3 9080 02618 0957 i 2 DEWEV HB31 .M415 Massachusetts Institute of Technology Department of Economics Working Paper Series The Causes and Consequences of Attending Historically Black Colleges and Universities Roland G. Fryer, Jr. Michael Greenstone Working Paper 07-1 April 9, 2007 Room E52-251 50 Memorial Drive Cambridge, MA 021 42 This paper can be downloaded without charge from the Network Paper Collection at Social Science Research http://ssrn.com/Gbstroct=979336 I The Causes and Consequences of Attending Historically Black Colleges and Universities* Roland G. Fryer, Jr. Harvard University and NBER Michael Greenstone MIT and NBER April 2007 We David Card, David Bryan Graham, Chang-Tai Hsieh, Lawrence Katz, Henry Louis Hoxby, Glenn Loury, Enrico Moretti, Andrei Shleifer, Lawrence Summers, colleagues at the Mellon Foundation, and participants in numerous seminars. Fryer is especially thankful to the Michor family in Kritzendorf, Austria for their support and generous hospitality while working on the paper. This paper makes use of the College and Beyond (C&B) database. The C&B database is a restricted-use database. Researchers who are interested in using the database may apply to the Andrew W. Mellon foundation for access. Sam Schulhofer-Wohl, Sheldon Bond, Jorg Spenkuch, Elizabeth Greenwood, and Paul Torelli provided exceptional research assistance. All errors are our own. Correspondence can be addressed to Fryer at the Department of Economics, Harvard University, 1805 Cambridge Street, Cambridge MA, 02138, (e-mail: rfryer@fas.harvard.edu); or Greenstone at Department of Economics, Massachusetts Institute of Technology, 50 Memorial Drive, E52-359, Cambridge MA, 02142 (e-mail: mgreenst@mit.edu) Gates, are grateful to Jr., Edward Cutler, Glaeser, Lani Gunier, Caroline Digitized by the Internet Archive in 2011 with funding from Boston Library Consortium IVIember Libraries http://www.archive.org/details/causesconsequencOOfrye The Causes and Consequences of Attending Historically Black Colleges and Universities ABSTRACT (HBCUs) were practically the only Using nationally representative data Until the 1960s, Historically Black Colleges and Universities institutions files of higher learning open to Blacks in the irom 1970s and 1990s college attendees, we find US. that in the 1970s HBCU matriculation was associated with higher wages and an increased probability of graduation, relative to attending a Traditionally White Institution (TWI). By the 1990s, however, there is a wage penalty, resulting wages of HBCU graduates between the two decades. We also analyze the College and Beyond's 1976 and 1989 samples of matriculates which allows us to focus on two of the most elite HBCUs. Between the 1970s and 1990s, HBCU students report statistically significant declines in the proportion that would choose the same college again, preparation for getting along with other racial groups, and development of leadership skills, relative to black students in TWIs. On the positive side, HBCU attendees became relatively more likely to be engaged in social, political, and philanthropic activities. The data provide modest support for the possibility that HBCUs' relative decline in wages is partially due to improvements in TWIs' effectiveness at educating blacks. The data contradict a number of other in a 20% decline in the relative pre-coUege credentials and expenditures per student at HBCUs. intuitive explanations, including relative decline in students attending HBCUs Roland G. Fryer, Jr. Haiward University Department of Economics 1805 Cambridge Street Cambridge, 02138 scores) of MIT ' Department of Economics 50 Memorial Drive, E52-359 Cambridge, 02142 MA MA ^^ andNBER email: rfryer@fas.harvard.edu SAT Michael Greenstone . ' (e.g., " • : , , andNBER ' email: mgreenst@mit.edu Introduction (HBCUs) have Historically Black Colleges and Universities and storied a proud role in the education and progress of Blacks in America. For nearly a century, Historically Black Colleges and Universities (HBCUs) were 20% Today, roughly of practically the only institutions of higher learning 22% of cuirent bachelor's degrees granted (Wilberforce), Ralph Ellison (Tuskegee), Martin Luther King, Simmons HBCU Among alumni. HBCUs, 50% state, and Blacks, 40%) of 80% of lawyers, and HBCUs' all to a lesser degree, local While numbers are smaller of famous at non- for private HBCUs, HBCUs public support still In the years 1999-2001, annual total public support of total revenues. et al. 2004). HBCUs are at a crossroads today. In U.S. U.S. 717 (1992)), the Supreme Court instructed state legislatures to find "educational justification" for the existence of mission. list governments. Between 1977 and 2001, 61%) to 73%o of public averaged roughly $2.65 billion (2005$) (Provasnik Fordice (Morehouse), Thurgood Marshall State) headline a long HBCU graduates.' Despite their past successes and historical importance, V. and these W.E.B. Dubois congressmen, 12.5%) of CEOs, 50%) of professors of judges are fiinds. accounts for nearly a third of ("505 HBCUs to Blacks. successes are in no small part due to their substantial financial support from federal, revenues came from public HBCUs Jr. and Oprah Winfrey (Tennessee (Dillard), Blacks in the US. to college going Blacks chooses to attend one of the 103 all institutions are responsible for (Lincoln), Ruth open In response, some HBCUs HBCUs or integrate them; the latter would completely alter their experienced declines in enrollment, others have pursued dramatic increases in the fraction of non-Black students, and a number of HBCUs have seen important declines in their financial positions.^ The Supreme Court's call for an "educational justification" is surely related to the absence of convincing evidence on the causes and consequences of attending an proponents claim that they provide an discrimination and racism. by engendering It is also argued that a strong sense of environment that idyllic learning communal HBCUs HBCU is for Blacks. help to build important social capital for Blacks responsibility and civic consciousness networking opportunities for high achieving Blacks (Drewry and Doennarm 2001). are correct, then HBCUs HBCU from the pressures of free offer unique opportunities for educafional and social and providing If these arguments development of Black students and the argument for remaining segregated seems justified. The data sources are Congressional Black Caucus (congressmen), Black Enterprise (CEOs), US Department of in enrollment and Education, Office of Civil Rights (professors), and Ehrenberg (1996) (Lawyers and Judges). The ruling had an adverse effect Mississippi Valley State a 20.1% on many HBCUs. Alcorn decrease. Other HBCUs, increases in integration. Elizabeth City State increased from State experienced a 9.9% decrease especially in North Carolina, have 1 1% white from in 1980 to 23. 7% 11. 9 to 22.2, North Carolina Central from 4.1 to 13.4, and Winston Salem from such as Florida, the ruling has been largely ignored; Florida remains 90% black. A&M 1 shown substantial in 1998, Fayetteville State 1.3 to 18.0. In other states, On (TWIs) in the other hand, it HBCUs possible that is preparing Blacks for post-college life. are inferior to Traditionally White Institutions If students are taking less challenging courses fi-om less distinguished faculty, have access to poor resources, or are not investing in the social skills necessary to of people, then graduates will perform poorly interact with diverse sets inferior non-labor market outcomes. In in the labor market and have HBCUs scenario, the case for supporting this with public resources appears weak. This paper empirically assesses the causes and consequences of future decisions HBCU attendance so that the by governments, students, and parents are based on evidence, rather than theories and We historical anecdotes. analyze thi^ee large data sets with adequate pre and post college infonnation for Blacks that identify the students' choice of college and whether it an is HBCU. The data sets are: the National Longitudinal Sui^vey of the High School Class of 1972 (NLS-72), Baccalaureate and (B&B), and the College and Beyond database (C&B). The nationally representative sample of dataset contains four look into the most HBCU students at two first two points HBCUs, Howard, Morehouse, Spelman, and elite HBCUs in Beyond datasets provide a snapshot of a in time: 1972 and 1992. The third Xavier, allowing us to take a focused 1976 and 1989. Although there are important limitations with each of these data sets, together they provide a rich portrait of the causes and labor market and non-labor market consequences of HBCU attendance. Importantly, these data and 1990s so it is possible to assess how sets sample college matriculates in the 1970s these causes and consequences changed during these two decades of dramatic social change. The between analysis uses HBCU and TWI four separate statistical approaches to adjust for pre-college differences attendees. We begin by using the rich and high school academic achievement (including the same pre-college SAT set scores) to fit of covariates on family backgroimd least squares models. then use covariates to implement a propensity-score matching estimator to assess the robustness of the results to functional form assumptions about the observables. These approaches are supplemented by methods that are designed to account for selection bias due observations We (Heckman 1979) and bias to missing outcome which emerges when colleges admit students based characteristics unobsei"ved in our data that are positively correlated with future partly on outcomes (Dale and Ki-ueger 2002). The results are robust across these four methods. However, lacking a randomized experiment or may credible quasi-experiment, thorny issues of selection remain. Consequently, we urge caution in interpreting the results as causal. Together the nationally representative HBCU attendance. In the 1970s, HBCU NLS and matriculation probability of graduation, relative to attending a B&B reveal an important change in the returns to was associated with higher wages and an increased TWI. By the 1990s, however, there is a substantial wage Overall, there penalty. is 20% a Interestingly, relative pre-college wages of decline in the relative HBCU graduates in just two decades. measures of student quality (SAT scores, HBCU improved among e.g.) attendees during this period, so higher achieving students were increasingly choosing these schools at the same time that the returns for attending these schools The underlying source of the decline in were falling behind. HBCU perfoiTnance is unlikely reasons, given the high court's stance. Nevertheless, understanding researchers and educational practitioners. support, the possibility that HBCUs' effectiveness at educating blacks. the decline in increased outcomes among more The at C&B HBCUs than at The data fail to decline relative is it attendees TWIs between — for policy interest to and contradict, to in at least one specification due to improvements in TWIs' partially In contrast, the data contradict a HBCU be important would be of considerable number of intuitive explanations for for example, educational expenditures per student the 1970s and 1990s. provides a rare opportunity to assess the most elite evidence of a wage decline between the 1976 and 1989 cohorts but suggestive because these estimates are imprecise. There is colleges. Here too, there is should only be considered it HBCU stronger evidence that the later matriculates were less satisfied with their choice of college and self-reported developing fewer leadership and social skills that are valuable in post-college later cohort was significantly The paper proceeds more Hkely to Section as follows. relative to life, TWI students. On the other hand, the be involved in political, social, and philanthropic role in the education of blacks in America. II activities. provides a brief history of HBCUS and their important Section reviews some theoretical explanations for III blacks might benefit (or be haniied) by attending a historically black college or university. presents the data and of HBCU summary statistics. Sections Section IV V and VI report results on the causes and consequences Section VII summarizes the differences between the results from the attendance. 1 970s and 1990s and assesses alternative explanations for these differences. Lastly, Section VIII concludes. : appendix describes the details of our sample constmction. II. A why A data .• Brief History of Historically Black Colleges and Universities A. Ante-Bellum Period The 1860 Census counted 4.4 million Black people in the United States, most of the Southern states and were held as slaves. Prior to the end of the Civil write was prohibited by law (or social custom) in colleges founded before the Civil War: University) was founded Wilberforce College in in Pennsylvania Ohio, 1856. students. Fonrial education for many the Institute in 1837; areas of the South. for Colored whom War, teaching slaves Still, there were three Black Youth (now known Lincoln College in lived in to read or Pennsylvania, as Cheney 1854, and All of these universities served secondary and post-secondai'y most blacks would not become available until after the Civil War, when the Freedmen's Bureau, black communities and their churches, and private philanthropists organized schools for Blacks (Donohue, Heckman, and Todd 2002). B. Post-Civil War and the Second Morrill Land Grant During the period immediately following the Civil War, there was a dramatic increase number of educational in the geared toward blacks, funded primarily through groups like the institutions American Missionary Association, the Freedmen's Bureaus, and southern state governments, especially during the Reconstruction period. Between 1865 and 1890, over two hundred private black institutions were founded in Very few of these early the south. American Missionary Association, the awarded bachelor's degrees. The institutions Freedmen's Bureaus, and other groups that were active education of freed blacks played a large role in establishing some standard of education literacy — that would be important when degree granting institutions for in the early —most notably Blacks opened en masse in the 1890s. Most public next decade, 16 HBCUs HBCUs trace their history to the opened their doors. second Moixill Act, passed The Morrill Act allowed of land grant universities, with southern and border states creating Florida well-known agricultural, mechanical, and technical August 1890. In the for the creation of a two-tier system HBCUs HBCUs funds to develop white land grant colleges. These federal" training; in principally to gain access to were largely limited institutions to vocational such as North Carolina A&T and A&M were founded during this period. By 1895, public education, as HBCUs was offered During the Jim Crow era at had awarded 1,100 college diplomas many in the to black students. Yet, liberal arts public white institutions, remained unavailable to black students. south that followed Reconstmction, educational opportunities for white students expanded and blacks were almost completely excluded from white institutions. 