Percent plans, automatic admissions, and college outcomes

Daugherty, Lindsay; Martorell, Paco; McFarlin, Isaac Article Percent plans, automatic admissions, and college outcomes IZA Journal of Labor Economics Provided in Cooperation with: IZA – Institute of Labor Economics Suggested Citation: Daugherty, Lindsay; Martorell, Paco; McFarlin, Isaac (2014) : Percent plans, automatic admissions, and college outcomes, IZA Journal of Labor Economics, ISSN 2193-8997, Springer, Heidelberg, Vol. 3, pp. 1-29, https://doi.org/10.1186/2193-8997-3-10 This Version is available at: https://hdl.handle.net/10419/152336 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by/4.0/ ORIGINAL ARTICLE Open Access Percent plans, automatic admissions, and college outcomes Lindsay Daugherty 1 , Paco Martorell 2 and Isaac McFarlin Jr 3* * Correspondence: [email protected] 3 Gerald R. Ford School of Public Policy University of Michigan, 735 S. State Street, Ann Arbor, MI 48109, USA Full list of author information is available at the end of the article Abstract Access to selective universities is highly coveted because of the perception that attending one provides opportunities otherwise difficult to obtain. To broaden access to the state’s flagship universities in a manner that does not rely on conventional affirmative action, Texas passed the Top Ten Percent Plan in 1997, which guarantees automatic admission to any public university in the state to students in the top decile of their high school class. We estimate the effect of eligibility for automatic admissions on college choice and persistence for students in a diverse urban school district. Regression discontinuity estimates show that eligibility for guaranteed admissions has a substantial impact on enrollments at Texas flagship universities and increases the number of semesters enrolled at flagships. The increase in flagship enrollments appears to displace enrollments in private universities but has no effect on overall college enrollment or the quality of college attended. The effects are concentrated in schools that have high college-sending rates (relative to other schools in the district), suggesting that automatic admissions may have little effect on students in the most disadvantaged schools. JEL: I2 1. Introduction An influential body of research shows that the college wage premium has risen in recent decades (Katz and Murphy, 1992; Juhn et al. 1993; Goldin and Katz, 2008). Recent studies have also found that the return to attending higher-resource, selective colleges may be larger than the return to attending less-selective institutions (Andrews et al. 2012; Hoekstra, 2009; Zimmerman, 2014). This differential return to selective college attendance may be especially large for lower-income black and Hispanic students (Dale and Krueger, 2002; 2011). Despite the relatively large returns to selective college attendance among disadvantaged and underrepresented minority students, they matriculate at much lower rates than do higher-income and white students (Bastedo and Jaquette, 2011; Carnevale and Rose, 2004; Reardon et al. 2012). 1 Thus, devising programs to close these gaps has considerable significance for economic policy. “Percent plans”represent an important type of policy aimed at enhancing diversity at selective public universities. 2 Currently three states, Florida, California, and Texas, have percent plans. These plans arose as a response to legal and political pressure against race-conscious affirmative action and give preferential treatment in the college admissions process to students in the top xpercent of their high school class (Marin © Daugherty et al.; licensee Springer. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. Daugherty et al. IZA Journal of Labor Economics 2014 2014, 3:10 http://www.izajole.com/content/3/1/10 and Lee, 2003; Holzer and Neumark, 2000). To the extent that there exists racial and economic stratification across high schools, percent plans could help provide students from underrepresented groups greater access to selective colleges since the admissions preference is given to the same proportion of students at every high school (Horn and Flores, 2003). However, percent plans are very controversial, with some critics claiming thattheyareinadequateforensuringdiversityandothersarguingthattheyunfairly disadvantage students from more competitive high schools (Barr, 2005). 3 Despite this controversy, it remains unclear how eligibility for these admissions preferences affects college outcomes. This partly reflects the fact that eligibility for automatic admission is not randomly assigned and also because it is difficult to assemble a dataset containing information on high school students that contains their class rank and is linked to college outcomes. The purpose of this paper is to evaluate the effect of eligibility for automatic college admissions. Specifically, we estimate the effect of being eligible for automatic admissions under the Texas state policy known as the Top Ten Percent (TTP) Plan. In Texas during our study period, if a student was above the 90 th percentile of their high school’s class rank distribution, they qualified for automatic admission to any Texas public college or university. 4 This makes it possible to use a regression discontinuity design that identifies the causal effect of eligibility for the admissions preference by comparing the outcomes of students on either side of the top ten percent class rank cutoff. To carry out this analysis, we have assembled a dataset of high school records for students from a large, diverse Texas school district that has the information necessary for calculating class rank and that is linked to information on college enrollment from the National Student Clearinghouse (NSC). Our approach has several important strengths. First, our data are drawn from students in a large urban school district where the majority of students are black or Hispanic and that historically has sent few students to the state’s leading universities. Thus this study sheds light on whether eligibility for automatic college admissions affects college outcomes for a demographic that the TTP Plan was designed to benefit. Second, the use of college enrollment data from the National Student Clearinghouse allows us to consider a wide set of college outcomes. In particular, we can assess what types of institutions are displaced when flagship enrollment increases as a result of being in the TTP. This is also the first study on the effect of eligibility for automatic admissions examining longer-run college persistence measures. Our findings show that eligibility for automatic admissions under the TTP Plan increases the likelihood that students in our sample enroll at a Texas flagship university. The magnitude of this effect is large; the smallest point estimates imply an increase of at least 60 percent above the baseline enrollment rate for students who barely miss being in the top decile. We also find some evidence that barely being in the top ten percent of one’s class increases the total number of semesters enrolled at a flagship university over the four years following high school graduation (although by less than would be implied if the enrollment effect in the fall following high school graduation persisted over the following four years). Our results also suggest that this effect on flagship enrollment leads to substitution away from private or out-of-state colleges rather than from lower-ranked public institutions. 