What Are The Returns to Attending a Non-Elite Private College or University?

Shreyasee Das University of Houston Scott A. Imberman1 University of Houston and NBER

March 6, 2011 Abstract: Consumers and governments have increasingly become concerned with the high cost of a college education. These concerns are especially large for students attending private colleges since sticker-price tuition far exceeds that for public schools. This leaves a question as to what benefit, if any, students get from these higher cost schools, particularly from non-elite colleges. To answer this we a use highly detailed and rich data set to assess whether there are benefits to attending private colleges over public ones. Through OLS regressions and propensity score matching using a large set of observable student characteristics we find that, conditional on observable school quality, males attending non-elite private colleges gain substantially in wages and earnings over those attending non-elite public colleges. However, we find no effect for females.

1 204 McElhinney Hall, Houston, TX 77204-5019. Correspondance should be made to Scott Imberman at [email protected]. Financial assistance for this project from the UCLA Center on Education Policy and Evaluation is greatly appreciated. We would like to thank Mark Duggan, Judy Hellerstein and Jeff Smith for their advice as well as participants at the American Education Finance Association Annual Meetings. Part of this project is from Scott Imberman’s dissertation conducted while at the University of Maryland.

1

Introduction

Average tuition, fees, room and board at a four-year public college or university was $14,870 in 2009-10. For a student attending a non-profit private university these cost more than double to $32,475. This large gap in costs between private and public colleges has been widening over the past 40 years. Figure 1 shows that the difference between public and private college tuition has increased from $7,467 to $17,605 from 1977 to 20102 . With these cost differences one would expect there to be significant benefits to attending a private college rather than a public college. However, little research has tried to quantify these returns or even to establish whether they actually exist. Indeed, while there is substantial evidence that private schooling is beneficial at the primary and secondary level (Altonji, Elder and Taber 2005, Grogger and Neal 2000, Rouse 1998, Neal 1997, Evans and Schwab 1995), it is not clear whether such a relationship holds as students move onto higher education, particularly at institutions that are not at the top of the college quality distribution - that is, non-elite colleges - as the little research there is on this topic focuses on the returns to attending elite private institutions (Brewer, Eide and Ehrenberg 1996, Brewer, Eide and Ehrenberg 1999, Dale and Krueger 2002) Understanding whether there are benefits to attending a non-elite private college has substantial policy implications. Large deficits are putting pressure on the Federal government to cut funding for financial aid. In the expectation that less Federal spending for higher education may be forthcoming, state governments will need to distribute their education funding more efficiently. One potential avenue to improve efficiency would be to divert funds from public colleges to financial aid programs. If this occurs, then more students would likely attend private colleges (Long 2004). This can be efficiency enhancing if attending a private college provides a higher return, in terms of the student’s labor market outcomes, per dollar of government funds spent. The fact that students are often willing to pay more to attend a non-elite private college than they would have to at a comparable public college suggests that private colleges are higher quality. However, there are some peculiarities in the higher education market that complicate this reasoning. One is that both supply-side and demand-side subsidies play a large role in this market. In FY2007 state and local governments spent $205 billion on higher education with the federal government providing another $25 billion, though some of the federal funding is already included in the state and local figure through intergovernmental grants3 . Hence tuition and related costs may reflect how much government support institutions receive in addition to being indicators of quality. 2 Digest

of Education Statistics, 2011 Abstract of the United States, 2011

3 Statistical

1

Another complication is that, while students select schools, schools also select students. Thus, not only do students need to agree to the price and the quality of education the school provides before they attend, but the school also has to approve of the quality of the students attending. Part of the reason for this is that students are both consumers of education and inputs in the educational process through mechanisms like peer effects (Lazear 2001, Rothschild and White 1995). Schools thus have an incentive to reduce prices overall to attract high quality students or they may charge lower prices to high quality students specifically through adjustment of institutional financial aid packages and through provision of scholarships (Singell 2002). Thus, we cannot use prices as an indicator of quality in higher education and hence, whether non-elite private colleges are better is an empirical question. However, empirical analysis also has complications due to selection of certain types of students into private schools. Unfortunately, a lack of natural experiments or feasible instruments makes these unobserved selection correction methods infeasible for our question. As such, in this paper we take advantage of the large set of control variables available in our data by providing both OLS and propensity score matching analyses. If conditional independence is satisfied - that is our control set is large enough to assume selection on observables - our analysis will provide causal effects. Nonetheless, we acknowledge the possibility that, despite our large set of covariates, there may be some characteristics of the selection process that we may not be able to sufficiently account for. If this is the case, we argue that any bias would likely be positive, allowing us to at the least provide upper-bound estimates of the returns. Our findings suggest that attending a non-elite private college with the same observed quality (proxied by mean SAT scores) of a comparable non-elite public college generates returns for males of a 14% increase in wages and an 18% increase in earnings. Using our preferred propensity-score matching model, however, our estimate of the average effects of treatment on the treated for females are statistically insignificant 5% for wages and -2% for earnings. Hence, while non-elite privates provide some benefits to males, they appear to provide little benefit to females.