1896 decision, Plessy In the v. Ferguson (163 U.S. 537 (1896)), the two-tier system of higher education, based on the incentive structure in the Second Morrill Act, result, HBCUs schools. The demand chain students into began to become World War finnly set. As a rapid expansion of black high schools in southern urban areas set in motion a supplyin which HBCUs availability to qualify II HBCUs, of teaching positions, supported by state treasuries, drew more black themselves for teaching positions (Roebuck and Murty 1993). There became an interdependence between C. became more institutions that primarily trained teachers to teach in segregated public the public school system and HBCUs. and the Higher Education Act of 1965 as well as other institutions of higher learning, faced a funding crisis in the 1940s due to the budget cuts in educational fimding associated with WWII. In 1944, the United Negro College Fund was established, raising $765,000 for been raised by the individual colleges The landmark decision developed to 1965, which implement Brown in to much campaign; three times as in its first fixnding as had previous three years. Board of Education (349 U.S. 294 (1955)) and v. improved the plight of many HBCUs. was devoted HBCUs. As refen-ing to it HBCUs in the Title III the legislation of the Higher Education Act of "Strengthening Developing Institutions" was interpreted as primarily a result many HBCUs benefited greatly from the federal funds provided under Title III, funds that could be used for faculty and student exchanges, faculty improvement programs, cumculum improvements, student services, visiting scholars programs, and administrative improvements. HBCUs Despite the material gains to the arising under the Higher Education Act, the NAACP continued a legal strategy of attacking the two-tier educational system, asking the U.S. Department of Health, Education, and Welfare to enforce the Civil Rights Act of 1 964 and prohibit southern operating a segregated higher education system. In the 1973 case of Adams 92 ( D.D.C.), modified and aff d, 480 F.2d 1 159 (D.C. Cir. 1973)), the v. NAACP from states Richardson (356 won, requiring F. Supp. that states develop desegregation strategies that would allow for a better racial mix of students, faculty, and staff public colleges, and increase the access and retention of minorities at ruling was primarily aimed at non-HBCUs, however, and mandate should not be accomplished at the improvement of black colleges' was that this the only way HBCUs and facilities as HBCUs for that they levels of higher education. all made expense of or detriment The Adams decision increased funding not meet their mandates by closing the court it black colleges to traditionally because the decision stated that states could must instead include "yardsticks" HBCUs The clear that fulfilling the ruling's well as their academic programs. possible to ensure that in The to measure the court's reasoning would become desirable was institutions for white students. D. The Unintended Consequences of U.S. On litigant Fordice - June 26, 1992, the Supreme Court decided U.S. v. . •. The number and plaintiff, quality of the that this resulted represented by the NAACP, court ruled that it would be wasteful to , HBCUs r staff, and physical plant :,y . .%'/ ^ ^ facihties, arguing of higher education." maintain the two-tier system that had been erected during an had eight institutions, five white and three black, and of them (two white and two black) were within 25 miles of one another. The decision was a victoiy for civil rights lawyers, ordering Mississippi and 18 other southern states to do its :. and traditionally white racially segregated dual systems era of de jure segregation, noting that Mississippi that four -^ brought the suit because of the disparity in the academic programs, instmctional from the "historically operated ,;•::;.'':; Fordice, a decision brought by a black with the chief aim of removing structural differences between institutions. The v. HBCUs and traditionally white institutions. more to integrate However, the mling had an adverse effect on HBCUs to find "educational justification for the continued existence" consequences of seem mling this for the future of possible: 1) a decision that HBCUs HBCUs is because the court ordered state legislatures of the two parallel education systems. The unclear at this point, but at least three outcomes are indispensable for the education of Blacks and an increase in public funding; 2) increased recruitment and matriculation of white students, which has the potential to undermine the unique mission and culture of these institutions; and 3) a decision that they no longer are necessary (or as necessary) and a commensurate reduction in public financial support. The remainder of HBCUs, which the paper assesses empirically the causes Conceptual Framework III. There are institutions and consequences of attending will help determine their "educational justification." at least three theories for of higher education. First, why blacks would benefit from racial segregation of Tatum (1997) argues that racial grouping process in response to racism. This argument goes that segregation by race that allows individuals to gather support is is a developmental a positive coping strategy through shared experiences and mutual understanding. Second, Wilson's (1987) pioneering study of the South Side of Chicago argues that the migration of talented blacks from black neighborhoods had adverse phenomenon may exist for on the individuals effects segregation across schools - low ability blacks may left behind. A similar benefit from segregation through more intensive interactions with their high ability peers. Third, segregated social connections within schools may Torelli (2006) shows in also reduce adverse peer interactions resulting that racial differences in the social price for from interracial contact. Fryer with academic achievement are exacerbated environments with more inteixacial contact. There are also several theories for harm blacks. A social interactions (Case may racial segregation across colleges and universities may and Katz 1991, Cutler and Glaeser 1997), and network externalities (Borjas 1995, Lazear 1999), especially for youths. skill why well developed literature emphasizes the importance of peer groups (Coleman 1966), Many argue that these effects are important in the fonnation of and values and the development of human and social capital. Moreover, segi-egation across schools lead to the development of an "oppositional culture" and the enforcement of other negative behavioral nonns (Ogbu 1989). Additionally, segregation across schools can prevent positive spillovers between racially defined peer groups (Lazear 1999). A final disadvantage of the separation of racial groups across universities concerns the importance of interracial contact in mediating stereotypes and promoting understanding and tolerance. Interracial interaction generally leads to positive sentiment (Hpmans "bonding" and "bridging" capital (Granovetter 1973, Putnam 2000) 1950), and fosters the creation of impossible to identify the separate impact of each of these chamiels on segregated Blacks' It is well being with the available data net impact of HBCU sets. The attendance. Instead, this paper's goal is produce reliable estimates of the to resulting "reduced forni" estimates will likely reflect a number of the channels specified in this section. IV. Data Sources and We Summary Statistics analyze three large datasets: The National Longitudinal Study of the High School Class of 1972 (NLS-72), Baccalaureate and Beyond Longitudinal Study (B&B), and the College and Beyond (C&B) database. These data sets have been chosen because of the enormous amount of information each contains on pre-college academic performance, family background, college entry decisions, perfoiTnance while in college, and later life outcomes.^ Throughout the analysis, the rich set of pre-college and family background variables are used HBCU and non-HBCU and present summary We matriculates in equations for the other variables. statistics Before proceeding Fall of They discuss each of these sources from them. to this material. Appendix Table 1 provides some 4-year historically black colleges and universities in the United States. institutions. between as conditioning variables to adjust for obsei^vable differences summary statistics on the 89 Foity-nine of them are private are predominantly located in the South. Together their undergraduate enrollment in the 2005 was 238,911 and there were an additional 37,151 graduate students The fourteen enrolled. historically black 2-year colleges are not included in this table. A. The National Longitudinal Survey of the High School Class of 1972 The National Longitudinal Survey of High School Class of 1972 (NLS-72) the representative sample of 23,451 high school seniors in 1972. the Spring of 1972, and in a supplemental^ sample and follow-up surveys in 1973, 1974, 1976, 1979, the sample, with an average of A wide range of data 1 parent's education characteristics of each student also detailed records Two other data sets Longitudinal Study 2000 the questions posed here. The data include a nationally in a base year survey, and 1986. Roughly 1,300 high schools are included in gathered on the students is in 1973. is were selected 8 students per school in the study. There website (http://nces.ed.gov/surveys/nls72). environment, drawn Participants in the sample (i.e. and is described in detail in the study, as detailed information on occupation, socio-economic status, at the and the high school grades, college admission scores, and so on). from post-secondary transcripts, collected in NLS-72 each student's family pre-college There are 1984, and high school records. by the National Center for Education Statistics, the National Educational the Beginning Postsecondary Study (BPS), are equipped to answer some of Unfortunately, however, these data sets do not track individuals long enough after college collected (NELS) and completion to be useful for understanding later life outcomes. Important for our purposes, a six digit identification number was assigned to educational institutions by Committee on Education (FICE) the Federal Interagency distinguish postsecondary historically, to, schools that qualified as institutions of higher learning from those that did not. These codes are cracial in HBCUs defining B. and ensuring that this definition is consistent across data sets. Baccalaureate and Beyond The Baccalaureate and Beyond Longitudinal Study (B&B) a nationally representative sample is of 11,192 degree completers from 648 American colleges and universities in the 1992-1993 academic year. To identify a Aid Study random sample of degree completers, as a basis. B&B uses the National Postsecondary Student The National Postsecondary Student Aid Study is a large nationally representative sample of colleges and universities, students, and parents. A B&B the considerable amount of data website characteristics is gathered on the students in the study, as described in detail http://nces.ed.gov/sui"veys/b&b. of each student, information about It contains their parents information detailed 1 997 data sui-vey, and which takes place after more extensively, but we found about. just employment and entry For example, 6% 25% to in the 1997 sample. And, is Whenever this difference to restrict the was this attrition largely attrited had hoped sample in the results if from to in and the other data we cared most to TWIs. sets HBCUs graduation rates between this use the 2003 from the 2003 sample, compared from black students B&B we employ. The NLS B&B samples degree and non-HBCUs survey differ with our other data, we differ be clear about will A convenient way to handle this is can be accounted for by differences in samples. our other data sets to be of degree completers, which we do throughout. The College and Beyond Database The College and Beyond Database contains student time students the We begin with samples of students that enrolled their freshman year. The substantially. C. These follow-up surveys focus on the responses to the be of generally poor quality on the dimensions one important difference between the completers, which can introduce bias whether are in the workplace. of the original black respondents There C&B it We into graduate school. most students on pre-coUege and home environment, and financial aid infonnation. Follow-up sui"veys were administered in 1994, 1997, and 2003. include information on at fall who level administrative data on on 93,660 full- entered (but did not necessarily graduate from) thirty-four colleges and universities in of 1951, 1976, and 1989. These institutional records were linked to an extensive survey conducted by the Andrew W. Mellon Foundation between 1995 and 1997 and Entrance Examination Board and the Higher Education Research database: Xavier, Morehouse, Spellman, and Howard - the most to files Institute. elite provided by the College There are four HBCUs. The 1976 HBCUs in the cohort contains data on four black colleges; the all consists of black students 1 989 cohort only includes Morehouse and The Xavier.'' from 34 colleges and universities including the four elite final data set HBCUs; the sample consists of 2,125 students in 1976 and 1,785 in 1989. C&B The data are remarkably rich, containing infoiTnation Scholastic Aptitude Test transcripts, drawn from students' applications (SAT) and American College Test (ACT) and as well scores, as infonnation on family demographic and socioeconomic status. This information was attained by linking the institutional files of the thii1y-four colleges and universities with data provided Entrance Examination Board and the Higher Education Research includes the responses to a questionnaire administered to information on post-college labor market, all life satisfaction, Institute. Importantly, the Bowen and Bok D. Summaiy and other outcomes. The response is who (1998). statistics for the variables in our core specification are displayed in Table Table 1 consists of five sets of columns. first college An year college a student attends. then attends an with B&B HBCU whose Across individual NLS-72, HBCUs C&B SAT are and C&B column provides summary first college is statistics for students defined as the first 4- attends a junior college, technical school, or the like and a bachelor's degree, allowing column 3. HBCU. The second column one to make direct comparisons Columns 4 and 5 provide means of the ACT TWIs tend to have substantially higher academic scores of Blacks in have slightly higher GPAs TWIs are roughly than their peers 1 who standard deviation attend TWIs (2.86 A likely to attend private high schools. Similar trends are apparent in the in academic credentials between portion of the difference between the HBCU students and NLS-72 and B&B non-HBCU students can be explained by the by restricting the 1976 cohort to Morehouse and Xavier to ensure which emerge cannot be explained by different samples in the two cohorts. 1 976 data used to construct the summary statistics in Table 1 contains all 4 HBCUs available. All forthcoming results have been run The TWI. database for the 1976 and 1989 cohorts, respectively.^ HBCUs more though the differences are less pronounced. ^ who our datasets, blacks attending all or to 2.83), suggesting that these students attend less academically challenging high schools. Students in differences first descriptive statistics are displayed in higher. Yet, Blacks attending B&B, The HBCU versus TWI, where sample to those who completed credentials. In the compared was an will be counted as having his first college as an variables for students in the for black 1 HBCU on race or which college they attended are dropped from the sample. are missing data NLS-72 whose restricts the rate to the Statistics Summary in the survey described in greater students in the three datasets described above, according to whether or not they attend a Students C&B three cohorts in 1996 that provides detailed 1996 questionnaire was approximately 80%. The College and Beyond Survey detail in by the College that any sample different restrictions. In the 1976 and 1989 cohorts of standard deviations to SAT less, respectively. deviation behind Black students in C&B, and non-HBCUs. GPAs the ACT of Black students scores of HBCU In these data, students in HBCUs students are who attend are .73 and .5 more than HBCUs 1 standard are less likely have attended a private high school. The "Pre-College Personal and Family Background" environments that students were reared in. variables provide measures for the These variables include family income (measured whether or not a student attended high school dollars), parental education, or students were asked, "What the approximate is were dependents of that income before taxes of your parents sources." For all to convert all It is income measures to 1982-84 is derived from the HERI raw data, blacks with life. HBCU In the who attend They their bachelor's degree, HBCUs tend to make more likely to major less 1991 for students that The CPI-U was who attend HBCUs to adjust for these differences. attended graduate school, or obtained a money employed are also less likely to be students are in student sui"vey. graduate degree, and variables designed to measure college experiences and job and exception; NLS-72.