5 In fact, we do not find that eligibility for automatic admissions affects the “quality”of colleges students attend, as measured by Barron’s rankings and the Daugherty et al. IZA Journal of Labor Economics Page 2 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 fraction of applicants who are admitted. We also do not find evidence of an effect on the likelihood of attending college or on enrollment at a four-year college. We find some evidence that these effects vary across student subgroups. We find statistically significant effects on measures of college choice for underrepresented minorities that are similar in magnitude to those for white and Asian students (the estimates for whites and Asians, however, are not precisely estimated). However, these effects are concentrated in schools that have high college-sending rates (relative to other schools in the district). In high schools that send relatively few students to college, we find little evidence that the automatic admissions guarantee affects any of the college outcomes we examine. Overall, our results suggest that in the urban school district we examine, offering students eligibility for automatic admissions may do little to improve access to selective colleges, especially among students who attend the most disadvantaged high schools. 2. Background 2.1 Rationale for percent plans Selective colleges admit students on the basis of academic achievement, ACT and SAT scores, and extracurricular activities. Consequently, students who attend low-quality schools, are financially constrained, and have limited information about how the admissions process works or about financial aid opportunities are at serious disadvantage with respect to enrolling in selective colleges (Winston and Hill, 2005; Owings et al., 1995; Pallais and Turner, 2006). In particular, the lack of diversity at selective colleges and flagship universities has been attributed to the underrepresentation of low-income and black and Hispanic students (Pallais and Turner, 2006; Bowen and Bok, 1998). One of the primary ways universities have attempted to increase the diversity of their student body is through “affirmative action”policies that grant preferences in admissions decisions to members of racial and ethnic minorities (Holzer and Neumark, 2000). However, political and legal pressure has led to rollbacks or bans in affirmative action, and as shown in Hinrichs (2010), these bans led to reductions in minority enrollment in selective colleges. 6 Percent plans, including the Texas TTP Plan arose as an attempt to formulate a policy that would help maintain diversity in competitive public universities in a way that did not explicitly give admissions preferences based on race (U.S. Department of Education 2003). The idea behind these plans is that the top students in schools with high concentrations of low-income and underrepresented minority students would be eligible for admissions preferences even if their other credentials (e.g., SAT scores) were lower than those of students in more competitive high schools who did not qualify for the admissions preference. Thus, the extent to which percent plans would increase diversity in selective public universities depends on there being considerable segregation across high schools by race and socioeconomic status (Horn and Flores, 2003). 2.2 Prior research and contributions of this study Much of the research on the TTP centers on how it compares to race-based affirmative action in creating diverse college campuses. Long (2004a, b) argues that there are too few minority students who would qualify for automatic admissions for the policy to offset the reductions in minority enrollment that result from banning affirmative action. Daugherty et al. IZA Journal of Labor Economics Page 3 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 Tienda et al. (2008) find that the perceived increases in diversity at the University of Texas –Austin under the TTP Plan are largely driven by an increase in the enrollment rates of Asian students, while enrollment rates of blacks and Hispanics declined. There are a number of explanations for why the TTP Plan may have smaller effects than affirmative action on the enrollment of black and Hispanic students. Niu et al. (2006a) find that black and Hispanic students are less likely to be ranked in the TTP, particularly in schools with high minority populations. Even among students in the top decile, minorities and graduates from schools serving a large share of low-income students are less likely to choose selective institutions as their most preferred school Niu et al. (2006b). In addition, there are students who may be uninformed of the policy. In a recent survey of Texas seniors, Hispanic students were significantly less likely to report that they know “a lot”about the TTP Plan, a pattern that holds even among TTP students with a preference to attend a four-year university (Niu et al., 2006a). Financial constraints are also likely to play a role, though there are no studies to date that look at the potential that unmet financial need is deterring TTP students from enrolling at the Texas flagships 7 . While a number of studies have examined the effects of the TTP Plan on overall enrollment rates, solid empirical evidence on how automatic admissions affect individual enrollment decisions remains limited. This is likely due to the lack of student-level data with the necessary class rank information needed to conduct such an analysis. One exception is a study by Niu and Tienda (2010) that uses survey data on a sample of high school graduates that includes administrative data on class rank at graduation. Overall, they do not find statistically significant estimates of effects on flagship enrollment, although they do find positive effects for Hispanics and those attending “typical”high schools. Fletcher and Mayer (2013) find some evidence of an increase in applications and enrollment resulting from being in the TTP. However, their sample is restricted to high school students that apply to Texas flagships, making their findings difficult to interpret. Our paper builds on prior research in several ways. Most importantly, since our data cover a broad range of public and private institutions both inside and outside of Texas, we can examine what types of institutions are displaced by increased flagship enrollment. We can also examine whether there are effects on longer-run measures of college persistence. Second, we have information on all high school graduates from the study district, not just applicants or matriculants. This allows us to quantify the effect of eligibility for automatic admission on college outcomes and to do so for key student subgroups (e.g., college-sending rate of the student’s high school). Finally, while our data are not representative of the entire state of Texas, we have large samples for a key population of interest –students in a large, diverse urban school district with traditionally low college enrollment and a large number of low-income and minority students. 