Winners and Losers: Changes in Texas University Admissions post-Hopwood

Abstract

This paper evaluates changes in the racial and ethnic composition of admissions at three Texas universities following the judicial ban on affirmative action imposed by the 1996 Hopwood decision. We estimate the extent to which universities practiced affirmative action before the judicial ban, and evaluate how admission officers at these universities changed the relative weights accorded to various applicant characteristics during the ban. After assessing whether changes in the relative weights favored minority applicants, we simulate the degree to which these new policies succeeded in maintaining minority admission rates at their pre-Hopwood levels. We find that these universities complied with the Hopwood ruling such that direct advantages given to black and Hispanic applicants disappeared (and, in some cases, became disadvantages). While we find some evidence that universities changed the weights they placed on applicant characteristics aside from race and ethnicity in ways that aided underrepresented minority applicants, these changes in the admissions process were insufficient to fully restore black and Hispanic applicants’ share of admitted students.

1. Introduction and Research Questions

During the past decade, several states have had affirmative action banned in their public university admissions, (including Texas, California, Florida, Washington, Georgia, and Michigan),¹ and ballot initiative efforts are being attempted in other states. In each of these states, the general public and policymakers have expressed concerns about maintaining minority enrollment in elite institutions.

In its July, 1996 Hopwood² decision, the Fifth Circuit Court of Appeals opined that the only legal justification for affirmative action is to rectify the present effects of past discrimination, concluding therefore, that the goal of achieving institutional diversity was not an acceptable rationale for considering race in admissions decisions. The Attorney General of Texas interpreted the Hopwood decision as a ban on race-based admissions, financial aid, and recruiting policies at public and private institutions in the state. This ban was in-force for the fall class of 1997, which registered appreciable declines in the representation of minority students at the state’s public flagship institutions.

Anticipating further declines in minority enrollment at public universities with selective admission policies, the Texas legislature passed H.B. 588, the “uniform admission law” (popularly known as the top 10% law), which guarantees admission to any public university in the state to Texas high school seniors who graduate in the top-10 percent of their class. Passed in May, 1997, the uniform admission law was fully in force for the fall of 1998 admission cohort. The uniform admission law also specified 18 factors that universities should consider in admitting students who do not graduate in the top-10% of their high school class, including socioeconomic status, second language ability, and indications that the student overcame adversity.³ Some have argued that under an affirmative action ban colleges will have an incentive to employ admissions practices that partially ignore an applicant’s “quality,” thereby placing more emphasis on characteristics that are correlated with race/ethnicity (Chan and Eyster, 2003). Among the criteria explicitly named in H.B. 588 for college admission are several non-traditional factors that could be used as proxies for race/ethnicity in order to achieve institutional diversity.

Whether a change from an explicit consideration of race to the use of proxy indicators of minority group status can effectively increase minority college admission rates, and whether the allowable proxies are as efficient as traditional race-sensitive admission criteria poses an empirical question with clear policy implications. Using unique administrative data from three Texas universities, this paper will answer the following questions. First, to what extent was affirmative action practiced in the admissions decisions of Texas universities before the Hopwood decision? Second, how did these universities change the weight they placed on various applicant characteristics and did compliance with the Hopwood decision reduce or eliminate the direct or indirect weight placed on an applicant’s race/ethnicity? Finally, did these universities add weight to characteristics that are correlated with an applicant’s race/ethnicity in ways that advantage underrepresented minority applicants? Assuming that the answer to the last question is “yes,” we then conduct simulations that estimate the extent to which the policy responses were able to maintain minority students’ share of admitted students.

We find evidence the Texas flagship public institutions (University of Texas at Austin and Texas A&M University) practiced affirmative action in the pre-Hopwood years. These universities complied with the Hopwood decision by eliminating the direct admission advantages given to minority applicants, but they also changed the weights placed on other applicant characteristics in ways the favored black and Hispanic applicants. However, these changes were not sufficient to restore minority representation to the level that would have occurred under the prior affirmative action policies.

Our analyses are based on data from Texas universities, but the results have broader policy implications because the use of race-sensitive admission criteria in college admissions remains a highly controversial issue.⁴ Following judicial and legislative bans on affirmative action, public universities in California, Florida, and Washington have used various strategies to increase minority representation on their campuses, including percent plans and proxy indicators for race and ethnicity.

Three years after the passage of the California Civil Rights Initiative (Proposition 209) in 1996, which prohibited the explicit use of race, ethnicity, national origin, and sex in university admissions, the University of California Regents approved a policy that guarantees admission to one of the UC campuses to the top-4 percent of graduates in each California high school.⁵ The University of California System issued additional guidelines for admission: for 25 to 50 percent of the freshmen admissions, decisions could consider “(a)cademic accomplishments in light of the applicant’s life experiences and special circumstances. These experiences and circumstances may include, but are not limited to, disabilities, low family income, first generation to attend college, need to work, disadvantaged social or educational environment, difficult personal and family situations or circumstances, refugee status, or veteran status” (Univ. of California 1996). Since 1999, the University of California has adopted a series of reforms that increase the consideration of non-academic criteria and expanded its comprehensive review in the admissions process to all students (Chan and Eyster 2003; Univ. of California 2001a, 2001b). For example, in 2001, the class rank criterion was amended by the UC Regents to offer students who were not in the top-4 percent, but were in the top-12.5 percent of their high school class, admission to one of the UC-campuses if they successfully completed first- and second-year requirements at a community college.

Changes in admission criteria implemented in Florida and Washington also allow for the consideration of non-academic factors. Florida Governor Jeb Bush announced the “One Florida” policy in 1999, which simultaneously eliminated affirmative action in admissions and guaranteed admission to one of the state’s public universities to the top-20 percent of graduates in each Florida high school. Subsequently the University of Florida added an essay requirement to their application and in the application solicited information about students’ “…extracurricular activities, work history, whether they were raised by a single parent, etc.” (Marin and Lee 2003, 33). According to James (James 2002: A1), the University of Florida also gives special consideration to “…students who are poor, attended a low-performing high school, or whose parents didn’t attend college.”