^ NLS-72, on many of the outcome variables. These include wages, major choice, whether or not a student received the In dollars. apparent that there are important observable differences between blacks third panel reports sets. (or guardian)? Include used family income and TWIs. The subsequent analysis uses a variety of statistical approaches The 1972 and the student's own taxed and untaxed income for students their parents were not dependents.^ For C&B, family income used B&B, we home in a rural area or the southern portion of the US. The definition of income differs slightly between data taxable and non-taxable income from in than blacks full-time who and more life satisfaction. attend In TWIs, with one likely to be dissatisfied physical sciences. in two nationally representative samples, black students at HBCUs are more likely to receive a bachelor's degree and attend graduate school (though they are less likely to graduate). Black students in the elite HBCUs are more likely to major in biological sciences (this is driven in large part by Xavier who has a storied reputation for pre-medical studies), business, less likely to receive a bachelor's degree or attend graduate school, and, in the 1989 cohort, less likely to report that their college experience helped For students who were their parents' dependents in 1991, total family taxed and untaxed income was obtained, in order of priority, from the student's financial aid application, a telephone interview of parents, a telephone interview of the student, the student's Pell grant dependent, the information was obtained, interview, the student's Pell grant ' In the NLS-72, file, file, in or the student loan that were not their parents' or the student loan file. wage is $12.82 ($14.46) for HBCU attendees (graduates) and $10.55 ($11. 38) The mean of the natural logarithm of hourly wages is about 5% (12%) higher for wages across HBCUs and TWIs is due to a single HBCU The influence of this obsei'vadon on the Table 1 entries is also standard deviation of wages among HBCU attendees and graduates. See the Data Appendix for difference in the rank of respondent with an average hourly wage of $494. details For students the average hourly for TWI attendees (graduates). TWI attendees (graduates). The evident file. order of priority, from the financial aid application, the student's phone in the larger on the sample selection ndes. 10 , develop an ability to get The sample. It final along with individuals of other races. panel in Table also provides of observations with some at least HBCU provides the total number of 1 and TWl observations in each on the incomplete observations. As a solution details one missing variable, we to the large number turn all of the explanatory variables into a series of indicator variables based on ranges of the values of these variables and include separate indicators for The bottom panel missing responses to each variable. also reports on the number of observations with missing wage information. The subsequent analysis implements a standard selection correction approach to account for these cases V. (Heckman 1979). The Causes and Consequences of Attending A. The Causes HBCUs in the NLS and B&B Data Files of Attending HBCUs Table 2 presents a series of estimates of the detenninants of HBCU attendance. The specifications estimated are of the form: HBCU, = a + pX.''""^ + yX.''^^-^''"^^^ + e, (1) where X °' X/™ HBCU " is a dichotomous variable that equals one if the student attends an denotes an aiTay of variables which proxy for a HBCU and zero if not, home environment, and student's ^° ''^''denotes pre-college of each characteristics student. In instances, all weighted Probit regressions are used and the coefficients reported are marginal effects evaluated at the sample mean. weights are the sample weights in the relevant data file. of that coefficient relative to the omitted category when The home environment variables we that The all interpretation of any coefficient is The the effect other variables are held at their sample mean. include are family income, mother and father's education, and whether or not a student lives in the South. Family income, measured in 1972 dollars, divided into 4 categories: <$3,000, S3000 - $6000, $6000-$9000, and $9000+ based on described in the previous section. Parental education (mother and father, independently) a survey question partitioned into is three categorical variables: less than a Bachelor's degree, a Bachelor's degree, and greater. not a student lives in the South is a Pre-college characteristics include attended a private high school. dummy SAT and Whether or variable that takes on the variable of one if the answer ACT Combined SAT scores, high school is GPA, and whether is yes. or not a student scores are divided into less than 600, between 600-800, ACT scores are divided up similarly, less than 11, between 11 and 15, and greater than 15. High school GPA is and greater than We and greater than 800. 3.5. In 1972, students measured on a four point scale and also include an indicator for who attend high school in the South. attend HBCUs is divided into less than 2.5, 2.5 to 3.5, whether the respondent have lower Moving from an SAT SAT and ACT is female. scores and are more likely to score below 600 to a score above 800 reduces the . probability of attending an HBCU by 32%. Similarly, moving from an score of more than 15 reduces the likelihood of attending an South are 43% more likely to attend HBCUs, education are not statistically related to holding HBCU all HBCU ACT score of less than 11 to a by 21%. Students who live in the Family income and parental else constant. attendance once our other covariates are taken into account. In the 1990s, things change. Standardized test scores are HBCU attendance. Parents with else equal, a student more education more are whose mother has more than HBCU. This legacies (Drewiy and Doermann 2001). Residing no longer such powerful predictor of a bachelor's degree is 28%) more likely to attend an not surprising, as these institutions have a long histoi-y of loyal alumni and familial is South continues to be an important deteraiinant of in the college choice, though the magnitude of the coefficient seventies. Private high school is negatively con'elated with is roughly a fourth of the magnitude HBCU The most obvious explanation of selects for high ability students Comparing colmnns these differences and among 2 and 3 in table 2 is this set, differences we can make direct in the attendance. There are marked differences between the determinants of 1990s. a have children who attend HBCUs. All likely to HBCU attendance in the 1970s and that conditioning on degree completion between the two periods will disappear. comparisons between the 70s and the 90s; each sample consists only of degi^ee completers. Intuition suggests that as resistance to black attendance at TWIs faded and the need for segregated schooling declines, those with the highest opportunity cost of such schooling will opt out and choose more tiaditional universities. In other words, HBCUs. Yet, the data predictive of HBCU likely to attend HBCU show it might be reasonable to expect a "brain drain" from the opposite to be true. Lower academic credentials are not as highly attendance in the 1990s as the 1970s. Students with higher educated parents are more HBCUs in the 1990s and higher academic credentials are not as strong a predictor of non- attendance. This provides evidence of a selection of talented Black students into HBCUs in recent years. B. Econometric Approach In the attendance, and TWI Estimating the Consequences of Attending HBCUs absence of a randomized experiment or a credible instrumental variable for we implement four statistical approaches to adjust for pre-college differences between HBCU HBCU attendees. This subsection details these strategies. The (2) to first and simplest model outcome, = p^+ pX .""'""" In all instances, the estimation sample weights provided is we estimate is a linear specification of the fonn: 5HBCU + f, + oXr"""'"' + done using weighted in the data. Equation (2) is 12 least squares, with weights coiTcsponding to the a simple and easily inteipretable way to obtain estimates of the effect of covariates HBCU and Xj"""""^ attendance on outcomes, but ^.'"^-"'^s^_ may This on a linear model relies it to control for the be unappealing since their true functional form is unknown. As a solution, we match propensity scores (p-scores) of HBCU HBCU TWI and attendance (Rosenbaum and Rubin 1983).^ The estimated p-scores compress the multi-dimensional vector of covariates score approach are twofold. manner than First, possible with is is it HBCU we model and is among attendees a real asset here. assumption that just as with linear regression, the identifying attendance) provides an opportunity to focus the it those with similar distributions of the covariates with indicator variables the former advantage all not so compelling in this setting, but the latter HBCU Finally, is important to emphasize assignment that is it is associated only with observable pre-period variables. is the propensity observables in a more flexible to control for Second, TWI The advantages of an index. into method a feasible regression. linear comparisons of outcomes between the of the obsei'vables. Since students with similar predicted probabilities or to the treatment (i.e., This is often called the ignorable treatment assignment assumption or selection on observables. We implement the p-score matching strategy obtained by fitting probit regressions for '^^'^ and HBCU attendance (identical to equation as explanatoiy variables. In other words, Xj'^'^'^" We selection rule with the obsei"ved covariates. For both suitable. in three steps. First, the estimated p-scores are tests, we and /or estimate a richer probit accepted for both the means of If the null hypothesis HBCU outcome between this in one two ways. The TWI HBCU first is and TWI alternative ' is to In is we two sets of students divide the qunitiles interactions.' Once is the null is calculated by comparing the difference in HBCU HBCU student for which there student's p-score. that individual (or possibly a few) crucial Govariate(s). TWI We is at In cases take the simple average of outcome across done with replacement so match on a single we students with similar or "matched" values of the p-score. multiple students have p-scores within 0.10, An we the next step. student with an estimated p-score within 0.10 of the Rubin (1977) In the first test, rejected for either test, calculates a treatment effect for each students. Further, this matching TWI the covariates are equal for the of equality Second, the "treatment effecf for a given outcome the and model by including higher order terms and we proceed to tests, average student's students within quintiles. divide the sample into quintiles based on their p-scores. we examine whether test, within each quintile. to replicate the try then conduct two tests to ensure that the p-scores are assess whether the estimated p-scores are equal across the the second we above), using X!""^""" 1 all do least where of these students can be used as See Angrist and Lavy (1998) or for applications. See Dehejia and Wahba (2002) and Rosenbaum (1984) for method. 13 more details on how to implement the propensity score controls for multiple fomi HBCU student but in calculating the average weighted average, where the weight We use score. The second matching approach uses students. HBCU a control for each is a Gaussian kernel with a Third, a single treatment effect which there was students for focuses the comparisons and TWI where there TWI of the inversely proportional to the distance to the HBCU students to use a kernel student's p- bandwidth of 0. 1 0. is estimated by averaging the treatment effects across all HBCU one suitable match. This approach has the desirable property is overlap in the distribution of propensity scores among the that it HBCU students so these students are "similar."" We First, at least all among them we we also implement two other econometric approaches estimate probits for whether the missing and non-missing wage values. wage We variable to account for potential selection issues. missing on the sample of obsei'vations with is then include the inverse Mill's ratio from these probits in equation (2) to account for the possibility that wages are not missing procedure identified is at random (Heckman 1979). This from a functional form assumption, since we are unaware of a valid exclusion restriction in this setting. Second, the available data sets may some not include measures of attributes (e.g., strength of essay, motivation, and teacher recommendations) that persuade admissions committees to select certain applicants for admission that are also rewarded in the labor market. across HBCU and TWI students. The least squares Further, these attributes may differ and propensity score approaches rely on "selection on observables" assumptions and will produce biased estimates in this case. To confront this source of misspecification, Dale and Krueger (2002) that NLS data are admitted. We approach can only be implemented with the sets of colleges to we implement which individuals file, accepted, we as find the midpoint of the 25th-75th percentile report only We use current ACT scores, we For each student, B&B student. We SAT This does not contain information on the SAT Among the colleges range reported in scores since scores from 1972 are unavailable. use an equivalence scale to convert to we method pioneered by operationalize the Dale and Krueger approach by detennining the identity of the colleges that accepted each student. Report (2006). a variant of the matches students based on the colleges where they were accepted. record the highest midpoint SAT SAT where they were US Nev.'s & World For colleges that scores. score of any college that accepted the divide the students into four groups by quartiles of the school with the highest midpoint among the schools groups in equation where they were admitted. (2). We SAT then include separate indicators for each of these This approach mitigates the impact of any confounding due to characteristics See Dehejia and Wahba (2002) and Heckman, Ichimura, and Todd (1998) on propensity score algorithms. " If there are heterogeneous treatment effects, this strategy produces an estimate of the average effect of the '" "treatment on the treated". 14 obsei-vable to admissions officers that are not measiu'ed in the data set. assumption is that after non-HBCU within a quartile Specifically, the identifying adjustment for the available covariates, the decision to attend a HBCU versus a "ignorable" or orthogonal to unobserved determinants of outcomes. is See Dale and Krueger (2002) for a more detailed discussion of this approach. we Finally, were unconvinced we note that distance to a student's nearest considered a number of candidate instrumental variables, such as HBCU, was very difficult to constmct an instmment with a powerful may C. Estimates still first stage. the and B&B data, Figures data files, respecdvely. who completed TWI is it To HBCUs evident that the mean B&B are more at an HBCU or TWI. In Figure p-scores exceeding roughly 0.9. students with p-scores greater than about 0.8. Thus, it 3 presents results graduation from a HBCU in 5). HBCU In Panel A, the treatment Panel B. For the B&B indicator. is For the NLS, wages are measured relevant sample. and roughly 10 years based on in number of is distributions of p- HBCU In the NLS aren't any meaningful on the natural we errors are entries in Panels HBCU, A same that attended the matriculation at a entries in Panel C, report the R-squared statistic, as well as the The The estimated standard while it is report standard en-ors that allow for unspecified heteroskedasticity in the variance-covariance matrix. we with the in this range. and B, the standard eiTors allow for clustering among observations from students column is However, there of the effect of attending a reported in parentheses below the point estimate for the in which 1, will be difficult to obtain logarithm of wages from the four approaches in the six columns. high school (except TWIs propensity score differs across the populations, but there similar throughout a broader range of the p-score. Wage Outcomes. Table and obtain these figures, weighted probits were estimated on the comparisons for the relatively small subset of HBCU students with p-scores eiTors, is and 2 present separate kernel density plots of the 1 their bachelor's degree at substandal overlap in the distributions, except scores in the it Consequently, thorny issues of of the Consequences ofAttending HBCUs of students NLS we cases remain. Distribution of Estimated P-Scores. set all With approximately 300 observations, estimated propensity scores for black students that received degrees from NLS but in were not powerful enough valid or the instruments for the relatively small samples in the available data files. selection HBCUs, residing in the South, or the closing of that the exclusion restriction Underneath the standard students in HBCUs and TWIs in the 1986 - fourteen years after high school graduation B&B, wages are in 1997, five years after after obtaining a bachelor's degree. In the completion of the bachelor's degree. Column 1 reports the mean difference in labor market without adjustment for any controls. In the NLS, 15 HBCU wages for individuals students earn roughly who 5% attend less HBCUs, when the treatment is the these estimates statistically different 1 on observable dimensions 1.5% when from zero HBCU graduates earn 1990s indicates that that college attended and first is HBCU 16.6% is it receipt of a bachelor's degree. less than students have TWI Column downward HBCU is 1970s in the is 11.1% when HBCU status is NLS dramatically.'" based on the defined as receiving a bachelor's degree. The foiTner estimate has an associated t-statistic less -13.8%) and the null of zero matching.'