3. Institutional details The Texas Top Ten Percent Plan was instituted in 1997 in response to Hopwood v. Texas, a case that banned the use affirmative action in college application decisions. Rather than using explicit race-based considerations in application decisions, the TTP Plan states that students who are ranked in the top ten percent of their class must be granted automatic admission to the public Texas college of their choice. The state Daugherty et al. IZA Journal of Labor Economics Page 4 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 grants flexibility to districts in how they choose to calculate GPA and class rank. However, to receive automatic admission, students must provide a transcript along with their application that verifies their class rank falls within the TTP. Students must also take either the SAT or the ACT, although for students in the TTP of their high school class, these tests are not used for admissions decisions. For students who are not in the TTP of their high school class, admissions decisions are based on standard criteria including GPA and class rank, admissions test scores, and non-academic factors such as socioeconomic status, personal statements, and extracurricular activities. 8 Following the Grutter v. Bollinger decision in 2003, the University of Texas at Austin (but not Texas A&M) reinstated race-conscious affirmative action. The aggressiveness of the new affirmative action policy, however, is unclear. There is evidence indicating that the Texas flagships did provide racial minorities preferential treatment in admissions decisions prior to the Hopwood decision and that this ended after the ban on affirmative action went into place (Long and Tienda, 2008). However, we are unaware of any research that examines the extent to which racial preferences are used in the post-Grutter context. Nonetheless, since the University of Texas at Austin uses affirmative action, in subgroup analyses we report estimates by race. A key consideration for our study is how class rank is defined for the purposes of determining eligibility for automatic admissions. The way in which class rank is calculated could vary across schools as state law does not specify how this is to be done. Moreover, both absolute class rank and the number of students used to determine percentile class rank are not constant over time. The relevant class rank for determining eligibility for automatic admissions is the one used at the time of application to college, which might vary across students. To better understand the process by which the relevant class rank for determining automatic admissions was calculated, we contacted counselors at each of the high schools in the district we examine as well as a representative from the district’s central office. Under the student information system used in the district during our study period, class ranks were calculated centrally. Both class rank and the total number of students enrolled were calculated at the end of each semester and given to the student records office, which then distributed transcripts to high school campuses. 9 However, this information is not stored by the district as part of its regular data system, and we instead calculate class rank based on course-level grade records (see Section 5 below for details). A related issue is that both absolute class rank and the number of students used to determine percentile class rank are not constant over time. The relevant class rank for determining eligibility for automatic admissions is the one used at the time of application to college, which might vary across students. Although students can choose when to apply to college, the applications to the University of Texas system are due December 1 st ,and applications to TAMU are due in the middle of January. At the same time, first semester high school grades are not released in the district until late January or early February. With senior grades not available in time for application to flagship colleges, most students will be accepted to colleges under the TTP Plan according to GPA and class rank measured at the end of 11 th grade. We confirmed with high school counselors in the district that 11 th grade class rank was most likely to be reported on applications to four-year universities, especially for the flagships which have relatively early application deadlines. Daugherty et al. IZA Journal of Labor Economics Page 5 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 4. How eligibility for automatic admissions can affect student outcomes In the college enrollment process, three distinct decisions take place: students determine which colleges they will apply to, colleges determine which students they will accept among the pool of applicants, and students must determine which college to enroll in (if any) among colleges that accept them. Automatic admissions policies could potentially affect enrollment and college choice through any of these processes (Card and Krueger, 2005). If all students at the TTP cutoff would likely be admitted even without the TTP Plan, and if acceptance rates for students just outside of the TTP are also high, then automatic admissions might have minimal effects on student outcomes. In fact, some research suggests that this may have been the case prior to the adoption of the TTP Plan (Walker and Lavergne, 2001). However, this reasoning fails to account for the increase in applications to the flagship universities since the TTP Plan went into effect. Today, the percentage of acceptances to UT and TAMU accounted for by TTP students increased from 36 percent to 61 percent, and the recent efforts of UT to cap the percentage of enrollees who are accepted under the TTP Plan provides further evidence that Texas flagships would be unlikely to accept all top decile applicants in the absence of the policy. Moreover, some research shows that the TTP Plan expanded the set of high schools that have students that apply to the flagship universities (Long et al. 2010). If students from these schools have lower college entrance exam scores (and other credentials) than did the typical pre-TTP Plan applicant, the admissions patterns from the period before the adoption of the TTP Plan might not reflect the importance of automatic admissions to current applicants. This is especially important for our study district, which has much lower college-sending rates than is typical in Texas. Eligibility for automatic admissions could also increase the likelihood of applying to a particular school. A number of studies find that as race-based affirmative action policies were eliminated, the decrease in probability of acceptance for minority students led to lower application rates to competitive colleges (Long, 2004b; Brown and Hirschman, 2006), even though highly qualified minority applicants were less affected (Antonovics and Backes, 2013). By increasing the probability of acceptance to 100 percent for students who are ranked in the top of their class, automatic admission policies should increase the number of applications from TTP students. Even if being in the TTP does not actually increase the probability of acceptance conditional on applying, the TTP Plan makes the admissions guarantee explicit, and this alone could change application behavior if students were not aware that they had a very high probability of being accepted without the highly visible TTP Plan. In addition to considering the ways in which automatic admission could affect enrollment and college choice, it is also important to think of how these effects could vary across student subgroups. One of the primary purposes of the TTP Plan was to ensure diversity at college campuses, particularly in Texas flagships where black and Hispanic students have historically been underrepresented and that accept students from a relatively small number of high schools throughout Texas (Barr, 2005). Ex ante, though, it is not clear whether effects of automatic admissions would be larger or smaller for students from more disadvantaged backgrounds. On one hand, minority students in the district tend to be concentrated in the high schools that have lower college-sending rates and lower academic achievement in general. Since these students may have difficulty being Daugherty et al. IZA Journal of Labor Economics Page 6 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 admitted without automatic admissions, effects on outcomes like flagship enrollment might be larger for them. On the other hand, factors such as credit constraints and lack of information could mitigate the effects of automatic admission for minority students and students from schools with lower college-sending rates. These considerations motivate our analysis of the effects by race and also by whether a student attends a high school that traditionally sends few students to college. 