This paper uses student-level data to investigate how the college application behavior of underrepresented minorities (URMs) changed in response to the 1998 end of affirmative action in admissions at the University of California (UC). We show that all URMs experienced a drop in their probability of admission to at least one UC campus. However, the relative decline in URM SAT score-sending rates—our proxy for application rates—was small and concentrated at Berkeley and UCLA among underrepresented minorities who experienced the largest relative drop in their predicted probability of admission. In addition, we find some evidence of a shift toward less-selective UC campuses rather than out of the UC system. Overall, our paper highlights the stability of URM application behavior in the face of substantial declines in their admission rates.

Affirmative action in college admissions continues to be a central issue in American higher education. Although the debate is often focused on abstract principles (e.g., should past inequities be corrected by differential treatment?), several empirical questions are prominent in the discussion. For example, it is often argued that the abolition of racial preferences sends a message of institutional hostility to potential minority students, discouraging applications and further lowering the already depressed college graduation rates of blacks and Hispanics.1 Indeed, later this year, the Supreme Court will hear Fisher v. University of Texas in which the University of Texas is defending the reintroduction of explicit racial preferences partly on the rationale that bans on affirmative action dissuade minority students from applying.2

In this paper, we explore the effect of California's statewide ban on racial preferences (also known as Prop 209) on the application rates of underrepresented minorities (URMs) to each of the eight campuses enrolling undergraduates that made up the University of California (UC) system at the time the ban went into effect.3 The simultaneous removal of racial preferences from the different UC campuses (which varied widely in their admissions selectivity and demographics) and the extensive publicity that accompanied this shift provide an excellent setting in which to study the effect of affirmative action bans on the application behavior of minority students. Using administrative data from the University of California on every first-year fall semester applicant to the University of California, we document the effect of the end of racial preferences in admission on URMs’ relative chances of admission to each campus, and highlight how these changes depended on students’ academic credentials. Then, using data from the College Board, we relate these changes in the probability of admission to changes in SAT score-sending behavior (our proxy for application behavior) to schools both inside and outside the UC System.4

Like us, Long (2004) examines the effect of affirmative action bans on SAT score-sending behavior and finds that changes in the probability of admission predict changes in score-sending. The paper argues that the affirmative action bans in California and Texas caused URMs to send scores to lower-quality colleges and non-URMs to send scores to higher quality colleges. A weakness of Long's work, however, is that he uses data from the National Education Longitudinal Study (NELS) to generate a student's predicted probability of admission to different schools. The NELS, however, reports admissions outcomes only for students’ top two college choices, which are likely to reflect students’ beliefs about their chances of admission. In contrast, we have data on every campus to which applicants to the UC system applied. Thus, we are able to more precisely gauge the change in the likelihood that students with different academic credentials were admitted to the various UC campuses. A second limitation of Long's analysis is that, rather than having data on score-sending to specific institutions, Long's data divide universities into quintiles based on academic quality. In contrast, our data on SAT score-sending contain institution names, allowing us to look at the change in score-sending rates to each UC campus. This turns out to be important because the effects of the ban on affirmative action varied considerably by campus.

Card and Krueger (2005) also investigate the effect of California's ban on the use of racial preferences on the SAT score-sending behavior of URMs.5 They find no evidence that minority students with SAT scores above 1150 or A/A+ GPA reduced the rate at which they sent scores to at least one of the three most selective UC schools (Berkeley, UC Los Angeles [UCLA], and UC San Diego). One drawback of their analysis, however, is that Card and Krueger did not have access to detailed information on how the end of affirmative action affected URMs’ chances of admission. Thus, although their paper provides strong evidence that highly qualified students were not discouraged from applying to the most selective UC campuses after the end of affirmative action, it is able to present only a partial picture of the changes brought about by the end of affirmative action. A major contribution of this paper is to link the relative changes in URMs’ predicted probability of admission at each of the eight UC campuses to their changes in score-sending. We also expand upon Card and Krueger by extensively exploring treatment heterogeneity: We show that the applicants whose score-sending to selective UCs dropped the most were the ones with the largest falls in admissions probability.

In addition to Long (2004) and Card and Krueger (2005), several other papers examine the effect of bans on affirmative action in college admission on other aspects of the college admission process. Backes (2012) and Hinrichs (2012) find that enrollment shares of URMs at selective universities fall in the wake of affirmative action bans but find little evidence of a change in the URM share of the total college-going population. Long and Tienda (2008) show how the removal of racial preferences in Texas affected URMs’ relative admissions probabilities, with a focus on how university admissions offices reweighted applicant characteristics to favor URMs. Arcidiacono et al. (2011) examine the effect of Prop 209 on the share of URMs enrolled in different types of colleges and URM graduation rates, and find evidence that URM graduation rates (the fraction of enrolled students who graduate) increased at the University of California system after Prop 209. Finally, Antonovics and Sander (2013) examine the likelihood that URMs enroll in UC schools conditional on being accepted and find evidence that URM enrollment rates increased after the end of affirmative action.

There are several ways in which this analysis adds to the previous literature. First, we use both admissions and score-sending data to contrast the relative changes in each following an affirmative action ban. Because admissions probabilities play an important role in the application decision, it is natural to motivate changes in score-sending by first tracing out changes in admissions probabilities. Second, we extensively explore treatment heterogeneity across both applicant quality and university selectivity. Each of these turn out to be highly salient and have been largely overlooked by previous studies, which have focused on highly qualified applicants to selective universities.

Our approach yields three important findings. First, we show that all URMs, regardless of their academic credentials, were affected by the ban on affirmative action. Relative to white applicants, even the most highly qualified URMs experienced a substantial fall in their chances of admission to the most selective UC schools (i.e., Berkeley and UCLA), and less qualified URMs became relatively less likely to gain admission to the least selective UC schools (i.e., UC Santa Cruz and UC Riverside). Second, the relative decline in URM score-sending rates after the end of affirmative action was small and concentrated at Berkeley and UCLA among URMs who experienced the largest relative drop in their predicted probability of admission. Third, for the group of URMs with the largest relative drop in their score-sending rates to Berkeley and UCLA, we find some evidence of a relative increase in the number of scores URMs sent to less selective UCs, though the magnitude of this shift appears to have been small. Overall, our paper highlights the stability of URM application behavior in the face of substantial declines in their admission rates.

This paper proceeds as follows. Section 2 discusses institutional background in California and the details of the affirmative action ban. Section 3 presents our theoretical model, and section 4 our empirical specification. Section 5 discusses our data sets. Section 6 shows results for admissions rates and section 7 for score-sending. Finally, section 8 concludes.

The University of California is one of the largest public university systems in the world, and with several campuses that do well in world rankings, it is unique in its combination of quality and scale. In the mid 1990s—the period on which this paper focuses—the UC system matriculated about 24,000 freshmen per year across eight campuses. Because California graduates more than 10 times that many students from high school each year, admission into the UC system has long been quite competitive.

If the University of California as a whole was constrained in admitting URMs even before the ban on affirmative action, the individual campuses were fairly unconstrained in competing for the modest number of URMs who made it into the eligible pool. The ironic result was that around half of all the blacks and Hispanics who matriculated at UC schools in the mid 1990s entered the two most elite and competitive campuses, Berkeley and UCLA. Blacks, for example, made up about 8 percent of the new freshmen at Berkeley, but only 2 percent of the new freshmen at UC Irvine, a less elite campus.

The first threat to affirmative action in California was in July 1995, when the Board of Regents of the University of California passed a resolution (SP-1) that stipulated that UCs would discontinue considering race in admissions by the beginning of 1997. The implementation of SP-1, however, was delayed following the passage in November 1996 of Proposition 209 (Prop 209), which banned the use of racial preferences in university admissions. Prop 209 underwent legal challenges until the Supreme Court denied further appeals in November 1997. Thus, the incoming class of 1998 was the first to be admitted under the statewide ban on affirmative action.

It is important to recognize that in an effort to minimize the effects of Prop 209 on minority enrollment, UC campuses increased minority outreach efforts.6 These efforts were widely viewed as ineffective, however, at least initially.7 Part of the reason for lack of effective programs was that in the immediate aftermath of Prop 209, there were concerns about whether race-specific outreach (as opposed to, for example, targeting low income areas) was permitted after Prop 209. In addition, outreach programs focusing on elementary and middle schools would not have an effect until those students reached college age. To the extent that increased outreach had an effect, our measured effects of Prop 209 will be the net effect of the ban and the change in outreach efforts.

In 2001, the UC implemented Eligibility in a Local Context, guaranteeing admission to the top 4 percent of students in each California high school (conditional on completing specified coursework). The new policy was designed to attract students from high schools which did not typically send many students to UCs, giving the UC a way to potentially increase URM enrollment. Because this policy implemented a new admissions rule for UC campuses, we do not use any data later than 2000 in an effort to isolate the effects wrought by Prop 209.

Conceptually, Prop 209 could have affected application rates to the UC through two primary channels. First, by directly lowering URMs’ chances of admission, Prop 209 is likely to have affected the expected benefit of sending applications to UC schools. Second, Prop 209 also may have changed the utility URMs expected to receive from attending a UC school because of, for example, the change in the racial composition of the student body. Ideally, we would like to be able to use the relative changes in URMs’ application rates after Prop 209 to say something about the latter channel because, as mentioned earlier, there has been considerable debate about whether bans on affirmative action lessen URMs’ interest in attending the schools directly affected by the bans. Our ability to do this, however, is complicated by the fact that it's not clear how changes in admission rates should influence application rates.

In a model in which the likelihood of admission falls at a single school, it is clear that the expected benefit of an application to that school will fall, so application rates should also fall. In a model where admissions rates fall for an entire set of schools, however, Long (2004) shows that the optimal number of applications sent to schools within that set could increase. The intuition is that if students place a relatively high value on attending the set of schools at which their admissions chances fall, then they may send more applications to those schools in an effort to avoid being pushed into less-attractive alternatives. Thus, we cannot take our finding that URM application rates remained stable after Prop 209 as evidence that Prop 209 did not negatively affect the utility URMs expected to receive from attending UC schools. Rather, the stability of URM application rates may partially reflect URMs’ responses to the fall in their likelihood of admission to UC schools. This is particularly problematic in our setting because we find that all URMs, regardless of their academic credentials, experienced a relative fall in their probability of admissions to at least several UC schools.

As we show subsequently, Prop 209 generally had a larger effect on admission rates at more selective UC schools than at less-selective UC schools (though this pattern varies somewhat depending on URMs’ academic credentials). Thus, Prop 209 may not only have altered the total number of applications sent to UC schools (as Long's model suggests), but also the rate at which URMs applied to the different UC campuses. In particular, we might expect Prop 209 to shift URMs’ applications away from more-selective UCs (where admissions rates fell considerably) and toward less-selective UCs (where admission rates were relatively stable). In our empirical analysis, we focus on application rates to individual campuses, but we also examine the total number of applications sent to UC schools.