^ 1.'^ would be The next two columns of these results than to the linear B&B, In the first is TWI their biased. The adjustment 2 reports the results from estimating equation (2). controls changes the results in the demonstrated 1 academic credentials than lesser Neither of estimate from the graduates. Recall, Table counterparts (especially in the NLS), so these raw gaps are likely home environment B&B The conventional levels. at The wage for the academic and benefit of attending a 6.0% when college attended and marginally significant, while the however, the wage effect for attending it latter HBCUs is rejected with conventional criterion.''' report on the implementation of the p-score method to test the sensitivity model. Column 3 uses kernel matching, while column 4 relies on radius Standard errors for both matching estimates were bootstrapped (200 with iterations), propensity scores recomputed for each bootstrap sample. Further, the p-score matching estimates are not weighted with the sample weights. The p-score estimates are remarkably similar to those from the linear regression in column 2. This finding shouldn't be teixibly sui"prising because equation (2) models the covariates flexibly, nevertheless it is Column full set of covariates. from the labor change reassuring that functional fomi issues don't appear important in this setting. 5 presents estimates that are selection corrected for in the force. It seems plausible This possibility is that HBCU attendance is missing wages and adjusted for the coiTelated with selective withdrawal not supported by the data as this approach produces unimportant estimated impact of HBCUs on wages.' '^ A '^ Results are similar '^ In the similar result was found in Constantine (1995). one implements a "fraction method," using individual transcripts a student's college experience that was spent at an HBCU. if B&B (B&B 2003) there is no wage gap between HBCU students and non-HBCU HBCU sample does not have valid wages in the later survey (some are in the while others were dropped completely). HBCU graduates have a 9.6% higher most recent wave of students. Unfortunately, 53% the of the survey and unemployed unemployment rate, and median regression techniques provide concentrate on the earlier wave with more complete data. '' Observations with estimated p-scores that are not observations in the of to calculate the fraction NLS when the treatment is first strictly identical between results to the and college attended and 21 1 1997 wave. Thus, are dropped. when Further, the treatment and 21 observations in the B&B when implementing the radius matching estimators due matches in the relevant range. is we we lose 7 degree college, to outliers that did not have " We also assessed the impact of labor market dropouts on our estimates with a simple re-weighted linear regression and median regression. In the first approach, we estimate a probit of whether or not we have valid wages on all of the covariates in Table 2. We then multiplied the sampling weight by the inverse of the predicted probability in the new weights. Linear regressions are then esrimated with these new weights. This approach led to remarkably similar conclusions as the selection correction approach. Median regressions were estimated by probit to get 16 Column student 6 implements the column 2 specification but adds controls for the "best" school that the was admitted and Krueger 2002). to in order to account for the richer data available to we Specifically, the best school that the student was admitted method is NLS HBCU to only possible in the 22.5%. to, SAT This doubles estimated impact of attending a it from HBCU attendance However, the estimate's imprecision makes by the other methods. scores of leaving the lowest quartile as the excluded group. data and in this sample Specifically, this approach suggests that the gains larger than indicated admissions committees (Dale include indicators for the three highest quartiles of may be definitive conclusions unwarranted. Additionally, we conducted HBCU, which to attending a home with students' region a number of tests are reported in for whether there was heterogeneity Appendix Table We 2. in the returns assessed whether returns differed South versus North), their estimated propensity score, (i.e., SAT Score, parental education, and gender. In general, there isn't substantial evidence of heterogeneity across these The lone exception subsamples of students. black women sample is too than for black men in the demanding of the data NLS is that the returns to attending a sample. It is 1970s. suffering a wage relative return HBCU subsamples are much larger. conferred remarkable advantages on Conventional estimates of the average return to college are (Heckman, Lochner and Todd 2003). Attending equivalent to one appear higher for also immediately evident that subdividing the as the standard errors in the Overall, these results suggest that attending an students in the HBCU more year of penalty. schooling. a In contrast, If the point estimates are of HBCU attendance Non-wage outcomes. Thus in just far single outcome: labor market wages. labor market considerations. HBCU taken versus a TWI more recent in the HBCU literally, there is 10% its per year 1970s was roughly attendees appear to be nearly a -25% swing in the two decades. we have concentrated on the effect of attending a The value of attending The conventional wisdom is HBCUs, HBCU on a however, likely extends well past that these institutions instill confidence in their students, a sense of responsibility, and provide environments free of racism and discrimination that allow for greater personal development. Such environments are likely to have many benefits beyond those captured in wages. Table 4 explores the effect of attending probability of full-time employment, measures of HBCUS on a life satisfaction, number of outcomes, including and a series These wide ranging outcomes were chosen because of their economic and social relevance as well as comparability across datasets. dummy variables imputing zeros were smaller to all for The academic outcomes index majoring in business, majoring in is the first principal physical the of academic outcomes. their component of the science/mathematics/computer missing wage observations. Qualitative conclusions were the same, though the coefficients as expected. 17 science/engineering, majoring in biological science/liealth, receiving a bachelor's degree, graduate school, and receiving a graduate degree. matching estimates and confrnn these gain 10% that students HBCU and from these outcomes striking finding is HBCU that weighted matriculation increase in the probability of receiving a bachelor's degree. 1970s in the coefficients reported in the table are from kernel least squares results. The most nearly The their associated bootstrapped standard eiTors. In all cases, attending due is who to the increased probability attend HBCUS TWI matriculates are modestly It is is of graduating from college. There more major likely to report similar degrees of associated with a evident that part of the some evidence is in physical sciences. life satisfaction. wage Interestingly, There are negligible effects on all other outcomes. VI. There TWIs. To is A Focused Look substantial quality variation we have analyzed this point, that include the full the NLS spectrum of HBCUs and more focused look four of the most at Most at the among and B&B, which TWIs from elite Elite HBCUs HBCUs, the set of 89 4-year compared to the thirty selective Table of C&B in all 4 that is 5 reports from estimating equations 1989 cohort which (2) drops parameters on the standardized (colunm 2) SAT Howard 1989 cohort. income variables (3). The test scores are the for elite so that most notable HBCU in the elite scores have a similar effect on HBCU HBCUs we take a students.) HBCUs One can only be for the 1976 is (1) and 1989 cohorts a sample of individuals can make direct comparisons with the patterns across the SAT ACT positively associated with the Howard and Spelman of attending an approach, these four Table 2." Column score of less than 600 and a in the likelihood we Xavier, Spellman, and Howard. sample. identical to those in displayed in column is difference between a in the on the detenninants of attendance HBCUs, column 3) in the TWIs the are nationally representative data files HBCUs: Morehouse, C&B's sampling due to the among the quality continuum. In this section, (Importantly, the 1989 data set does not include infonnation on Spellman and limitation of this exercise as well as results. columns are quite Evaluated at the similar. The sample mean, the score above 800 implies a 40.1% decrease 1976 cohort and a 63.1% decrease (column HBCU attendance. Mother's education is attendance in 1989. After adjustment for the academic characteristics, are not reliable predictors of elite HBCU Residing attendance. portion of the United States continues to be a strong predictor of HBCU in the southern attendance; increasing the probability of attendance by roughly 40%.'^ '^ Throughout the C&B analysis, the treatment consistency with the other tables, we is defined as enrollment at a HBCU, not graduation from one. refer to the treatment as "first college", although due to the For C&B's sampling scheme it is possible (but unlikely) that a respondent was initially enrolled at a different college or university. '* The large negative coefficient on female in 1989 is related to the fact that the women's college Spellman is not in HBCU Figures 3 and 4 plot the distributions of the estimated propensity scores for the attendees from the 1976 and 1989 make this analysis to two C&B the samples consistent. It is This finding confirms the impression from the probit academic credentials of these students The poor overlap of the the propensity score exercise student, it is apparent that we differ in important distributions poses challenges for the where we require the will rely selection on observables assumption TWI ways. outcomes For example, in analysis. students to have p-scores within 0.1 of the on a small subset of the may TWI apparent that there isn't substantial overlap across the distributions in either year, especially in 1989. results that the and Howard and Spelman were dropped from classes, respectively. HBCU In this subset of the data, the data. be especially unlikely to be valid. On the other hand, we can use least squares functional forni assumptions to infer counterfactuals in parts of the distribution where there is little support. infeiTing the impact of Neither approach HBCU be interpreted with these important caveats We now measures of life the C&B especially appealing, which underscores the difficulties of is matriculation in this sample. Consequently, the forthcoming resuhs should in mind. turn to an exploration of the effect of attending elite satisfaction, database is the HBCUs and academic outcomes from the 1976 and 1989 availability of detailed questions about life on labor market outcomes, C&B. One major benefit of outcomes, beliefs, college experiences, labor market outcomes and experiences, political and civic engagement, and more. These rich questions can help to shed light HBCUs. For HBCUs on the overall experience of students attending both cohorts, data on the majority of outcomes was obtained 1995, in and non- 15 years after graduation for the early cohort and 2 years after graduation for the later cohort, though some data such as wages in the 1989 cohort were collected in 1996. Individuals without valid wage obsei"vations are dropped from the sample. Because the number of potential dependent variables indices to better understand the experiences of HBCU appendix describes the specific questions used first principal First, principal set component of to C&B is so numerous, we construct five students along the following dimensions: objective academic, subjective academic, labor market, leadership and the in make up lifestyle, and social interactions. these indices. Each index the set of variables described. This approach has is The data obtained by taking two main advantages. components analysis reduces the dimensionality of problems by extracting the portion of a of variables that explain the most variance within the measured on the same scale. The cost is that the result's set. Second, meaning it ensures that isn't transparent all variables are and cannot easily be applied to different settings. Table 6 reports the results of estimating the effect of HBCU non-wage outcomes from kernel matching, which can be compared the 1989 sample. 19 attendance on our set of wage and to the results in column 3 of Tables 3 and 4. Column 1 includes four all HBCUs Column 1976 sample. in the 2 also reports results from the 1976 sample but only includes students who attended Morehouse and Xavier with the 1989 cohort, whose HBCUs In the 1976 cohort, and 11.1% more on attending, 13.6% significantly major likely to 10% been written on these statistically significant likely to more racially institutions and substantively HBCU experience are large, which homogenous importance of HBCUs is that, would choose The 1989 cohort the zip codes. controlling for same college is HBCU The labor market experiences And, HBCU HBCU HBCU students are quite positive. much interactions are also are positive, but negligible. graduates are no more likely graduates are no other factors, all is consistent with (Drewey and Doennan 2001). Social large. political, religious, civil rights, social service, or philanthropic activities. report they in the biological sciences objective academic outcomes, Students do not seem to possess a particular taste for segregation, as to live in major though their subjective view of the academic experience negative, comparisons less likely to attend graduate school and, conditional On less likely to receive a degree. to facilitate 3. 9% more graduates were in business, but Leadership and lifestyle components of the that has column results are reported in more The likely to be engaged in clearest evidence of the students are 1 8% more likely to again. reports different experiences. Students are more likely to major in physical and biological sciences and business. Students continue to benefit from leadership and lifestyle components of HBCUs, but the magnitudes of these effects are less than one-fourth of their previous levels. other positive elements turn negative in the outcomes are negative and social interactions index, cohort. The negative, and they code that has a HBCU more which was positive 10% in the is less likely to receive a bachelor's degree. 