5. Data and descriptive statistics 5.1 Data This paper uses data from a large urban school district in Texas. We focus on graduates from the 2002 through 2008 graduating cohorts. Data files include administrative data on student demographics and high school membership, semester course files with grades, high school exit exam scores, and graduation information. Full student-level data are available beginning in 1999, allowing us to follow all students in our sample throughout high school. District files are supplemented with college data from the National Student Clearinghouse (NSC), a non-profit organization that is now the nation’s leading source for postsecondary degree and enrollment verification. District student records are matched to NSC data via students’social security numbers. Our data do not include a class rank measure, and we construct it from the district student-level data. As noted above, during our study period the class rank included on a student’s transcript is calculated centrally by the district, and we follow the procedure used by the district to determine the absolute and percentile rank. First, we compute cumulative GPA at a given time using grades received in courses taken up to that point. These data come from course enrollment files that include course numbers and titles for each course in which a student is enrolled, the grade earned, and course entry and withdrawal dates (where applicable). 10 Second, students are ranked within a school on the basis of cumulative GPA to determine absolute class rank at a particular time. Finally, the percentile class rank is calculated as the ratio of absolute class rank to the number of students in a school with a valid cumulative GPA. As described above, the class rank relevant for automatic admissions to UT-Austin and Texas A&M is the rank at the end of 11 th grade. Therefore, we use this measure in our analysis. NSC data includes semester-level observations for each NSC-reporting institution that a student attends, including date of enrollment and completion and level of enrollment (e.g. part-time, full-time). More than 92 percent of higher education institutions in the United States report to the NSC. We construct several outcomes using the NSC data. The NSC data are supplemented with data from the Integrated Postsecondary Education Data System (IPEDS) that contains other information on the intuitions students attend. We also use information on public or private funding structure and state in which the college is located to determine whether students go to college inor out-of-state as well as whether they attend a public or private institution. We also use the IPEDS data to construct measures related to college “quality”and cost. The first quality measure is the Barron’s ranking of how competitive admission is at a particular college. We focus on using this information to classify whether or not a school is ranked by Barron’s as either a “highly”or “most”competitive institution. 11 The second is the selectivity defined as the fraction of applicants who are admitted. Because these variables reflect the characteristics of applicants and students enrolled Daugherty et al. IZA Journal of Labor Economics Page 7 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 at a given institution, they should be thought of as indicators for how prestigious a school is rather than as a proxy for some output-based measure of effectiveness. However, the admission rate closely corresponds to what “selectivity”is commonly understood to mean, as are the Barron’s rankings (Hoxby, 2001). This notion of “quality”also corresponds with what is meant in public discussions about access to “elite”universities. For cost, we use the tuition “sticker price”students face at a particular institution (for which we use the 2009 tuition as listed in IPEDS). It is important to recognize that this is a limited proxy for the actual tuition students will have to pay since many of the students in our sample come from fairly disadvantaged backgrounds and would qualify for substantial amounts of financial aid. Nonetheless, because selective institutions generally charge a higher “sticker price”, tuition is useful for characterizing the type of institutions students attend. Moreover, other studies examining college choice have found that tuition influences the college students attend (Long 2004a; Jacob et al. 2013). Using the information from NSC and IPEDS, we first look at short-run measures of college enrollment and choice. These measures are based on the schools in which students enroll in the fall following graduation. Next, we look at persistence measures including year-by-year enrollment up to four years after high school graduation, total semesters enrolled in college four years after graduation, and transferring to a more (or less) selective college. The sample used in the analysis consists of students who graduated from a district high school between 2002 and 2008. We limit the analysis to graduates because the district only obtained college enrollment data from NSC for high school graduates. One concern with doing so is that it might impart selection bias if TTP status at the end of 11 th grade affects the probability of graduation. However as we discuss below, there is no evidence for this type of effect. We also exclude students who did not have valid GPA’s at the end of 11 th grade (for instance, students who transfer into the district in 12 th grade) since our empirical strategy relies on using 11 th grade percentile class rank as the running variable in the regression discontinuity estimation. Our final sample includes 17,057 students across the 7 cohorts. 