We begin by using administrative data from the UC Office of the President (UCOP) to document how the end of affirmative action affected admission rates for URMs relative to other racial groups. We then use data from the College Board to relate these changes in admission rates to changes in SAT score-sending patterns.

Throughout the paper, we use a simple differences-in-differences specification, which compares the minority–white gap after the implementation of Prop 209 to the gap before the implementation of Prop 209. For admissions, a basic equation for individual i applying to campus c in year t can be written as:
1
In this specification, Ban is equal to one for the 1998–2000 cohorts and equal to zero for the 1995–97 cohorts. For each campus, the regression sample is restricted to those who applied to the campus. The coefficient of interest, Θc, is intended to capture the effect of the ban on the URM–white admission gap at campus c. The term consists of controls for an applicant's SAT, GPA, parental income, and parental education. In response to Prop 209, universities may have changed the weight placed on demographic characteristics in the admissions process in an effort to partially undo the fall in URM admission rates (we are currently working on a paper further exploring this issue). As a result, we control for the interaction of demographic characteristics and the affirmative action ban.8 For SAT score-sending, we use the same specification above, but use Sent, a dummy indicating whether individual i sends SAT scores to campus c in year t as the left-hand side variable. Finally, we use robust standard errors in all regressions.

As discussed subsequently, the administrative data we use only allow us to identify the three-year window in which a student applied (1995–1997 versus 1998–2000). By design, the break in the time periods corresponds to the implementation of Prop 209. Nonetheless, one drawback of our data is that it does not allow us to include time trends in our analysis. On the other hand, the fact that our data are grouped by the three-year period before and the three-year period after the implementation of Prop 209 alleviates concerns about the impact of serial correlation in our outcome variable on our ability to make valid statistical inference.9

We draw on two data sets for our analysis: a data set obtained from the UCOP and a data set produced by the College Board and provided to us, with the College Board's permission, by David Card.10 The two data sets each contain individual-level data on hundreds of thousands of students, cover periods both before and after the implementation of Prop 209 in California, and fill in an important weakness of the other.

### UCOP

Our UCOP data include all applicants who applied for freshman admission to any of the UC's eight undergraduate campuses from 1995 through 2000.11 The data contain individual-level information on students’ race, adjusted high school GPA, SAT scores, parental income, and parental education. In addition, the data report the campuses to which each student applied, the campuses that accepted the applicant, and the campus at which the student enrolled, if any.12

Despite its strengths, the UCOP data have some significant drawbacks, most of which are related to its release as a publicly accessible data set. In an effort to protect student privacy, UCOP collapsed many important descriptive categories.13 Thus, for example, SAT scores and high school grades are reported in categories rather than discrete units, and year of application and enrollment is grouped into three-year cohorts (1995–97 and 1998–2000). As mentioned earlier, by design, the second-period cohort begins the year the ban on racial preferences was implemented. Race is also collapsed, from ten categories into four: white, Asian, URM, and other/unknown. The URM category includes American Indians, blacks, Chicanos, and Latinos; these are the primary groups that received preferential treatment based on race before Prop 209. The other/unknown category includes both students who indicate that their race falls outside the categories used by the university, as well as students who choose not to reveal their race (a group that grew substantially after Prop 209 went into effect). In our empirical analysis, we compare admissions rates of URMs with the combined set of white and other/unknown, but our results are not sensitive to the choice of dropping the other/unknown group or including them with white applicants.14 In addition, throughout the paper, we use white applicants rather than Asian applicants as our primary control group because the application patterns of Asians differ substantially from those of both whites and URMs (Asians apply at very high rates to UC schools).

Because our UCOP data contain every applicant to the UC system along with information on the campuses to which they were accepted, we use these data to estimate the change in the admission rate of URMs relative to non-URMs after Prop 209. One potential issue with this approach is that, to the extent that the composition of applicants changed in unobservable ways after Prop 209, this could affect our estimates of the change in URMs’ relative chances of admission. As we discuss in greater detail subsequently, however, our results suggest that the relative drop in URM application rates was small. In addition, to the extent that Prop 209 led to an increase in the relative quality of URM applicants (along dimensions that are observable to admissions officers but not to us), then this will lead us to understate the true drop in the admission rate of URMs relative to non-URMs. Given the magnitude of our estimates of the decline in applications rates, however, we would expect the size of this bias to be quite small.

### College Board

The College Board data include every California and Texas high school student who took the SAT I as part of the cohorts that graduated from high school between 1995 and 2000, inclusive.15 It also includes all black and Hispanic students who took the SAT elsewhere in the United States as part of the 1995–2000 cohorts and a 25 percent sample of all other SAT-takers who were part of those cohorts, but were neither black nor Hispanic nor residents of California or Texas. Demographic information comes from a background questionnaire that students fill out when they apply to take the SAT I. Students report their high school grades, their parents’ income and education level, their race or ethnicity, their high school coursework and activities, and their academic and career aspirations. The College Board adds to these data information on student performance on SAT I and SAT II exams.