1976 cohort, sharply negative tiuTis seem 16% to have a taste for segregation. HBCU attendance Most telling, HBCU higher fraction of blacks. students to report they would choose in the The 1989 also substantial in three of the variables that are used to 1976 and 1989. The labor market experiences of the in of the The objective and subjective academic recent cohort. matriculates are table demonstrates that this decline constmct the index zip HBCU Many the same is later cohort are even more associated with living in a students are less likely than non- college again, although the difference isn't statistically significant. Interestingly, activities wages. which HBCU students in the later cohort are significantly are associated with civic consciousness. This may likely to be VII. Reconciling the Differences between the 1970s We are grateful to between the 1970s and 1990s Results Lani Gunier for pointing out this possibility. 20 engaged in partly explain the divergence in 19 A. Assessing the Difference " more and 1990s Panel A of Table 7 summarizes the difference we between the 1970s and 1990s. in the results examined above. The difference reports regression results for five of the key dependent variables is It that use the stacked 1970 and 1990 data sets to estimate the following equation: outcome!, = (3) Po + y, X/"^="" + a, x,r-'°"'<'' + ei(HBCUi,)-l(1990i,) + where the i + X l(HBCUi,) + 5 Si,, subscript indexes an individual and the t subscript reveals whether the observation 1970s or 1990s college student. The parameter vectors y and a have HBCU interest is 9, which and whether the observation observation from the 1990s. attendance and is This parameter also includes separate intercepts HBCU indicator and the indicator for an (DD) estimate of a difference in differences is from a from a 1990s college attendee. The parameter of is associated with the interaction between the is is subscripts indicating that they are t The equation allowed to differ for 1970s and 1990s college attendees. for attending a 1(1990,,) equal to the difference of the cross-sectional HBCU estimates (e.g., the natural logarithm of wages. For column 2 HBCU in Table 3).^" In column (1), the nationally representative dependent variable NLS and Howard and Spellman The decades. first attendees are dropped fi-om the sample, just as in (i.e., same college 20% and TWIs. (2) HBCU attendance over the two of Table 6. decline in wages, a HBCU 13% attendance have declined. Specifically, depline in the fraction of students who would again, and substantial declines in the leadership and social interactions indices. other three "objective" outcomes, which measure political participation, social/civic sei-vice, and donations to national charities, HBCU all show an increase between the became measure indicates that noteworthy that of these estimates are economically and B. HBCUs column those where a higher value of the dependent variable a positive) suggest that the returns to the point estimates suggest a The in the relative returns to four "objective" outcomes would be considered attend the data files are used. this regression, the The other seven dependent variables are samples, which focused on a subset of elite summarize the change results The B&B C&B taken from the 1976 and 1989 is all Robustness of the Result that attendees HBCUs ' two C&B classes. less likely to live in integrated The non-objective neighborhoods. It is statistically significant. Performance Worsened between the 1970s and 1990s This subsection reports on some checks that aim to explore the robustness of the basic result that the economic returns differences. We that the definition of ^^ The C&B table reports to attending a HBCU declined. Many obvious explanations fail to explain the have ensured, through the use of Federal Interagency Committee on Education codes, HBCUs is consistent across data sets and over time. Moreover, the addifion of results are not identical to the difference on the kernel matching results, between the 1989 and 1976 results in Table 6 because that while Table 7 relies on least squares adjustment for the covariates. 21 more control variables, such as occupational choice indicators or a richer set of academic variables, fail to explain the differences. Further, differential labor market dropout rates cannot explain the change. Selection of students into (or by) colleges but the available data fail to support it. a potential explanation with is For example, Table 2 demonstrates that, if some intuitive appeal, anything, selection on observables appears to work in the opposite direction. Students with higher scores on aptitude tests and home environments better Nevertheless, are we more likely to attend the observations with non-missing wages from the whether the obsei-vation was from the probabilities that the B&B NLS OLS NLS The B&B and the is 1990s, relative to the 1970s. on observables. Specifically, we stacked B&B and estimated a weighted probit for We 1970s)."' 1990s) observations are from the (i.e., NLS. used as a weight then calculated the predicted These predicted probabilities are in the fitting of the Table 3 column is similar and B&B."" result of procedure this is that the estimated decHnes from -13.8% (standard eiTor of 0.050) that this in the This weighting scheme aims to ensure that the distribution of observables specification. across the NLS (i.e., multiplied by the sampling weight and this product 2 HBCUs further explored the role of selection approach only increases the relative worsening of Ahhough we cannot between the 1970s and 1990s. HBCU impact of attendance on wages in the -18.7% (standard error of 0.056). to HBCU It is evident attendees' labor market performance rule out selection on unobservables, seems it reasonable to conclude that the results aren't due to differences in observable characteristics. Another TWIs possibility represented in the due to the relatively small sample is that NLS and B&B To explore differed. this, (2006) to obtain the current midpoint of the 25th-75th percentile represented by the valid obser\.ations in the data the observations."^ HBCUs and quaUty of the SAT range for This inforaiation is the respondents in our analysis decreased in our sample, SAT scores in HBCUs and all TWIs were the HBCUs and TWIs only available for a subset of SAT Nevertheless, these data indicate that the gap in average TWIs chosen by and universities the colleges files. sizes, the we used US News and World Report scores between the between the two decades. 236.1 points higher than Among HBCUs in the 1970s and 226.2 higher in the 1990s. Thus, the available evidence suggests that the results are not due to a " change in the The weights composition of HBCUs and are adjusted sampling weights, observations and within the throughout the paper (e.g., in "" This method is -^ We matched 92.3% of SAT of the HBCU TWl colunm NLS and B&B. where the adjustment ensures that they sum to 1 within the NLS The covariates in the probit are the standard controls used (2) of Table 3). Lemieux (1996) TWIs represented in the NLS. For the B&B, these numbers many of the colleges represented in these data files do not US News and World Report (2006). Specifically, we obtained SAT information for 29.1% the and 84.0% of the respectively. UnforUinately, information in attendees and attendees in the HBCUs the observations. a variant of Dinardo, Fortin, and were 88.9%) and 82.7%), report B&B TWIs between 49.0% of the TWI attendees in the B&B. 22 NLS and for 33.0% of HBCU attendees and 49.2% C Did HBCUs Decline or Did TWIs Improve? If the estimated decline is causal, the source unlikely to be important for public policy purposes. Nevertheless, researchers and educational practitioners to understand whether of HBCUs One potential explanation 1990s leading answer to this it it is would be of considerable due is interest to performance to a decline in the question could lead to the design of more programs for blacks. effective educational some An or an improvement in TWIs. HBCUs' performance of the relative decline in is that because HBCUs weakened between the 1970s and financial position of the educational environment to a decline in the quality intuitive appeal, HBCUs' at experienced enrolhnent declines HBCUs. in this This explanation has period and Hoxby (2000) suggests that the nationalization of the higher education market has hurt smaller colleges and universities. However, Specifically on a host of input measures, the data don't appear to support this explanation. including financial informafion, faculty compensation, research grants received, and composition of faculty with various degrees, there substantial ways (Provasnik, evidence that the relative quality of is little HBCUs has declined in Shafer, and Snyder 2004). Expenditures per student are perhaps the best measure of the quality of the education that these schools provided. Data on educational expenditures the Higher Education General Infonnation Survey at HBCU and non-HBCU schools Education Data System from 1985-2004 through the ICPSR web site. In 2000 expenditures per student increased from $9,423 to $1 1,996 between 1974 and 1991 at to $10,560 at TWIs."'' (The years 1974 and 1991 are chosen, because they are roughly prime college attendance years of the students in the educational expenditures per student increased by 24.7% is at TWIs. It is due to a decline An alternative explanation is that at the civil rights struggle, campus many Southern TWIs. sit-ins Further, HBCUs HBCUs in the it respectively.) Put another A Educational expenditures is research, public the a greater sum of service, way, real between 1974 and 1991, compared It is HBCU attendance also notable that the level of effective at educating blacks After all, the 1970s are not far many TWIs were related theory that expenditures between the removed from is made from 23 to not convivial places for empirically indistinguishable is that cross-racial connections. the current funds that relate to the functions of academic support, student services, maintenance. to both the 1970s and 1990s. has been noted that premium on is and $8,468 middle of the and boycotts, and battles over even allowing black students society has changed so that there instruction, B&B, HBCUs in in spending. black students in the 1970s (Patton 2005). ^'^ and at TWIs became more 1970s and 1990s in ways that aren't reflected enroll in NLS 27.3% measured by expenditures. education expenditures per student was greater available from dollars, educational evident that data reject the hypothesis that the decline in returns to in their quality as is from 1970-1984 and the Integrated Postsecondary institutional support, operation and To explore this possibility credibly, experiences at TWIs. After all it is We students between the 1970s and 1990s. essential to have a counteifactual for blacks' changing TWIs became more possible that all, it is effective educational institutions for use whites' experiences at TWIs as a counterfactual in the following equation: outcome,, (4) = + where (again) the + Po Tn X;, '^•"'"'' may indicating that they is 0, which and t (Black,,) + 51(1 990,,) subscript reveals whether the observation references race. and whether the observation on whether the returns for All relative to whites. Panel r 1 s„, The vectors y and a have is from "rt" subscripts, is from a 1 990 college attendee. The parameter of interest associated with the interaction between the Black indicator and the indicator for an is B only, since this + vaiy by race, time period, or both. The equation also includes separate intercepts observation from the 1990s attendees. reports +X x/'"'""'^' subscript indexes an individual, the i a 1970s or 1990s college student, for black students + an 9 l(Blacki,)-l(1990„) HBCU This parameter TWI is a difference in differences (DD) estimator and attendance increased between the 1970s and 1990s for blacks, attendees are dropped from the estimating samples in all specifications. of Table 7 reports on the estimation of equation (4) for the natural logarithm of wages is the variable of interest available in the nationally representative data files. The column The column (1) specification constrains the y and a vectors to be constant across races and periods. (2) one allows them to differ across decades but holds them constant across whites and blacks within a decade (i.e., in the same survey). In column (3), they are allowed to vai-y across races but are restricted to be constant across decades. The column among TWI (3) specification suggests that by 13.4% more than wages of whites between 1970 and 1990. and would account for roughly two-thirds of the relative decline Panel A. The point estimates are also positive statistically indistinguishable The difference parameter vectors. It from in the other two attendees, the This estimate in the so is is HBCU attendees in and due to the choices about the y and a important to note that F-tests lead to rejections of the hypotheses that the just too parsimonious. the standard error statistically significant zero. parameter vectors are equal across decades (column 2) and race (column specification is wages of black specifications, but they are smaller in the results across the three specifications is is wages of blacks increased The R-squared on the parameter of interest. statistic is largest in the We 3), column also experimented with a so the column (1) (2) specification but model that allowed the parameter vectors to vary with indicators for the interaction of decade and race, but the resulting estimates had little empirical content.'^^ " This model is too demanding of the data. multicollinearity are an especial concem. As Recall, all of our controls are indicator variables so issues of a measure of 24 this problem, the standard error on the parameter of We we are left unaware of good reasons are to favor either the column (2) or thirds of the decline in the relative wages of HBCU attendees is due to educating Blacks. The other produces an imprecise estimate that explanation has sober conclusion (3) specifications, so suggests that roughly two- improvements in TWIs' efficacy that the data fail to reject the possibility that blacks between the 1970s and 1990s, but they also fail to at taken literally implies that this if empirical backing. Both have confidence intervals with significant overlap. little is column One with two specifications that appear to have equal standing. TWIs became more The effective at educating provide decisive evidence in favor of this possibility. VIII. Conclusion Historically Black Colleges and Universities are an integral and proud part of Black histoid and For generations, these institutions have educated Blacks and produced leaders culture. government, in business, entertainment, and the academy. Yet, their reliance on public funding and the Fordice decision mean that it more important than ever is to understand the causes and consequences of matriculation at HBCUs. Existing evidence on the effects of attending Historically Black Colleges and Universities has typically concentrated on approach - analyzing three large data sets with either degree attainment or future wages. In this paper, we take a more holistic adequate pre and post college infonnation, a student's choice of college, and myriad social and economic outcomes to paint a rich portrait of the experiences of black students to attend at Historically Black Colleges and Universities, relative non-HBCUs now and to their counterparts over time. Consistent with the charge from the high court, who choose we search for "educational justification." The nationally representative Several important results from this search have emerged. reveal an important change in the returns to associated with higher student quality (e.g., attendance. is a substantial wage penalty. wages of HBCU graduates between SAT scores) improved among In fact, there the fallen behind. HBCUs Finally, there is also some evidence of a HBCU is datasets matriculation relative to attending a relative was TWI. By a statistically significant two decades. Notably, HBCU achieving students were increasingly choosing these schools have In the 1970s wages and an increased probability of graduation, the 1990s, however, there decline in the relative HBCU 20% measures of attendees during this period, so higher at the same time that the schools appear to relative decline in the performance of elite from the College and Beyond. The analysis has unearthed some important clues declined in this period. The data provide interest (9) is more than 9 times some as to why HBCUs' support for the possibility that larger than the one in column 25 (1). relative HBCUs' perfonnance relative decline is partially due to improvements in TWIs' efficacy in educating Blacks, but this evidence certainly isn't number of seemingly In contrast, the data reject a decisive. declines in the pre-coUege credentials of students attending student HBCUs. at This question of why the identification of the exact channel summary, In is HBCUs' performance to and in educational expenditures declined merits further research, although the evidence presented in this paper suggests that relative to have disappeared on per unlikely to be important for policy puiposes. have provided unique educational services seems intuitive explanations, including relative HBCUs for blacks in the 1970s. However by many dimensions and by some measures retard black progress. 