With a 1998 ballot initiative (I-200) similar to California’s Proposition 209, voters in the state of Washington prohibited the use of race and ethnicity in college admissions. Concerned about the potential drop in minority enrollment as experienced in Texas and California following the ban on affirmative action, admission officers at the University of Washington also modified their college admissions policies and recruitment strategies. UW added essays and solicited additional information in their application materials that could be used to signal ethnic group membership (McCormick 2000). A broader survey of the efforts being attempted by these universities can be found in Long (2007a).

Given that efforts to use proxy indicators for race/ethnicity are being implemented at public universities in all of the states that have eliminated preferential admissions, the results in this paper have national importance, as they are the first to show whether such efforts have succeeded in replacing traditional affirmative action in actual practice.⁶

2. Administrative Data

For this analysis, we use administrative records from three public Texas universities: University of Texas at Austin (UT), Texas A&M University (TAMU), and Texas Tech University (TTU). The time span of these records include several years before and after the judicial ban on affirmative action.⁷ Barron’s Profiles of American Colleges (1996) classifies UT as Very Competitive, TAMU as Highly Competitive, and TTU as Competitive.⁸ In the years for which we have admissions data, the average SAT scores of admitted students were 1,233 at UT, 1,182 at TAMU, and 1,094 at TTU.⁹

The administrative records, which contain a wealth of information about the applicant pool, have been transformed to machine readable format, standardized to the extent feasible, and verified for consistency.¹⁰ While specific data elements vary across the universities, the student records for all of the universities include test scores (e.g., SAT/ACT), class rank percentile, and high school identifiers that allow us to append high school characteristics to the applicant records. Unfortunately, the data generally lack information about students’ high school coursework and their admission essays. We take special note of these data limitations in interpreting the results, and provide a sensitivity analysis of the likely effects of omitted variable bias in the appendix.

3. Methods

Using a probit regression, the following equation is estimated for student i applying to college j in year t:

$$Pr({\mathit{Admitted}}_{\mathit{ijt}}=1)=\mathrm{\Phi }({\beta }_0+{\mathit{U}}_i{\mathit{\beta }}_{jt}+{\mathit{X}}_i{\mathit{\theta }}_{jt}+{\epsilon }_{\mathit{ijt}})$$ (1)

where U is a vector of race and ethnicity indicator variables, X is a vector of other applicant characteristics including measures of high school quality, and ε is a normally distributed error term.

Equation 1 is estimated separately for each college in each year. β_jt represents the added advantage (if any) given to racial and ethnic group applicants at college j in year t, controlling for other applicant characteristics. This method of identifying the degree of affirmative action in admissions has been used previously in several prior studies including Kane (1998a), Long (2004b, 2007b), and Espenshade, Chung, and Walling (2004). Because these studies were based on cohorts of applicants in years prior to the elimination of affirmative action, their results can only suggest how such a policy change would affect admissions decisions. Notably, prior studies could not simulate whether and how universities might shift the weights placed on other applicant attributes. Estimating Equation 1 across successive cohorts of institution-specific applicants enables us to evaluate these policy responses directly. Specifically, three hypotheses are tested:

1. β_jt > 0 for black, Hispanic, and Native American applicants in the years prior to the Hopwood decision (i.e., the colleges practiced affirmative action in their admissions decisions.)

2. In the years prior to the Hopwood decision, β_jt is larger for the more selective colleges.

3. β_jt = 0 in the years after the Hopwood decision (i.e., the colleges did not practice affirmative action in their admissions decisions).¹¹

Additionally, we test whether the universities changed the weights placed on applicant characteristics in such a way as to favor underrepresented minority applicants. To conduct this test, we simulate the admissions decisions that would have occurred in the absence of the Hopwood decision and the top-10% policy. This counterfactual estimation begins by estimating Equation 1 using all applicants from the years 1996 and earlier.¹² We then apply the resulting parameter estimates to each applicant and estimate their probability of admission.¹³ A simulated class of admitted students is constructed by assuming that the university would accept the students with the highest probabilities of being accepted. We assume that university j would accept Z_jt students in year t, where Z_jt is set equal to the actual number of students accepted by university j in year t.¹⁴ We then compare the composition of the simulated class to the students actually accepted to infer the net effect of the Hopwood decision, the top-10% policy, and any other changes to the university’s admissions system.

Next, we estimate the consequences on the ethno-racial composition of admitted students of the judicial ban on affirmative action resulting from Hopwood decision and the top-10% policy by holding the pre-Hopwood admission weights constant, but setting the race-ethnicity coefficients to zero and admitting all in-state applicants who are in the top-10% of their high school class. By comparing the resulting composition of this simulated admission class to that of students who were actually accepted, we can infer the net effects of the university shifting the weights placed on applicant characteristics. These simulations permit us to evaluate the effectiveness of the changing admission policies in restoring minority applicants’ share of the admitted class that would have existed in the absence of the Hopwood ban and the top-10% policy.¹⁵

4. Applicant Characteristics

Before turning to the statistical results, this section discusses various details of the data and the definitions used for several applicant characteristics. In the admissions probit regressions, we use each piece of information that is available for at least 20% of the applicants to university j in year t. Race and ethnicity variables are taken as labeled by the universities. The percentage of students with missing race/ethnicity is below 0.4 percent for each institution.¹⁶ We treat students with missing race/ethnicity as if they were white, which renders our estimates of policy effects conservative.¹⁷

