshall Center, funded by the U.S. Department of Labor. The research presented here utilizes conﬁdential data from the State of Texas supplied by the Texas Education Research Center (ERC) at the University of Texas at Austin. The authors gratefully acknowledge the use of these data. The views expressed are those of the authors and should not be attributed to the ERC or any of the funders or supporting organizations mentioned herein, including the Univer- sity of Texas, Texas A&M University, the State of Texas, or the study’s sponsor. Any errors are attributable to the authors. Special thanks for helpful comments from Daniel Hamermesh, Dar- rick Hamilton, Jonathan Meer, Lori Taylor, as well as seminar and conference participants at the Association for Public Policy Analysis and Management, Southern Economic Association, Amer- ican Economic Association, University of Texas at Austin’s Economics Department, Texas A&M’s Bush School’s Quantitative Brown Bag Series, Teachers College Columbia University, and Stan- ford Graduate School of Education. The authors thank Jenna Cullinane, Matt Farber, Katherine Keisler, Chester Polson, Emily Weisburst, and Alan Blanch who provided outstanding research assistance. Institutional support was provided by the University of Texas at Austin, Texas A&M University, Tulane University, University of Maryland Baltimore County, and Stanford’s Graduate School of Education Center for Education Policy Analysis. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d . / f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 235 Apply Yourself REFERENCES Andrews, Rodney J., Scott A. Imberman, and Michael F. Lovenheim. 2016. Recruiting and sup- porting low-income, high-achieving students at ﬂagship universities. NBER Working Paper No. 22260. Andrews, Rodney J., Vimal Ranchhod, and Viji Sathy. 2010. Estimating the responsiveness of college applications to the likelihood of acceptance and ﬁnancial assistance: Evidence from Texas. Economics of Education Review 29(1): 104–115. Arcidiacono, Peter, and Michael Lovenheim. 2016. Aﬃrmative action and the quality-ﬁt trade-oﬀ. Journal of Economic Literature 54(1): 3–51. Bettinger, Eric, and Bridget T. Long. 2004. Shape up or ship out: The eﬀects of remediation on students at four-year colleges. NBER Working Paper No. 10369. Black, Sandra E., Kalena E. Cortes, and Jane Arnold Lincove. 2015. Academic undermatching of high-achieving minority students: Evidence from race-neutral and holistic admissions policies. American Economic Review: Papers & Proceedings 105(5): 604–610. Card, David, and Alan B. Krueger. 2005. Would the elimination of aﬃrmative action eﬀect highly qualiﬁed minority applicants? Evidence from California and Texas. Industrial and Labor Relations Review 58(3): 416–434. Clotfelter, Charles T., Helen F. Ladd, and Jacob L. Vigdor. 2015. Public universities, equal op- portunity, and the legacy of Jim Crow: Evidence from North Carolina. NBER Working Paper No. 21577. Cortes, Kalena E., and Jane Arnold Lincove. 2016. Can admissions percent plans lead to better collegiate ﬁt for minority students? American Economic Review: Papers & Proceedings 106(5): 348– 354. Cortes, Kalena E., and Jane Arnold Lincove. 2019. Match or mismatch? Automatic admissions and college preferences of low- and high-income students. Educational Evaluation and Policy Anal- ysis, 41(1): 98–123. Daugherty, Lindsay, Paco Martorell, and Isaac McFarlin. 2014. Percent plans, automatic admis- sions, and college outcomes. IZA Journal of Labor Economics 3(1): 10. Dickson, Lisa M. 2006. Does ending aﬃrmative action in college admissions lower the percent of minority students applying to college? Economics of Education Review 25(1): 109–119. Dillon, Eleanor Wiske, and Jeﬀrey Andrew Smith. 2017a. The consequences of academic match between students and colleges. CESifo Working Paper Series No. 6344. Dillon, Eleanor Wiske, and Jeﬀrey Andrew Smith. 2017b. The determinants of mismatch between students and colleges. Journal of Labor Economics 35(1): 45–66. Domina, Thurston. 2007. Higher education policy as secondary school reform: Texas public high schools after Hopwood. Educational Evaluation and Policy Analysis 29(3): 200–217. Dynarski, Susan M., Steven Hemelt, and Joshua Hyman. 2015. The missing manual: Using Na- tional Student Clearinghouse data to track post-secondary outcomes. Educational Evaluation and Policy Analysis 37(1): 53S–79S. Griﬃth, Amanda, and Donna Rothstein. 2009. Can’t get here from there: The decision to apply to a selective institution. Economics of Education Review 28(5): 620–628. 