26 TWIs, HBCUs may the 1990s, this advantage HBCU attendance appears to References Angrist, Josh and Victor Lavy. Comparisons "Does Teacher Training Affect Pupil Learning? Evidence from Matched Jemsalem Public in Journal of Labor Economics, Vol. Schools."'' 19, 2 (Apr. 2001), 343-369. Borjas, George Ethnicity, Neighborhoods, and J. Human-Capital Externalities. American Economic Review, June 1995, 85(3), pp. 365-90. Bowen, William and Bok, Derek. 1998. The Shape of the River: Long Term Consequences of Considering Race Case, in College and University Admissions. Princeton, NJ: Princeton University Press. Anne C and Katz, Lawrence F. "The Company You Keep: The Effects of Family and Neighborhood on Disadvantaged Youths." National Bureau of Economic Research (Cambridge, MA) Working Constantine, Paper No. 3705, 1991. 1995. "The Effect of Attending Historically Black Colleges and Universities on Future Jill. Wages of Black Coleman, James SiuAQnis." Industrial S., et al. and Labor Relations Review, Vol. 48, 3, 531-546. Equality of Educational Opportunity. Washington, D.C.: Govermnent Printing Office, 1966. Cutler, David M and Glaeser, Edward L. "Are Ghettos Good or Bad," Quarterly Journal of Economics, August 1 997, 2(3), pp. 827-72. 1 1 Dale, Stacey and Alan Krueger. 2002. "Estimating The Payoff Of Attending An Application Of Selection On Obsei"vables And A More Selective College: Unobsei"vables," Quarterly Journal of •,,-•, fconow/c^, 2002, 107(4,Nov), 1491-1527. Dehejia, Rajeey H. and Sadek Studies." The J. Wahba. "Propensity Score -Matching Methods Review of Economics and Statistics. Vol 84, 1 for Nonexperimental Causal (Feb. 2002), 151-161. DiNardo, N. Fortin and T. Lemieux, 1996, "Labor Market Institutions and the Distribution of Wages, 1973-1992 : A Semiparametric Analysis", Econometrica, Vol. 64, September, pp. 1001-1044 Donohue, James Heckman, and Petra Todd. 2002. "Social Action, Private Choice, and Philanthropy: Understanding the Sources of Improvements in Quarterly Journal of Economics, 225-268 1 17(1), pp. Black Schooling in Georgia, 1911-1960," The Drewry, Heniy and Humphrey Doermann. 2001 Stand and Prosper: Private Black Colleges and Their . Students. Princeton, NJ: Princeton University Press. Ehrenberg, Ronald. "Are Black Colleges Producing Today's African American Lawyers?" The Journal of Blacks in Higher Education, ]4(Winter 1996-1997), pp. Ehrenberg, Ronald and Donna Rothstein. 1994. "Do Confer Unique Advantages on Black Students? Consequences: Contemporary Policy Issues 1 17-119. Historically Black Institutions of Higher Education in 27 An Initial Analysis." In Choices and Education, Ronald Ehrenberg, ed. Ithaca, NY: ILR Press. Fordham, Signithia and Ogbu, John. "Black Students' School Successes: Coping with the Burden of 'Acting White.'" The Urban Review, 1986, 18, pp. 176-206. Fryer, Roland G., Jr. with Paul Torelli. "An Empirical Analysis of Acting White.'" National Bureau of Economic Research (Cambridge, Mark Granovetter, S. MA) Working Paper No. "The Strength of Weak I'lzs." 1 1334, 2005. American Journal of Sociology, 1973, 78, pp. 1360- 80. Heckman, James. 1979. "Sample Selection Bias as a Specification Error", Econometrica, 47(1), pp. 153- 161. Heckman, James, Hidehiko Ichimura, and Petra Todd. "Matching Estimator." Review of Economic Studies. Vol. 65, 2 Homans, George C. The Human Group. New ( as an Econometric Evaluation 1998), 261-294. York: Harcourt Brace, 1950. Hoxby, Caroline. 2000. "The Effects of Geographic Integration and Increasing Competition for College Education," NBER Working Paper No. Jencks, Christopher and David Reisman. 1968. The in the Market 6323. Academic Revolution. Garden Lazear, Edward. 1999. "Culture and Language," Journal of Political Economy City, NY: Doubleday. 107:6 (December 1999): S95-S126. Manski, Charles and David Wise. 1983. College Choice America. Cambridge: Harvard University in Press. Patton, Lori D. "Reflecting Back, Looking Forward: Civil Rights and Student Affairs (review)" Journal of College Student Development, Volume 46, Number Provasnik, Stephen, Linda Shafer, and 1976 to Thomas Snyder. 2001. National Center for Educational 1, January/February 2005, pp. 104-108 Historically Black Colleges and Universities, Statistics. NCES 2004-062. Putnam, Robert. 2000. Bowling Alone: The Collapse and Revival ofAmerican Community. Simon and Schuster: Roebuck, J. New York, 2000. B. and K. S. Murty (1993). Historically Black Colleges and Universities, Praeger Publisher. Rosenbaum, Paul R. and Donald B. Rubin. "The Central Role of the Propensity Score Studies for Causal Effects." Biometrika. Vol 70, 1 in Observational (Apr. 1983), 41-55. Rosenbaum, Paul R. and Donald B. Rubin. "Reducing Bias in Observational Studies Subclassification on the Propensity Score." Journal of the American Using Statistical Association. 79, 387 (Sept. 1984), 516-524. Rubin, Donald B. "Assignment to a Treatment Group on the Basis of a Covariate." Journal of Educational Statistics. Vol. 2, 1 (Spring 1977), 1-26. Tatum, Beverly D. Wliy are All the Black Kids Sitting Together in the Cafeteria? 28 And Other Vol Conversations about Race. US News New York: Basic Books, 1997. and World Report. (2006). America's Best Colleges and Universities. U.S. Dept. of Education, National Center for Education INFORMATION SURVEY Statistics. HIGHER EDUCATION GENERAL (HEGIS) XVI: FINANCIAL STATISTICS HIGHER EDUCATION FOR FISCAL YEAR 1981 [Computer OF INSTITUTIONS OF file]. Washington, DC: U.S. Dept. of Education, National Center for Education 1981. Ann ICPSR version. Statistics [producer], Arbor, MI: Inter-university Consortium for Political and Social Research [distributor], 2005. U.S. Dept. of Education, National Center for Education Statistics. EDUCATION DATA SYSTEM file]. ICPSR Version. Washington, Statistics [producer], 1994. Research Wilson, William [distributor], J. (IPEDS): Ann INTEGRATED POSTSECONDARY FALL ENROLLMENT ANALYSIS, 1989 [Computer DC: U.S. Dept. of Education, National Center for Education Arbor, MI: Inter-university Consortium for Political and Social 2003. The Truly Disadvantaged: The Inner Chicago, IL: University of Chicago Press. 29 City, the Underclass, and Public Policy. 1987. Appendix: Data Description General structure of dummy variables used on right-hand side of regressions Family income • dummy Five variables: family income in 1972 dollars $8,999, greater than or equal to $9,000, and missing. $3,000 to $5,999, $6,000 to omitted the less than $3,000 categoi'y. Parents' education • dummy For each parent, four variables: less than bachelor's degree, bachelor's degree, bachelor's degree, and missing. We omitted the less than bachelor's more than degree category. Test scores • For combined SAT scores, four missing. For composite missing. We dummy ACT scores, variables: less than 600, four omitted the lowest category dununy in 600 to 800, greater than 800, and variables: less than 11, 11 to 15, greater than 15, and each case. High school grades • Four variables: GPA on a four-point We omitted the lowest categoi'y. dummy missing. Binary variables • dummy Three • NLS1972 • Race The (sex, rural. dataset contains 974 • The data, scale less than 2.5, numerous race variables collected because We identified black this variable is from 2.5 to 3.5, greater than 3.5, and South, private high school) variables: each of the binary possibilities, not mutually consistent. 1 less than $3,000, We missing and missing. in different years. In 29 cases, these variables are and white students using the composite race variable in the in the fewest cases. Sex dataset contains several sex variables collected in different years. In 22 cases whom we identified as black, these variables are not mutually consistent. or female according to the composite sex variable in the • HBCU attendance We investigated the effect of a student's 1 974 We among the students identified students as male data. first college being a historically black institution. In each wave of were asked whether they had attended any institution of higher education since the previous wave. If they had, they reported the name of the institution; whether it was a vocational school, a two-year college, or a four-year college; and the first date on which they attended this school. We the survey, students 30 compared the dates of attendance at all four-year colleges reported to find the The survey includes a unique numerical code detemiine whether the first college was historically black. We first for each college; attended, if any. restricted our we four-year college used these codes to sample to students who one four-year college. We did not use the transcript data in the NLS to determine whether a student went to college and, if so, what college the student attended first because transcripts are sometimes missing. reported attending at least • Wages and employment We computed hourly wages only 30 hours per week. status for full-time workers, whom we defined as individuals who worked at least Respondents were asked in 1986 be reported in hourly, to report their salary in their current or most recent job. The wage could how many we weekly, bi-weekly, monthly or yearly increments. Respondents also said hours they worked per week in this job. For full-time workers who reported a weekly salary, We divided biweekly by two times the hours worked per week, monthly salaries by (52/12) times the hours worked per week, and yearly salaries by 52 times the hours worked per week. converted to an hourly wage by dividing the salai^ by hours worked per week. salaries We did not create the wage variable for individuals whose calculated wages were extremely low or high wages. We considered a calculated wage to be too low if was less than half the federal minimum wage in 1986 ($3.35). We also rejected as an outlier one individual whose calculated wage was more than $14,000 per hour; all other calculated wages were less than $1,000 per hour. We converted wages to it 1982-1984 dollars by dividing by the 1986 CPI-U. Family income In 1972, students were asked, "What is the approximate guardian)? Include taxable and non-taxable from • (or composite father's education and composite mother's education variables in the 1974 data. Test scores The combined SAT score composite • income before taxes of your parents sources." Parents' education We used the • all is the sum of the reported SAT math and verbal scores. The data include the ACT score. High school grades In 1972, students were asked, "Which of the following best describes your grades so far in high school?" The options were: 1) mostly A, 2) about half A and half B, 3) mostly B, and so on up to 7) mostly D, and 8) mostly below D. Call the reported number X. We approximated the student's grade point average on a four-point scale by (9-X)/2, which works out to 4.0 for students reporting mostly A's, 3.0 for students reporting mostly B's, and so on. • The Rural variable is one if the student reported living in a mral or fanning community. 31 • South The variable is one • Private high school The variable is one student's reported region of residence in high school if the if was in the South. the student's high school principal or counselor described the school as private or Catholic. • Post-secondary education Respondents were asked, "As of the first week of Febraary 1986, what v/as your highest level of education?" In addition, the data include transcripts from most post-secondary institutions that respondents attended. We considered a respondent to have a bachelor's degree if the person reported completing college or receiving a master's degree, doctorate or advanced professional degree, or the transcripts showed a bachelor's degree. We if any of considered a person to have received a graduate degree if the person reported receiving a master's degree, doctorate or advanced professional degree, or if any of the transcripts showed such To detennine whether a degree. a person had attended graduate school, on post-secondaiy attendance that we had we began with the self-reported information used to identify the fust college attended. Respondents were were you studying for?" We considered the any time, he or she reported studying for a master's degree, a doctorate or an advanced professional degree. In addition, any respondent who never reported studying for a graduate degi'ee, but whose transcripts showed such a degree, was coded as having attended asked, "What kind of certificate, license, diploma, or degree respondent to have attended graduate school if, at graduate school. • College majors We considered the student's major for the the student's transcripts. first The data include 661 bachelor's degree received, based on the dates six-digit codes for college majors. shown on We categorized any the 74 codes beginning with 06, 07 and 08 as a business major; any of the 53 codes beginning with 14, 27 and 40 as a major in of 1 1, computer science, mathematics, physical science or engineering; and any of the 108 codes beginning with 17, 18 and 26, plus code 190503, as a major in health or biological science. The categorization was chosen to match as nearly as possible the much less specific categories used in Baccalaureate and Beyond data. If a student received two bachelor's degrees on the same date, we counted the student as having a major in a particular field if either degree showed a major in that field. • Utility the measures were asked "how do you feel about" the following statements: "I take a positive "I feel I am a person of worth, on an equal plane with others"; and "I am able to do things as well as most other people." Respondents could say they agreed strongly, agreed, disagreed, disagreed strongly, or had no opinion. We converted these questions to binai^y variables by coding "agree strongly" and "agree" as a positive response, "disagree" and "disagree strongly" as a negative response, and "no opinion" as missing. In 1986, respondents attitude • toward myself; Weights 32 We used the • B&B • Race nonresponse-adjusted weight for participants in the 1986 fifth follow-up survey. 