Students tell the College Board to which schools they would like their SAT scores sent, and although the UC has a centralized admission process, if students wish to apply to more than one UC campus, they must send their SAT scores to each of those campuses. During the registration process for the SAT (before students know their scores), students are allowed to send their scores to up to four schools for free. If students elect to send scores to more than four schools or if they choose to send scores after the registration process is over, then there is an additional fee for each score sent (currently, the fee is $10.50 per score sent). Although the fee to send additional scores is seemingly small, work by Pallais (2009) suggests that students’ score-sending behavior is sensitive to small changes in price. Thus, students are likely to focus their score-sending on schools to which they are likely to apply. Indeed, as mentioned previously, like Card and Krueger (2005), we find a strong correlation between SAT score-sending rates and application rates for various race-SAT-year bins.16 Thus, we use the set of schools to which students send their scores as a proxy for student interest in a particular college and whether the student will eventually apply. In our data, the College Board reports up to thirty schools to which students send their SAT scores, though only 10 percent of students send to ten or more schools. In order to capture the population of students interested in attending college, in our main analysis we restrict our sample to Californians who send their scores to at least one school. Race information in the College Board data is more detailed than that in the UCOP data, with race being divided into the categories American Indian, Asian, black, Hispanic, white, other, and missing. To maintain consistency with the UCOP data, we condense the race categories into the following three groups: URM, Asian, and white (which, as with the UCOP data, includes the categories other and missing). As mentioned earlier, we do not group together Asian and white applicants in our control group because the score-sending patterns of Asians differ substantially from those of whites and URMs. One advantage of using the College Board sample is that nearly all of the SAT takers are interested in going to college, so they should be the ones most readily affected by the affirmative action ban. On the other hand, a potential problem with using the College Board is that a student's decision about whether to take the SAT could be affected by Prop 209, leading to possible sample selection bias. Indeed, Dickson (2006) finds that removal of affirmative action in Texas led to a decline in the share of minority students taking either the ACT or SAT. If Prop 209 led to a similar drop in the fraction of URMs who took the SAT, then to the extent that these students would have applied to the UC, this could lead us to understate the true drop in the relative score-sending rate of URMs. To address this issue, we examine the trend in the number of SAT takers from California as a fraction of the number of California public high school graduates from 1995 to 2000 using data from the 1997–2002 Digest of Education Statistics.17 We find no evidence of a trend break for either URMs or non-URMs around the implementation of Prop 209. Indeed, it seems unlikely that students on the margin of taking the SAT would have been affected by a policy change that primarily affected the most selective UC campuses. Finally, Furstenberg (2010) shows that the demographic characteristics of SAT takers are generally uncorrelated with the introduction of the bans in California and Texas. Thus, it seems unlikely that our results are affected by a change in the composition of SAT takers brought about by Prop 209. ### College Board Data and the Timing of Prop 209 Defining the pre and post periods in the UCOP data is straightforward, because by construction there are only two time periods which fall on either side of the change in the admissions regime. Defining the pre and post period in the College Board data, however, is more complicated because the timing of when students took the SAT and sent out their SAT scores will affect whether their observed score-sending choices should best be thought of as occurring before or after the end of the use of racial preferences. Table 1 shows a timeline of important events in the passage of Prop 209. Because Prop 209 began receiving media attention in the summer of 1996 and then was passed by voters in November 1996, it is difficult to know whether students in the 1997 cohort, who likely took the SAT between Spring 1996 and Fall 1996, should be counted in the pre or post Prop 209 period.18 Thus, we drop all observations from the 1997 cohort.19 Whether or not 1997 is included does not change our main conclusions—most of the effects on score-sending are concentrated in 1999 and 2000—but including the 1997 cohort muddies the interpretation of our estimates. We include the 1998 cohort with the treatment (post-Prop 209) group because their SAT score-sending decisions were likely to have been made after passage of Prop 209, and although most students in the 1998 cohort are likely to have taken the SAT (and made their score-sending decisions) during the period in which Prop 209 was still being challenged in court, a sampling of news articles from this time suggests that it was believed Prop 209 would eventually be upheld.20 Table 1. Proposition 209 Timeline DateEvent July 1995 SP-1 Spring 1996 1997 cohort begins taking SAT July 1996 First mention of Prop 209 in media Summer 1996 Prop 209 reported as being likely to pass November 1996 Prop 209 passed by voters Fall 1996 End of observation period for 1997 cohort Spring–Fall 1997 1998 cohort takes SAT November 1997 Supreme Court denies appeals, Prop 209 becomes law Fall 1998 First affected cohort (1998) enrolls DateEvent July 1995 SP-1 Spring 1996 1997 cohort begins taking SAT July 1996 First mention of Prop 209 in media Summer 1996 Prop 209 reported as being likely to pass November 1996 Prop 209 passed by voters Fall 1996 End of observation period for 1997 cohort Spring–Fall 1997 1998 cohort takes SAT November 1997 Supreme Court denies appeals, Prop 209 becomes law Fall 1998 First affected cohort (1998) enrolls Notes: See text for description. We first document changes in admissions probability for URM applicants relative to white applicants, and then in the following section show how score-sending patterns changed after Prop 209. Summary statistics for UC applicants are shown in table 2. Relative to white applicants, minority applicants tend to have lower SAT scores, lower GPAs, and are more likely to come from disadvantaged backgrounds. Our UCOP data do not provide continuous measures of parental income and education; for example, the given income measure is a set of ten income intervals. Rather than showing the percentage of our sample falling into each of the ten intervals of parental income, we collapse these intervals into the three categories shown in table 2. For education, we use a dummy variable for whether an applicant has a college-educated parent.21 The definitions of these categories have little impact on the predicted effects of Prop 209. Table 2. UCOP Summary Statistics AllWhiteURMAsian BeforeAfterBeforeAfterBeforeAfterBeforeAfter SAT Math 597 602 608 612 528 534 622 625 (94) (93) (82) (82) (93) (93) (90) (91) SAT Verbal 569 572 597 597 525 527 558 562 (97) (96) (85) (86) (93) (93) (102) (100) Adj HS GPA 3.6 3.6 3.6 3.7 3.4 3.5 3.7 3.7 (0.5) (0.49) (0.5) (0.49) (0.48) (0.49) (0.5) (0.49) Parent Income <$40,000 0.33 0.29 0.19 0.17 0.54 0.49 0.39 0.36
Parent Income $40,000–99,999 0.43 0.42 0.48 0.44 0.37 0.38 0.42 0.4 Parent Income >$100,000 0.24 0.29 0.33 0.39 0.098 0.13 0.19 0.24
Parent Educ: At Least 4 Year College 0.66 0.65 0.79 0.77 0.37 0.36 0.65 0.64
Total UC applications 2.9 3.2 2.6 2.9 2.7 3.3 3.6
(1.4) (1.5) (1.4) (1.5) (1.2) (1.3) (1.6) (1.6)
Observations 136,766 149,305 63,258 70,448 26,694 27,707 46,814 51,150
AllWhiteURMAsian
BeforeAfterBeforeAfterBeforeAfterBeforeAfter
SAT Math 597 602 608 612 528 534 622 625
(94) (93) (82) (82) (93) (93) (90) (91)
SAT Verbal 569 572 597 597 525 527 558 562
(97) (96) (85) (86) (93) (93) (102) (100)
Adj HS GPA 3.6 3.6 3.6 3.7 3.4 3.5 3.7 3.7
(0.5) (0.49) (0.5) (0.49) (0.48) (0.49) (0.5) (0.49)
Parent Income < $40,000 0.33 0.29 0.19 0.17 0.54 0.49 0.39 0.36 Parent Income$40,000–99,999 0.43 0.42 0.48 0.44 0.37 0.38 0.42 0.4
Parent Income > $100,000 0.24 0.29 0.33 0.39 0.098 0.13 0.19 0.24 Parent Educ: At Least 4 Year College 0.66 0.65 0.79 0.77 0.37 0.36 0.65 0.64 Total UC applications 2.9 3.2 2.6 2.9 2.7 3.3 3.6 (1.4) (1.5) (1.4) (1.5) (1.2) (1.3) (1.6) (1.6) Observations 136,766 149,305 63,258 70,448 26,694 27,707 46,814 51,150 Notes: Standard deviations in parentheses. Before includes years 1995–1997; After includes 1998–2000. White includes white and other/unknown, and URM includes blacks, Hispanics, and American Indians. To better understand the predictors of admission to UC campuses, table 3 shows basic difference-in-difference regression coefficients for each campus and the UC system as a whole. Going from left to right, the campuses are ordered from most selective, defined by the average math SAT score of incoming freshman, to least selective. Relative to white applicants, URM applicants were significantly less likely to gain admission to the top schools after Prop 209. At Berkeley and UCLA, the likelihood of admission fell by 30 and 25 percentage points, respectively, for URM applicants compared with white applicants. Given that about 50 percent of URM applicants were accepted to Berkeley and UCLA before Prop 209 (authors’ calculation), these declines are substantial. Relative to Berkeley and UCLA, there is a somewhat smaller fall in the relative admission rates of URMs at UCSD and UC Davis. Then, at the remaining four campuses, the effect of Prop 209 on URMs’ relative probability of admission is smaller still. Thus, the relative fall in URMs’ chances of admission was the largest at the most selective UC schools. Table 3. Probability of Admission Difference-in-Difference BerkeleyUCLAUCSDUCDUCIUCSBUCSCUCR URM 0.42** 0.37** 0.27** 0.28** 0.09** 0.14** 0.07** 0.02* (0.01) (0.00) (0.00) (0.01) (0.01) (0.00) (0.01) (0.01) Asian −0.00 0.02** 0.03** −0.04** −0.03** 0.01** −0.02** −0.02** (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.01) (0.01) URM*Ban −0.30** −0.25** −0.20** −0.21** −0.05** −0.02** −0.06** −0.04** (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Asian*Ban 0.01 0.01* 0.00 0.02** −0.00 −0.00 0.00 −0.00 (0.01) (0.00) (0.00) (0.01) (0.01) (0.01) (0.01) (0.01) R2 0.27 0.38 0.50 0.36 0.40 0.39 0.18 0.17 Obs 133589 152268 134536 101649 97102 112949 65976 63985 Non-URM mean 0.30 0.34 0.49 0.68 0.67 0.64 0.83 0.85 URM mean 0.38 0.36 0.42 0.72 0.59 0.64 0.80 0.79 BerkeleyUCLAUCSDUCDUCIUCSBUCSCUCR URM 0.42** 0.37** 0.27** 0.28** 0.09** 0.14** 0.07** 0.02* (0.01) (0.00) (0.00) (0.01) (0.01) (0.00) (0.01) (0.01) Asian −0.00 0.02** 0.03** −0.04** −0.03** 0.01** −0.02** −0.02** (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.01) (0.01) URM*Ban −0.30** −0.25** −0.20** −0.21** −0.05** −0.02** −0.06** −0.04** (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Asian*Ban 0.01 0.01* 0.00 0.02** −0.00 −0.00 0.00 −0.00 (0.01) (0.00) (0.00) (0.01) (0.01) (0.01) (0.01) (0.01) R2 0.27 0.38 0.50 0.36 0.40 0.39 0.18 0.17 Obs 133589 152268 134536 101649 97102 112949 65976 63985 Non-URM mean 0.30 0.34 0.49 0.68 0.67 0.64 0.83 0.85 URM mean 0.38 0.36 0.42 0.72 0.59 0.64 0.80 0.79 Notes: Robust standard errors in parentheses. The dependent variable is a binary indicator of whether the student was accepted to a given school conditional on applying to that school. The excluded race category includes white and other/unknown. Controls include SAT math and verbal scores, high school GPA, parental income and education, and their interactions with the Ban dummy. The sample includes data from 1995–1997 (pre Prop 209) and 1998–2000 (post Prop 209). *p < .05, **p < .01. Interestingly, at the top six campuses, the coefficient on Ban*URM is significantly smaller in magnitude than the coefficient on URM, suggesting that even though Prop 209 significantly reduced the admissions advantage enjoyed by minority applicants, it did not eliminate it entirely. Of course, this continued admissions advantage may have been indirect. For example, the campuses may have given admissions advantages to students with characteristics that are correlated with race but that are not available in our data (such as whether the student is from a single-parent family). Indeed, although not shown in table 3, we find that parental income and parental education are negatively correlated with the likelihood of admission, suggesting that admissions preferences are given to students from disadvantaged backgrounds. Finally, the coefficient on URM tends to be the largest at the most selective schools, suggesting that before Prop 209 the extent of racial preferences was the highest at the most selective UC schools. Although having higher test scores and a higher GPA is positively related to the likelihood of admission (coefficients not displayed but available from authors), the relationship between the change in URMs’ relative likelihood of admission after Prop 209 is likely to be non-monotonically related to their academic credentials. Students with sufficiently high or low academic credentials would