236 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / f e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d . f / f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Sandra E. Black, Kalena E. Cortes, and Jane Arnold Lincove Hoﬀman, Saul D., and Greg J. Duncan. 1998. Multinomial and conditional logit discrete-choice models in demography. Demography 25(3): 415–427. Hoxby, Caroline M., and Christopher Avery. 2013. The missing “one-oﬀs”: The hidden supply of high-achieving, low-income students. Brookings Papers on Economic Activity 2013 (1): 1–65. Hoxby, Caroline M., and Sarah Turner. 2013. Informing students about their college options: A proposal for broadening the Expanding College Opportunities Project. Washington, DC: The Hamilton Project Discussion Paper No. 2013-3. Jenness, Jeﬀ. 2004. Distance and bearing between matched features (distbyid.avx) extension for ArcView 3.x, v. 2. Flagstaﬀ, AZ: Jenness Enterprises. Long, Bridget T. 2004. How have college decisions changed over time? An application of the conditional logistic choice model. Journal of Econometrics 121(1-2): 271–296. Long, Mark C. 2004a. College application and the eﬀect of aﬃrmative action. Journal of Econo- metrics 121(1-2): 319–342. Long, Mark C. 2004b. Race and college admission: An alternative to aﬃrmative action? Review of Economics and Statistics 86(4): 1020–1033. McFadden, Daniel. 1973. Conditional logit analysis of qualitative choice behavior. In Frontiers in econometrics, edited by Paul Zarembka, pp. 105–142. New York: Academic Press. National Center for Education Statistics (NCES). 2016. Digest of Education Statistics Table 302.20: Percentage of recent high school completers enrolled in 2- and 4-year colleges by race/ethnicity, 1960 through 2015. Available https://nces.ed.gov/programs/digest/d16/tables/dt16_302.20.asp. Accessed 22 September 2019. Niu, Sunny X., and Marta Tienda. 2008. Choosing colleges: Identifying and modeling choice sets. Social Science Research 37(2): 416–433. Niu, Sunny X., Marta Tienda, and Kalena Cortes. 2006. College selectivity and the Texas top 10% law: How constrained are the options? Economics of Education Review 25(3): 259–272. Page, Lindsay C., and Judith Scott-Clayton. 2016. Improving college access in the United States: Barriers and policy responses. Economics of Education Review 51(1): 4–22. Pallais, Amanda. 2015. Small diﬀerences that matter: Mistakes in applying to college. Journal of Labor Economics 33(2): 493–520. Papay, John P., Richard J. Murnane, and John B Willett. 2011. How performance information aﬀects human-capital investment decisions: The impact of test-score labels on educational out- comes. NBER Working Paper No. 17120. Reardon, Sean F., Rachel Baker, and Daniel Klasik. 2012. Race, income, and enrollment patterns in highly selective colleges, 1982–2004. Unpublished paper, Stanford University. Smith, Jonathan, Matea Pender, and Jessica Howell. 2013. The full extent of student-college aca- demic undermatch. Economics of Education Review 32(2): 247–261. U.S. News & World Report. 2007. America’s best colleges, 2008. Washington, DC: U.S. News & World Report, L.P. Walton, Gregory M., and Geoﬀrey L. Cohen. 2007. A question of belonging: Race, social ﬁt, and achievement. Journal of Personality and Social Psychology 92(1): 82–96. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / f / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d / f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 237 Apply Yourself A P P E N D I X Table A.1. Summary Statistics by College Application Behavior College application Submitted ApplyTexas Submitted ApplyTexas or enrolled at a 4-year university Applied to a flagship university Student demographics Female Age, years White (non-Hispanic) Black Hispanic Asian Free or reduced-price lunch (FRPL) eligible Limited English proficient (LEP) High school coursework Number of AP/IB courses completed (semesters)a Took AP English language arts Took AP mathematics Took AP science Total semesters failed High school exit exam (z-score)b High school characteristics Campus size % FRPL eligible % Black % Hispanic % Earning AP credit % SAT tested Average SAT score Number of observations Submitted ApplyTexas Submitted ApplyTexas or Enrolled at a Four-Year University Applied to a Flagship University All Graduates 1.000 1.000 0.306 0.547 17.03 (0.346) 0.478 0.145 0.311 0.063 0.293 0.008 4.82 (5.36) 0.447 0.299 0.235 1.14 (2.53) 0.435 (0.877) 0.874 1.000 0.267 0.545 17.03 (0.350) 0.497 0.147 0.293 0.059 0.280 0.008 4.86 (5.41) 0.448 0.301 0.235 1.13 (2.52) 0.444 (0.886) 1.000 1.000 1.000 