1997 Respondents were asked selection in the 1 their race in the 994 survey. We 1993 survey; they were then asked to confirm or correct limited the sample to students who this identified themselves as black in the 1994 survey. • Sex Respondents were asked gender when their it was not obvious to the interviewer. • HBCU attendance We investigated the effect of receiving a bachelor's degree from a indicating whether the student received a bachelor's degree from a HBCU. The data contain a variable HBCU, transferred from a HBCU to variable is missing either if the student is known never to have no information on whether the student ever attended a HBCU. We checked this variable by merging the school codes for bachelor's degree institutions to our list of HBCUs. Among black students, the only discrepancies were in 10 cases where the HBCU variable in the data was missing but our list showed that the school was a HBCU. We used the results from merging tlie school codes to The the degree institution, or both. attended a HBCU or if there is our list. • Wages and employment We divided the annual salary recorded for the respondent's April 1997 job by 52 times the average status number of hours worked per week. The data define ftiU-time workers as non-teachers who worked at least 30 hours per week and teachers who described themselves as working full time. We rejected as outliers three cases where the calculated hourly wage was less than half the federal minimum wage in 1997 ($5.15). We found no implausibly high wages. We converted wages to 1982-1984 dollars by dividing by the 1997CPI-U. Family income We used family income students who were in 1991 and converted to 1972 dollars by dividing by the annual CPI-U. For their parents' dependents in 1991, total family taxed and untaxed income was obtained, in order of priority, from the student's financial aid application, a telephone interview of parents, a telephone interview of the student, the student's Pell grant students who were not their parents' dependents, the variable income. This was obtained, in file, or the student loan the student's own file. For taxed and untaxed order of priority, from the financial aid application, the student's phone interview, the student's Pell grant • is file, or the student loan file. Parents' education The data contain variables from 1994 and 1997 indicating each parent's highest level of educafion. The only cases in which these variables disagree are where one of them is missing. We combined them. 33 • Test scores The data list • South The variable combined is one SAT scores and composite if the parents' state ACT scores. of legal residence was Alabama, Arkansas, Delaware, the District of Columbia, Florida, Georgia, Kentucky, Louisiana, Maryland, Mississippi, North Carolina, Oklahoma, South Carolina, Tennessee, Texas, Virginia or West Virginia. • Private high school The variable is one if the student said he or she graduated from a Catholic, other religious, or non- religious private high school. • Post-secondary education All students in this dataset received a bachelor's degree. The data contains variables listing the highest degree a student reported receiving after the bachelor's and the highest-level degree program student enrolled after receiving the bachelor's. doctorate, first professional, post-master's certificate, certificate or license. program We classify the student as if the variable lists a doctorate, first master's. We The degrees verified that all students who that MBA, in can be listed in these variables which the are: master's, bachelor's, associate's, and other having received a graduate degree or attended a graduate professional degree, post-master's certificate, MBA or received graduate degrees are listed as having attended a graduate program. • The College majors dataset contains a composite variable indicating the student's major that surveys, transcripts and financial aid records. This variable a physical science major if this variable lists lists when We record the major as engineering or "mathematics and other sciences," which consist of computer science and physical sciences. science major was compiled from student 12 broad categories of majors. We record a health and biological the variable indicates a major in either "health professions" or "biological science." We record a business major when the variable indicates "business and • Weights We weighted by the nonresponse-adjusted weight for respondents • College and Beyond • Race The institutional data files contain to students listed as black in either two variables indicating of these variables. 34 management." to the 1997 survey. the student's ethnicity. We limited the sample Sex • From the institutional data. HBCU attendance • The four HBCUs in the College and Beyond sample are Howard University, Morehouse College, Spelman College and Xavier University of Louisiana. Wages and employment • Respondents reported status their annual pre-tax earned income, including income from jobs; net income from businesses, farms or rent; pensions; and Social Security. They were instructed not to include income from dividends and interest and not to count the income of other family members. They reported income We take the midpoint of each bracket except the top bracket, which in 10 S200,000 or more. For the top bracket, we follow Dale and Kj'ueger (2002) and use the average income shown in the 1990 census for college graduates in the appropriate cohort who earned more than $200,000 per year. We included all individuals with reported income in our regressions. categories. We is employment as an outcome. Respondents were asked, "Are you currently answered yes, they were asked whether they were working "only or mostly full-time," "some full-time and some part-time," or "only or mostly part-time." We coded only the first of these categories as full-time employment. also considered full-time working for pay or profit?" If they Family income We used the family income variable from the HERI student survey; this is a categorical variable. We converted to 1972 dollars using the CPI-U. Parents' education • When a parent's education is listed as "attended or graduated from college (can't tell)", we categorized the parent as having less than a bachelor's degree. Test scores • We used SAT scores from the Educational Testing Service file and ACT scores from the institutional data. • High school grades We obtained high school grades from the HERI student survey. This variable lists the student's average A or A-t-. We treat D as a GPA of 1, C as a GPA a GPA of 3, B-^ as a GPA of 3.33, A- as a GPA of high school grade as either D, C, C+, B-, B, B-h, A-, and of 2, 3.67, • C+ GPA of 2.33, B- as a GPA A or A-^ as a GPA of 4. as a and of 2.67, B as South variable is one if, when he or she applied to college, the student lived in Alabama, Arkansas, Delaware, the District of Columbia, Florida, Georgia, Kentucky, Louisiana, Maryland, Mississippi, North The 35 Carolina, Oklahoma, South Carolina, Tennessee, Texas, Virginia or • Private high school The variable • Post-secondary education is one West Virginia. student attended a parochial or private high school. if the Respondents were asked in the follow-up sui"vey whether they had received a bachelor's degree. For each of more than 20 kinds of graduate and professional degrees, respondents were asked whether they had studied for the degree and whether they had actually received it. We combined these responses into binary variables indicating whether the respondent studied for or received any graduate degree. College majors • The data list up to two majors for each student. We recorded a major in physical science if either major was engineering, computer and infonnation sciences, mathematics, astronomy/atmospheric sciences, chemisti'y, geological sciences, physics, or other physical sciences. We recorded a major in health or was biological sciences, pre-med, nursing, dentistry, or health business if either major was business and management. biological sciences if either major We recorded a major In 1989, five total in major codes — 345, 734, 905, 920 and 980 - appear in the data but not in the codebook. We assumed none of these codes represented a Development as an at all," how much field. undergraduate Respondents indicated on a one-to-five "not A of 17 black students have one of these codes. physical science, a health or biological science, or a business • sciences. scale, where five represented "a great deal" their undergraduate experiences helped them develop in and one represented each of 15 areas. We kept these variables in continuous fonn. • Job satisfaction Respondents were asked about 13 aspects of job satisfaction. In each categoi'y, they could indicate they were "vei^ satisfied," "somewhat satisfied," or "not satisfied." We converted these variables to binary forni by coding "vei^y satisfied" as one and "somewhat" or "not satisfied" as zero. We examined each of the 13 binary variables separately. variables and • Utility examined We also computed the first principal component of the 13 binaiy the resulting continuous variable. measures Respondents were asked: "In general, how satisfied would you say you are with your life right now?" and "Overall, how which you satisfied first have you been with the undergraduate education you received at the school at enrolled?" The possible responses were "vei7 satisfied," "somewhat satisfied," "neither satisfied nor dissatisfied," "somewhat dissatisfied," and "very dissatisfied." We converted these to binai-y variables by coding "veiy" and "somewhat satisfied" as one and the remaining categories as zero. 36 Respondents were also asked, "Imagine that you had your life to live over again and were graduating from high school. Knowing what you do now, how likely is it that you would choose the same undergraduate school?" Possible responses were "veiy likely," "somewhat likely," and "not converted this to a likely. We binary variable by coding "very" and "somewhat likely" as one and "not likely" as zero. Academic outcomes index The first principal component of the dummy variables for majoring in business, majoring in physical science/mathematics/computer science/engineering, majoring in biological science/health, receiving a bachelor's degree, attending graduate school, and receiving a graduate degree. Subjective academic index The component of the dummy variable for being satisfied with the undergraduate duinmy variable for being likely to choose the same undergraduate school again; the variable for reporting that anyone took a special interest in one's work at the undergraduate first principal education; the dummy school; a variable that rates on a one-to-five scale how much the undergraduate education helped develop and a variable that rates on a one-to-five scale how much the undergraduate education helped develop knowledge of a particular field. analytical and problem-solving skills; Subjective labor market index first principal component of duminy variables for whether the person reported being vei7 satisfied with the following characteristics of the cuaent job: intellectual challenge, flexible schedule, high level of The responsibility, low stress, working environment, job security, employer-supported child care, fair treatment of women and minorities, high income, good benefits, promotion cpportunides, and service to society. Leadership/lifestyle index The first principal component of variables rating, on a one-to-five education helped develop each of the following: leadership scale, how much the undergraduate community sei'vice, work cooperatively, ability abilities, interest in religious values, ability to cominunicate orally, competitiveness, ability to to relax or enjoy leisure, ability to write clearly and effectively, and ability to adapt to change. Social interactions index The first principal component of variables rating, on a one-to-five scale, how much the undergraduate education helped develop each of the following: ability to fonn and retain friendships, ability rapport with people of different beliefs, and ability to • Other outcome variables The other outcome variables • Weights we work and to have get along with people of different races. studied are binary in the data. Because the College and Beyond surveys included 100% of minority students is no need to use weights. 37 at the sample schools, there 38 1 . s . ^ ° ^"^^o S P ^T-r^Tmifiin-a-ocn STcDr^incNjoco ^r^03in--^'3-tDOT-OI^ iOT--.-cNLnino^'Tcomt^ j-_;r-^'^J-OCOCM'3-T- ^^dddddd dddt^dddddd "" 33" O s 1 ^8 S O ^—— 2 vz Z) ^"^•^03 I g"^o ddddoddoodob g „ So T-^CO-^COCNTj-U^ OQOrocNCNICOT-.0(DCOOtO T-2-o^ <-i-U3COCDCOOt—-cNicMc^'T'^co — oicnTj-coh-TrT-vrr^roo>i^ ^^dddddd ^ r^ ;= C= S; a r^ n ° d dddddddddddd mC-CNrl-h-rfCOCn ,(OCN-^T-_tD-^tOCNIO>h- gcNidddddd ' ^ r«- d So" — Ss2? g^^SSSSfeS^SS^ ^SgS SdoS5od§SS2S . co.x r*- cncDmiDu^eMT-t*-OTtD'* _1^ ^5 d dddddddddddd 1 £ csi m "° "^ £ [\ D '^^cniounor-o -r ^^dddddf-^ d c2,oa3-3-'Ta>cno,io, ^ cri , , , , , , , , 5 5^ og " CTi_-; m ^Sx fN o d S °^minr»T-ujo -* cnocnfv.T-tDmo,t- ^tDrrmiTicNr— ^focvjino) CNJtriN.COCOr-'a-COP-.CTJC'^O, rv,CNi^-t-r5incN^in^cMr-' (N CD r-S — rr Q) a coi-inrococn^cDcocD-^Tr-in^^cocor-^ocooomo gSgd o) „ " q CO rn o Tcnf;^ooooof-ooo *-_ tNj cji ^ en ^ ^ LD ^Sd^tn cncocNi^T-r--c\ior>-^T-f^cD 1" g^gd ado(DciQd>a':D'd(6':ic3 5 tri^^-^I^T-CNrOO'3-CNO ^oifii^mcncoo^cDtD. — — ooT-'q-T-orMr-CD' ^j^OOOOO^OOO ^ , . ' ' . , , . ' ' ' ' s "^ ^ tsdg SS?SS§§8SS§S§ ^"gd ddddddddddddd 2^:;°°°°°°°°° 0) " ^ S 3 "" ^ ^ CN ro g *" » ^ ^ °Dcj 111 h^ r^ [c-O'3-ocNj'^rsjoincncN d dd d d d d d d d d d d ^ § " *° ^ r^ <- T-Sooo — -—o g g ^ doddddddoddd ^ s r^ ^si^ - 3 ipddddo dod|dodddd gs^ ' en 1- ^ ST" , P) S" CO in CO SS ' r^ -fi i^'::ic6'^C3cic5cicic6 m ^ ^"--. I --^ g IN T- CMtDoioincDinaiT-rjtDtD oco^cncootD(N'.-r---r-co t£)O^CNintOCN^CDCSr-CM ^ Sl^ ss S —. gr- ^ "^ 5 h-_V.fM t-Sot-; — — -S c>':6<::>^c>ci<^c:icic> 3" S ^ " i ?J r^ OOCO'^CNi"<rr*--r-COa3 ^COT-Ln->-r>mCNjT-T-CD *tnooco— LncDf^r^cocD s § ^ ro ml !X 2 = u - ?^" ^— X f; 5g2 S'— T- • . r~ — ^ooT-fn-t-toojcor^ ddddddddddddd S g d ' ' ^'^c>c><6cicicic:>c>(6 CN CN S^'^ „ S 0) = 1 - 2 s s m T3 C D 1 ^ m ^ te «° o o o S V -^ -^ 3 1 s O o - 1 < 3 < 9 1^ o S ° o m e e 8 S ^ E = < o o) Sa<S ^ < Z ? S.Q. w w 0^^| s iLlilii . gsli ft iiii^ liiiiljii » ° E B s s II > > 1 1 1 1 1 Sis D X i 1 i f 1 i 1 III li llllii ? S i i o e r e E e e III § m » g s » S 1 8 £ EEEEE5££ooog2 1 ,cca.3i^S£5S"sSiaa2.2.D.m ^ro^^ro^raraEEE^B z z fc >. s; (5 3 g I E m 2 a > w M « o o 5 ! 1 ? s S m 1 1 •5 1 re "1 2 g- :i o S S Table 2: Determinants of Dataset HBCU HBCU Definition Among Attendance NLS ' first college SAT: >800 ACT: 11-15 ACT:>15 family income: family income: family income: $3000-$5999 $6000-$8999 >=$9000 father's education: bachelor's father's education: >bachelor's mother's education: bachelor's mother's education: >bachelor's south private high school female ObsePv/ations pseudo R-squared in NLS and BB BB (1972) bachelor's degree college (1997) bachelor's degree college (1) 12] (3) -0.170 -0.012 -0.032 (0.072) (0.112) (0.128) -0.320 -0.226 -0.067 (0.049) (0.081) (0.120) ., SAT: 600-800 Blacks NLS (1972) -0.116 -0.306 0.130 (0.112) (0.058) (0.200) -0.213 -0.296 -0.167 (0.083) (0.062) (0.109) 0.018 0.010 0.064 (0.079) (0.117) (0.089) -0.069 -0.174 0.073 (0.077) (0.092) (0.099) 0.062 -0.014 -0.058 (0.088) (0.113) (0.063) -0,031 -0.154 0.165 (0.144) (0.130) (0.087) -0.130 -0.318 0.127 (0.112) (0.036) (0.110) 0.097 0.230 0.144 (0.145) (0.182) (0.086) 0.082 0.035 0.280 (0.133) (0.156) (0.115) 0.429 0.473 0.126 (0.049) (0.060) (0.054) -0.065 0.099 -0.139 (0.132) (0.153) (0.059) 0.033 0.024 0.047 (0.053) (0.069) (0.057) 496 308 424 24 0.31 0.15 Notes: The tables reports the results from the estimation of probits for the probability of enrolling at a HBCU (column (noted in 1 ) and graduating from a HBCU (columns 2 and 3). All of the explanatory variables the row headings are indicators, so the entries report the discrete change associated with a change in the probability one and the associated heteroskedastic-consistent standard error. (The standard errors in the columns (1) and (2) are clustered by high school.) The omitted categories are: SAT<600, ACT<1 1, family income <$3000, father's education < bachelor's, mother's education < bachelor's. The models also include dummy variables for each category whose value is one when the categorical variable is missing. The samples are limited to Blacks with measured 1986 wages (NLS) or measured 1997 wages (B&B). In the B&B, 2 observations were dropped because a regressor perfectly predicted the outcome. See the text for further details. in value from zero to ; J ID i^ v> O CD ' t: 2rco CD T3 en Q £ .9 5 (D £ CO "2 O 9 j;? n 00 CM o r-^ * O CO CO * CD CO '^ rsi ly: o c=i CO -5, E 5 0) o o CD .^ a> '-' =1 S CD .9 2 D) o E o jz C/3 o .; > O o E ? — IS 2 >- CD <u ro en in en CX3 en CM en CO en J^ _ CO "^ en £5 CD J3 &• O ro o en CD O) CO TO in "5 CO => CD z: c O ^ CO o -E, D O CO I.U '-' § £ '^ CD .7. o < < H en < w S b^ E ro iz: L_ ,S in " 0) - <D 03 . en e" 9J (D fli ID F — 4 CO .ro -D C CD ii 2 E , -d E o ^ — -o-— g-l^ o-i CD I.U ^ Q) Q..!5 CD CL (/I -a CO o o p ° "cD c " ro CD CO 2 -c "cO (U C/) Q. Q^ £ .y CO 9 d o ^ _O = ^ S h_ CM en CO en o Q. -a > o c to -S .o) "O CD CD CO CD CO CM CM o X3 (u 5 CO o o8 ^c ci O CD Z3 siO ^u CO CD T- g 4S< S C\l CM CO <— a T3 -2 ^ E 5 en <D cn c W CD "5^ ,- £ a? £ o •.^ ii " C UJ b E T3 CD t= "2 O CD E E P C CP i; a: u LO CO CO CD C\J CM o > 3 CD CO "^ £^ f° 5 0) q: ro c O 3 O o o 9 ° O O ^ ^ T o CD CD in t 3 S ^ ^ 1 C3) h~ CO MCM O o O CO C in c E D CD I P a. Q. E to en > cn CD 'o it o o "D O OQ I 3 — O ^°° 5 S^t CO =J 0) CD JD JD _] CD Z) "c a m X SI _l 'o "(5 0) o O o CD -^ CO en (D to CD Z) E £ m D CT U) tr b b J3 In r) S m z z Z X E E D 5 (D >. oj o en O « ±: "o CO != (D (D -O CD P CD b -^ 2 C3. CO c en CD ^ .en D Scn cn CO g O) Q. J3 -?? ,n m CO £ g O (D in ro g 0 CO (D "•' ci -Q Q. s t: o S ^-2 ID E j::^ en u .