SAT Dataset Data and Estimation Strategy

her freshman year of college; have graduated from high school or received a GED on or after January 1, 1998; have begun an undergraduate course of study within three years of high school graduation or receipt of the GED extended for periods of mili- tary or Peace Corps service; be enrolled at least half-time in a course of study leading to an accredited degree; and not already hold a bachelor’s degree. In 2002 the eligibil- ity requirements were liberalized to allow older high school graduates to benefit from the program. The DCTAG program is not means tested; indeed, those who apply are not required to provide any information about their family’s income or assets. Benefits may be applied only to tuition and fees, and benefit grants are to be made directly to the eligible institution, which must agree to comply with the program’s administrative requirements. In the DCTAG program’s first year, 1,900 individuals used the grant to attend 152 participating institutions U.S. General Accounting Office 2002, and the number of students receiving awards had doubled by the program’s fourth year Williams 2004. The number of students taking advantage of the program is large rel- ative both to the number of District residents of college age and to the number of District residents who graduate from high school: The Census Bureau estimates that, as of July 1, 2000, there were 5,765 17-year-olds resident in the District. In that same year, 2,695 stu- dents graduated from the District’s public high schools. Comparable data on the number of District residents graduating from private high schools are not available.

III. Data and Estimation Strategy

Previous research on the effects of variation in college costs typically has focused on how costs affect enrollments. Enrolling in college, however, is the end result of a sequence of steps. Students must identify a set of colleges they might like to attend; apply to some or all of those schools; be accepted for admission at one or more schools; and finally choose a particular school at which to enroll. Because we have access both to data on applications as proxied by sending SAT scores to a school and to data on enrollments, we are able to explore more fully than has previous research the effects of changes in costs on students’ decisions about college attendance.