0.529 17.02 (0.330) 0.608 0.068 0.206 0.116 0.155 0.002 8.45 (5.96) 0.677 0.586 0.442 0.32 (1.04) 0.966 (0.805) 0.331 0.379 0.101 0.502 17.12 (0.486) 0.435 0.137 0.385 0.040 0.379 0.036 2.50 (4.34) 0.241 0.145 0.117 2.97 (4.54) 0.009 (0.990) 1,992.44 (943.93) 1,991.08 (945.22) 2,167.56 (890.66) 1,836.93 (1012.26) 0.415 (0.263) 0.134 (0.167) 0.410 (0.303) 0.131 (0.130) 0.728 (0.177) 987.59 (98.08) 162,271 0.403 (0.259) 0.135 (0.165) 0.395 (0.297) 0.136 (0.133) 0.730 (0.178) 0.325 (0.236) 0.123 (0.134) 0.332 (0.258) 0.178 (0.149) 0.764 (0.172) 0.439 (0.249) 0.141 (0.163) 0.414 (0.292) 0.109 (0.125) 0.667 (0.211) 991.76 (97.83) 185,742 1022.09 (92.52) 49,580 976.00 (96.18) 490,561 Notes: Summary statistics for all graduates of Texas public high schools from 2008 and 2009. Standard errors are in parentheses for continuous variables. aTotal number of Advanced Placement (AP) and International Baccalaureate (IB) courses completed (semesters). bHigh school exit exam scores are a composite z-score of both English language arts and mathematics. Sources: Authors’ calculations from Texas Workforce Data Quality Initiative Database, graduating student cohorts from spring 2008 and 2009. National Student Clearinghouse data from the 2008—09 and 2009—10 academic years. 238 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . f / / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d / f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 Sandra E. Black, Kalena E. Cortes, and Jane Arnold Lincove Table A.2. Ordinary Least Squares Regression Results for Applying to College, Excluding from the Analysis Historically Black Colleges and Universities Dependent Variable: Any Four-Year University (1) (2) (3) (4) (5) Race and ethnicity Black Hispanic Asian Other demographics Free or reduced-price lunch (FRPL) eligible Limited English proficient (LEP) High school coursework (college readiness) Number of AP and IB semesters Took AP English language arts Took AP mathematics Took AP science Total semesters failed High school exit exam (z-score)a Geographic proximityb Distance to nearest university (100 miles) Distance squared −0.012 (0.007) −0.077* (0.009) 0.177* (0.009) −0.114* (0.006) −0.203* (0.006) 0.092* (0.006) −0.003 (0.007) 0.017* (0.009) −0.047* (0.005) −0.095* (0.006) 0.014* (0.001) 0.132* (0.007) 0.122* (0.005) 0.051* (0.006) −0.019* (0.001) 0.079* (0.002) 0.106* (0.004) −0.061* (0.003) −0.004 (0.007) −0.052* (0.002) −0.096* (0.006) 0.011* (0.001) 0.130* (0.006) 0.123* (0.004) 0.047* (0.005) −0.018* (0.000) 0.071* (0.001) 0.109* (0.004) −0.060* (0.003) −0.019* (0.006) −0.053* (0.002) −0.096* (0.005) 0.013* (0.001) 0.138* (0.004) 0.117* (0.004) 0.048* (0.004) −0.020* (0.000) 0.067* (0.001) 0.106* (0.004) −0.061* (0.003) −0.004 (0.007) −0.052* (0.002) −0.096* (0.006) 0.011* (0.001) 0.130* (0.005) 0.123* (0.004) 0.048* (0.005) −0.018* (0.000) 0.071* (0.001) −0.049 (0.031) 0.018 (0.028) Number of observations 427,300 427,300 427,300 427,300 427,300 Other controls High school characteristicsc High school fixed effects Yes Yes Yes Notes: Robust standard errors (shown in parentheses) are clustered at the high school level. Regressions also control for graduation year, gender, and age. Linear probability models for high school graduates from 2008 and 2009. Dependent variable is equal to one if the student applied to any Texas four-year university or enrolled at any four-year university within one year of graduation. Enrollment data are available for all U.S. universities through the National Student Clearinghouse. AP = Advanced Placement; IB = International Baccalaureate. aHigh school exit exam scores are a composite z-score of both English language arts and mathematics. bThe distance variables are generated using longitude and latitude to compute the distance between all high schools and the flagship public institutions in Texas. cCharacteristics include logged enrollment, percent Free/reduced-price lunch eligible, percent black, percent Hispanic, lagged measures of percent of graduates entering four-year colleges, percent earning AP credit, percent SAT tested, and Flagship scholarship program. *Statistical significance at p < 0.05 level. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . f / / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d . / f f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 239 Apply Yourself Table A.3. Factors Associated with Applying to College by Family Income and Race/Ethnicity—–Results of a Conditional Logit Estimation (Expressed in Odds Ratios) Panel A: Low Income (<$40k) Panel B: High Income (>$80k)