have been unaffected by the end of racial preferences if they would have been accepted (high credentials) or rejected (low credentials) under either regime. To better understand which students bore the brunt of the change in admissions regime, we use their index score, which is a weighted average of high school GPA (40 percent) and SAT scores (30 percent each for math and verbal). The index score variable was created for us by UCOP, where the weights were determined by regressing students’ first-year UC GPA on high school GPA and SAT scores. Thus, by construction, the index score is designed to predict students’ performance in college. From our standpoint, a primary reason for using the index score is that it provides a nice summary of students’ academic credentials (as measured by high school GPA and SAT test scores). In addition, the index is calculated from students’ actual high school GPA and actual SAT scores (rather than the discretized versions of those variables available in the rest of the data). Thus, it is the only continuous measure of academic credentials available to us. To characterize the effect of Prop 209 at various points of the academic credential distribution, we divide URM applicants to the UC into percentiles based on their index score. We then categorize non-URMs into corresponding bins based on their index scores. Thus, for example, we can compare URM applicants at a given index score percentile with non-URM applicants whose index score falls in a similar range. For a summary of average SAT and GPA within each percentile, see figure 1a. Note that because white applicants tend to have higher GPA and SAT scores, there are relatively more white applicants at the higher end of the URM distribution. Figure 1b shows the share of white applicants who fall into each decile of the URM distribution; about 30 percent of white applicants have index scores that would place them in the top decile of the minority distribution. Figure 1a. Average SAT and GPA by Credential Decile, All Score-senders Figure 1a. Average SAT and GPA by Credential Decile, All Score-senders Close modal Figure 1b. Distribution of non-URM at URM Credential. Notes: Index generated by weighted average of SAT (30% each for verbal and math) and GPA (40%). URM Percentile axis consists of percentile of distribution of URM UC applicants. Figure 1b. Distribution of non-URM at URM Credential. Notes: Index generated by weighted average of SAT (30% each for verbal and math) and GPA (40%). URM Percentile axis consists of percentile of distribution of URM UC applicants. Close modal An overview of how Prop 209 affected the admissions rates of applicants at various points of the credential distribution is shown in figure 2a, which plots average admissions rates to Berkeley (conditional on applying) for URMs and non-URMs at each index percentile. Prior to Prop 209, URMs at nearly every point of the credential distribution were more likely than whites to gain admission to Berkeley, and this advantage was the largest for applicants in the middle of the distribution. After Prop 209, much of the gap in admission rates between URMs and whites was closed, although conditional on index, URMs remained more likely to be admitted. Figure 2a. Admission to Berkeley by Credential Distribution Figure 2a. Admission to Berkeley by Credential Distribution Close modal To get a sense of how Prop 209 affected URMs’ relative chances of admission at different points of the credential distribution, figure 3 plots the coefficient on Ban*URM (using the same basic specification outlined in equation 1) at each percentile of the URM credential distribution for each of the eight UC campuses. The figure is generated using a three-step process. First, coefficient estimates are obtained for each percentile of the index distribution. Second, the set of point estimates is smoothed using local polynomial smoothing. Finally, standard errors are obtained using a bootstrap method. The resulting plot shows the change in admissions probability for URM applicants relative to white applicants at different percentiles of the academic credential distribution. Figure 3 indicates that even for students at the top of the URM credential distribution, there was a large and statistically significant drop in URMs’ relative chances of admission to Berkeley, though at both Berkeley and UCLA the largest relative drop in URMs’ chances of admission occurred for those in the middle of the URM credential distribution. At UC San Diego and UC Davis, the largest relative drop in URMs’ likelihood of admission occurred for students in the 30th–60th percentiles of the URM credential distribution. At UC San Diego, for example, students at the 60th percentile of the URM credential distribution saw their predicted probability of admission fall by close to 50 percentage points (relative to a pre Prop 209 admission probability of about 60 percent). At the bottom four campuses, URMs in the lowest deciles experienced a significant relative drop in their chances of admission. At UC Santa Cruz, for example, URM applicants in the bottom 10 percent of the credential distribution experienced a 20-percentage point drop in their probability of admission relative to similarly qualified white applicants. Thus, due to the wide range of selectivity of the various UC campuses, all URMs, regardless of their academic credentials, were affected in some way by Prop 209. URMs at the top of the credential distribution saw their relative chances of admission fall at the most selective UCs, whereas those at the bottom of the credential distribution saw their relative chances of admission fall at the least selective UCs. In addition, the fact that different groups of students were affected differently at each campus suggests that URMs may have responded to Prop 209 by shifting applications away from some schools and toward others. The next section investigates these changes in score-sending patterns. We now turn to the effect of the ban on affirmative action on the SAT score-sending behavior of URMs using data from the College Board. As with the UCOP data, we begin by reviewing summary statistics, showing predictors of score-sending, and finally breaking students into credential groups. ### Main Results Summary statistics by race and time period are shown in table 4 and show similar patterns as the UCOP data. The main difference is that in the College Board data set we are able to differentiate between race, so we show black students and Hispanic students separately. Table 4. College Board Summary Statistics AllWhiteBlackHispanicAsian BeforeAfterBeforeAfterBeforeAfterBeforeAfterBeforeAfter SAT Math 520 527 543 553 428 436 456 459 554 560 (114) (116) (102) (102) (100) (103) (102) (103) (117) (120) SAT Verbal 502 507 538 543 437 441 450 450 490 498 (115) (115) (102) (102) (103) (104) (105) (104) (123) (121) Adj HS GPA 3.3 3.3 3.4 3.4 2.9 3.1 3.2 3.4 3.4 (0.63) (0.61) (0.6) (0.58) (0.62) (0.62) (0.62) (0.62) (0.61) (0.59) Parent Income <$40,000 0.45 0.39 0.28 0.22 0.62 0.57 0.68 0.63 0.54 0.49
Parent Income $40,000–99,999 0.4 0.39 0.5 0.47 0.31 0.32 0.26 0.28 0.35 0.35 Parent Income >$100k 0.11 0.14 0.17 0.22 0.038 0.055 0.033 0.044 0.082 0.1
Parent Educ: At Least 4 Year College 0.57 0.56 0.7 0.7 0.47 0.47 0.28 0.26 0.59 0.57
Number of Schools to Which Sent Scores 5.6 5.7 5.6 5.6 5.2 5.5 5.1 5.2 6.3 6.2
(2.8) (2.9) (2.9) (3) (2.6) (2.8) (2.3) (2.4) (2.9) (2.9)
Sent to UC 0.67 0.68 0.62 0.65 0.52 0.52 0.64 0.64 0.83 0.84
Number of UCs to Which Sent Scores 1.8 1.5 1.7 1.1 1.2 1.5 1.5 2.7
(1.8) (1.9) (1.6) (1.8) (1.4) (1.5) (1.5) (1.6) (2) (2)
Sent to More Than 4 Schools 0.54 0.55 0.52 0.54 0.46 0.5 0.45 0.47 0.66 0.67
Observations 196,405 314,308 95,645 152,306 14,006 20,907 38,749 63,276 45,624 74,966
AllWhiteBlackHispanicAsian
BeforeAfterBeforeAfterBeforeAfterBeforeAfterBeforeAfter
SAT Math 520 527 543 553 428 436 456 459 554 560
(114) (116) (102) (102) (100) (103) (102) (103) (117) (120)
SAT Verbal 502 507 538 543 437 441 450 450 490 498
(115) (115) (102) (102) (103) (104) (105) (104) (123) (121)
Adj HS GPA 3.3 3.3 3.4 3.4 2.9 3.1 3.2 3.4 3.4
(0.63) (0.61) (0.6) (0.58) (0.62) (0.62) (0.62) (0.62) (0.61) (0.59)
Parent Income < $40,000 0.45 0.39 0.28 0.22 0.62 0.57 0.68 0.63 0.54 0.49 Parent Income$40,000–99,999 0.4 0.39 0.5 0.47 0.31 0.32 0.26 0.28 0.35 0.35
Parent Income > $100k 0.11 0.14 0.17 0.22 0.038 0.055 0.033 0.044 0.082 0.1 Parent Educ: At Least 4 Year College 0.57 0.56 0.7 0.7 0.47 0.47 0.28 0.26 0.59 0.57 Number of Schools to Which Sent Scores 5.6 5.7 5.6 5.6 5.2 5.5 5.1 5.2 6.3 6.2 (2.8) (2.9) (2.9) (3) (2.6) (2.8) (2.3) (2.4) (2.9) (2.9) Sent to UC 0.67 0.68 0.62 0.65 0.52 0.52 0.64 0.64 0.83 0.84 Number of UCs to Which Sent Scores 1.8 1.5 1.7 1.1 1.2 1.5 1.5 2.7 (1.8) (1.9) (1.6) (1.8) (1.4) (1.5) (1.5) (1.6) (2) (2) Sent to More Than 4 Schools 0.54 0.55 0.52 0.54 0.46 0.5 0.45 0.47 0.66 0.67 Observations 196,405 314,308 95,645 152,306 14,006 20,907 38,749 63,276 45,624 74,966 Notes: Standard deviations in parentheses. Sample includes Californians who sent their SAT scores to at least one school. Before includes years 1995–1996; After includes 1998–2000. White category includes white and other/unknown. Table 4 helps explain why we do not group Asian students with white students as part of our control group; their overall score-sending rates are much higher than those of other groups. In addition, in the 1990s, there was a general increase in Asian interest in UC campuses that began before Prop 209 and continued afterwards (the starkest example is at UC Irvine). Plotting score-sending rates shows a steady upward climb in the Asian score-sending rate, rather than a break from trend around the time of Prop 209, and this pattern holds after controlling for student-level characteristics.22 As a result, we drop Asians from the remainder of our score-sending analysis. When Asians are added to the excluded category, there is little change in the estimates for the most selective schools, where Asian score-sending rates were relatively stable. At schools such as Irvine, however, where Asian score-sending increased dramatically, URM score-sending estimates become smaller. Because students can send their SAT scores to up to four schools for free, one possible explanation for the stability of URM score-sending rates after Prop 209 is that most students send scores to exactly four schools. Of course, even if this were true, students can still change the set of schools to which they send scores, and our data allow us to capture this. In addition, as table 4 shows, approximately 50 percent of SAT takers sent their SAT scores to more than four schools both before and after Prop 209. Table 5 shows the estimation results of our basic difference-in-difference specification. We see that after the ban on affirmative action, black students and Hispanic students experienced the largest relative declines in score-sending rates at the most selective UC campuses, and the relative declines in score-sending rates were generally smaller at the other UC campuses. At UC Riverside the score-sending rates of black students and Hispanic students actually increased by about 1.5 percentage points relative to white students. This increase likely reflects the fact that URMs responded to the end of racial preferences by shifting their applications toward the less-selective UCs. We discuss this phenomenon in greater detail subsequently. Table 5. Score-sending Difference-in-Difference BerkeleyUCLAUCSDUCDUCIUCSBUCSCUCR Black 0.174*** 0.179*** −0.002 0.009** 0.014*** −0.087*** −0.053*** 0.009*** (0.004) (0.004) (0.003) (0.003) (0.003) (0.003) (0.002) (0.003) Hispanic 0.116*** 0.172*** 0.054*** 0.002 0.064*** 0.001 0.009*** 0.043*** (0.003) (0.003) (0.003) (0.002) (0.002) (0.003) (0.002) (0.002) Ban*Black −0.019*** −0.007 0.004 −0.010* −0.005 0.001 −0.002 0.017*** (0.005) (0.005) (0.004) (0.004) (0.004) (0.004) (0.003) (0.003) Ban*Hispanic −0.010** −0.004 −0.000 −0.015*** −0.004 −0.006 −0.006* 0.014*** (0.003) (0.004) (0.004) (0.003) (0.003) (0.004) (0.003) (0.003) R2 0.14 0.10 0.09 0.07 0.02 0.06 0.02 0.01 Observations 384,889 384,889 384,889 384,889 384,889 384,889 384,889 384,889 Non-URM mean 0.23 0.27 0.29 0.23 0.13 0.29 0.15 0.08 URM mean 0.21 0.32 0.21 0.13 0.15 0.17 0.10 0.12 BerkeleyUCLAUCSDUCDUCIUCSBUCSCUCR Black 0.174*** 0.179*** −0.002 0.009** 0.014*** −0.087*** −0.053*** 0.009*** (0.004) (0.004) (0.003) (0.003) (0.003) (0.003) (0.002) (0.003) Hispanic 0.116*** 0.172*** 0.054*** 0.002 0.064*** 0.001 0.009*** 0.043*** (0.003) (0.003) (0.003) (0.002) (0.002) (0.003) (0.002) (0.002) Ban*Black −0.019*** −0.007 0.004 −0.010* −0.005 0.001 −0.002 0.017*** (0.005) (0.005) (0.004) (0.004) (0.004) (0.004) (0.003) (0.003) Ban*Hispanic −0.010** −0.004 −0.000 −0.015*** −0.004 −0.006 −0.006* 0.014*** (0.003) (0.004) (0.004) (0.003) (0.003) (0.004) (0.003) (0.003) R2 0.14 0.10 0.09 0.07 0.02 0.06 0.02 0.01 Observations 384,889 384,889 384,889 384,889 384,889 384,889 384,889 384,889 Non-URM mean 0.23 0.27 0.29 0.23 0.13 0.29 0.15 0.08 URM mean 0.21 0.32 0.21 0.13 0.15 0.17 0.10 0.12 Notes: Robust standard errors in parentheses. The dependent variable is a binary indicator of whether the student was sent a score to a given school. The excluded race category includes white and other/unknown, and URM includes blacks and Hispanics. Controls include SAT math and verbal scores, high school GPA, parental income and education, and their interactions with the Ban dummy. Sample includes Californians who sent a score