A. SAT Dataset

To examine the college application decisions of individual students affected by the DCTAG program, we use data from the College Board on high school students who took the SAT. These data cover the 1994 through 2001 cohorts, where a cohort is defined as the group of students who on a normal schedule would graduate from high school in the indicated year. The database includes all black and Hispanic test takers, all test takers residing in Washington, D.C., and a 25 percent random sample of non- black, non-Hispanic test takers in other states. Most of the variables included in the database were derived from information supplied by the test takers on the Student Descriptive Questionnaire. Variables that we use in our analysis include colleges to which test scores were sent four-digit code, up to 20 recorded in 1994–98 and up to 30 recorded in 1999–2001, state of residence, high school attended, type of high school attended public, private, or other, raceethnicity, highest level of education attained by mother and by father, and SAT math and verbal scores. The Journal of Human Resources 582 We merge the data on each college or university to which students sent their scores to information on various characteristics of this institution provided by the College Board—whether it is a two- or four-year institution, whether it is public or private, and whether it is foreign or domestic. We also merge the data to measures of the selec- tivity of each institution. We define an institution as “top-tier” if it is ranked among the top-50 national universities or top tier of national liberal arts colleges according to the U.S. News and World Report 1998 College Rankings U.S. News and World Report 1997. 3 We treat a student’s decision to have SAT scores sent to a particular college or uni- versity as a proxy for the decision to apply to that institution. We expect that most four-year-college-bound high school seniors in D.C. and the other cities in our analy- sis will take the SAT and will send their SAT scores to every four-year college or uni- versity to which they apply. Sending SAT scores to a particular college is not a perfect proxy for actually applying to that college—some students may send their scores to colleges to which they decide ultimately not to apply or they may apply to colleges to which they do not send SAT scores for instance, if the college does not require SAT scores for admission. As documented by Clark 2003, the vast majority of college- bound high school students in the District of Columbia and the other states repre- sented in our analysis take the SAT as opposed to the ACT. We are not aware of general information about the probability that a student who sends his or her SAT scores to a school will complete an application to the school, but Card and Krueger 2005 report that the number of students of a given raceethnicity who send their SAT scores to particular public universities in the state of California is a good proxy for the number of students of a given raceethnicity who actually apply to those same cam- puses. As most community colleges and other two-year colleges do not require SAT scores for admission, we limit our analysis of the SAT data to students applying to four-year colleges and universities. 4 We adopt a difference-in-differences approach to study District students’ decisions about where to send their SAT scores, using students from large cities in nearby states Baltimore, Maryland; Newark, New Jersey; Norfolk, Virginia; and Philadelphia, Pennsylvania as a comparison group. To select these cities, we first identified nearby states with SAT-taking rates similar to those of students in the District of Columbia. The District has a population of 572,059 that is 60 percent African-American, and for our comparison group we identified cities in these states with populations of 200,000 or more that were at least 40 percent African-American, based on 2000 Census data 3. As an alternative measure, we define an institution as being in the top-tier if it is located in the United States; at least 500 students sent SAT scores to the institution in the year 2000; and the average SAT score of these applicants, as computed from the SAT database, exceeded the 75th percentile of scores in that year 1,160. As expected, there is considerable overlap between the lists of schools defined as top-tier institu- tions according to the two measures, and the two measures have a correlation of 0.74. Results are not qualitatively sensitive to which measure of selectivity we use, and we present only those using the selectiv- ity measure based on the U.S. News and World Report rankings. 4. An important consequence is that we are not able to examine DCTAG’s effects on students’ applications to two-year colleges. If the program induced students who otherwise would not have applied to college at all to apply to an eligible two-year college, that would not be reflected in this analysis. It would, however, be captured in our analysis of students’ enrollment decisions. We also note that the University of the District of Columbia does not require applicants to submit SAT scores. Abraham and Clark 583 as reported in U.S. Census Bureau 2001. 5 None of the states in which the compari- son cities are located instituted a college assistance program on anything like the scale of DCTAG over the period of interest. As the SAT dataset has information on stu- dents’ state of residence and high school which can be matched to school address information, we identify SAT-takers by city based on their state of residence and their attendance at a high school located in the relevant city. The D.C. figures include D.C. residents who attend suburban high schools located in Maryland and Virginia. Throughout our analysis, we exclude students without a valid SAT score and students who did not send their score to at least one four-year college or university. For consis- tency across years, we also exclude students who did not report an ethnicity these stu- dents were missing from the 1994–98 samples provided by the College Board. Because we use high school address to match students to cities, students not reporting a valid high school code or students reporting a high school code that could not be matched to an address are excluded as well. Our results are not sensitive to any of these exclusions. Appendix Table A1 reports the means of selected characteristics for the SAT sample, both overall and then separately for D.C. versus the comparison cities. We begin our analysis by examining trends in the number of students who send their SAT scores to at least one college or university, comparing the trend in the District of Columbia to that for the comparison cities. If the DCTAG program has had a positive effect on college applications, we would expect the share of D.C. students who send their SAT scores to at least one college to have grown relative to the share in the comparison cities. Among those sending SAT scores to at least one school, we use the following model to identify the effect of the DCTAG program on where students choose to send their scores: 1 . Y X DC ijt o ijt t t t j j j t t t ijt 1 1995 2001 1995 2001 = + + + + + c c a d n z a i y = = In this model, Y ijt is the outcome variable of interest for student i in city j graduating from high school in year t; X ijt is a vector of individual-level characteristics; α t repre- sents a set of year dummies, described below; µ j represents a set of city dummy vari- ables; DC is a dummy variable equal to one if the city is D.C., zero otherwise, and υ ijt is a stochastic error term. Outcomes examined using this model include whether student i graduating from high school in year t sent SAT scores to any DCTAG-eligible public Maryland or Virginia institution, whether the student sent scores to any DCTAG-eligible public institution anywhere in the United States, and various measures of the selectivity of the schools to which students sent scores. The X ijt vector includes individual-level characteristics that are likely to be correlated with college application decisions, including race, parental education, and type of high school public, private, or other that the student attends. The α t year dummy variables are defined in the following way: α 1995 equals one if the year in which the student is expected to graduate from high school is 1995 or later 5. We also repeated the analysis using all test-takers from Virginia, Maryland, Connecticut, Delaware, New Jersey, North Carolina, and Pennsylvania as a comparison group. The results of this exercise are nearly iden- tical to those we report. The Journal of Human Resources 584 and zero otherwise; α 1996 equals one if the student is expected to graduate in 1996 or later and zero otherwise; and so forth, for each year from 1995 through 2001. Defined in this way, the coefficients on the year dummies represent the average change in the dependent variable conditional on the covariates X ijt in the comparison cities from year t-1 to year t. The coefficient on the interaction between each α t and the D.C. dummy variable represents the change in the dependent variable from year t-1 to year t in D.C. relative to the comparison cities. For example, in a model of applications to DCTAG-eligible colleges, if DCTAG had a positive impact on the likelihood that a student applied to an eligible college, we would expect to see this reflected in posi- tive, significant coefficients on α 2000 DC andor α 2001 DC that is, in positive, sig- nificant values of θ 2000 and θ 2001 . These two coefficients represent the change in the probability that a D.C. student applied to an eligible college between 1999 and 2000 and between 2000 and 2001, over and above any change in the probabilities for stu- dents in the comparison cities. The sum of the two coefficients θ 2000 plus θ 2001 rep- resents the total change in the probability of a D.C. high school graduate applying to an eligible college between 1999 and 2001.

B. IPEDS Dataset