2

Prior Literature The study of how private colleges affect labor market and educational outcomes is closely linked

to the literature on college quality. Since natural experiments and instruments in higher education are hard to find, most of this literature has relied on selection on observables techniques to study whether attending a college of higher observable quality increases earnings. Loury and Garman (1995) and Black and Smith (2005) control for large sets of observable characteristics and run

2

OLS analyses of the relationship between college quality and wages. Black and Smith (2004) use propensity score matching. Long (2008) and Black and Smith (2006) test a few strategies. All of these papers find that higher quality colleges increase wages. Only a handful of papers have addressed the returns to attending a private college specifically. The paper closest to ours is Brewer, Eide, and Ehrenberg (1999; hereafter BEE). BEE separate colleges into 6 groups based on ratings in Barron’s Profiles of American Colleges: Private top, middle, and bottom and public top, middle, and bottom. Using data from the National Longitudinal Study of the High School Class of 1972 (NLS72) and the High School and Beyond (HSB) study they find that attending a top private college increases wages and earnings over a bottom public college, although there is no significant difference between bottom privates and bottom publics. Our work differs from BEE in a few important ways. First, they focus on the differences in wage returns to attending an elite private college over other types of colleges. Our focus is on how attending a non-elite private college compares to a non-elite public. Second, their wage and earnings data are from 1986, at most 10 years after college graduation. Our data allows us to look at wages and earnings much later in life - up to 23 years after college completion - providing evidence of impacts on lifetime wages and avoiding the potential that some students would still be in school. Third, although they separate their set of colleges into broad categories they do not control for school quality in their models. We do this by controlling for mean SAT scores in a school (or ACT scores converted via concordance tables to the corresponding SAT score). This is potentially important as there is a potential for substantial variation within the categories. Additionally, from the point of view of a student deciding between private and public colleges the relevant comparison is between colleges with the same quality. Fourth, their empirical methodology is not able to address potential violations of common support across their college types, which Black and Smith (2004) find to be a very important factor in proper identification. Our propensity score matching analysis addresses the common-support issue. Finally, BEE do not separate their model by gender. A distinction that we find to be very important as the returns for males and females differ substantially. One possible reason for such a difference is that males may benefit more from non-academic aspects of attending a public college such as attaining jobs through networking. Another possibility is that males benefit from smaller class sizes than female students. Dale and Krueger (2002) also look at the return to attending an elite private college using an alternative selection correction model that uses students who applied to similar sets of colleges but where some are accepted and some are not to a higher quality school. They do not find significant returns to attending a more selective college. However, their sample is limited only to a small set 3

of elite schools and hence tells us little about the returns for students attending non-elite schools. We also note that Dale and Krueger repeat BEE’s analysis on the NLS72 data using their selection correction mechanism and do not find any return to attending an elite private college there as well.4