to at least one school and includes data from 1995–1996 (pre Prop 209) and 1998–2000 (post Prop 209). *p < .1, **p < .05, ***p < .01. ### Magnitude of Findings The declines in score-sending rates for black applicants and Hispanic applicants reported in table 5 are small relative to the drop in their admissions rates reported in table 3. At Berkeley, for example, the admission rate of URMs fell by 30 percentage points relative to white applicants, whereas the relative score-sending rate fell by only 2 percentage points for black applicants and 1 percentage point for Hispanic applicants. When grouping black applicants and Hispanic applicants together to form one URM category, the estimated drop in score-sending is 1.3 percentage points (authors’ calculation). Of course, this may just reflect a small elasticity of score-sending rates with respect to admission rates, but the relationship between score-sending rates and admissions rates in the cross section does not support this. For example, a 30 percentage point fall in URMs’ chances of admission to UC Berkeley is roughly equivalent to moving from the 60th to the 20th percentile of the URM index score distribution prior to Prop 209 (see figure 2a). Given that the gap in URM score-sending rates between these two percentiles is about 15 percentage points (see figure 2b), the 1.3 percentage point fall in the score-sending rates of URM applicants relative to white applicants at Berkeley after Prop 209 is comparatively small. Figure 2b. Score-sending to Berkeley by Credential Distribution. Notes: Index generated by weighted average of SAT (30% each for verbal and math) and GPA (40%). URM Percentile axis consists of percentile of distribution of URM UC applicants. Figure 2b. Score-sending to Berkeley by Credential Distribution. Notes: Index generated by weighted average of SAT (30% each for verbal and math) and GPA (40%). URM Percentile axis consists of percentile of distribution of URM UC applicants. Close modal Interestingly, the relative decline in score-sending rates to Berkeley and UCLA is also small relative to URMs’ initial higher tendency to send their SAT scores to those schools. For example, row 1 of table 5 suggests that prior to Prop 209, the black score-sending rate to Berkeley was 17 percentage points higher than the white score-sending rate (conditional on observables), and after Prop 209 this gap fell by only 2 percentage points, suggesting that Prop 209 only closed 11 percent of the gap in score-sending rates. In contrast, Prop 209 greatly reduced the gap in admissions rates. Table 3 shows that, prior to Prop 209, URMs were 42 percentage points more likely to be admitted to Berkeley than were whites with similar observable characteristics. After Prop 209 this gap fell by 30 percentage points, suggesting Prop 209 closed almost 71 percent of the gap in admissions rates. Another way to gauge the size of the effect of Prop 209 on score-sending rates is to assume that the change in URMs’ score-sending rate equals the change in their application rate and then determine the annual fall in the number of URMs who would have been admitted due to this change in application behavior. A rough estimate of the yearly change in the number of students admitted to Berkeley due to changes in application behavior is given by: In this calculation, all of the values are taken at their post Prop 209 levels. Next, we compare this number to the yearly fall in the number of admitted URMs due to the fall in the admission rate after Prop 209: To obtain the change in eventual enrollment, we then multiply the numbers obtained here by the enrollment rate (the fraction of admitted students who eventually enroll), which was approximately 0.41 for URMs in the post period. Doing so implies that six fewer URMs went to Berkeley because of changes in application behavior and 567 fewer URMs went to Berkeley because of changes in the admissions rate. These estimates ignore potential selection effects (for example, those who were no longer admitted may have been more likely to enroll, conditional on acceptance), but regardless of how the calculation is performed, it appears that Prop 209's effect on URM enrollment came largely through admissions probabilities rather than application rates. ### Effects Over Time The coefficients shown in table 5 represent the average change in score-sending rates for URM applicants relative to white applicants over all three post-policy-change years. It is possible, however, that the effects vary over time. Thus, in table 6, we show the relative change in score-sending rates for each year in the post-policy-change period. The coefficients represent the change in score-sending rates for URM applicants relative to white applicants between the pre-policy-change period and each year in the post-policy-change period. For both black and Hispanic applicants, there is evidence that the relative fall in score-sending rates after Prop 209 was concentrated in 1999 and 2000. For example, relative to white applicants, black applicants did not send significantly fewer scores to Berkeley in 1998 compared to the pre-policy-change period, but their relative score-sending rate was 2.3 percentage points lower in 1999 and 2.9 percentage points lower in 2000. Similarly, the relative decline in Hispanic score-sending rates to Berkeley was the largest in 1999, though the difference between 1998 and 2000 is not statistically significant. Table 6. Score-sending Difference-in-Difference by Year BerkeleyUCLAUCSDUCDUCIUCSBUCSCUCR Black 0.174*** 0.179*** −0.002 0.009** 0.014*** −0.087*** −0.053*** 0.009*** (0.004) (0.004) (0.003) (0.003) (0.003) (0.003) (0.002) (0.003) Hispanic 0.116*** 0.172*** 0.054*** 0.002 0.064*** 0.001 0.009*** 0.043*** (0.003) (0.003) (0.003) (0.002) (0.002) (0.003) (0.002) (0.002) Black*1998 −0.005 0.006 0.008 −0.004 −0.001 0.021*** −0.000 0.013** (0.007) (0.007) (0.006) (0.006) (0.005) (0.005) (0.004) (0.005) Black*1999 −0.023*** −0.015* 0.003 −0.011* −0.004 −0.003 −0.004 0.020*** (0.006) (0.007) (0.006) (0.005) (0.005) (0.005) (0.004) (0.005) Black*2000 −0.029*** −0.011 0.002 −0.015** −0.011* −0.013* −0.003 0.019*** (0.006) (0.007) (0.006) (0.005) (0.005) (0.005) (0.004) (0.005) Hispanic*1998 −0.007 0.006 −0.002 −0.012** 0.001 0.006 −0.001 0.009* (0.004) (0.005) (0.005) (0.004) (0.004) (0.005) (0.004) (0.003) Hispanic*1999 −0.016*** −0.011* −0.000 −0.016*** −0.003 −0.010* −0.012*** 0.015*** (0.004) (0.005) (0.005) (0.004) (0.004) (0.004) (0.004) (0.003) Hispanic*2000 −0.008* −0.006 0.001 −0.018*** −0.009* −0.012** −0.004 0.018*** (0.004) (0.005) (0.005) (0.004) (0.004) (0.004) (0.004) (0.003) R2 0.14 0.10 0.09 0.07 0.02 0.06 0.02 0.01 Observations 384,889 384,889 384,889 384,889 384,889 384,889 384,889 384,889 Non-URM mean 0.23 0.27 0.29 0.23 0.13 0.29 0.15 0.08 URM mean 0.21 0.32 0.21 0.13 0.15 0.17 0.10 0.12 BerkeleyUCLAUCSDUCDUCIUCSBUCSCUCR Black 0.174*** 0.179*** −0.002 0.009** 0.014*** −0.087*** −0.053*** 0.009*** (0.004) (0.004) (0.003) (0.003) (0.003) (0.003) (0.002) (0.003) Hispanic 0.116*** 0.172*** 0.054*** 0.002 0.064*** 0.001 0.009*** 0.043*** (0.003) (0.003) (0.003) (0.002) (0.002) (0.003) (0.002) (0.002) Black*1998 −0.005 0.006 0.008 −0.004 −0.001 0.021*** −0.000 0.013** (0.007) (0.007) (0.006) (0.006) (0.005) (0.005) (0.004) (0.005) Black*1999 −0.023*** −0.015* 0.003 −0.011* −0.004 −0.003 −0.004 0.020*** (0.006) (0.007) (0.006) (0.005) (0.005) (0.005) (0.004) (0.005) Black*2000 −0.029*** −0.011 0.002 −0.015** −0.011* −0.013* −0.003 0.019*** (0.006) (0.007) (0.006) (0.005) (0.005) (0.005) (0.004) (0.005) Hispanic*1998 −0.007 0.006 −0.002 −0.012** 0.001 0.006 −0.001 0.009* (0.004) (0.005) (0.005) (0.004) (0.004) (0.005) (0.004) (0.003) Hispanic*1999 −0.016*** −0.011* −0.000 −0.016*** −0.003 −0.010* −0.012*** 0.015*** (0.004) (0.005) (0.005) (0.004) (0.004) (0.004) (0.004) (0.003) Hispanic*2000 −0.008* −0.006 0.001 −0.018*** −0.009* −0.012** −0.004 0.018*** (0.004) (0.005) (0.005) (0.004) (0.004) (0.004) (0.004) (0.003) R2 0.14 0.10 0.09 0.07 0.02 0.06 0.02 0.01 Observations 384,889 384,889 384,889 384,889 384,889 384,889 384,889 384,889 Non-URM mean 0.23 0.27 0.29 0.23 0.13 0.29 0.15 0.08 URM mean 0.21 0.32 0.21 0.13 0.15 0.17 0.10 0.12 Notes: Robust standard errors in parentheses. The dependent variable is a binary indicator of whether the student was sent a score to a given school. The excluded race category includes white and other/unknown, and URM includes blacks and Hispanics. Controls include SAT math and verbal scores, high school GPA, parental income and education, and their interactions with the Ban dummy. Sample includes Californians who sent a score to at least one school and includes data from 1995–1996 (pre Prop 209) and 1998–2000 (post Prop 209). *p < .1, **p < .05, ***p < .01. Unfortunately, we are unable to identify the mechanism behind this trend. One possibility is that it took time for potential applicants to learn about the effect of Prop 209 on admission rates. Alternatively, to the extent that potential applicants care about their own-race enrollment shares, it would have taken time for the full effects of Prop 209 on URM enrollment shares to play out because Prop 209 only affected newly admitted students. Regardless of the explanation, however, the effects in later years are still extremely small relative to the changes in admissions probabilities. In section 7's rough calculations of the change in the number of admitted students, even a doubling of the score-sending effect would result in very few URMs displaced from Berkeley due to application decisions. ### Effects by Academic Credentials To get a sense of how Prop 209 affected the average score-sending rate of students with different levels of academic preparation, figure 2b shows the average score-sending rate to Berkeley for URM students and white students both before and after Prop 209 by index score percentile. As with admissions probabilities (see figure 2a), there was a large gap in score-sending rates between URM students and white students before Prop 209, and this gap narrowed after Prop 209, especially for students in the middle of the index score distribution (though again we see that the narrowing is much smaller for score-sending rates than admissions rates). To more fully explore the heterogeneity in the response of application behavior across the range of student quality, we repeat our score-sending analysis for the index percentiles used in figure 3. Because the response of black and Hispanic students shown in table 5 is similar, we combine them into one URM group. This simplifies our results and has the added benefit of easing comparison with our estimates of the changes in admission rates obtained from the UCOP data where black and Hispanic students could not be separately identified. In general, the effects of Prop 209 are similar for black and Hispanic students, but any changes that did occur were more pronounced for blacks. Figure 3a. Relative Change in URM Admissions at Berkeley Figure 3a. Relative Change in URM Admissions at Berkeley Close modal Figure 3b. Relative Change in URM Admissions at UCLA Figure 3b. Relative Change in URM Admissions at UCLA Close modal Figure 3c. Relative Change in URM Admissions at UCSD Figure 3c. Relative Change in URM Admissions at UCSD Close modal Figure 3d. Relative Change in URM Admissions at UCD Figure 3d. Relative Change in URM Admissions at UCD Close modal Figure 3e. Relative Change in URM Admissions at UCI Figure 3e. Relative Change in URM Admissions at UCI Close modal Figure 3f. Relative Change in URM Admissions at UCSB Figure 3f. Relative Change in URM Admissions at UCSB Close modal Figure 3g. Relative Change in URM Admissions at UCSC Figure 3g. Relative Change in URM Admissions at UCSC Close modal Figure 3h. Relative Change in URM Admissions at UCR. Notes: The dependent variable is a binary indicator of whether the student was accepted to a given school conditional on applying to that school. Each point shows OLS coefficient estimate on the interaction between post Prop 209 and URM for a decile of the academic credential distribution of URM UC applicants. Each estimate also includes controls for SAT math and verbal scores, high school GPA, parental income and education, and their interactions with the Ban dummy. The excluded race category includes white and other/unknown. Sample includes 1995–1997 (pre Prop 209) and 1998–2000 (post Prop 209). Plots smoothed with local polynomial regressions; bootstrapped standard errors displayed. DD = difference-in-difference. Figure 3h. Relative Change in URM Admissions at UCR. Notes: The dependent variable is a binary indicator of whether the student was accepted to a given school conditional on applying to that school. Each point shows OLS coefficient estimate on the interaction between post Prop 209 and URM for a decile of the academic credential distribution of URM UC applicants. Each estimate also includes controls for SAT math and verbal scores, high school GPA, parental income and education, and their interactions with the Ban dummy. The excluded race category includes white and other/unknown. Sample includes 1995–1997 (pre Prop 209) and 1998–2000 (post Prop 209). Plots smoothed with local polynomial regressions; bootstrapped standard errors displayed. DD = difference-in-difference. Close modal Figure 4 plots the coefficient on Ban*URM (using the basic specification outlined in equation 1) at each percentile of the URM index distribution for each of the eight UC campuses. Consistent with the findings in