5.2 Descriptive statistics Table 1 provides descriptive statistics for our analysis sample by TTP status. We also show results for students close to the TTP cutoff (i.e., above the 80 th percentile) and stratified by whether a student attended a high school that sent a relatively high fraction of its graduates to college (we define a high college-sending high school as one where at least 25 percent of the graduates enroll in college in the fall following graduation). The results indicate that TTP students are more likely to be white and female and are less likely to be economically disadvantaged. As expected, TTP students are higher performing across all measures of academic achievement. Students in the TTP are more likely to graduate with a recommended or distinguished diploma, take a college entrance exam, and have much higher exit exam scores. These patterns are found in both types of high schools, although higher college-sending schools serve students from higher socioeconomic backgrounds and have much better academic achievement. TTP students have substantially higher college-going rates than students outside the TTP. Overall, almost 60 percent of students in the TTP attend college in the fall Daugherty et al. IZA Journal of Labor Economics Page 8 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 7.2 Effects on college enrollment and choice Table 3 presents the estimates of the effect of being in the TTP on college enrollment and college choice outcomes. As mentioned previously, all of these outcomes refer to enrollment in the fall following graduation from high school. The results indicate that membership in the TTP does increase flagship enrollment. The estimates range from about 5 to 7 percentage points in columns 1-4 (local linear regressions with bandwidths of 5 or 10 percentage points) and 11 to 14 percentage points in the IK specification, which uses a very narrow bandwidth. These results provide strong evidence that eligibility for automatic admissions increases the likelihood of flagship university attendance, although the magnitude of the effect is somewhat sensitive to the model specification. When examining effects on flagship enrollment separately for the two flagship universities, we find positive estimates for both Texas A&M and UT-Austin, although these are not always statistically significant. Graphical evidence consistent with these results Table 3 Effect of being in top 10% at end of 11th grade on college enrollment and choice Enrolled in Flagship 0.057** 0.053** 0.074* 0.066* 0.136** 0.110** (0.021) (0.020) (0.029) (0.027) (0.048) (0.042) Enrolled UT Austin 0.042** 0.040** 0.032 0.027 0.077 0.063 (0.016) (0.016) (0.023) (0.022) (0.041) (0.039) Enrolled Texas A&M 0.016 0.014 0.039* 0.037 0.058 0.048 (0.015) (0.014) (0.020) (0.020) (0.031) (0.030) Enrolled Public Non-Flagship -0.016 -0.010 -0.017 -0.013 -0.097 -0.089 (0.023) (0.023) (0.033) (0.032) (0.056) (0.049) Enrolled Private or Out-of-State -0.081** -0.079** -0.028 -0.030 -0.075 -0.076 (0.024) (0.023) (0.033) (0.031) (0.055) (0.050) Enrolled Out-of-State -0.012 -0.013 0.005 0.005 0.013 0.010 (0.015) (0.015) (0.021) (0.021) (0.036) (0.034) Enrolled Private -0.069** -0.066** -0.033 -0.035 -0.090 -0.085 (0.019) (0.019) (0.027) (0.026) (0.052) (0.049) Enrolled 4 Year -0.041 -0.036 0.017 0.010 -0.050 -0.064 (0.030) (0.027) (0.043) (0.038) (0.067) (0.054) Enrolled any College -0.040 -0.036 0.029 0.022 -0.023 -0.042 (0.031) (0.028) (0.044) (0.039) (0.070) (0.059) Enrolled Highly or Most Competitive College -0.009 -0.010 0.055 0.043 0.065 0.035 (0.027) (0.024) (0.038) (0.033) (0.060) (0.049) Selectivity (% Admitted) 0.015 0.014 -0.016 -0.013 -0.000 0.003 (0.012) (0.011) (0.017) (0.015) (0.029) (0.024) Tuition (Sticker Price), zero if not enrolled -1751.048** -1650.040** -243.655 -329.141 -589.470 -632.018 (575.676) (524.495) (792.473) (720.169) (788.656) (692.873) Bandwidth 10 10 5 5 IK opt bw IK opt bw Baseline covariates? No Yes No Yes No Yes Note: Outcomes refer to enrollment and choice in the semester following high school graduation. Columns 1 and 2 use local linear regression with bandwidths of 10 and 5 percentage points, respectively; robust standard errors in parentheses. Column 3 uses the Imbens and Kalyanaraman procedure for point estimates and standard errors. N for bw=10 is 4,196 and for bw=5 is 2107 (N varies by covariate in column 3). **, * denote statistically significant estimate at the 1% and 5% level, respectively. Daugherty et al. IZA Journal of Labor Economics Page 15 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 can be seen in Figure 3, which shows local averages (in bins 1 percentile point wide) and local linear regression fits (separately for bandwidths of 5 and 10 percentage points). The results for flagship attendance (upper-left panel) clearly show that the likelihood of attending a flagship drops when class rank falls above the ten percent cutoff. Three other points about the flagship enrollment results are noteworthy. First, the estimates are not very sensitive to the inclusion of baseline covariates, which is consistent with the baseline covariates being balanced on either side of the TTP cutoff. Second, the fact that we find a discontinuity in flagship enrollment suggests that measurement error in our class rank measure might be minimal. This is because even small amounts of error in either the GPA calculation or the count of students included in the denominator could wipe out any discontinuity that exists at the actual TTP cutoff when using the noisy class rank measure (Pei, 2011). 17 Third, the effects are large in magnitude. Only about 9 percent of students who barely miss being in the TTP enroll in a flagship, so even the smallest point estimates imply being in the TTP increases the likelihood of flagship enrollment by almost 60 percent. A key question is whether this effect on TTP enrollment reflects an overall increase in the likelihood of enrolling in college. The estimates in Table 3 offer little indication that this is the case. Across all specifications, the largest point estimate of the effect of being in the TTP on enrollment in a four-year institution is 2 percentage points. In the specification with the most precise estimates (bandwidth of 10 controlling for baseline covariates), we can rule out effects larger than about 2 percentage points. These estimates are consistent with the lack of a discontinuity in the probability of 4-year college enrollment at the TTP cutoff as seen in the lower-right panel of Figure 3. We find very similar results when examining the effect of enrolling in any college, which includes enrollment at two-year community colleges. A positive effect on flagship enrollment but no effect on overall college enrollment implies that the increase in flagship enrollment must be displacing enrollment at some other type of institution. One possibility is that students who would go to a flagship if 0.1 .2 .3 .4 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 0.05 .1 .15 .2 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 0.05 .1 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 .2 .3 .4 .5 .6 .7 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 Figure 3 Flagship and 4-year college enrollment by end of 11 th grade class rank. Daugherty et al. IZA Journal of Labor Economics Page 16 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 they barely make it into the TTP would go to a lower-ranked public university if they barely missed it. We find little evidence of this type of displacement, as we see no discontinuity in the probability of attending a public non-flagship institution at the TTP cutoff. The estimates in Table 3 from the models with a bandwidth of 10 and 5 are very close to zero. The estimates from the IK estimator are negative and larger in magnitude, but imprecisely estimated due to the very narrow bandwidth (1.96 percentage points). We also do not find evidence that TTP affects enrollment in any college, suggesting that it does not increase community college enrollment either, but this is to be expected since these schools admit anyone with a high school degree. We do find some evidence that being in the TTP displaces enrollment at private or out-of-state colleges. The upper-right panel of Figure 3 shows that enrollment in these institutions trends downward with class rank but that it appears