Noir

Hispanic

Asian

Blanc

Noir

Hispanic

Asian

(1)

(2)

(3)

(4)

(5)

(6)

(7)

5.458* 19.365*
(1.327)
(0.395)
4.688*
3.246*
(0.253)
(0.155)

8.378* 11.779*
(0.785)
(0.858)
3.469*
3.319*
(0.139)
(0.262)

2.761*
(0.256)
2.352*
(0.128)

6.581*
(0.518)
3.036*
(0.141)

Geographic proximitya

Can commute (distance (cid:2) 60 miles)

Can visit home (distance between 61 et

250 miles)

College preparedness: Measures of SAT Matchb

Own SAT 100—149 pts above campus mean

Own SAT 150—199 pts above campus mean

Own SAT greater than 200 pts above

campus mean

Own SAT 1—99 pts below campus mean

Own SAT 100—149 pts below campus mean

Own SAT 150—199 pts below campus mean

Own SAT 200 or more pts below campus

mean

0.441*
(0.021)
0.255*
(0.015)
0.069*
(0.007)
2.279*
(0.074)
3.196*
(0.132)
3.896*
(0.176)
4.776*
(0.275)

% Black enrolled on campus

Campus racial demographics (prior year’s freshman class)
7.646*
(0.664)
0.396*
(0.040)

% Hispanic enrolled on campus

% Asian enrolled on campus

0.693
(0.160)

Feeder patterns of student’s high school to college campus
1.029*
(0.005)

Number enrolled from same high school in

prior year

Number graduated from same high school in

prior year

Number enrolled of same race from same

high school

Number graduated of same race from same

high school

0.991
(0.005)

1.000
(0.009)
1.105*
(0.019)