table 5, figure 4 suggests that Prop 209 may have led URMs to shift their applications away from the most selective UC schools (like Berkeley) toward the less selective UC schools (like Riverside). In addition, at Berkeley and UCLA (the only two schools at which we find a statistically significant drop in URMs’ relative score-sending rates for a large part of the credential distribution), the fall in score-sending rates is the largest for URMs who experienced the largest relative drop in admission rates—those in the middle of the credential distribution. Nonetheless, even for this group, the fall in score-sending rates is small, never exceeding 3 percentage points. Overall, figure 4 suggests that Prop 209 had the largest impact on URMs who experienced the largest drop in their chances of admission to UCLA and Berkeley. Figure 4a. Relative Change in Score-sending to Berkeley Figure 4a. Relative Change in Score-sending to Berkeley Close modal Figure 4b. Relative Change in Score-sending to UCLA Figure 4b. Relative Change in Score-sending to UCLA Close modal Figure 4c. Relative Change in Score-sending to UCSD Figure 4c. Relative Change in Score-sending to UCSD Close modal Figure 4d. Relative Change in Score-sending to UCD Figure 4d. Relative Change in Score-sending to UCD Close modal Figure 4e. Relative Change in Score-sending to UCI Figure 4e. Relative Change in Score-sending to UCI Close modal Figure 4f. Relative Change in Score-sending to UCSB Figure 4f. Relative Change in Score-sending to UCSB Close modal Figure 4g. Relative Change in Score-sending to UCSC Figure 4g. Relative Change in Score-sending to UCSC Close modal Figure 4h. Relative Change in Score-sending to UCR. Notes: The Dependent variable is a binary indicator of whether the student sent their SAT scores to a given school. Each point shows OLS coefficient estimate on the interaction between post Prop 209 and URM for a percentile of the academic credential distribution of URM UC applicants. Each estimate also includes controls for SAT math and verbal scores, high school GPA, parental income and education, and their interactions with the Ban dummy. The excluded race category includes white and other/unknown. Sample includes 1995–1996 (pre Prop 209) and 1998–2000 (post Prop 209). Plots smoothed with local polynomial regressions; bootstrapped standard errors displayed. DD = difference-in-difference. Figure 4h. Relative Change in Score-sending to UCR. Notes: The Dependent variable is a binary indicator of whether the student sent their SAT scores to a given school. Each point shows OLS coefficient estimate on the interaction between post Prop 209 and URM for a percentile of the academic credential distribution of URM UC applicants. Each estimate also includes controls for SAT math and verbal scores, high school GPA, parental income and education, and their interactions with the Ban dummy. The excluded race category includes white and other/unknown. Sample includes 1995–1996 (pre Prop 209) and 1998–2000 (post Prop 209). Plots smoothed with local polynomial regressions; bootstrapped standard errors displayed. DD = difference-in-difference. Close modal We next investigate further the nature of the changes in score-sending patterns by academic credentials. To simplify our analysis, we combine students into three groups: students in the top 40 percent of the URM credential distribution, those in the middle 40 percent (where the relative decline in URMs’ score-sending rates to Berkeley and UCLA was concentrated), and those in the bottom 20 percent. The top panel of table 7 shows how the average number of SAT scores sent to Berkeley and UCLA changed for URM applicants relative to white applicants after the ban on affirmative action. Because some of the results are sensitive to whether we control for Ban*Demographics, we display the results both ways. For the middle 40 percent of the academic credential distribution, we find a statistically significant drop in the number of scores sent to Berkeley and UCLA, though the size of the point estimate is extremely small and suggests that URMs reduced the number of scores sent to Berkeley and UCLA by at most 0.05, relative to a mean of 0.39 scores sent. We also find evidence of a small relative drop in the number of scores sent to Berkeley and UCLA for URMs in the bottom 20 percent of the academic credential distribution, but we find no relative change in the number of scores for those in the top 40 percent. Table 7. Score-sending Difference-in-Difference Coefficients by Academic Credentials Bottom 20%Middle 40%Top 40%Bottom 20%Middle 40%Top 40% Panel 1: Number of scores to Berkeley and UCLA URM 0.179** 0.337** 0.347** 0.184** 0.347** 0.342** (0.005) (0.007) (0.009) (0.005) (0.007) (0.009) Ban*URM −0.018* −0.037** −0.019 −0.026** −0.053** −0.012 (0.006) (0.008) (0.011) (0.007) (0.009) (0.011) Dep var mean 0.26 0.39 0.78 0.26 0.39 0.78 Panel 2: Number of scores to UC other than BER/UCLA URM 0.121** 0.162** −0.019 0.132** 0.183** −0.019 (0.008) (0.012) (0.014) (0.009) (0.012) (0.015) Ban*URM −0.003 0.049** −0.005 −0.023 0.014 −0.003 (0.011) (0.015) (0.018) (0.012) (0.016) (0.019) Dep var mean 0.52 1.02 1.52 0.52 1.02 1.52 Panel 3: UC senders – average SAT of UCs sent to URM 12.192** 16.102** 13.550** 11.827** 16.331** 13.837** (0.983) (0.688) (0.595) (1.042) (0.732) (0.617) Ban*URM −1.186 −4.940** −1.905* −0.578 −5.322** −2.378* (1.234) (0.843) (0.722) (1.371) (0.938) (0.764) Dep var mean 1,245 1,241 1,259 1,245 1,241 1,259 Ban*Demographics Bottom 20%Middle 40%Top 40%Bottom 20%Middle 40%Top 40% Panel 1: Number of scores to Berkeley and UCLA URM 0.179** 0.337** 0.347** 0.184** 0.347** 0.342** (0.005) (0.007) (0.009) (0.005) (0.007) (0.009) Ban*URM −0.018* −0.037** −0.019 −0.026** −0.053** −0.012 (0.006) (0.008) (0.011) (0.007) (0.009) (0.011) Dep var mean 0.26 0.39 0.78 0.26 0.39 0.78 Panel 2: Number of scores to UC other than BER/UCLA URM 0.121** 0.162** −0.019 0.132** 0.183** −0.019 (0.008) (0.012) (0.014) (0.009) (0.012) (0.015) Ban*URM −0.003 0.049** −0.005 −0.023 0.014 −0.003 (0.011) (0.015) (0.018) (0.012) (0.016) (0.019) Dep var mean 0.52 1.02 1.52 0.52 1.02 1.52 Panel 3: UC senders – average SAT of UCs sent to URM 12.192** 16.102** 13.550** 11.827** 16.331** 13.837** (0.983) (0.688) (0.595) (1.042) (0.732) (0.617) Ban*URM −1.186 −4.940** −1.905* −0.578 −5.322** −2.378* (1.234) (0.843) (0.722) (1.371) (0.938) (0.764) Dep var mean 1,245 1,241 1,259 1,245 1,241 1,259 Ban*Demographics Notes: Robust standard errors in parentheses. Each set of estimates shows OLS coefficient estimate on the URM constant term and interaction between post Prop 209 and URM. Each estimate also includes the controls used in table 2: SAT math and verbal scores, high school GPA, parental income and education, and their interactions with the Ban dummy. The excluded race category includes white and other/unknown, and URM includes blacks and Hispanics. Sample includes Californians who sent a score to at least one school and includes data from 1995–1996 (pre Prop 209) and 1998–2000 (post Prop 209). Groups are created by using the academic credential distribution of URM UC applicants. Bottom panel restricted to students who sent to at least one UC. *p < .05, **p < .01. In panel 2, we find some evidence that the relative fall in the number of scores sent to Berkeley and UCLA by URMs in the middle of the credential distribution was offset by a relative increase in the number of scores sent to the other UC campuses, though this increase is only statistically different from zero when we do not control for Ban*Demographics.23 For URMs in the top 40 percent and bottom 20 percent of the credential distribution, we find no relative change in the number of scores sent to other UCs. A direct implication of the fact that students sent fewer scores to Berkeley and UCLA and no fewer scores to other UCs is that the average quality of the set of UC schools to which URMs sent their scores must have declined. To examine this, the bottom panel of table 7 uses the average SAT score of incoming freshmen at the schools to which students sent their SAT scores as the outcome variable (for example, 1,344 at Berkeley and 1,111 at Riverside). This allows us to gauge the extent to which there was a fall in the average quality of the schools to which URMs sent their SAT scores after the end of affirmative action. Not surprisingly, we find there was a relative decline in the average SAT score of the schools to which URMs applied after Prop 209, and this relative drop was the largest in the middle credential group, though the magnitude is still extremely small; for students in the middle 40 percent of the academic credential distribution, the average combined math and verbal SAT scores at the UC schools to which URM applicants sent their scores fell by only 5 points relative to white applicants after Prop 209.24 We test whether the basic patterns revealed in table 7 by our score-sending data from the College Board—that URMs in the middle of the credential distribution shifted applications away from the top tier toward the remaining UCs—also appear in our application data from the UCOP. This helps reassure us that changes in application behavior are reflected in score-sending patterns. Note, however, that whereas table 6 focuses on Californians who sent their SAT scores to at least one school, our analysis of the UCOP data focuses on Californians who applied to at least one UC school (because those are the only individuals in the UCOP data). Results are shown in table 8. Unlike table 7, results are not sensitive to including Ban*Demographic interactions, so we only display our results with these interaction terms included. Although the magnitude of the findings is somewhat larger for our UCOP data than our College Board data, the basic patterns hold: Relative to similarly qualified white applicants, URM applicants in the middle of the credential distribution had a small but statistically significant drop in the number of applications sent to Berkeley and UCLA, and a statistically significant increase in the number of applications sent to the remaining UCs.25 Consistent with the College Board data, the UCOP data also show that the most highly qualified URM applicants did not send fewer applications (relative to whites) to Berkeley and UCLA after Prop 209. Like the College Board data, table 8 also shows a small drop (less than 7 points) in the average combined SAT score at the schools to which URMs applied. Unlike the College Board data, however, the UCOP data show that for the most highly qualified applicants, there was a marginally significant, small relative increase in the number of applications URMs sent to the remaining UCs and in the number of applications URMs sent to UC schools in general. Table 8. Application Difference-in-Difference Coefficients by Academic Credentials Bottom 20%Middle 40%Top 40% Panel 1: Applications to Berkeley and UCLA URM 0.25* 0.27* 0.09* (0.01) (0.01) (0.01) Ban*URM −0.06** −0.07*** 0.01 (0.02) (0.01) (0.01) Dep var mean 0.34 0.71 1.17 Panel 2: Applications to UC other than BER/UCLA URM −0.24* −0.34* −0.30* (0.03) (0.02) (0.01) Ban*URM 0.07 0.11* 0.04* (0.04) (0.02) (0.02) Dep var mean 2.12 2.30 1.88 Panel 3: Average SAT of UCs applied to URM 11.59* 14.10* 7.72* (1.12) (0.57) (0.47) Ban*URM −5.80* −6.63* −2.80* (1.60) (0.80) (0.64) Dep var mean 1221 1229 1266 Bottom 20%Middle 40%Top 40% Panel 1: Applications to Berkeley and UCLA URM 0.25* 0.27* 0.09* (0.01) (0.01) (0.01) Ban*URM −0.06** −0.07*** 0.01 (0.02) (0.01) (0.01) Dep var mean 0.34 0.71 1.17 Panel 2: Applications to UC other than BER/UCLA URM −0.24* −0.34* −0.30* (0.03) (0.02) (0.01) Ban*URM 0.07 0.11* 0.04* (0.04) (0.02) (0.02) Dep var mean 2.12 2.30 1.88 Panel 3: Average SAT of UCs applied to URM 11.59* 14.10* 7.72* (1.12) (0.57) (0.47) Ban*URM −5.80* −6.63* −2.80* (1.60) (0.80) (0.64) Dep var mean 1221 1229 1266 Notes: Conditional on applying to at least one UC. See notes from table 7. *p < .01. Nonetheless, on balance, the changes in application patterns in the UCOP data tell a similar story to the changes in score-sending patterns in the College Board data. That is, it appears that URMs, especially those in the middle of the index score distribution where the relative drop in URMs’ probability of admission to Berkeley and UCLA was the largest, shifted their applications toward less-selective UC schools, though the magnitude of this shift is extremely small. ### Other Schools This section explores whether there were changes in score-sending patterns to the most popular UC alternatives. #### Number of Scores to Non-UCs To assess whether URMs changed their score-sending patterns to schools outside the UC, we count the total number of scores sent to non-UC campuses.26 Results for the total number of scores sent outside the UC system are shown in Panel 1 of table 9. For these regressions, we show black and Hispanic students separately because their results are quite different. Black students in all credential groups and Hispanic students in the bottom group increased the number of scores sent outside the UC system. To further investigate, figure 5 plots the residuals from a regression of score-sending on student background characteristics (excluding race) and their interactions with the Ban dummy. For black students in the bottom two groups and Hispanic students in the bottom group, it appears that the results in panel 1 of table 9 are driven by an upward trend in scores sent that predated Prop 209.27 Thus, it is unclear what to