to jump up at the TTP cutoff. The estimates in Table 3 with the bandwidth of 10 and the IK specification are negative and about 7-8 percentage points as well as statistically significant with the wider bandwidth. The estimates for the bandwidth of 5 are also negative but smaller in magnitude and not statistically significant. When examining enrollment in private and out-of-state enrollment separately, the estimates in Table 3 suggest that any displacement seems to be operating through a reduction in enrollment in private institutions rather than out-of-state institutions since the estimates for out-of-state enrollment are small and not statistically significant. Next, we examined whether TTP status affects measures of college “quality”. This is important, because eligibility for automatic admissions is designed to help improve access to elite universities by guaranteeing access to the state’s flagship institutions. On the other hand, much of the opposition to the TTP Plan centers on the claim that students admitted the TTP Plan’s automatic admissions guarantee will take spots from better-qualified students who are not admitted under the TTP Plan because they attend more competitive high schools. To examine this issue, we first estimate the effect of eligibility for automatic admission on the probability of enrolling in a college ranked by Barron’saseithera“highly”or “most”competitive institution. The point estimates in Table 3 are not statistically significant, but the magnitudes are sensitive to the choice of bandwidth. With a bandwidth of 10, the estimates are very close to zero and statistically insignificant. The estimates with narrower bandwidths are positive and larger in magnitude but are too imprecisely estimated to be statistically significant. This is consistent with the graphical evidence in the lower-left panel of Figure 4, where there is not clear evidence of a sharp change in this outcome at the TTP cutoff. Next, we use a continuous measure of selectivity defined as the fraction of applicants to a college who were admitted (so that a lower valueofthemeasureindicatesamoreselective institution). 18 The lower-right panel of Figure 4 shows that selectivity is strongly related to class rank but that there is no discontinuous change in average college selectivity at the TTP cutoff. Similarly, the point estimates in Table 3 are small in magnitude and statistically insignificant. Taken together, the above results provide very little indication automatic eligibility for admission under the TTP Plan affects the “quality”of institutions students attend. However, since it appears that automatic admissions increases flagship enrollment at the expense of private university attendance, it may be that students attend more expensive colleges. To address this question, we estimated effects on annual tuition charged by Daugherty et al. IZA Journal of Labor Economics Page 17 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 the colleges students attend. 19 As noted earlier, the sticker price tuition is likely to severely overstate the actual tuition faced by many students in our sample, but it still is useful for characterizing the type of institution students attend, and also because it has been shown to be correlated with college choice (Long 2004a; Jacob et al., 2013). The estimates in Table 3 offer some evidence that barely being in the TTP leads to attending a college with lower sticker price tuition, but the estimate is only statistically significant with the bandwidth of 10. The estimates for narrower bandwidths are smaller and not statistically significant. 7.3 Effects on college persistence We now turn to outcomes that characterize student persistence in college. Figure 5 shows flagship enrollment by class rank in years 2-4 following high school graduation as well as total semesters spent in a flagship 4 years after graduation. For enrollment in years 2-4, it appears flagship enrollment is higher just to the left of the TTP cutoff, although the magnitude of any discontinuity is smaller than it is for year 1 enrollment (Figure 3). This is confirmed in Table 4, where the point estimates are smaller in magnitude than they are for year 1 flagship enrollment and not statistically significant. The average of total semesters spent in a flagship does appear to change discontinuously at the TTP cutoff. The estimates in Table 4 are positive and statistically significant with the bandwidth in the IK specification (and 3 out of the 4 other estimates are also statistically significant at the 10 percent level). The magnitude of the estimates is about 4 to 5 times larger than the effect on year 1 flagship enrollment in Table 3. This is consistent with the effects of being in the TTP on flagship enrollment in subsequent years being positive but fading out relative to the effect in year 1. In contrast if the year 1 enrollment effect were persistent throughout the following four years, the effect on total semesters at a flagship would be 8 times as large as the year 1 enrollment effect. Figure 6 shows results for enrollment in private or out-of-state schools. For year-byyear enrollment and for total semesters, there is some indication that students in the .1 .15 .2 .25 .3 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 .1 .2 .3 .4 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 0.2 .4 .6 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 .7 .8 .9 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 Figure 4 Non-flagship enrollment and college quality by end of 11 th grade class rank. Daugherty et al. IZA Journal of Labor Economics Page 18 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 TTP are less likely to be enrolled (or enroll in fewer semesters) in private or out-ofstate colleges. The point estimates in Table 4 for a bandwidth of 10 are negative and statistically significant for year 2 enrollment and total semesters (the estimates for year 3 and 4 enrollment are at the margin for statistical significance), and the estimates for the IK specification are similar in magnitude but less precisely estimated. However, the estimates from the model with a bandwidth of 5 are smaller in magnitude and not statistically significant, although they are still negative. Overall, these results suggest that being in the TTP may lead to persistent displacement of private or out-of-state enrollment, although the estimates are somewhat sensitive to the regression specification. Figure 7 shows similar graphs for enrollment in any 4-year college. The visual evidence does not suggest that there is any effect of being in the TTP on either year-by-year enrollment nor on total semesters enrolled in college. The point estimates in Table 4 again exhibit sensitivity to the specification but are never statistically significant and are not consistently positive or negative. Given that we find that the effect on flagship enrollment declines over time, an important question is whether this is being driven by students admitted under the TTP dropping out at a higher rate than other students. For instance, it may be that students admitted because of TTP are not able to do well in the rigorous academic environment of the flagship universities. 20 To investigate this possibility, we examined whether being in the TTP affected the likelihood of transferring to a more (or less) selective institution (measured by the Barron’s competitive admissions ratings) and the probability of dropping out of college. The evidence in Figure 8 provides no indication of any effects on these outcomes, which is consistent with the point estimates in Table 4. 