ACT test scores were converted into their equivalent SAT test score values using a conversion table provided by the College Board (Dorans, 2002), and for students who took both tests we use the higher of the two scores.¹⁸ This conversion is valid for SAT scores after the College Board “re-centered” scores upwards in 1996, therefore we have re-centered prior years SAT scores prior to using this conversion.¹⁹

The statistical models include several variables to measure the type and quality of the applicants’ high school, namely whether the applicant’s high school was a private, a “feeder”, a “Longhorn”, and/or “Century” high school. “Feeder” high schools are defined as the top 20 high schools based on the absolute number of students admitted to UT-Austin and Texas A&M in the year 2000, which yields a combined pool of 28 campuses due to considerable overlap between the sets (Tienda and Niu, 2004, 2006a). “Longhorn” high schools are defined as those ever targeted by the University of Texas for the Longhorn Opportunity Scholarships (LOS) for students who qualify for the admission guarantee. According to UT’s Office of Student Financial Services (2005), “these schools were included based on criteria that takes into account their students’ historical under-representation, measured in terms of a significantly lower than average percentage of college entrance exams sent to The University by students from this particular school, and an average parental income of less than $35,000.” “Century” high schools are the LOS counterparts at Texas A&M, namely campuses ever targeted for Century Scholarships for applicants who graduate in the top decile of their senior class. Finally, we include the average SAT/ACT score for the student’s high school, converted into ACT-score equivalent points,²⁰ and the sum of the share taking the SAT and the share taking the ACT in the student’s high school.²¹

Students with missing values for their SAT/ACT score, grade point average, class rank percentile, or their high school’s average SAT/ACT score are assigned the average value for that characteristic among all applicants to university j in year t, and dummy variable flags indicating missing values are included for each of these characteristics. At TAMU, courses taken and participation in various extracurricular activities are taken from SAT surveys. If the SAT survey data are unavailable, each of these variables is set equal to zero, and an indicator variable for missing SAT survey data is added. Students lacking high school identifiers are assumed not to attend a private, feeder, LOS, or Century high school, or high school in the state of Texas. Missing data for advanced placement (AP) course testing also is set equal to zero.²²

5. Results

5.1 University of Texas at Austin

Table 1 reports the admissions probit regression results for UT. To facilitate interpretation, rather than show the parameter estimates for Equation 1, we present marginal effects for each student attribute for a hypothetical applicant with mean characteristics. The first column of Table 1 presents estimates for all applicants between 1990 and 1996, which was before the Hopwood decision. During these years, the likelihood of admission for black and Hispanic applicants was 13 to14 percentage points higher than comparable non-Hispanic white applicants. Year-by-year, cohort-specific estimates (not shown) also reveal consistently positive admission advantages for black and Hispanic applicants before 1996; these range from 9 to 17 percentage points, with no obvious pre-Hopwood trends.²³ Under the affirmative action regime, the admission probability for Asian applicants was indistinguishable from white students. Surprisingly, American Indian applicants were 3 percentage points less likely to gain admission than comparable white students (this estimate, which is statistically insignificant, is based on 468 American Indian applicants).

Table 1.

UT-Austin -- Admission Probit Regression Results^a

Applicant Characteristic	Admission Entry Year
	1990–96	1990–96	1997	1998–03
Race/Ethnicity
Black	0.132***	0.309***	−0.017	0.070***
Hispanic	0.142***	0.302***	−0.023	0.016**
Asian	−0.007	−0.028***	−0.015	0.047***
American Indian	−0.032	−0.072**	−0.042	−0.043
Ethnic=International	0.116***	0.221***	−0.069	−0.257***
U.S. Citizen	0.084***	0.091***	0.069***	0.061***
Female	−0.001	−0.003	0.046***	0.052***
Test Scores and Class Rank
SAT/ACT (00s)	0.152***	0.286***	0.127***	0.153***
SAT/ACT = Missing	−0.680***	−0.543***	−0.776***	−0.566***
Class Rank Percentile (0s)	0.101***	0.175***	0.158***	0.116***
Class Rank PCT = Missing	−0.257***	−0.384***	−0.331***	−0.203***
Top 10%	0.079***	−0.110***	−0.223***	0.166***
AP Exams
Took Test	0.141***	0.258***	0.137***	0.348***
Passed Math Test			−0.005	0.171***
Passed Science Test			0.050	0.096***
Passed Foreign Language Test			0.028	0.092***
Passed Social Science Test			0.010	0.020
Passed Other Test			0.096***	0.011
High School Characteristics
SAT/ACT Average Score (in ACT points)	0.007***	0.017***	0.006	0.003*
SAT/ACT Average Score = Missing	−0.241***	−0.349***	−0.205***	−0.017
% Took SAT + % Took ACT	0.015**	0.051***	0.039	0.018
% Took SAT + % Took ACT = Missing	0.013*	0.021*	0.003	0.066***
Feeder	0.007	0.019***	0.014	0.020***
LOS	−0.033***	−0.050***	−0.054	−0.001
Century	0.006	−0.013	0.034	0.006
Private	0.037***	0.075***	0.048**	−0.019*
In-state	0.096***	0.079***	−0.040**	0.060***
Including In-state Top-10% Applicants?	Yes	No	No	No
Number of Observations	103,547	72,073	10,592	72,553
McFadden’s Psuedo-R2	51.9%	46.8%	51.1%	37.4%
Joint Significance of Race-Ethnicity Variables (p-value)	0.0%***	0.0%***	45.6%	0.0%***

^a“***”, “**”, and “*” indicate two-tailed significance at the 1%, 5%, and 10% levels.

Note: Table displays the marginal effect of each dependent variable for an applicant with mean characteristics. When multiple years are included in the regression, the specification also includes a vector of year of application dummy variables. Standard errors (which are omitted here for space concerns) are available upon request.