0.495*
(0.021)
0.308*
(0.017)
0.102*
(0.009)
1.960*
(0.059)
2.795*
(0.123)
2.984*
(0.163)
3.139*
(0.231)

0.350*
(0.031)
1.333*
(0.127)

0.752
(0.151)

1.043*
(0.005)

0.992
(0.005)
0.985*
(0.007)
1.050*
(0.014)

0.535*
(0.048)
0.350*
(0.040)
0.078*
(0.012)
1.524*
(0.109)
1.599*
(0.185)
1.601*
(0.230)

1.255
(0.238)

0.402*
(0.061)
1.669*
(0.302)
257.298*
(117.235)

1.038*
(0.006)
1.015*
(0.006)

1.009
(0.016)

0.990
(0.012)

0.613*
(0.023)
0.388*
(0.019)
0.113*
(0.008)
1.589*
(0.054)
1.822*
(0.099)
1.618*
(0.106)
1.297*
(0.111)

0.079*
(0.007)
0.190*
(0.012)
0.003*
(0.001)

1.026*
(0.005)
1.015*
(0.008)

1.005
(0.007)

0.993
(0.010)

0.450*
(0.032)
0.269*
(0.026)
0.056*
(0.008)
2.045*
(0.119)
2.730*
(0.224)
2.694*
(0.280)
2.316*
(0.308)

3.790*
(0.447)
0.441*
(0.051)

1.325
(0.397)

1.026*
(0.004)

1.004
(0.004)

1.017
(0.011)
1.128*
(0.024)

0.579*
(0.029)
0.375*
(0.024)
0.102*
(0.011)
1.499*
(0.073)
1.641*
(0.130)
1.537*
(0.142)

1.003
(0.127)

0.169*
(0.016)
0.721*
(0.062)
0.316*
(0.081)

1.030*
(0.006)

1.008
(0.005)
0.969*
(0.007)
1.079*
(0.014)

Blanc

(8)

3.491*
(0.226)
2.283*
(0.065)

0.675*
(0.018)
0.515*
(0.019)
0.171*
(0.010)
1.115*
(0.036)

0.972
(0.057)
0.743*
(0.056)
0.314*
(0.033)

0.068*
(0.003)
0.101*
(0.006)
0.031*
(0.009)

1.012*
(0.005)
1.018*
(0.007)
1.021*
(0.006)

0.999
(0.010)

3.888*
(0.455)
2.928*
(0.218)

0.897
(0.068)

0.926
(0.094)
0.427*
(0.057)
0.625*
(0.056)
0.391*
(0.054)
0.261*
(0.051)
0.063*
(0.017)

0.142*
(0.017)
0.550*
(0.102)
291.763*
(120.375)

1.024*
(0.006)
1.035*
(0.007)
1.026*
(0.013)

0.972
(0.016)

Number of observations

323,857 711,202

99,417

388,678 87,451 187,364

106,640

1,191,392

Remarques: Conditional logit estimates are reported in odds ratios (with student fixed effects) of the probability of application to thirty-one state
universities. Robust standard errors for clustering within high schools. Sample includes students who graduated from a Texas public high school
dans 2008 et 2009 and applied to at least one Texas public university for admission in the fall following high school graduation, and provided
income data on ApplyTexas. A small number of students (<1%) are omitted who were missing distance to college or SAT scores. aDistance is measured from the student’s high school to each college campus. bDichotomous measures of SAT match (omitted group has SATs 0—99 points [pts] above campus mean). Specifications also control for in- state tuition. All college campus variables (i.e., enrollment, graduation, SAT, and tuition) are lagged one year to information the student would have as she entered her senior year of high school. *Statistical significance at p < 0.05 level. 240 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . f / / e d u e d p a r t i c e - p d l f / / / / 1 5 2 2 0 9 1 6 9 3 4 7 2 e d p _ a _ 0 0 2 7 3 p d / f . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 APPLY YOURSELF: RACIAL AND ETHNIC image

Télécharger le PDF