3

Data

The data for this paper comes the geocoded National Longitudinal Survey of Youth 1979. This is a restricted access, nationally representative survey of all persons aged 14-22 in 1979. After an initial interview in 1979, the survey follows people every year until 1994 and every 2 years thereafter. This survey is very useful in our context because it has a large set of pre-treatment observable characteristics and it has college identifiers that allow us to match students to the colleges they attend. We focus on the 2002 wave of the NLSY. While the dataset covers later years, due to attrition our sample size gets considerably smaller over time. Hence 2002 is the latest date of the survey that provides a large enough sample size to generate reasonably precise estimates. The initial survey contains 12,686 observations. Since we are looking at a specific subset of the NLSY population this number falls considerably to 4,595 when we restrict to people who ever attended a four-year college or university. Due to attrition, by 2002 our sample falls to 1,140 observations with earnings data and 1,223 observations with wage data. To account for the attrition, we weight all regressions using the NLSY sample weights for 2002. One problem in defining our sample is how to treat students who attend multiple colleges. We believe that the most reasonable strategy is to use whichever school the student last attended as this will capture sheepskin effects - the effect of the degree itself - for those students who complete college. We use a large set of control variables in our regressions. Nevertheless, there are several variables that warrant extra attention, especially those that serve as measures of student ability and motivation. The key motivation variable is the number of clubs the student participated in during high school. Arguably, a student who participates in more clubs in high school cares more about his or her education. The main student ability variable is his or her performance on the Armed Services Vocational Aptitude Battery. This is a test given by the US Armed Forces to new recruits. As part 4 Also see James, Alsalam, Conaty, and To (1989), who look at college quality using a small set of covariates, Bowman and Mehay (2002) who look at job performance and promotions of naval officers, Eide, Brewer, and Ehrenberg (1998) who look at the impact of college quality on graduate school attendance, and Brewer and Ehrenberg (1996) who followup BEE using the 1986 HSB Seniors cohort. All four papers find positive returns to attending an elite private college. Behrman, Rosenzweig, and Taubman (1996)(Behrman, Rosenzweig and Taubman 1996) consider an indicator for attending a private college when looking at the returns to different college qualities between female twins. Their estimate is strongly positive.

4

of a renorming process, the military gave the test to NLSY participants. The battery includes questions on 10 subjects. In addition to the ASVAB, the NLSY also has some information on the quality of the student’s high school which provide additional measures of students’ pre-college educational environment.5

4

Empirical Strategy

A large concern in the college quality literature that there is substantial selection of students into schools of differing quality based on ability and motivation. These concerns remain when considering the choice of whether to attend a public or private college. For example, it is possible that students (and parents) who take a more active interest in their (children’s) education may be willing to pay more for college, thus they would be more likely to attend private schools. The ideal way to solve the ability selection problem would be to find an instrument that is correlated with attending a private college but uncorrelated with other factors that are related to the outcome variables of interest. Unfortunately, such a strategy is infeasible in this context.6 Thus, in this paper, we are unable to fully account for unobserved ability bias. However, this does not mean that we cannot garner some accurate information from the estimates. First of all, we make use of very detailed data with unique variables. No work has been done on this question previously that takes into account as many potentially important control variables. One particularly useful variable is the number of extracurricular activities or clubs the students in which the students participated during high school. Arguably, more motivated students would participate in more clubs, thus I use this variable as a proxy for motivation. We also include a number of measures of the quality of the student’s family life as a child. In addition, we will make use of a propensity score matching analysis. Although this does not address unobserved omitted variable bias, it does ensure that - at least on observables - private students are only compared to similar public students; that is, that the sample has common support. Also, since we also match on average SAT score of the college we ensure that students are only 5 In our primary estimates we include the motivation measure described below, ASVAB scores, gender of the individual, mean SAT score of the college, race, religion, highest degree of parents, urbanization of residence at age 14, age in 1979, whether the individual is in the military in 1979, unemployment rate at current residence as of 2002, whether individual knew their parents, number of books at high school library, whether household member had library subscription, magazine subscription, newspaper subscription, drug usage of the individual, occupation of parents, whether individual is employed fulltime or in 2002 (except in employment regressions), and number of siblings. We do not include prior earnings as we only have the earnings of the individual available to us and this variable is zero for most of our sample. 6 One potential exclusion restriction is to use relative supply of public and private colleges near student residences. This is similar to the strategy used by Card (1995) and Kling (2001). Unfortunately this potential instrument ends up being too weak in our data.

5

compared to those who attend colleges with similar quality levels. Black and Smith (2004) use propensity score matching in an analysis of the impacts of college quality and show that it is superior to standard OLS techniques. Of course, even with all of these additional covariates and the matching strategy we may still be left with residual bias. However, there is a strong reason to believe that unobserved ability and motivation would be positively correlated with attending a non-elite private college relative to nonelite public. Later, we provide some evidence that supports this assertion. Thus, we interpret our estimates as upper bounds of the impact of attending a non-elite private college or university on labor market outcomes. Our baseline OLS model we will use for wages and earnings in this paper is as follows:

Yij = α + βP rivateij + δSATj + ΓXi + ij

(1)

where Yis an outcome variable for individual iwho attended school j. We focus on log of hourly wage and log of annual earnings,but also consider marital status, unemployment status, spousal earnings and educational attainment. Privateis an indicator for whether or not a student’s most recent school is a private school, and Xis a vector of individual, high school, and family characteristics. We estimate both pooled models and models separated by gender and whether the student attended an elite or non-elite college. We define the college’s status using the Barron’s Profiles of American Colleges, 1988 edition. Barron’s uses a 7 tiered system with colleges ranked, in descending order of competitiveness, as most competitive, very competitive, highly competitive, competitive, least competitive, non-competitive, and special. The last category is for non-traditional schools which we do not include in regressions that split the sample by elite status. We define our non-elite samples to be any schools with a competitive, least-competitive, or non-competitive ranking and the elite schools as any other ranking besides special. Our propensity score model is defined similarly. First we estimate the following probit model:

P r(P rivateij ) = Φ(α + δSATj + ΓXi + ij ).

(2)

We then match students attending private colleges to those attending public colleges using an epanechnikov kernel with a bandwidth of 0.06 and restricting to observations where the propensity score for a private student is lower than the maximum score for public students and larger than the minimum score for public students; that is we limit to common support.

6

5

Results

Table 1 provides summary statistics for select outcome and control measures by public and private college attendance. In terms of outcomes, we see that public students garner significantly lower wages and earnings. The average public college student earns $4 an hour and $9,386 a year less than the average private college student. We also see that public students are less likely to obtain a graduate or professional degree, but no less likely to finish college. Finally, we also note that public college students are less likely to be white, live in a town or city, and have had a college preparatory curriculum in high school. Table 2 provides the baseline pooled OLS results. In column (1) we provide an model with no controls. Without any conditions private students see wages and earnings that are 15% and 17% higher than public students. In column (2) we add a control for gender and see little change. In column (3) we add controls for the average SAT score at the college the student last attended. In this case the estimate drops substantially, suggesting that much of the difference in column (1) is due to the private schools being higher quality on observable dimensions. In columns (4) we add ASVAB scores, which surprisingly have little impact on the estimates. In column (5) we add the rest of the student controls listed above. While these do not appear to matter much for earnings, they reduce the estimate for wages by more than half to the point where the estimated wage returns are small and insignificant. Earnings returns are also insignificant, although the point estimate is larger at around 10%. Finally, in column (6) we add the student’s choice of major. NLSY provides detailed information on students’ majors separated into 26 categories. If students with different preferences for certain fields of study sort into public schools then we would expect the major choice to affect the estimates. On the other hand, public or private schools may be better in certain majors and hence push their students towards them. To the extent the former is true, this is a variable we would want to control for. To the extent the latter is true, however, then this is part of the return to attending a private college and controlling for majors would be inappropriate. Nonetheless, when we look at column (6) we find that including majors as a control makes little difference in the estimates. One important aspect of this table is it provides some evidence for our assertion that any bias that remains from unobservables is likely to be positive bias. This is because as we add variables to the regressions the estimates generally get smaller. This is clearer for wages rather than earnings, where the estimates increase slightly in columns (4) and (5). Nonetheless, the increase is very small to the point where one can argue that the impact of adding more variables beyond the school’s SAT score is negligible. Hence, if we assume that unobservable bias is in the same direction as observable

7

bias, then this model provides some evidence in support of positive rather than negative bias, if any bias remains. In Table 3 we provide estimates of the returns for earnings when we split the sample by gender and elite status of the colleges. Panel A provides estimates without controls for majors and panel B provides estimates with indicator variables for each major. Although we provide the estimates for elite schools for comparison purposes, the sample sizes are very small. Hence we focus on the pooled and non-elite estimates. The primary finding in this table is that there is a marginally significant impact on earnings for students attending non-elite private colleges compared to non-elite publics with the same SAT scores. Private students earn 17% more than publics if we do not control for majors and 16% if we do. These returns are similar for males and females. Table 4 repeats the analysis in Table 3 but for wages instead of earnings. In general there is no statistically significant effect of attending a private college on wages regardless of elite status or gender, although the point estimates for male non-elites are somewhat larger than for females. Table 5 provides the estimates from our preferred model - the propensity score matching model split by elite status and gender. In all models we include the choice of major in the propensity score.7 Once again, due to small sample sizes we focus on the pooled and non-elite models. When we turn to the propensity scores we see some stark differences between males and females. For males attending a non-elite college there is a significant return to earnings of 18% and to wages of 14%. These estimates are slightly larger for the full pooled sample. For females on the other hand, the estimates are insignificant and close to zero. Taken at face value, the earnings estimates show a negative return of 2% and the wage estimates show a positive return of only 5%. Hence, the propensity score estimates suggest that men benefit much more from a private college environment than women conditional on attending a non-elite college. We also note that the differences between OLS and matching for females suggest that the functional form restrictions and lack of common support for OLS invalidate it as an appropriate identification strategy. In Table 6 we consider the potential impact of attending a non-elite private college on other outcomes. In particular, we look at matching estimates for marital status, unemployment status, spousal earnings and degree attainment. In general, we find little evidence of differences between attending non-elite private or public colleges along these measures. The only marginally significant estimates are a negative effect on marital status for females and a positive effect on spousal earnings for males. However given that this is only two of 15 coefficients, these results could merely be 7 Note that the sample sizes for the subsamples do not add up to the full sample due to the common-support condition we impose.