7.4 Subgroup analyses Table 5 shows estimates by subgroups. The first panel shows results stratified by race, with whites and Asians compared to underrepresented minority students (i.e., Hispanics, blacks, and Native Americans). The estimated effects on flagship enrollment are more 0 1 2 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 0.1 .2 .3 .4 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 0.1 .2 .3 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 0.1 .2 .3 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 Figure 5 Flagship enrollment through year 4 after HS graduation by end of 11 th grade class rank. Daugherty et al. IZA Journal of Labor Economics Page 19 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 precisely estimated for underrepresented minority students, but the magnitude of the estimates is similar for both groups. This pattern also holds for the other outcomes, although the negative effect on private or out-of-state enrollment is not robust to changing the bandwidth for underrepresented minority students. Next, we examine effects stratified by whether students attend a high school with a relatively high (for the district) college-sending rate. The results strongly suggest that the effects of eligibility for automatic admission are concentrated in the higher collegesending schools. The estimated effects on flagship enrollment range from 8.5 percentage Table 4 Effect of being in top 10% at end of 11th grade on persistence outcomes Enrolled Flagship in: Year 2 0.026 0.022 0.031 0.025 0.081 0.057 (0.020) (0.019) (0.028) (0.026) (0.043) (0.038) Year 3 0.036 0.031 0.039 0.033 0.082 0.060 (0.020) (0.019) (0.027) (0.025) (0.043) (0.038) Year 4 0.020 0.015 0.029 0.021 0.074 0.057 (0.019) (0.018) (0.027) (0.025) (0.043) (0.039) Enrolled Priv/OOS in: Year 2 -0.061** -0.058** -0.032 -0.032 -0.051 -0.051 (0.023) (0.022) (0.032) (0.030) (0.054) (0.049) Year 3 -0.043 -0.040 -0.009 -0.008 -0.044 -0.043 (0.023) (0.021) (0.031) (0.030) (0.051) (0.046) Year 4 -0.044 -0.040 -0.022 -0.022 -0.023 -0.022 (0.022) (0.021) (0.031) (0.029) (0.051) (0.046) Enrolled in 4-Yr: Year 2 -0.044 -0.040 -0.008 -0.012 -0.011 -0.029 (0.030) (0.025) (0.042) (0.035) (0.064) (0.052) Year 3 -0.023 -0.020 0.024 0.021 0.030 0.013 (0.030) (0.026) (0.042) (0.036) (0.063) (0.051) Year 4 -0.021 -0.017 0.025 0.019 0.069 0.050 (0.030) (0.026) (0.042) (0.036) (0.067) (0.056) Transfer to More Selective School -0.003 -0.001 -0.032 -0.029 0.006 0.007 (0.016) (0.016) (0.022) (0.022) (0.038) (0.033) Transfer to Less Selective School 0.006 0.008 0.013 0.010 -0.033 -0.034 (0.018) (0.018) (0.026) (0.026) (0.038) (0.037) Dropped Out (=0 if did not enroll) -0.003 -0.002 0.002 -0.001 -0.049 -0.057 (0.016) (0.016) (0.023) (0.023) (0.038) (0.037) Total Semesters: Flagship 0.279 0.245 0.349 0.293 0.589* 0.459* (0.150) (0.141) (0.208) (0.193) (0.261) (0.233) Total Semester: 4 Year -0.188 -0.161 0.242 0.202 0.229 0.130 (0.209) (0.174) (0.295) (0.240) (0.344) (0.267) Total Semesters: Highly or Most Competitive -0.104 -0.119 0.249 0.178 0.338 0.217 (0.196) (0.165) (0.276) (0.229) (0.320) (0.259) Total Semesters: Private or Out-of-State -0.396* -0.376* -0.122 -0.125 -0.211 -0.232 (0.165) (0.155) (0.227) (0.214) (0.303) (0.272) Bandwidth 10 10 5 5 IK opt bw IK opt bw Baseline covariates? No Yes No Yes No Yes Note: Enrollment outcomes refer to enrollment in the indicated year following high school graduation. Total semesters enrolled refers to enrollment over the four years following high school graduation. The specifications are the same as in Table 3. N for bw=10 is 4,197; bw=5 is 2110. Robust standard errors in parentheses. **, * denote statistically significant estimate at the 1% and 5% level, respectively. Daugherty et al. IZA Journal of Labor Economics Page 20 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 points to 15.7 percentage points in the IK specification. The estimates for the total semesters of flagship enrollment are only statistically significant when the bandwidth is 10, but the point estimates are similar in magnitude for the other specifications. In contrast, we do not find any evidence of impacts on flagship enrollment for students from low college-sending high schools, and the estimated effects on total semesters in a flagship are not statistically significant and change signs across specifications. Despite the lack of effects on flagship enrollment, we find negative effects on private or out-of-state enrollment and total semesters in a four-year college. However, these estimates are only statistically significant with a bandwidth of 10 and are much 0 1 2 3 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 .1 .2 .3 .4 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 .1 .2 .3 .4 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 .1 .2 .3 .4 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 Figure 6 Private or out-of-state enrollment through year 4 after HS graduation by end of 11 th grade class rank. 2 3 4 56 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 .2 .3 .4 .5 .6 .7 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 .2 .3 .4 .5 .6 .7 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 .2 .3 .4 .5 .6 .7 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 Figure 7 4-year college enrollment through year 4 after HS graduation by end of 11 th grade class rank. Daugherty et al. IZA Journal of Labor Economics Page 21 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 smaller in magnitude and not statistically significant in the other two specifications. When interpreting these results, it is important to remember that even the higher college-sending high schools in the district have relatively low college-sending rates relative to the state as a whole. Nonetheless, these results indicate that eligibility for automatic admissions may have little effect on college enrollment and choice for the most disadvantaged urban high schools. The third panel shows results by socioeconomic status. The estimated effects on flagship enrollment are larger in magnitude for students who are not economically disadvantaged and range from 9-13 percentage points. In contrast, they are only statistically significant for the economically disadvantaged sample in the IK specification. The estimates for total semesters of flagship enrollment also are larger for the not economically disadvantaged sample, and the estimates for the economically disadvantaged sample are small in magnitude and statistically insignificant. For both groups, we find negative and statistically significant effects on private or out-of-state enrollment in the bandwidth of 10 specification, and for the non-economically disadvantaged sample the estimate for the IK specification is even larger in magnitude (although not statistically significant). Finally, we examined results by gender. The estimated effects on flagship enrollment are larger for women than they are for men. However, the difference in magnitudes is not especially large, and the estimated effects on total semesters on flagship enrollment are larger for men in some specifications than for women. Overall, these results do not offer clear evidence of differential effects by gender. 