The second column of Table 1 again focuses on applicants in the years 1990–96, but excludes those who were in the top-10% of their Texas high school class. Among this group of applicants, race/ethnicity was an even stronger factor in admission. Specifically, the admission likelihood for Hispanic and black applicants was 30 to 31 percentage points higher than comparable non-Hispanic white applicants. This result is consistent with evidence that affirmative action largely involved students ranked in the second decile of their class because UT had a de facto practice of admitting applicants who graduated in the top 10 decile of their high school class (Alon and Tienda, 2007). Asian and American Indians were at disadvantages of 3 and 7 percentage points, respectively. The bottom row of Table 1, which shows the P-value for the joint test of significance for group status, reveals that jointly, race and ethnicity was a highly significant determinant of an applicant’s likelihood of admission during this period. These results accord with claims that UT practiced affirmative action for black and Hispanic applicants in the years prior to the Hopwood decision (THECB, 1998).²⁴

To comply with the judicial ban imposed by the Hopwood decision, UT immediately eliminated the admission advantages given to black and Hispanic applicants. Among students who did not graduate in the top-10% of a Texas high school, namely the students for which UT had discretion in the admission decision, the marginal effects on the likelihood of admission for black and Hispanic applicants fell to an insignificant -2 percentage points in 1997, the year the judicial ban on affirmative action was in force, but before the uniform admission law went into effect. During the years 1998–03, a period covered both by the judicial ban and the top 10% law, the marginal effects on the likelihood of being admitted in individual years ranged from +4 to +13 percentage points for blacks and from −3 to +6 percentage points for Hispanics relative to comparable white applicants.²⁵ The last column of Table 1 shows that over this whole period blacks and Hispanics enjoyed significant admission advantages of 7 and 2 percentage points, respectively. It is noteworthy that the magnitudes of the estimated advantages given to minority applicants in the years 1998–03 were substantially smaller than the advantages given in the pre-Hopwood years. Moreover, in the year 2003, the treatment of black and Hispanic youth strongly diverge, with black applicants receiving a significant positive boost of 13 percentage points and Hispanic applicants receiving a significant 3 percentage point reduction in their likelihood of admission (relative to whites not in the top-10%).²⁶

It is crucial to underscore that these results do not necessarily imply that UT was using the applicant’s race or ethnicity in making their admissions decisions. Rather, UT may have changed the weights they placed on other applicant characteristics that are not available to us (e.g., essays or high school coursework), or used the observed applicant characteristics in nonlinear or interactive ways (e.g., class rank × took AP exam, etc.) that favor black and Hispanic students. The Office of Admissions at UT published criteria considered in admission since 1997, explicitly noting that in addition to the academic index based on the high school record (class rank, completion of the required curriculum, completion of additional academic units, and standardized test scores), students would be evaluated on a personal achievement index (PAI) that included scores on two essays, leadership and extracurricular activities, awards/honors, work experience, community service, and special circumstances (University of Texas at Austin, Office of Admissions, 2005). The latter includes both family and school socioeconomic status, as well as home language, domestic responsibilities, and students’ test scores relative to the average for their high school. In short, the PAI scores enabled admissions officers to take a “holistic approach” in reviewing applications.²⁷

Although informative, the marginal effects can not answer whether UT admission officers changed the weights accorded to observed applicant characteristics in ways that favored minority applicants. A few results are suggestive, however. First, the positive weight placed on an applicant’s SAT/ACT test score declined by half post-Hopwood. Second, the admission advantages enjoyed by applicants from high schools with high average SAT/ACT scores nearly disappeared in the post-Hopwood years.²⁸ Black and Hispanic students comprise relatively small shares of the student bodies at affluent high schools with high average SAT scores (Tienda and Niu, 2006b).²⁹ Finally, attending a high school that was targeted for the Longhorn Opportunity Scholarships lowered applicants’ likelihoods of admission in the years 1990–1996, but had an insignificant effect on their likelihood of admission in the post-Hopwood years. Likewise, attending a private high school, which had been a positive factor before 1997, became a liability in the post-Hopwood years, Each of these changes is likely to benefit minority applicants. The simulation presented in section 5.4 illustrates how these changes, along with other components of the admissions policy, altered the composition of the admitted class.

5.2 Texas A&M University

Table 2 reports the regression results for TAMU, which gave similar admission boosts to black and Hispanic applicants as UT in the pre-Hopwood years. From 1992 to 1996, a black or Hispanic applicant’s admission probability was 12 to 13 percentage points higher than a comparable non-Hispanic white applicant at TAMU, with some annual variation. Cohort-specific estimates reveal that preferences for black and Hispanic applicants rose from 8–10 percentage points in 1992 and 1993 to 15–18 percentage points in 1994 and 1995, and subsequently declined to 9–10 percentage points in 1996 (perhaps suggesting that TAMU responded in advance to the anticipated Hopwood ruling).³⁰ When restricted to applicants not in the top-10% of their Texas high school class, black and Hispanic admission preferences were higher still, averaging 28–29 percentage points for the pre-Hopwood period, with considerable variation across years (23–26 percentage points in the years 1992 and 1993, 32–36 percentage points in 1994 and 1995, and 18–19 percentage points in 1996). The large fluctuation in the size of the admission boosts for black and Hispanic students across years suggests that the university did not use pre-established quotas to make their decisions.³¹

Table 2.