8

spurious.

6

Conclusion

In this paper we estimate whether attending a non-elite private college or university has an impact on wages relative to attending a non-elite public college or university of similar quality (as proxied by mean SAT scores). Given that private universities often cost considerably more to the student than a public university, it is important that students know whether paying the higher costs are worthwhile in terms of future earnings. To answer this question we use highly detailed data from the National Longitudinal Survey of Youth in 1979. Using both OLS and propensity-score matching methods we control for a large set of observable characteristics. Our preferred matching estimates suggest that males gain returns to earnings and wages of 18% and 14%, respectively. However, for females there is little evidence of impacts. While it is unclear why there is a difference by gender, one possibility is that males benefit more from networking provided by private colleges, e.g. the ”old boy’s network.” Another possibility is that male students benefit more from some of the inputs into education often provided by private colleges such as smaller classes and more access to faculty. We also note that the matching estimates for females are smaller than OLS estimates, suggesting that OLS does not satisfy the functional form and common support assumptions necessary for appropriate identification. Nonetheless, we acknowledge that our methodologies do not account for unobservable differences between public and private non-elite college students. While our exhaustive set of control variables help remove a substantial amount of bias, we cannot rule out the possibility of residual bias. Given this, we argue that it is likely any omitted variable bias would lead us to over-estimate the impacts. We provide some evidence that selection on observables are in this direction, and hence this statement would be true if selection on unobservables is in the same direction as selection on observables. In this case, we can interpret the results above as upper-bound estimates of the returns. This suggests that we can at least rule out positive returns for females to attending a private, non-elite college.

References Altonji, Joseph G., Todd E. Elder, and Chris R. Taber, “An Evaluation of Instrumental Variable Strategies for Estimating the Effects of Catholic Schools,” Journal of Human Resources, 2005, 40 (4), 791–821. Behrman, Jere R., Mark R. Rosenzweig, and Paul Taubman, “College Choice and Wages: Estimates Using Data on Female Twins,” The Review of Economics and Statistics, 1996, 78 (4), 672–685. 9