^ cz ;^ Q- o E b. c c c 9? .f? in E CO g b in 2 N ^ZJ t: c CO > = E ™ ^ N s a? c cn c S-t 5; E Q. c Q) D cn g E D '!! c d) c en e b CD I « "> g 0) in b CD ID CO E i5 en CD ID "en CD 03 -0 CT D "5 ID 0) CD ro ,c 0) \_ (U Z o CO z z CQ I £ E m 'P 3 Zl q: o ^ E -Q en c -— a> CL 5f|5 ~ I •" ID JZ C iJ 1- ro CO Z CD 1 CO CO §^ CM CD CO CO CD CM CM cr P E C .— un pr <^ c -"2 -S S en to -Q 8| -C OQ c <D en fS CJ -2 t: <D -O O cr> 03 Oi Q- 0) E £ "S oj O o 0^ 0) '^ — CQ CD ooooooPoRo OCDCDOCDCJOoOo CD -— QJ CO "O ^^ C2. a; _(/) O en CD 0) i: -C o 03 03 J3 -^ 5 Q) >> H c 1- o o TO 2 f^ Q> a> TD ^ CDOqcDOCDOOCDO CNJ ^^ CDOCDOOCDCDCJOCD 0> 2 CO ^ O TD a> o •S 03 Z3 ^' C35 O O c .i g ^ ^ CO ~ - Ei; OOCDCOCMCDOCO-*J2 cr>-:]-co^^roLnro2 * cDOCDCDpqqoO OCDOOCDC3C300 o. h Q. n^ H 2 H W £ a> oj "" CO '" H g CO "o -S i =3 o ->- 03 "3 o TO 5 sz E c O o o SZ o TO c-J CO UJ "O o Q. .5 -c ~ 03 o X3 CO 03 >. -C 00 3 ^ TO > o o 03 o F C (U o " ^ ^ I- -o CO CO 2 o ^ E E w " 03 TD !_ -o .2 Q- "co -' « ^ - 03 > n u ^ m O "^ 2 o _ CM Cl) Q. Q. S ^ O ^ ^ O 03 r c — CO 03 .»-< —' TO CO CD C O O O C CD _2 QT3 O in _i "2 ^^^ o 3 E _(/) s: CM CO CU CO ^ 00 en en ^(U Uj _i CL i^ Q- o o m o E 8 S? CD O) _gj CM r- a. o) 2 n E o rod) CO r 0) -i_ ^^ I ™ r-- O) OT ^^ TO 05 B £ en CO c o 1 C 1 •0 c 03 c o c CO 03 c o o c .° E c o -+-' 'c S^ ^ i2 ro 3 O T3 ^ 03 E XI "5 3 ^-^R 0) Q ID O To CQ Q X 5 ro .^ 0) CO CQ J= < 03 O Q. -Q £ c to CO 3 03 1 o ^ 15 0" c E "co o o q: O c 'q_ --2 s"5 0000 O O CO 3 03 -Q CO . Table 5: Determinants of HBCU Attendance Dataset C&B HBCU first Definition Include Howard, Spelman? SAT: 600-800 SAT: >800 ACT: 11-15 ACT:>15 family income: family income: family income: >=$9000 father's education: bachelor's father's education: >bachelor's college first C&B (1976) college C&B (1989) college first no no (2] (3) -0.138 -0.620 (0.074) (0.035) (0.021) -0.584 -0.512 -0.998 (0.056) (0.075) (0.001) -0.315 -0.099 -0.124 (0.124) (0.052) (0.180) -0.489 -0.167 -0.336 (0.039) (0.025) (0.108) 0.005 -0.013 -0.089 (0.067) (0.056) (0.106) -0.059 -0.091 -0.264 (0.064) (0.035) (0.068) -0.033 -0.105 -0.271 (0.060) (0.042) (0.073) -0.064 0.152 0.011 (0.053) (0.062) (0.063) -0.042 -0.085 -0.065 (0.068) (0.046) (0.070) 0.034 0.097 0.144 (0.052) (0.055) (0.068) 0.101 -0.113 0.316 (0.068) (0.039) (0.070) mother's education: >bachelor's 0.350 0.357 0.327 (0.027) (0.035) (0.034) south female in C&B m -0.212 mother's education: bachelor's private high school Blacks yes $3000-$5999 $6000-$8999 Among (1976) -0.051 0.093 -0.031 (0.032) (0.033) (0.039) -0.077 -0.337 -0.343 (0.028) (0.034) (0.034) Observations 1982 1152 1414 pseudo R-squared 0.33 0.55 0.48 Notes: The tables reports the results from the estimation of probits for the probability of enrolling at a samples HBCU among of Blacks. All the C&B 1976 (columns and 1 of the explanatory variables (noted 2) in and C&B 1989 (column 3) the row headings) are indicators, so the entries report the discrete change in the probability associated with a change in value from zero to one and the associated heteroskedastic-consistent standard error. The omitted categories are: SAT<600, ACT<1 1 family income <$3000, father's education < bachelor's, mother's education < bachelor's. The models also include dummy variables for each category whose value is one when the categorical variable is missing. The samples are limited to individuals with measured 1995 annual income (C&B 1976) or measured 1996 annual income (C&B 1989). In the C&B 1976 sample without Howard and Spelman colleges, 202 observations were dropped because the regressors perfectly predicted the outcome. In C&B 1989, 345 observations were dropped because the , regressors perfectly predicted the outcome. See the text for further details. Table 6: Effects of HBCU Attendance on Labor Market and Other Outcomes C&B Dataset HBCU Definition Include Howard, first Spelman? (1976) college C&B first (1976) college employed full time subjective labor market index C&B first (1989) college m (1) Labor Market Outcomes -0.074 C&B no yes Ln (Annual Income) in -0.032 -0.114 (0.069) (0.145) (0.160) 963/1019 611/711 0.099 335/468 0.513 (0.112) (0.363) (0.202) 110/123 24/10 100/109 0.059 0.104 0.032 (0.061) (0.164) (0.123) 972/1135 339/532 550/655 -0.181 Life Satisfaction satisfied with college was first -0.046 -0.013 (0.026) (0.044) (0.044) 977/1137 341/536 621/716 -0.009 life choice would choose same college again leadership/lifestyle index social interactions index college developed ability to form and retain friendships college developed ability to have rapport w/ people of college developed ability to get along with other races different beliefs -0.084 -0.079 -0.063 (0.034) (0.079) (0.058) 932/1108 314/545 0.219 613/712 0.177 (0.035) (0.086) (0.054) 981/1132 340/534 0.417 0.633 622/717 0.152 -0.048 (0.069) (0.157) (0.109) 977/1132 618/717 0.220 341/530 0.357 (0.071) (0.169) (0.123) 976/1131 341/530 0.423 0.530 619/715 0.129 -0.177 (0.088) (0.142) (0.124) 966/1100 337/520 0.386 613/710 0.254 (0.088) (0.205) (0.116) 966/1121 0.034 337/526 0.255 614/709 (0.093) (0.211) (0.170) 961/1124 334/525 613/713 -0.220 -0.445 Academic Outcomes received bachelor's degree attended graduate school received graduate degree academic outcomes index subjective academic index physical science major biological science major -0.105 -0.047 -0.105 (0.025) (0.068) (0.037) 978/1138 339/539 623/718 -0.101 -0.116 -0.019 (0.033) (0.086) (0.059) 983/1142 342/537 623/718 -0.136 -0.149 -0.007 (0.033) (0.082) (0.057) 983/1142 342/537 623/718 -0.301 -0.296 -0.123 (0.066) (0.185) (0.110) 983/1142 342/537 0.356 623/718 0.165 (0.072) (0.204) (0.096) 982/1138 341/539 623/718 -0.027 -0.107 0.074 (0.028) (0.080) (0.042) 784/771 0.087 269/459 0.120 552/636 0.196 -0.201 business major (0.026) (0.051) (0.040) 784/771 269/459 0.234 552/636 0.111 0.214 (0.021) (0.030) (0.028) 784/771 269/459 552/636 Other fraction black in zip code 0.022 0.081 0.156 (0.025) (0.074) (0.041) 902/669 315/350 602/593 Participation political religious civil rig hits social service alumni national cliarity 0.019 -0.032 0.216 (0.032) (0.078) (0.065) 983/1142 342/537 0.039 0.085 623/718 0.108 (0.036) (0.087) (0.062) 983/1142 342/537 623/718 0.004 0.071 0.105 (0.035) (0.081) (0.063) 983/1142 342/537 623/718 0.116 -0.024 -0.059 (0.029) (0.069) (0.063) 983/1142 623/718 -0.007 342/537 0.000 (0.036) (0.086) (0.056) 983/1142 342/537 623/718 -0.098 -0.168 0.148 (0.038) (0.079) (0.062) 983/1142 342/537 623/718 0.030 Notes: The entries report the results from propensity score matching routines. The row headings report the dependent variables. The samples and treatment are noted in the column headings. The entries report the unweighted impact of HBCU matriculation (graduation), relative to TWI matriculation (graduation). The propensity score is estimated with a probit and all of the covariates used in Table 2 are included as explanatory variables. The matching method uses a gaussian kernel with a bandwidth of 0.1. The standard errors (reported in parentheses) are computed by bootstrapping, with propensity scores recomputed for each bootstrap sample. For each outcome, all observations with data on that outcome are used. Indexes are first principal components of sets of variables, normalized to have mean and variance 1. In constructing each index, we include every observation that has data on at least one variable in the set, by replacing any missing data with the mean of the corresponding variable. Numbers underneath standard error are number of HBCU and TWI students, respectively. Observations are dropped if the propensity score is not strictly between and 1. See the text and notes to Table 3 for more details. — S c — CO cQ "D 5; ra CD tn "' "D ?i '8 en ?5 I-' CD O) tn ?; en MX- Si ^ s ^ £ I « ^ -oQ.ilO0oboCU ^^ So? tn O E o en O " 9 c o Q. o So 05 CD p ^ -G -^ " .ii' ° £ -w i= . 5 iE ,- ? o S Q-.h="3m^ § 0) -o .5 = -o .£ ^ en ,E i^cuS.cn^gg-o o "'ISa'aii^ScTO = jr £ a> as o c o o V) CD CNI CM o Ln CNi d 1 1 CO o CD ^ O " X C35 o" CM 1 1 ^ s i5 0) en (D -a CM .^ -D CO Sod CO CX CD 03 CO 53 0) s s > > (35 O) to — O 0) J- o ^^ .2 i5 §-| /rt" > s >% ^ en q - ° S o ^Oo3CO<'3o°°S CO CD tCD d o <co 03 CJ) CD ,_ . CO in » < § g,"o .b CD ^^ (D T^ O "oT eu o' CD ^a r §-0 D.g £~. CO -1 ,n " Q> g O "t ^ O CD ^^ £ — „ ^O 03 --- en m 3; o o o g -o ofo ?S o ^ -o CO 0^ 0-C„^ £ D) '". c S2 Di o 9 -c o g CO C CD en ^ -Q <I> o "*- o " S J= E §|eDEp2cD-§,t E = O O id o g £: -o 5 m en 2 CO •o •a 0) 0) 0) u o c ra ra _ m aj Qi CD -S S J3 a, p-£ £ c ^ ? S E I c £ O) ^-oiw^ii-So-sC. c ^— coco c co^ ra en -ri eu 03 - E|S&>™g<D«'. -Sooc-sEco-cc -Q>i=C03™=>".9. ^, CD„(D-D£enp£en 2 5 So-- -coSl S.g|2. 1— CD go -fe -S -5 .5 fe O ^ ?) 03 o 5 I 13- o Q a: °iT<^"Sgeu=co Z £ ^ g O - c C 1 l CO c o o roJ=EStS° c O 5 5 ^ " S,£ S ^= CD -5 °? g £ S ^ en > ^ o :^ CO E X <'' 03 en 0) (D 5 ^ S t £ -S (0 Q. (D 0! ra , I Appendix Table 1: Summary Basic Statistics Nurriber College or University on the Historically Black Colleges and Universitie Location Normal, 2 Alabama A&M University Alabama State University 3 Albany State University Albany, 4 Alcorn State University Alcorn State, 1 AL Montgomery, AL GA 5 Allen University 6 Arl<ansas Baptist College 7 Benedict College 8 Bennett College MS Columbia, SC Little Rock, AR Columbia, SC Greensboro, NC 9 Bethune-Cookman College Daytona Beach, FL 10 Bluefield State College Bluefield, 11 Bowie State University Bowie, 12 Central State University Wilberforce, WV MD OH Cheyney, PA Orangeburg, SC Atlanta, GA 13 Cheyney 14 Claflin University 15 Clark Atlanta University 16 Coppin State College Delaware State University Baltimore, 17 18 Dillard University New University of Pennsylvania Dover, MD DE LA Orleans, 19 Edward Waters College Jacksonville, 20 Elizabeth! City State University Elizabeth City, 21 Fayetteville State University Fayetteville, 22 23 24 25 26 27 28 29 30 Fisk University Nashville, Florida A&IVI University Tallahassee, FL Florida IVIemorial University Miami Gardens, FL Type Undergraduate Total Enrollment Enrollment 5,047 6,182 Public 4,485 5,469 Public 3,228 3,649 Public 2,962 3,544 Public 624 287 624 287 Private 2,552 2,552 572 Private 572 3,090 3,090 Private Public Private Private 1,708 1,708 4,020 5,319 Public 1,617 1,623 Public 1,401 1,560 Public 1,678 1,728 Private 3,667 4,469 Private 3,451 4,306 Public 3,440 3,722 Public 1,993 1,993 Private Private FL 839 839 NC NC TN 2,604 2,664 5,029 6,072 Public 864 920 Private 10,552 12,154 Public 1,945 2,004 Private 1,997 2,174 GA Public Fort Valley State University Fort Valley, Grambling State University Grambling, LA 4,573 5,164 Public Hampton University Hampton, VA 5,325 6,309 Private Harris-Stowe State University St. Louis, MO 1,662 1,662 Public Howard Washington, 7,164 10,930 Private Private University DC Public Huston-Tillotson College Auston, TX 31 Interdenominational Theological Center Atlanta, 32 33 34 35 36 37 38 39 40 Jackson State University Jackson, GA MS Jarvis Christian College TX Charlotte, NC Frankfort, KY Knoxville, TN Jackson, TN 1,404 1,404 2,228 2,386 Public 300 300 Private 1,213 1,213 Private Johnson C. Smith University Kentucky State University Knoxville College Lane College Langston University 675 Hawkins, 706 447 8,416 Public 572 572 Private Private Private Langston, OK 3,001 3,151 Public Lemoyne-Owen College Memphis, TN 809 809 Private Lincoln University Jefferson City, 2,953 3,180 1,714 2,278 Public 895 895 707 Private Private MO 41 Lincoln University Lincoln University, 42 43 44 45 46 47 48 49 50 Livingstone College Salisbury, Meharry Medical College Nashville, Miles College Fairfield, Mississippi Valley State University Itta Morehouse College Morehouse School of Medicine Morgan State University Morris Brown College Atlanta, Morris College Sunter, 51 Norfolk State University 52 53 54 55 56 57 58 59 60 North Carolina 61 - 6,660 A&T State University North Carolina Central University Oakwood College Atlanta, Public 3,029 3,029 Private - 272 Private 5,747 6,438 Public 66 863 66 863 Private MD SC VA University Rust College AL GA TX Rock, Prairie NC NC Huntsville, Little A&M AR View, TX MS Lawrenceville, VA Savannah, GA Holly Springs, Saint Paul's College Savannah State 1,758 3,165 Durham, Dallas, View 1,758 Greensboro, Philander Smith College University Private 2,748 Georgia Paul Quinn College - Public MS GA GA Baltimore, Norfolk, PA AL Bena, Augusta, Paine College Prairie NC TN Private 5,337 6,096 Public 9,735 11,103 Public 6,353 8,219 Public 1,751 1,751 Private 828 790 785 828 790 785 Private 5,702 7,912 Public 970 717 970 717 Private 2,975 3,091 Public Private Private Private Selma, AL Selma University 287 287 Private 2,565 2,762 Private 3,888 4,446 Public Baton Rouge, LA 8,969 10,364 Public New 2,824 3,647 Public 62 63 64 65 66 67 68 69 70 Stillman College Tuscaloosa, 71 Talladega College Talladega 72 73 74 75 76 77 78 79 80 Tennessee State University Texas College Texas Southern University Tyler, Tougaloo College Tougaloo, Shaw Southern University A&M Southern University at College New Orleans Orleans, Terrell, Spelman College Atlanta, Augustine's College Tuskegee University University of Arkansas SC Orangeburg, Southwestern Christian College St. NC Raleigh, University South Carolina State University LA TX Raleigh, GA NC AL TN Nashville, TX Houston, TX IvIS Tuskegee, AL at Pine Bluff Pine Shore Columbia Bluff, AR University of Maryland Eastern Princess Anne, University of the District of Washington, MD DC 251 251 Private 2,318 2,318 Private 1,163 1,163 Private 804 368 804 368 Private 7,036 8,880 Public 794 807 Private 9,760 11,903 Public 933 933 Private 2,510 2,880 Private 3,132 3,231 Public 3,448 3,870 Public 5,169 5,363 Public Public Private University of the Virgin Islands Charlotte Amalle, VI 2,185 2,392 81 Virginia State University Petersburg, 4,332 5,055 Public 82 Virginia 1,344 1,700 Private 83 84 Virginia University of Private Voorhees College VA Richmond, VA Lynchburg, VA Denmark, SC 85 86 87 88 89 West Institute, Union University Lynchburg Virginia State University Wilberforce, Wilberforce University TX Marshall, WInston-Salem State University WInston-Salem, Xavier University New Notes: The is OH Wiley College list of schools is taken from the US Department http ://www. ed.gov/about/inits/list/whhbcu/edlite- list.html. 2005) WV from the National Center for Education Orleans, NC LA of Education's The web location data 122 709 178 709 3,455 3,491 Public 1,157 1,170 Private Private Private 827 827 5,264 5,566 Public 2,343 3,091 Private site: and enrollment Information (current as Statlstlcs's site: http://nces.ed.gov. : of Fall CD O CD -^ ^ ~ C/) 9 o >, J= T 51 o ^ (n o to b "> o -a ro "5 > > 13 o -5 SS I- ra .a § _ =o 0) ui _ , Is CO 1 ° S g ° s I s s £- .9 • 2 ° S •O CD ri ™ c lis E i2 S ° a. CO —tn § ,^ -a £ -H Q. '5 —> CO enc -^„ S -oc g E ro If III 0) .^ £ Ul CO M E -C = a> 0) "D E o o Td b <o t: C3> h- ^ o o ^ 2| b b CO CD Ol CD ai CO --f T- CM ^ ^ csi p '^ £ 3 . w 1 "> ^ ° § § a o (0 to 2 CD -o ca O) " 2 Q. i^ • m £ o en I g-EI g ° - ™ ?E w t5 m D. <s> 5) o cj " (3) JO "C S o o ° 21 Q b 0)1 i" 00 .& == (D C t C (0 ?> O 9^ S H £ .1= 0) -D If I Ol i= CO "- p ^ en CO 1) p E (!) tJ) Q i? -S ^ § § £ ~^ ? CL " S Z £ § m o " £ - <D 0) X g H S ^ CO E E Q F '^ m ra 0) E I g H ^ 1 3— O CD E E £ o w 1- ?? <D _a> CO E c JD -Q Zl o — (D 0) > E E o i CO 'ro S g §,-S .Q 13 ±= ,-C .5 >, io -a en 0) -D X) E E ^ " o m o != 0) o .£ ^ E O ro c p -o X Is >, CD o S m ™ -o c/) J: " W c) |||2 ^ CM ^^ ?^ CD :d CO >- ^3^ P 5 o ^ (B > ^ CD ™ m CL J) I I I CCJ .£ " -O CU O O C CD £ CO -t^ CD -^ § ? t - t0) O g ^ i^ § CD £ r L. 0) m T ^ -a T- ra 0) 5 3 c 1O) >» u. ^ 9-e 9'Z Z 9-1 Ai!SU9p CN c ™ = I- Ll. ^ c Q. CO •a 3 O DQ X re "D re E a> c T3 (D re "D C o X -t— -t— D) C 03 3 I ^- I re a CD c -t— 03 3 o J2 o to CO hOi Q 0) r E o o H- +-< >4— (/) (/) u iJ Q. >. 3 O re 1— (Ti s Q3 o CQ T V •oc iQ) CO 3 CT U. re 5 ^ n> 2 9" A^isuep I. J-^"*^ 1 / r \ K. 3 O CD X r - 00 / \ "D \ J "D / ^ C / \ \ / / / ^ \ CO - V 1 1 CD o o 1 if) \ \ \ ! - >^ / Y J c CD V / ^ c 1 / 1 / 03 1 - CM ^ ]^^ 1 ^...'-^''"'"^ \ y^^^ \ / \ I \ x^^ \ - ^''"'^""-^ ^"'""^"^-^ \ ^ II 91 \ ^""-^^.^^ oc re 1 9e 1 9"^ 1 S A^jsuep .It O 1 / ' 1 / -\ Q. 1 1 9- O 5 2! '59 Date Due ]ii!il;|lliilliillllilllli;iliil!i!lllll|!|il;!lJl!;illlll!lKiillliil!!ii!i;: iiiiiii«^^^^^^^