Texas A&M -- Admission Probit Regression Results^a

Dependent Variable	Admission Entry Year
	1992–96	1992–96	1997	1998–02
Race/Ethnicity
Black	0.124***	0.286***	−0.040	−0.021
Hispanic	0.132***	0.280***	0.014	−0.013
Asian	−0.014**	−0.029**	−0.110***	−0.141***
American Indian	0.008	0.020	−0.071	0.039
Ethnic=Other	−0.065***	−0.102***	−0.063**	−0.078***
U.S. Citizen	0.062***	0.113***	0.085***	0.071***
Female	−0.006**	−0.011**	0.043***	−0.011**
Test Scores
SAT/ACT (00s)	0.102***	0.183***	0.122***	0.125***
SAT/ACT = Missing	−0.224***	−0.312***	−0.334***	−0.313***
Class Rank Percentile (0s)	0.084***	0.147***	0.152***	0.141***
Class Rank PCT = Missing	−0.037***	−0.074***	−0.083***	−0.126***
Top 10%	0.100***	0.026*	0.130***	0.156***
Curriculum
Took AP Test	−0.024***	−0.043***	−0.042***	−0.060***
Took 4 Years of HS English	0.001	0.006	−0.013	−0.003
Took 2 Years of HS Foreign Language	0.026***	0.032***	−0.007	0.030**
Took 3 Years of HS Math	0.024***	0.037**	0.004	0.002
Took 2 Years of HS Science	−0.042***	−0.068***	−0.042	−0.081***
Activities
HS Band	0.015***	0.030***	0.032	0.002
HS Athlete	0.021***	0.040***	0.076***	0.040***
HS Drama			0.009	0.030***
HS Student Government	0.016***	0.031***	0.074***	0.071***
HS National Honor Society	0.048***	0.088***	0.058***	0.047***
HS Year Book	−0.028***	−0.041***	0.050**	−0.010
HS Hobby Club	−0.003	−0.005	0.004	−0.006
SAT Survey Data Missing	0.098***	0.202***	0.150***	0.090***
High School Characteristics
SAT/ACT Average Score (in ACT points)	0.005***	0.010***	−0.008	−0.020***
SAT/ACT Average Score = Missing	0.041***	0.072***	0.082**	0.177***
% Took SAT + % Took ACT	0.036***	0.059***	−0.039	−0.024**
% Took SAT + % Took ACT = Missing	−0.013**	−0.028**	−0.039*	0.042***
Feeder	0.003	0.010	−0.012	0.018**
LOS	−0.036***	−0.055**	−0.064	−0.012
Century	0.021**	0.050***	0.092**	0.193***
Private	0.022***	0.047***	0.018	−0.005
Instate	0.202***	0.245***	0.195***	0.182***
Including In-state Top-10% Applicants?	Yes	No	No	No
Number of Observations	69,691	46,701	10,016	52,525
McFadden’s Psuedo-R2	40.4%	0.5%	29.8%	24.7%
Joint Significance of Race-Ethnicity Variables (p-value)	0.0%***	0.0%***	0.0%***	0.0%***

^a“***”, “**”, and “*” indicate two-tailed significance at the 1%, 5%, and 10% levels.

Note: Table displays the marginal effect of each dependent variable for an applicant with mean characteristics. When multiple years are included in the regression, the specification also includes a vector of year of application dummy variables. Standard errors (which are omitted here for space concerns) are available upon request.

From 1998 to 2002, black and Hispanic applicants were slightly less likely to be accepted than comparable white youth, although this difference was insignificant. This average effect conceals substantial year to year fluctuations, however. The estimated penalty (boost) accorded to black and Hispanic applicants, respectively, ranged from −8 to +11 and from −9 to +9 percentage points, with no clearly discernable trends.

Among applicants not in the top-10% of their Texas high school class, the admission probability for Asian Americans was 3 percentage points lower than comparable white applicants between 1992 and 1996. After the top 10% law was in force, this disadvantage widened to 14 percentage points. This admission disadvantage, which ranged from 6 to 19 percentage points over the post-Hopwood years, was statistically significant in each year. Again, it bears emphasizing that this result does not necessarily indicate that Texas A&M discriminated against Asian American applicants. Among Texas students who did not graduate in the top decile of their class, it is possible that TAMU considered other unobservable applicant characteristics or used the observed applicant characteristics in ways that favor white students.³²

In every year during both the pre- and post-Hopwood periods, race and ethnicity were highly influential in determining the likelihood that an applicant to TAMU would be admitted, although the winners and losers differed by period and demographic group. By law, the university could not directly consider these ascribed characteristics in their admission decisions after 1996, therefore the apparent disadvantages experienced by black, Hispanic, and Asian applicants likely reflect either the weight placed on other applicant characteristics that are not available in the administrative data, or interactive combinations of the observed characteristics that favor white applicants. While these results show that university admission officers did not place weights on unobserved applicant characteristics in ways that favored black or Hispanic applicants after the judicial ban on affirmative action, it is both permissible and plausible that they changed the weights placed on observed applicant characteristics in a manner that boosted minority students’ likelihood of admission. For example, when we restrict the sample to students not in the top-10% of their Texas high school class, we observe declining weight placed on the student’s SAT/ACT and their high school’s average SAT/ACT score, and increasing weight placed on attending a high school that was targeted for the Longhorn or Century Scholarships. The effects of these changes are simulated in Section 5.4.

5.3 Texas Tech University

Apparently Texas Tech did not mirror UT-Austin and Texas A&M in giving sizeable admission boosts to black and Hispanic applicants in the years before Hopwood (see THECB, 1998). In fact, as shown in Table 3, from 1991 through 1996 black applicants were slightly less likely (−1 percentage points) to be accepted than comparable white applicants, while Hispanic applicants were slightly more likely to be accepted (+1 percentage points). Asian American applicants were significantly less likely to be accepted than comparable white applicants, by 9 percentage points.

Table 3.