Black, Dan A. and Jeffrey A. Smith, “How Robust is the Evidence on the Effects of College Quality? Evidence from Matching,” Journal of Econometrics, 2004, 121 (1-2), 99–124. and , “Estimating the Returns to College Quality with Multiple Proxies for Quality,” Journal of Labor Economics, 2006, 24 (3), 701–728. , Kermit Daniel, and Jeffrey A. Smith, “College Quality and Wages in the United States,” German Economic Review, 2005, 6 (3), 415–443. Bowman, William R. and Stephen L. Mehay, “College Quality and Employee Job Performance: Evidence from Naval Officers,” Industrial and Labor Relations Review, 2002, 55 (4), 700–714. Brewer, Dominic J., Eric R. Eide, and Ronald G. Ehrenberg, “Does It Pay to Attend an Elite Private College? Evidence from the Senior High School Class of 1980,” Research in Labor Economics, 1996, 15, 239–271. , , and , “Does It Pay to Attend an Elite Private College? Cross-Cohort Evidence on the Effects of College Type on Earnings,” The Journal of Human Resources, 1999, 34 (1), 104–123. Card, David, “Using Geographic Variation in College Proximity to Estimate the Returns to Schooling.,” in Louis N. Christofides, Kenneth E. Grant, and Robert Swidinsk, eds., Aspects of Labor Market Behaviour: Essays in Honor of John Vanderkamp, University of Toronto Press, 1995, pp. 201–222. Dale, Stacy Berg and Alan B. Krueger, “Estimating the Payoff to Attending a More Selective College: An Application of Selection on Observables and Unobservables,” Quarterly Journal of Economics, 2002, 117 (4), 1491–1526. Eide, Eric, Dominc J. Brewer, and Roland G. Ehrenberg, “Does it pay to attend an elite private college? Evidence on the effects of undergraduate college quality on graduate school attendance,” Economics of Education Review, 1998, 17 (4), 371–376. Evans, William N. and Robert M. Schwab, “Finishing High School and Starting College: Do Catholic Schools Make a Difference?,” Quarterly Journal of Economics, 1995, 110 (4), 941–974. Grogger, Jeffrey and Derek Neal, “Further Evidence on the Effects of Catholic Secondary Schooling,” Brookings-Wharton Papers on Urban Affairs, 2000, 1, 151–193. James, Estelle, Nabeel Alsalam, Joseph C. Conaty, and Duc-Le To, “College Quality and Future Earnings: Where Should You Send Your Child to College?,” American Economic Review - Papers and Proceedings, May 1989, 79 (2), 247–252. Kling, Jeffrey R., “Interpreting Instrumental Variables Estimates of the Returns to Schooling,” Journal of Business and Economic Statistics, 2001, 19 (3), 358–364. Lazear, Edward, “Education Production,” Quarterly Journal of Economics, August 2001, 116 (3), 777–803. Long, Bridget Terry, “Does the Format of a Financial Aid Program Matter? The Effect of State In-Kind Tuition Subsidies,” Review of Economics and Statistics, 2004, 86 (3), 767–782. Long, Mark, “College Quality and Early Adult Outcomes,” Economics of Education Review, October 2008, 27 (5), 588–602. Loury, Linda D. and David Garman, “College Selectivity and Earnings,” Journal of Labor Economics, 1995, 13 (2), 289–308.

10

Neal, Derek, “The Effects of Catholic Secondary Schooling on Educational Achievement,” Journal of Labor Economics, 1997, 15 (1), 98–123. Rothschild, Michael and Lawrence J. White, “The Analytics of the Pricing of Higher Education and Other Services in Which the Customers Are Inputs,” Journal of Political Economy, 1995, 103 (3), 573–586. Rouse, Cecelia E., “Private School Vouchers and Student Achievement: An Evaluation of the Milwaukee Parental Choice Program,” The Quarterly Journal of Economics, 1998, 113 (2), 553–602. Singell, Larry D., “Merit, Need, and Student Self Selection: Is There Discretion in the Packaging of Aid at a Large Public University?,” Economics of Education Review, 2002, 21 (5), 445–454.

11

Figure 1 - Tuition, Fees, Room and Board Four-Year Colleges in 2008-09 Dollars $35,000

$30,000

$25,000

$20,000

$15,000

$10,000

$5,000

$0

Year

Private

Source: Digest of Education Statistics, 2010

Public

Table 1 : Summary Statistics for Selected Outcomes and Covariates Public

Private

Description

Mean

N

Mean

N

Difference

Hourly Rate of Pay (2002) Earnings as of 2002 Married as of 2002 Bachelors Graduate and Professional Degree Gender US Born Age as of 1979 Age as of 2002 Unemployment in Current Residency (2002) White Black Hispanic Lived in Town/City Lived with both parents No religion Protestant Catholic Jew College prep for high school curriculum General prep for high school curriculum

24.45 47966 0.67 0.41 0.16 0.51 0.97 17.64 41.43 2.57 0.54 0.27 0.09 0.81 0.83 0.02 0.54 0.33 0.01 0.44 0.50

1147 1270 1270 1239 1239 1270 1270 1270 1270 1270 1270 1270 1270 1270 1270 1270 1270 1270 1270 1270 1270

28.37 57352 0.71 0.46 0.22 0.51 0.94 17.71 41.50 2.53 0.62 0.24 0.07 0.86 0.82 0.03 0.44 0.33 0.02 0.55 0.39

476 537 536 518 518 537 537 537 537 537 537 537 537 537 537 537 537 537 537 537 537

-3.92 -9386 -0.04 -0.05 -0.05 0.00 0.02 -0.07 -0.07 0.32 -0.08 0.04 0.02 -0.05 0.00 0.00 0.04 0.00 -0.01 -0.12 0.11