8. Discussion and conclusion Attending a selective college is widely seen as providing students, particularly those from disadvantaged backgrounds, access to opportunities that are otherwise difficult to obtain. Consistent with this view, evidence from recent studies suggests that selective college attendance has sizable returns that are especially large for lower-income and minority students. Nonetheless, there remain large gaps in selective college attendance 0.05 .1 .15 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 0.05 .1 .15 .2 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 0.05 .1 .15 0 5 10 15 20 25 Class Rank (0=Highest) BW=10 BW=5 Figure 8 Transfer and dropout by end of 11 th grade class rank. Daugherty et al. IZA Journal of Labor Economics Page 22 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 and completion by race and socioeconomic status. Policies to address these gaps are therefore an important social and economic issue. The Texas Top Ten Percent Plan is an example of such a policy. In this paper, we estimated the effects of eligibility for the automatic admissions guarantee this policy provides to students in the top decile of their high school class. We did so using administrative data from a large urban school district in Texas linked to college Table 5 Effects of being in top 10% at end of 11th grade by subgroups Race Whites & Asians Under-represented minorities Enrolled in Flagship 0.054 0.048 0.113 0.053** 0.085** 0.075** (0.037) (0.054) (0.084) (0.020) (0.025) (0.027) Enrolled Private or Out-of-State -0.083* -0.045 -0.135 -0.076** -0.010 0.001 (0.038) (0.054) (0.094) (0.027) (0.038) (0.059) Total Semesters: Flagship 0.255 0.209 0.560 0.260* 0.429** 0.371* (0.269) (0.393) (0.439) (0.129) (0.166) (0.156) Total Semester: 4 Year -0.138 0.157 0.256 -0.209 0.367 0.119 (0.283) (0.414) (0.458) (0.219) (0.286) (0.319) HS College-Sending Level High college sending Low college sending Enrolled in Flagship 0.085** 0.095* 0.157* -0.004 0.013 -0.002 (0.030) (0.042) (0.062) (0.019) (0.025) (0.025) Enrolled Private or Out-of-State -0.065* -0.039 -0.091 -0.099** -0.022 -0.039 (0.032) (0.045) (0.071) (0.028) (0.041) (0.057) Total Semesters: Flagship 0.423* 0.428 0.589 -0.070 0.047 0.118 (0.216) (0.303) (0.339) (0.118) (0.132) (0.107) Total Semester: 4 Year 0.058 0.402 0.384 -0.503* -0.085 -0.222 (0.242) (0.344) (0.376) (0.230) (0.300) (0.359) SES Not econ. Disadvantaged Economically disadvantaged Enrolled in Flagship 0.089** 0.097* 0.128 0.006 0.025 0.056* (0.032) (0.046) (0.065) (0.020) (0.024) (0.028) Enrolled Private or Out-of-State -0.075* -0.031 -0.101 -0.076** -0.026 -0.015 (0.035) (0.048) (0.079) (0.028) (0.040) (0.053) Total Semesters: Flagship 0.472* 0.428 0.584 -0.041 0.139 0.111 (0.233) (0.329) (0.374) (0.127) (0.144) (0.138) Total Semester: 4 Year -0.004 0.266 0.127 -0.342 0.195 0.014 (0.256) (0.363) (0.400) (0.233) (0.307) (0.323) Gender Women Men Enrolled in Flagship 0.061** 0.083** 0.127** 0.042 0.038 0.107 (0.023) (0.031) (0.044) (0.038) (0.055) (0.073) Enrolled Private or Out-of-State -0.094** -0.029 -0.085 -0.056 -0.028 -0.163 (0.027) (0.038) (0.056) (0.040) (0.057) (0.091) Total Semesters: Flagship 0.191 0.287 0.595* 0.393 0.363 0.458 (0.161) (0.217) (0.260) (0.271) (0.389) (0.386) Total Semester: 4 Year -0.303 0.046 -0.027 0.086 0.559 0.491 (0.215) (0.294) (0.325) (0.299) (0.439) (0.460) Bandwidth 10 5 IK opt bw 10 5 IK opt bw Note: See notes to Tables 4and 5. All estimates are from models that include controls for baseline covariates. Robust standard errors in parentheses. **, * denote statistically significant estimate at the 1% and 5% level, respectively. Daugherty et al. IZA Journal of Labor Economics Page 23 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10 enrollment records and a regression discontinuity approach that compared the outcomes of students who just made the cutoff for the TTP and those who just missed it. The analysis discussed in this paper generated several notable findings that have important policy implications. First, our results demonstrate that eligibility for automatic admissions does appear to increase enrollment at flagship universities for students in an urban school district that sends relatively few students to college who are at the margin of being in the top decile of their class. In particular, we also find these effects for underrepresented minority students. These findings are noteworthy because a key goal of the TTP Plan, and percent plans in general, is to increase access to the top public universities for traditionally under-served populations. Despite these effects on flagship enrollment, we find little evidence that the automatic admissions guarantee leads to increases in the “quality”of the colleges students attend. Instead, the increases in flagship enrollment appear to “crowd out”enrollment at comparably ranked private or out-of-state institutions. Thus, offering eligibility for automatic admissions does not appear to increase access to selective colleges in general even if it increases access to the best public universities in Texas. We also find that the effects on flagship enrollment are concentrated in the most advantaged schools in the district. Indeed, when we estimated effects by the fraction of students at a high school that attend college, we found no evidence of effects on college choice in the schools with the lowest college-sending rates. While the college-sending rates of the highest performing high schools are not very high relative to Texas as a whole, our findings suggest that offering eligibility for automatic admissions may not be effective at accomplishing even the narrow goal of increasing access to the top public universities for students in the most disadvantaged settings. These results complement other research showing that ending affirmative action decreased enrollment of underrepresented minority students at selective colleges and descriptive analyses showing that percent plans do not fully offset these reductions. While we do not examine the effect of introducing the TTP Plan, we find little evidence that the centerpiece of this policy –the automatic admissions guarantee –increases enrollment at competitive universities or that it increases access to public flagship universities for students at the most disadvantaged schools. These results are consistent with the claim that percent plans are unlikely to have large aggregate impacts on selective college enrollment among students from backgrounds that send relatively few students to these schools. There are also some important questions that our paper does not address. Since we use data from an urban school district, our findings may not reflect how automatic admissions guarantees affect students in rural or suburban schools. While the effects on students in urban school districts have clear importance for policy, the effects in other settings do as well. The effects in rural areas are important because a goal of the TTP Plan is to increase geographic diversity at the state’s public universities. Likewise, the effects in suburban areas have significance because many of the criticisms about the policy’s fairness stem from concerns that students in suburban districts are at a disadvantage because it is harder to get into the top decile. Nonetheless, it is plausible that the lack of effects on college selectivity that we find may hold in higher-income districts as well since students in these districts who are at the margin for being in the top decile of their class likely face smaller Daugherty et al. IZA Journal of Labor Economics Page 24 of 29 2014, 3:10 http://www.izajole.com/content/3/1/10