Texas Tech -- Admission Probit Regression Results^a

Dependent Variable	Admission Entry Year
	1991, 1993, 1995, 1996	1991, 1993, 1995, 1996	1997	1998–03
Race/Ethnicity
Black	−0.013*	−0.008	0.030	−0.105***
Hispanic	0.013***	0.024***	0.032	−0.046***
Asian	−0.091***	−0.084***	0.024	−0.067***
American Indian	−0.033	−0.055	0.008	−0.003
Ethnic=International	−0.376***	−0.408***	0.085	0.158***
U.S. Citizen	0.001	0.002	−0.066	0.064***
Female	−0.018***	−0.026***	0.030	−0.003
Test Scores and Class Rank
SAT/ACT (00s)	0.012***	0.022***	0.105***	0.123***
SAT/ACT = Missing	−0.286***	−0.357***	−0.963***	−0.650***
Class Rank Percentile (0s)	0.047***	0.063***	0.004	0.065***
Class Rank PCT = Missing	−0.844***	−0.852***	−0.627***	−0.733***
Top 10%	−0.027***	−0.025	0.051	−0.169***
High School Characteristics
HS SAT/ACT Average Score (in ACT points)	0.009***	0.015***	−0.004	0.018***
HS SAT/ACT Average Score = Missing	0.005	−0.001	0.050	0.036**
% Took SAT + % Took ACT	0.012	0.017	−0.075	−0.008
% Took SAT + % Took ACT = Missing	−0.003	0.008	0.008	0.000
Feeder HS	0.014**	0.017**	−0.108***	−0.024***
LOS HS	−0.018*	−0.014	−0.168*	−0.134***
Century HS	−0.017	−0.007	−0.102	−0.020
Private HS	0.020*	0.027*	−0.077	−0.008
Instate HS	0.006	0.012	0.002	0.003
Including In-state Top-10% Applicants?	Yes	No	No	No
Number of Observations	25,509	20,892	6,046	50,661
McFadden’s Psuedo-R2	51.3%	51.0%	84.8%	45.9%
Joint Significance of Race-Ethnicity Variables (p-value)	0.0%***	0.0%***	76.8%	0.0%***

^a“***”, “**”, and “*” indicate two-tailed significance at the 1%, 5%, and 10% levels.

Note: Table displays the marginal effect of each dependent variable for an applicant with mean characteristics. When multiple years are included in the regression, the specification also includes a vector of year of application dummy variables. Standard errors (which are omitted here for space concerns) are available upon request.

This practice continued after the judicial ban on affirmative action. Between 1998 and 2003, black, Hispanic, and Asian American applicants who did not qualify for automatic admission under the uniform admission law were significantly less likely to be accepted than comparable white applicants. Moreover, the admission disadvantages for black applicants to TTU who did not graduate in the top decile of their Texas high school were even larger compared with pre-Hopwood levels, ranging from 9 to 12 percentage points over the period and statistically significant in each year. Likewise, beginning in 2000, Hispanic applicants witnessed a significant admission disadvantage relative to comparable white students, which ranged from 3 to 6 percentage points through 2003. Asian American applicants also were 4 to 8 percentage points less likely to be accepted at TTU than comparable white students who applied for admission between 1998 and 2003; moreover, their acceptance disadvantages reached statistical significance after 2001. Thus, there does not appear to be any evidence that TTU’s admission officers used unobserved applicant characteristics in ways that boosted the admissions probabilities of black and Hispanic applicants when the ban on affirmative action was in force – rather, it appears that Texas Tech placed such applicants at a disadvantage post-Hopwood.

With respect to observable applicant characteristics, the results are mixed. On the one hand, the positive weights placed on attending a feeder or private high school declined, which would tend to favor minority applicants. On the other hand, the weight placed on the student’s SAT/ACT scores increased, and attending a Longhorn Opportunity High School became a disadvantage to the applicant. The following section evaluates the relative impact of changing admission policies on the ethno-racial composition of these three universities.

5.4 Simulated Effects of Hopwood, the Top-10% policy, and University Responses

Because the weights assigned to several individual attributes appear to have changed at each of the three universities, we disaggregate the net impacts using a simulation exercise. The composition of the admitted class is simulated by computing the hypothetical probability of admission (as shown in Equation 1, with a standard normal error term added to β̂₀ +U_iβ̂_jt + X_iθ̂_jt) and “accepting” Z_t number of students with the highest probabilities of admission. Z_t is set equal to the number of students who were actually admitted by each institution in year t.^{33, 34}

Table 4 reports the shares of white, black, Hispanic and Asian students who would be admitted (with American Indian and International shares not shown) along with the mean SAT/ACT of admitted students. Each panel shows four counterfactuals:

Table 4.