Differences that are significant at the 5% level are starred

T-Stat Total N * *

* *

*

*

* *

-2.7 -3.0 -1.8 -1.9 -2.5 0.1 2.2 -0.6 -0.6 0.8 -3.2 1.6 1.6 -2.9 0.0 -0.2 1.6 0.0 -1.4 -4.6 4.5

1623 1807 1806 1757 1757 1807 1807 1807 1807 1807 1807 1807 1807 1807 1807 1807 1807 1807 1807 1807 1807

Table 2 :OLS Regressions of Earnings and Wages on Private College/University Attendance A. Log of Earnings as of 2002 Private N

(1)

(2)

(3)

(4)

(5)

(6)

0.167** (0.072)

0.151** (0.070)

0.078 (0.069)

0.0923 (0.0691)

0.100 (0.0799)

0.0832 (0.0795)

1,174

1,174

1,174

1,174

1,174

1,174

B. Log of Wages as of 2002 Private

N

(1) 0.154** (0.0565)

(2) 0.149** (0.069)

(3) 0.0826 (0.0701)

(4) 0.0820 (0.0683)

(5) 0.0366 (0.0675)

(6) 0.0332 (0.0687)

1,223

1,223

1,223

1,223

1,223

1,223

Standard errors in parentheses. ***, **, * denotes significance at 1%, 5% and 10% respectively. Identification of college is based on the most recent college attended by the individual and regressions are weighted using the survey's probability weights for 2002. (1) is a univariate regression on private school attendance and a constant. (2) is (1) + gender of the individual. (3)is (2) + SAT score of the school .(4) is (3) + ASVAB Scores of the individual. (5) is (4) + race of the individual, religion of the individual, highest degree of parents, urbanization of residence at age 14, unemployment at current residence as of 2002, whether individual knew their parents, number of books at high school library, whether household member had library, magazine, newspaper subsciption, drug usage of the individual, occupation of parents, number of siblings, #clubs participated in. In (6) we add the dummies for the student's choice of major.

Table 3 : OLS Regressions of Earnings on Private College/University Attendance by Elite Classification (1)

A. No Majors (2)

(3)

All

Non-Elite

Elite

(4)

B. Majors (5)

(6)

All

Non-Elite

Elite

i. Full Sample Private N

0.100 (0.0799)

0.172* (0.0967)

-0.103 (0.158)

0.0832 (0.0795)

0.164* (0.0966)

-0.261 (0.192)

1,174

900

272

1,174

900

272

ii. Males Private N

0.172** (0.0761)

0.159* (0.0850)

0.0979 (0.191)

0.154** (0.0760)

0.139 (0.0901)

0.128 (0.232)

595

436

158

595

436

158

iii. Females Private N

0.0491 (0.137)

0.155 (0.171)

-0.364 (0.584)

0.0375 (0.133)

0.151 (0.166)

-0.315 (1.418)

579

464

114

579

464

114

Standard errors in parentheses. ***, **, * denotes significance at 1%, 5% and 10% respectively. Identification of college is based on the most recent college attended by the individual and regressions are weighted using the survey's probability weights for 2002. Controls include gender of the individual, mean SAT score of the school, ASVAB Scores of the individual, race, religion, highest degree of parents, urbanization of residence at age 14, unemployment at current residence as of 2002, whether individual knew their parents, number of books at high school library, whether household member had library, magazine, newspaper subscription, drug usage of the individual, occupation of parents, number of siblings, #clubs participated in. Regressions with majors include dummies for the student's choice of major. Elite Classification based on the competitiveness of the schools as reported by Barrons. Non Elite - LC, NC, C+ , C. Elite - HC , HC+ , VC, VC+, MC.

Table 4 : OLS Regressions of Wages on Private College/University Attendance by Elite Classification A. No Majors

B. Majors

(1)

(2)

(3)

(4)

(5)

(6)

All

Non-Elite

Elite

All

Non-Elite

Elite

i. Full Sample Private N

0.037 (0.068)

0.107 (0.081)

-0.105 (0.135)

0.033 (0.069)

0.106 (0.079)

-0.174 (0.140)

1,223

939

282

1,223

939

282

ii. Male Private N

0.124 (0.091)

0.141 (0.111)

-0.030 (0.201)

0.113 (0.090)

0.148 (0.098)

-0.067 (0.196)

619

454

164

619

454

164

iii. Female Private N

0.031 (0.085)