make of these results. Figure 5. Score-sending Residuals. Notes: Score-sending residuals to non-UC schools. The residuals were obtained from regressing number of scores sent on the demographic characteristics shown in table 2: SAT math and verbal scores, high school GPA, and parental income and education, and their interactions with the Ban dummy. The figure divides students into groups based on the academic credentials of URM applicants to the UC; see text for description. Figure 5. Score-sending Residuals. Notes: Score-sending residuals to non-UC schools. The residuals were obtained from regressing number of scores sent on the demographic characteristics shown in table 2: SAT math and verbal scores, high school GPA, and parental income and education, and their interactions with the Ban dummy. The figure divides students into groups based on the academic credentials of URM applicants to the UC; see text for description. Close modal Table 9. Score-sending to Non-UCs, Difference-in-Difference Coefficients Bottom 20%Middle 40%Top 40% Panel 1: Number of scores sent to Non-UC Black 0.513*** 0.383*** 0.615*** (0.027) (0.042) (0.068) Hisp −0.012 −0.049* 0.058 (0.020) (0.024) (0.031) Ban*Black 0.226*** 0.243*** 0.175* (0.037) (0.055) (0.087) Ban*Hisp 0.113*** 0.034 0.065 (0.027) (0.030) (0.039) Dep var mean 3.64 3.66 3.89 Panel 2: Number of scores sent to Cal State system URM 0.116*** −0.102*** −0.186*** (0.014) (0.013) (0.011) Ban*URM 0.050** 0.007 −0.000 (0.019) (0.017) (0.014) Dep var mean 1.62 1.38 0.77 Bottom 20%Middle 40%Top 40% Panel 1: Number of scores sent to Non-UC Black 0.513*** 0.383*** 0.615*** (0.027) (0.042) (0.068) Hisp −0.012 −0.049* 0.058 (0.020) (0.024) (0.031) Ban*Black 0.226*** 0.243*** 0.175* (0.037) (0.055) (0.087) Ban*Hisp 0.113*** 0.034 0.065 (0.027) (0.030) (0.039) Dep var mean 3.64 3.66 3.89 Panel 2: Number of scores sent to Cal State system URM 0.116*** −0.102*** −0.186*** (0.014) (0.013) (0.011) Ban*URM 0.050** 0.007 −0.000 (0.019) (0.017) (0.014) Dep var mean 1.62 1.38 0.77 Notes: Robust standard errors in parentheses. Each set of estimates shows OLS coefficient estimate on the URM constant term and interaction between post Prop 209 and URM. Controls include SAT math and verbal scores, high school GPA, parental income and education, and their interactions with the Ban dummy. The excluded race category includes white and other/unknown, and URM includes blacks and Hispanics. Sample includes Californians who sent a score to at least one school and includes data from 1995–1996 (pre Prop 209) and 1998–2000 (post Prop 209). Groups are created by using the academic credential distribution of URM UC applicants. *p < .1, **p < .05, ***p < .01. #### Less-selective Non-UCs We also examine the number of scores sent to the twenty-two campuses in the CSU system, because these schools form the overwhelming majority of non-UC schools to which low-credential URMs (those in the bottom 20 percent of the academic credential distribution) sent SAT scores. It is worth noting that the CSU system was also affected by the statewide ban on affirmative action. Interestingly, however, at the time the ban went into effect (1998) virtually none of the CSU campuses was selective; twenty out of the twenty-two were able to admit every student who was eligible to apply (those with a GPA in select courses and test scores above a given threshold). Of the two schools that did have selective admissions, Cal Poly San Luis Obispo and Sonoma State, the only discernible break in enrollment numbers for URMs occurred at Cal Poly San Luis Obispo, suggesting that the ban on affirmative action had very little de facto effect on the CSU system.28 For the CSU system, table 9 reports there is no statistically significant relative change in the number of scores sent by URMs to schools in the CSU system, except for an increase for URMs in the bottom 20 percent of the credential distribution. ### Score-sending Relative to the Rest of the Country Thus far, our analysis has focused on describing the effects of Prop 209 on URM applicants relative to white applicants in California. Some of the effects of Prop 209, however, such as lower levels of diversity, could readily affect both URMs and non-URMs. To the extent that white students cannot be considered a valid control group, our estimates simply capture the change in the racial gap in admissions rates and score-sending rates. In this case, we can strengthen our analysis of score-sending rates by using data from the College Board on score-sending patterns in other states to control for nationwide trends in the racial gap in score-sending rates. Unfortunately, we cannot make a similar adjustment to our analysis of admission rates because we lack information on admissions rates outside the UC. In order to conduct as valid a comparison as possible, we focus on score-sending rates to Berkeley compared with score-sending rates to other flagship public universities that also have an enrollment of at least 2,500, average SAT math scores of at least 700, and a URM enrollment share of at least 10 percent. These restrictions yield a comparison group consisting of the top public universities in Illinois, Maryland, Michigan, North Carolina, and Virginia. We then estimate score-sending difference-in-difference-in-difference (DDD) regressions for SAT-takers who reside in one of these states and who send their SAT scores to at least one school. Results are shown in table 10. The coefficients on Black*Ban*CA and Hispanic*Ban*CA capture the change after Prop 209 in the racial gap in score-sending rates to Berkeley relative to the change in the gap at the other flagship universities in our sample. For Hispanic applicants, we find no statistically significant change in the racial gap in score-sending rates to Berkeley relative to the change in the gap at the other flagship universities in our sample after Prop 209. On the other hand, the coefficient on Black*Ban*CA suggests that the relative score-sending rate of black applicants fell by approximately 6 percentage points at Berkeley compared with the other flagship universities in our sample. Note that this is three times larger than the 2-percentage-point relative drop implied by table 5. As the coefficient on Black*Ban in table 10 suggests, this finding is driven by a relative increase in black applicant score-sending rates to the other flagship universities in our sample after Prop 209. Because this increase was not present in California, the relative drop in black applicant score-sending rates shown in table 10 is larger than the differences-in-differences coefficient in table 5. Table 10. Change in the Racial Gap in Score-sending Rates, Berkeley versus Other Flagships Flagship CA −0.212*** (0.023) CA*Ban 0.015 (0.009) Black 0.046* (0.018) Hisp 0.102* (0.039) CA*Black 0.130** (0.022) CA*Hisp 0.015 (0.048) Black*Ban 0.040* (0.013) Hisp*Ban 0.004 (0.013) Black*Ban*CA −0.061** (0.010) Hisp*Ban*CA −0.014 (0.011) R2 0.15 Observations 591,101 Dep var mean 0.32 Dep var mean, CA 0.22 Dep var mean, non-CA 0.40 Flagship CA −0.212*** (0.023) CA*Ban 0.015 (0.009) Black 0.046* (0.018) Hisp 0.102* (0.039) CA*Black 0.130** (0.022) CA*Hisp 0.015 (0.048) Black*Ban 0.040* (0.013) Hisp*Ban 0.004 (0.013) Black*Ban*CA −0.061** (0.010) Hisp*Ban*CA −0.014 (0.011) R2 0.15 Observations 591,101 Dep var mean 0.32 Dep var mean, CA 0.22 Dep var mean, non-CA 0.40 Notes: Score-sending to flagship universities in California, Illinois, Virginia, Michigan, Maryland, and North Carolina. Control group of flagship universities defined to be public universities with average SAT math scores of at least 700, enrollment of at least 2,500, and blacks and Hispanics constituting at least 10% of enrollment. The outcome variable is a dummy variable indicating whether the individual sent a score to the home state flagship university. Controls include SAT math and verbal scores, high school GPA, parental income and education, and their interactions with the Ban dummy. The excluded race category includes white and other/unknown. Sample includes students who sent a score to at least one school and includes data from 1995–1996 (pre Prop 209) and 1998–2000 (post Prop 209). *p < .1, **p < .05, ***p < .01. In terms of the imputed change in enrollment calculated in section 7, even if one were to use the DDD results to calculate the fall in URM enrollment, the magnitude of the change still would be extremely small. A tripling of the estimated change in score-sending would lead to an estimated forty-two URMs being displaced from Berkeley's admitted student pool per year, rather than fourteen. Compared with the estimated fall of nearly 1,400 URM admits per year due to the drop in admissions probability, it appears that changes in score-sending patterns explain a very small portion of the fall in URM enrollment after Prop 209—the importance of score-sending changes remains at least an order of magnitude less important than admissions rate changes. It appears that black applicants’ relative score-sending rates to Berkeley failed to keep pace with black applicants’ relative score-sending rates to similar flagship universities in other parts of the country, however. Our analysis of admission rates and SAT score-sending patterns reveals a number of interesting findings. First, all URMs, regardless of their academic credentials, experienced a substantial relative drop in their chances of admission to at least one UC school. Second, the relative decline in URM score-sending rates after the end of affirmative action was concentrated at Berkeley and UCLA among students who experienced the largest drops in their predicted probability of admission to those schools, but the fall in score-sending rates for these students was still small relative to the drop in their predicted probability of admission. Third, for the group of students for whom URMs’ relative score-sending rates to Berkeley and UCLA fell the most, we find some evidence that, relative to white applicants, URM applicants increased the number of scores they sent to less-selective UCs. This had the effect of lowering the average quality of the set of UC schools to which URMs sent their SAT scores (relative to non-URMs), though the magnitude of this drop was extremely small. An important issue in the debate surrounding Prop 209 (and bans on affirmative action more generally) is whether the end of affirmative action lowered the value URMs placed on attending UC schools. For example, URMs may have become less interested in applying to UC schools after the ban because they feared they would have fewer same-race peers if they were admitted. Unfortunately, our results do not allow us to make definitive conclusions about this kind of “chilling effect” because all URMs experienced a fall in their probability of admission to at least one UC school, and in theory this fall could have either increased or decreased URMs’ incentives to apply to UC schools (even holding constant the value they placed on attending UC schools). Thus, URM application rates to the UC could have remained high even if the value URMs placed on attending the UC fell after Prop 209. Nonetheless, regardless of the mechanism, our paper highlights the stability of URM application behavior in the face of substantial declines in their admission rates. In terms of evaluating why the total enrollment numbers of URMs fell dramatically at schools like Berkeley and UCLA after the end of racial preferences, the results of this paper suggest that this fall was brought about by the direct effect of the drop in URM applicants’ chances of admission and not because URM students were discouraged from applying. Conditional on observables, URM students were more likely than white students to apply to UC schools both before and after the ban on affirmative action. 1 “[UC Regent William] Bagley said […] that many qualified minority students do not apply to the UC system because of some students’ perception of its ‘anti-affirmative action’ stance” (Chung 2000). 2 Fisher v. University of Texas at Austin (2011, par. 238): “UT was keenly aware that by sending a message that people of all stripes can succeed at UT, the University would attract promising applicants from once-insulated communities, over time narrowing the credentials gap between minority and non-minority applicants. After Hopwood [Texas’ affirmative action ban], such applicants were dissuaded from applying to UT.” 3 Because UC San Francisco enrolls only graduate students and UC Merced did not open until 2005, neither is included in this analysis. 4 Card and Krueger (2005) also use SAT score-sending rates as a proxy for application rates. In their paper, they investigate how well score-sending rates predict application rates by comparing the number of score-senders with the number of applicants in race-SAT-campus-year bins. They show that the number of score-senders in a bin is highly correlated with the number of applicants. Because we also have data on race, SAT scores, applications, and score-sending, we were able to conduct our own parallel analysis. Our findings are similar, although the correlation decreases as the number of bins increases. 5 Their study also examines Texas's ban on affirmative action. 6 For example, “In an attempt to improve minority access to UC without the help of affirmative action, the university's investment in kindergarten-through-12th-grade outreach has rocketed from about$60 million in 1995 to $180 million last year and a planned$250 million this year” (Schevitz 2000).