Effects of University Responses to Hopwood and the Top-10% Policy

Period & Place	Counterfactual	%white	%black	%Hispanic	%Asian	SATACT
UT-Austin
1990–96: Pre-Hopwood	A: Actual Admits	65.1	4.0	15.9	12.8	1,229
	B: Predicted Admits	65.2	4.0	15.9	12.7	1,228
	C: AA Ban	68.0	3.1	13.2	13.3	1,232
	D: AA Ban + Top 10%	67.2	3.2	14.0	13.3	1,228
1997: Hopwood	A: Actual Admits	65.7	3.2	12.4	16.0	1,234
	B: Predicted Admits	63.4	3.9	13.8	15.5	1,235
	C: AA Ban	65.1	3.1	12.2	15.9	1,238
	Policy Change (C-B)	1.9	−1.0	−1.8	0.4	3
	Discretionary (A-C)	0.0	0.2	0.1	0.1	−4
1998–03: Top-10%	A: Actual Admits	61.7	3.9	14.5	17.1	1,236
	B: Predicted Admits	60.3	4.3	15.6	16.0	1,247
	C: AA Ban	62.8	3.3	12.9	16.7	1,251
	D: AA Ban + Top 10%	61.7	3.7	14.0	16.6	1,244
	Policy Change (D-B)	2.1	−0.8	−2.2	0.5	−3
	Discretionary (A-D)	−0.9	0.4	0.7	0.4	−8
Texas A&M
1992–96: Pre-Hopwood	A: Actual Admits	73.4	5.1	14.6	5.9	1,172
	B: Predicted Admits	73.3	5.1	14.6	5.9	1,172
	C: AA Ban	76.7	3.9	12.0	6.3	1,175
	D: AA Ban + Top 10%	76.4	4.0	12.2	6.3	1,174
1997: Hopwood	A: Actual Admits	75.1	3.7	11.5	6.7	1,185
	B: Predicted Admits	72.4	4.7	13.1	6.9	1,188
	C: AA Ban	75.1	3.5	10.9	7.3	1,191
	Policy Change (C-B)	3.4	−1.4	−2.6	0.5	3
	Discretionary (A-C)	−0.6	0.3	0.7	−0.6	−6
1998–02: Top-10%	A: Actual Admits	76.8	3.4	11.2	6.5	1,187
	B: Predicted Admits	74.1	4.2	13.0	6.8	1,194
	C: AA Ban	77.3	3.1	10.4	7.2	1,197
	D: AA Ban + Top 10%	76.9	3.2	10.7	7.1	1,195
	Policy Change (D-B)	3.7	−1.3	−2.3	0.5	1
	Discretionary (A-D)	−0.5	0.3	0.4	−0.7	−8
Texas Tech
1991–96: Pre-Hopwood	A: Actual Admits	78.6	4.4	13.6	2.3	1,057
	B: Predicted Admits	79.0	4.3	13.8	2.3	1,057
	C: AA Ban	78.8	4.4	13.5	2.6	1,057
	D: AA Ban + Top 10%	78.8	4.4	13.5	2.6	1,057
1997: Hopwood	A: Actual Admits	81.8	3.9	10.3	2.8	1,097
	B: Predicted Admits	82.1	3.7	10.6	2.7	1,097
	C: AA Ban	82.0	3.8	10.3	3.0	1,098
	Policy Change (C-B)	−0.1	0.1	−0.3	0.3	0
	Discretionary (A-C)	−0.2	0.1	0.0	−0.2	−1
1998–03: Top-10%	A: Actual Admits	80.3	3.5	11.4	3.3	1,106
	B: Predicted Admits	79.8	4.0	12.6	3.0	1,098
	C: AA Ban	79.6	4.1	12.3	3.4	1,099
	D: AA Ban + Top 10%	79.5	4.1	12.3	3.4	1,099
	Policy Change (D-B)	−0.3	0.1	−0.3	0.4	1
	Discretionary (A-D)	0.8	−0.5	−0.9	−0.1	7
Conterfactual:
A: Actual Admits =	Actually admitted.
B: Predicted Admits =	Simulated by applying the 1990–96 admissions formula in each year.
C: AA Ban =	Simulated by applying the 1990–96 admission formula, but setting race-ethnicity coefficients to zero (i.e., automatic effect of Hopwood).
D: AA Ban + Top 10% =	Simulated by applying the 1990–96 admission formula, but setting race-ethnicity coefficients to zero and admitting all in-state top-10% students (i.e., automatic effect of Hopwood and Top-10% policy).
Policy Change (D-B) =	Effect of policy changes in the University’s admissions required by Hopwood and the Top-10% Policy for the years after 1997.
Discretionary (A-D) =	Net effect of discretionary changes in the University’s admissions policy for the years after 1997.

1. “Actual Admit Class” = Students actually admitted;

2. “Predicted Admit Class with AA” = Admitted students simulated by applying the estimated 1990–96 admissions formula in each year;

3. “Predicted Admit Class with AA Ban” = Admitted students simulated by applying the estimated 1990–96 admission formula, but setting race-ethnicity coefficients to zero; and

4. “Predicted Admit Class with AA Ban and Automatic Top-10%” = Admitted students simulated by applying the estimated 1990–96 admission formula, but setting race-ethnicity coefficients to zero and admitting all in-state top-10% students.

Table 4 shows the simulation results. By comparing the distributions of admitted students between counterfactuals A and B (i.e., A – B), we obtain an estimate of the total effect of Hopwood, the top-10% policy, and other changes to the universities’ admissions policies. For the years after 1997, this total effect can be decomposed into A – D and D – B, where D – B reveals the admission consequences of Hopwood and the top-10% law, and A – D shows the effects of discretionary changes in the admissions policy implemented by each institution.

For UT, the row designating net effects of the policy change shows that the combined effects of Hopwood and the top-10% law shifted the composition of the admission class towards whites and Asian-Americans, and away from blacks and Hispanics. For blacks and Hispanics, the net effect of these policy changes lowered their combined shares of admitted students from 19.8% to 17.7%. However, other changes in UT’s admissions decisions offset some of these effects. The discretionary changes in UT’s admissions decisions shifted the composition of UT’s admitted students towards blacks, Hispanics, and Asian-Americans, and away from whites. This policy response helped recover the combined share of admitted black and Hispanic students to 18.4%. That is, UT’s response produced a 32% rebound in black and Hispanic students’ combined share of admitted students (i.e., (18.4% - 17.7%)/(19.8% - 17.7%)).³⁵ Nonetheless, despite a re-weighting of applicant characteristics in a legally compliant manner, the university was unable to maintain the share of black and Hispanic students that would have been admitted under a regime that allowed explicit consideration of race.

Finally, it is worth noting that the combined effects of Hopwood, the top-10% policy, and UT’s response led to a modest 11-point reduction in the average SAT/ACT score of admitted students. Most of this decrease resulted from discretionary changes in the weights placed on other applicant characteristics by UT’s admission officers rather than the mandatory policy changes. This result might be surprising to those who expect the elimination of affirmative action at selective institutions to substantially raise the average ability level of admitted students, and it challenges critics who might attribute the drop in average test scores to the top 10% law.