0.083 (0.10)

-0.452 (0.449)

0.032 (0.089)

0.073 (0.105)

-0.104 (0.853)

604

485

118

604

485

118

Standard errors in parentheses. ***, **, * denotes significance at 1%, 5% and 10% respectively. Identification of college is based on the most recent college attended by the individual and regressions are weighted using the survey's probability weights for 2002. Controls include gender of the individual, mean SAT score of the school, ASVAB Scores of the individual, race, religion, highest degree of parents, urbanization of residence at age 14, unemployment at current residence as of 2002, whether individual knew their parents, number of books at high school library, whether household member had library, magazine, newspaper subsciption, drug usage of the individual, occupation of parents, number of siblings, #clubs participated in. Regressions with majors include dummies for the student's choice of major. Elite Classification based on the competitiveness of the schools as reported by Barron's. Non Elite - LC, NC, C+ , C. Elite - HC , HC+ , VC, VC+, MC.

Table 5 : Propensity Score Matching of Earnings and Wages on Private college/University Attendance (1)

A. Log of Earnings (2)

(3)

(4)

B. Log of Wages (5)

(6)

All

Non Elite

Elite

All

Non Elite

Elite

i. Full Sample Private N

0.107 (0.0691)

0.108 (0.0753)

-0.0720 (0.158)

0.108** (0.0479)

0.107* (0.0576)

-0.0698 (0.112)

1,140

868

248

1,192

912

255

0.218*** (0.0689)

0.183** (0.0921)

0.175 (0.186)

0.163** (0.0704)

0.139* (0.0838)

-0.00740 (0.170)

559

384

112

581

399

122

ii. Males Private N

iii. Females Private N

-0.0416 (0.101)

-0.0206 (0.114)

0.0617 (0.407)

0.0550 (0.0730)

0.0528 (0.0858)

-0.104 (0.272)

542

422

76

563

440

83

Standard errors in parentheses. ***, **, * denotes significance at 1%, 5% and 10% respectively. Identification of college is based on the most recent college attended by the individual. Propensity scores are estimated using an unweighted probit regression that includes gender of the individual, mean SAT score of the school, ASVAB Scores of the individual, race, religion, highest degree of parents, urbanization of residence at age 14, unemployment at current residence as of 2002, whether individual knew their parents, number of books at high school library, whether household member had library, magazine, newspaper subsciption, drug usage of the individual, occupation of parents, number of siblings, #clubs participated in, and student's choice of major. Standard errors are bootstrap standard errors using Elite Classification based on the competitiveness of the schools as reported by Barron's. Non Elite - LC, NC, C+ , C. Elite - HC , HC+ , VC, VC+, MC.

Table 6 : Marital Status, Unemployment Status and Spousal Earnings - Non-Elite Colleges, Propensity Score Matching Models Married in 2002 Unemployed in 2002 (1)

(2)

Spouse Earnings in 2002

Highest Degree Obtained Some College vs. Bachelors Bachelors vs. Graduate

(3)

(4)

(5)

i. Full Sample Private N

-0.024 (0.028)

0.000 (0.018)

0.038 (0.110)

-0.0376 (0.0290)

-0.0051 (0.0337)

825

1021

532

1,471

798

ii. Males Private N

-0.001 (0.045)

0.020 (0.028)

0.138* 0.082

-0.0342 (0.0471)

0.0242 (0.0583)

338

424

397

649

283

iii. Females Private N

-0.075* (0.045)

-0.020 (0.022)

-0.039 (0.154)

-0.0518 (0.0381)

-0.0464 (0.0472)

414

530

280

760

441

Standard errors in parentheses. ***, **, * denotes significance at 1%, 5% and 10% respectively. Identification of college is based on the most recent college attended by the individual. Propensity scores are estimated using an unweighted probit regression that includes gender of the individual, mean SAT score of the school, ASVAB Scores of the individual, race, religion, highest degree of parents, urbanization of residence at age 14, unemployment at current residence as of 2002, whether individual knew their parents, number of books at high school library, whether household member had library, magazine, newspaper subsciption, drug usage of the individual, occupation of parents, number of siblings, #clubs participated in, and student's choice of major. Standard errors are bootstrap standard errors using Elite Classification based on the competitiveness of the schools as reported by Barron's. Non Elite - LC, NC, C+ , C. Elite - HC , HC+ , VC, VC+, MC.