7

“[In the last five years] minority admissions have dropped significantly and outreach expenditures have almost quadrupled to nearly \$300 million with minimal results” (Kudo 2000).

8

In general, results are not sensitive to whether these interaction terms are included. When they are sensitive, we provide regression results with and without the interaction terms.

10

This data set was used by David Card and Alan Krueger in the 2005 paper we build upon here.

11

The eight UC campuses are Berkeley, Los Angeles, San Diego, Santa Barbara, Davis, Irvine, Santa Cruz, and Riverside.

12

Additional information about this rich, publicly available data set can be found in Antonovics and Sander (2013).

13

Public institutions releasing individual-level data are generally cautioned to suppress or combine data cells containing fewer than five observations (where cells are represented by combinations of information that can be established from public sources other than the database). UCOP generally combined data cells containing fewer than one hundred observations—a very cautious stance.

14

The magnitude of the change in admissions probability for URMs is so large that it shows up no matter how the excluded group is defined.

15

In other words, a student taking the SAT in May 1994 as a high school junior would be part of a cohort that generally finished high school in 1995 and was reported by College Board as part of the 1995 cohort.

16

Results available upon request.

17

Published by the National Center for Education Statistics (NCES, various years).

18

Searching the LexisNexis article database gives the first mention of Prop 209 on 22 July 1996 (Lynch and Burdman 1996).

19

A portion of the 1996 cohort made score-sending decisions after the passage of SP-1. Our main results are similar when dropping the 1996 cohort, although dropping 1996 leaves 1995 as the only pre-policy change year.

20

When dropping 1998, score-sending effect sizes generally become larger. We add further discussion in the following.

21

Categorizing education in this way is natural because many schools give an admissions preference to first-generation college students.

22

As a formal check, we run a “placebo” test, comparing score-sending in 1995 with that in 1996, both pre-policy change years. Nonsignificant coefficients would be evidence of similar trends before the passage of Prop 209. Whereas coefficients for URMs are generally not significant, many campuses have significant coefficients for Asians, further evidence of pre-existing differential trends between Asian students and white students. Results available from authors.

23

As discussed subsequently, when using actual application data, we find a relative increase in the number of applications URMs sent to other UCs regardless of specification.

24

Another avenue by which students gain entrance into UCs is by initially attending another college and then transferring; this would not be captured by our score-sending data. StatFinder releases data on the number of transfer students by year and race (available at http://data.universityofcalifornia.edu/). A visual inspection of a plot of transfer students by race over time reveals no obvious trend break after Prop 209, with the possible exception of an increase for white applicants at UCLA.

25

The results of table 7 remain relatively stable when restricting the sample to UC score-senders. Thus, it is likely that the difference between tables 7 and 8 is due to differences between the application and score-sending data, rather than sample selection.

26

We exclude score-sending to scholarship organizations.

27

Although it would be possible to add race-specific time trends to these regressions, the limited number of years causes standard errors to be quite large.

28

The CSU system has become much more selective over time, so the ban on affirmative action is likely to bite at many schools in the CSU system now.

We thank the University of California Office of the President, the College Board, and David Card for assistance in obtaining our data. In addition, we are grateful to Prashant Bharadwaj, Julie Cullen, Gordon Dahl, Karthik Muralidharan, Joel Sobel, UC San Diego seminar participants, two anonymous referees, and especially Richard Sander for helpful comments and discussion.

Antonovics
,
Kate
, and
Richard
Sander
.
2013
.
Affirmative action bans and the ‘chilling effect'
.
The American Law and Economics Review
. In press.
Arcidiacono
,
Peter
,
Esteban
Aucejo
,
Patrick
Coate
, and
V.
Joseph Hotz
.
2011
. The effects of Proposition 209 on college enrollment and graduation rates in California. Unpublished paper, Duke University.
Backes
,
Ben
.
2012
.
Do affirmative action bans lower minority college enrollment and attainment? Evidence from statewide bans
.
Journal of Human Resources
47
(
2)
:
435
55
. doi:10.1353/jhr.2012.0013
Bertrand
,
Marianne
,
Esther
Duflo
, and
Sendhil
Mullainathan
.
2004
.
How much should we trust differences-in-differences estimates?
Quarterly Journal of Economics
119
(
1
):
249
75
. doi:10.1162/003355304772839588
Card
,
David
, and
Alan B.
Krueger
.
2005
.
Would the elimination of affirmative action affect highly qualified minority applicants? Evidence from California and Texas
.
Industrial & Labor Relations Review
58
(
3
):
416
34
.
Chung
,
Juliet J.
2000
. Minority admissions rise in U. of California. The Harvard Crimson, 6 April.
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
19
. doi:10.1016/j.econedurev.2004.11.005
Fisher v. University of Texas at Austin
.
2011
.
631F 3d 213 (2011), Court of Appeals, 5th circuit
.
Furstenberg
,
Eric
.
2010
. College admissions reform and racial SAT score gaps: An empirical analysis. Unpublished paper, University of Virginia.
Hinrichs
,
Peter
.
2012
.
The effects of affirmative action bans on college enrollment, educational attainment, and the demographic composition of universities
.
Review of Economics and Statistics
94
(
3
):
712
22
. doi:10.1162/REST_a_00170
Kudo
,
Timothy
.
2000
. Effects of outreach may take time: Increasing minority numbers will be long-range goal for UCs. The Daily Bruin, 8 June. Available http://www.dailybruin.com/article/2000/06/effects-of-outreach-may-take-t/. Accessed 21 August 2012.
Long
,
Mark C.
2004
.
College applications and the effect of affirmative action
.
Journal of Econometrics
121
(
1–2
):
319
42
. doi:10.1016/j.jeconom.2003.10.001
Long
,
Mark C.
, and
Marta
Tienda
.
2008
.
Winners and losers: Changes in Texas University admissions post-Hopwood
.
Educational Evaluation and Policy Analysis
30
(
3
):
255
80
. doi:10.3102/0162373708321384
Lynch
,
April
, and
Pamela
Burdman
.
1996
. UC regent blasts special graduations: Connerly questions ceremonies sponsored by some ethnic groups. The San Francisco Chronicle, 22 July.
National Center for Education Statistics (NCES)
. Various years. Digest of education statistics. Available http://nces.ed.gov/programs/digest/. Accessed 18 May 2012.
Pallais
,
Amanda
.
2009
. Small differences that matter: Mistakes in applying to college. Unpublished paper, Massachusetts Institute of Technology.
Schevitz
,
Tanya
.
2000
. UC regents urged to step up minority outreach at schools. The San Francisco Chronicle, 21 January.