The second panel of Table 4 shows the collective effects of changes in TAMU’s admission regime. As occurred at UT, the winners from the combined effects of Hopwood and the top-10% policy were white and Asian American applicants, whose share of the admission pool rose. Again, blacks and Hispanics were the losers in the shuffle produced by the affirmative action ban and the top 10% law combined; the net effect of these policy changes was to lower their combined shares of admitted students from 17.2% to 14.0%. As occurred at UT, changes in the weights placed on other applicant characteristics by TAMU’s admissions officials offset some of these losses. The discretionary changes in TAMU’s admissions decisions increased black and Hispanic students’ combined share of admitted students to 14.5%. Thus, TAMU’s discretionary response in admissions led to a 18% rebound in black and Hispanic students’ combined share of admitted students (i.e., (14.5% - 14.0%)/(17.2% - 14.0%)).³⁶ Further, the collective impact of changes in policy and weights assigned to applicant attributes produced a 7-point decline in the average SAT/ACT score of students admitted to TAMU, a drop due entirely to changes in the weights placed on applicant characteristics. Our results showing that the strategies to restore campus diversity pursued at UT and TAMU did not restore minority representation to the levels that would have occurred with traditional affirmative action are consistent with the findings in Bucks (2003) and Kain and O’Brien (2004).

The third panel of Table 4 presents the simulation results for Texas Tech. The policy change row shows the combined effects of TTU complying with both the Hopwood ruling and the top 10% law. For this simulation, compliance with the Hopwood ruling would require Texas Tech to eliminate pre-Hopwood admissions advantages for Hispanic applicants, and disadvantages for black and Asian American applicants. This simulation thus predicts a modest decline in the share of white and Hispanic students admitted, combined with increases in the shares of black and Asian American students. The final row shows the effect of discretionary changes in TTU’s admissions policy, which benefited white students and lowered other students’ admission shares. On balance, there is no evidence that TTU sought to boost the admission prospects of black and Hispanic students in the wake of the Hopwood decision. Finally, in contrast to the experiences of the two public flagships, the shift in admission regimes coupled with changes in the institution-specific admission system raised the average SAT/ACT score of students admitted to TTU by 7 points.

Figure 1 diagrams the net effects on the composition of admitted students of the Hopwood decision, the Top-10% policy, and other discretionary changes in the admission decisions of these three universities. This figure corresponds to a plot of the difference between the “Actual Admits” and “Predicted Admits” under the admission regime that permitted affirmative action, but disaggregated by individual years. At all three universities, which differ appreciably in the selectivity of their admissions, white students were the clear winners in that their rising shares among the admitted were accompanied by corresponding declines for black and Hispanic applicants. Not surprisingly, the impact of the change in admission regime was greatest at TAMU and UT, which gave admission boosts to minority applicants before the judicial ban imposed by the Hopwood decision. Using the simulation results shown in Table 4, and given the total numbers of students admitted in the post-Hopwood years, we find that the combination of the policy and discretionary changes reduced the number of accepted black and Hispanic applicants by 1,288 at UT (1997–03), 1,813 at TAMU (1997–02), and 746 at TTU (1997–03).

Figure 1.

Net effect on the ethnic and racial composition of admission cohorts of Hopwood, the Top 10% law, and allowable discretionary changes in university admission policies: UT-Austin, Texas A&M, and Texas Tech, 1991 – 2003.

6. Conclusion

The Hopwood decision and the top-10% policy had sizable effects on the racial and ethnic composition of public universities in Texas—but the winners and losers differed across institutions and over time. We find evidence that the public flagships (UT-Austin and Texas A&M) offered significant advantages to black and Hispanic applicants prior to the Hopwood decision. Both universities responded to changes in admission policies by shifting the weights they placed on applicant characteristics in ways that boosted the admissions probabilities of black and Hispanic applicants. However, these efforts did not fully compensate for the effects of the Hopwood decision in lowering the odds of admission for blacks and Hispanics. Public universities were unable (or did not sufficiently attempt) to proxy race and ethnicity using other applicant attributes, although UT’s Personal Achievement Index (PAI) sought to weight extracurricular and extraordinary circumstances in their admission decisions in ways that could have boosted minority applicants’ admission probabilities. Further, the two flagships targeted fellowships to high schools with low college going traditions. Although minority students comprised large shares of the student body at the Longhorn and Century high schools, they are less likely than whites to qualify for the admission guarantee, and more likely to report financial barriers as reasons for not enrolling (Tienda and Niu, 2006b). This suggests the need for financial aid as a necessary adjunct to race-sensitive admission criteria as an ingredient for diversifying college campuses. Finally, we find no evidence that Texas Tech University gave preferences to minority applicants in the pre- or post-Hopwood period, and changes in TTU’s post-Hopwood admissions policy lowered the probability of acceptance for minority applicants.

Also unclear is whether public universities that did not use full file review could have used proxies in ways that allowed them to maintain campus diversity achieved before the Hopwood decision. Simulations produced by Kane (1998b) suggest the folly of using proxies in this manner; in order to maintain the same admissions rates for black and Hispanic applicants, the new admissions rules imply that colleges in the top-quintile of the SAT/ACT distribution would have to include a lower chance of admission for students with higher SAT scores! Thus, our findings showing only modest shifts in the weights used by these colleges in their admissions systems are highly plausible.

Minority applicants were the “winners” of the institutional responses by UT and Texas A&M. Despite their intentions in restoring and maintaining campus diversity, however, these institutional efforts were unable to offset the deleterious effects of the judicial ban on affirmative action imposed by the Hopwood decision. In the end, minority applicants were the net “losers” of the changing admission regimes while whites continued to maintain their admission advantage. This outcome is all the more remarkable in a state where black and Hispanic students are rapidly moving towards being the majority of high school graduates.