Quantile Treatment Effects of College Quality on Earnings Evidence - PowerPoint PPT Presentation

Quantile Treatment Effects of College Quality on Earnings Evidence from Administrative Data in Texas Rodney Andrews The University of Texas at Dallas and The Texas Schools Project ‘ Jing Li ‘ Michael Lovenheim Cornell University January 24, 2014

Disclaimer “The conclusions of this research do not necessarily reflect the opinions or official position of the Texas Education Agency, the Texas Higher Education Coordinating Board, or the State of Texas.”

Outline Introduction and Motivation Data Methodology Results Conclusion

Introduction There is a large literature that examines the labor market returns to college quality. Brewer, Eide, and Ehrenberg (1999): 20-25% higher wages from attending an elite public college or university. Black and Smith (2004, 2006): Find sizable returns to graduating from a college of higher quality. Hoekstra (2010): Finds returns of 24% in earnings from attending the state flagship. Dale and Krueger (2002): Payoff to attending a more costly college; payoff to attending more elite college is greater for students form disadvantaged background. Dale and Krueger (2011): Returns are sizeable until they are adjusted for unobserved ability by controlling for the SAT score of other colleges the student applied to.

Introduction What does the literature miss? Basic Statistics: The mean is only one feature of any distribution ,and it may not offer an accurate characterization of the returns students face. If the distribution of returns is highly skewed, then the expected return may have considerable risk associated with it. Lack of focus on the most prevalent college choice margins. Some colleges examined in previous national studies: Amherst, Yale, UPenn, Oberlin, Princeton... Only 5.8% of first-time college attendees in NELS:88 attend a highly selective private school.

Introduction What we do in this paper: 1 We estimate the “distribution” of returns to college quality; that is, we estimate unconditional quantile treatment effects (QTE) of college quality on earnings in Texas that show the effect of college quality on earnings at each percentile of the marginal earnings distribution 2 We measure college quality by sector: UT-Austin 1 Texas A&M – College Station 2 Non-flagship public four-year (e.g., UTD, UT-Pan American, 3 UTSA, Texas Tech, TAMU-Commerce) Community college (e.g., CCCC, Austin Community 4 College,Tarrant County College)

Introduction: What is college quality? UT Texas A&M Other Public Two Austin College Station Four-year Year 25 th Percentile Math SAT 535 520 440 75 th Percentile Math SAT 650 630 549 Faculty-Student Ratio 0.045 0.041 0.034 0.023 Expend. Per Student 25081 27449 10981 5756 Inst. Expend. Per Student 6900 8931 3648 2317 Graduation Rate 0.710 0.750 0.338 In-state Tuition 3212 3187 2001 1217

Introduction Preview of the Findings. Mean estimates do a poor job of characterizing the returns to college quality for most students. There are large differences in the distributions of the unconditional quantile treatment effects across school types and by race/ethnicity.

Data We make use of administrative data housed at the University of Texas at Dallas’s Education Research Center.

Data: The Sources Texas Education Agency (TEA) administrative records on K-12 student outcomes, including test scores and high school attended. Texas Higher Education Coordinating Board (THECB) records on all students attending public colleges and universities in Texas. Texas Workforce Commission (TWC) data on earnings that come from unemployment insurance records in 2007-2009. Use individual-average residuals from regressions of log quarterly earnings on cohort, year, and quarter-of-year dummies.

Data: The Sample We focus on male graduates from Texas’ public colleges and universities who graduated from high school during the years 1996–2002. The total sample size includes 94,071 male graduates. Sample Inclusion Criteria: No missing data for any of the covariates 1 The student must graduate before the age of 25 2 The graduate’s earnings for a given year are included only if he 3 worked for at least three consecutive quarters The student must not be currently enrolled in graduate school 4 when the earnings are measured. Table 3 and Figure 1 provide evidence that our sample 5 selection criteria do not produce large differences in observable characteristics.

What we estimate? We estimate the distribution of returns to graduating from either UT-Austin, Texas A&M-College Station, or one of Texas’s public community colleges relative to one of Texas’s Non-Flagship public universities.

The Problem We do not observe an individual graduating from two institutions.

The Method We estimate the propensity score-the likelihood that a student graduates from a particular college sector. We use the propensity score to re-weight the distribution of earnings such that the characteristics of the quality sector under consideration and the Non-Flagship Public Universities are balanced. This method is analogous to using survey weights to estimate population quantities using a specific sample.

Overlap of Propensity Scores Panel A: UT−Austin 2328 2031 1644 1625 1187 649 329 35 0 84 10 345 886 2146 4587 −10000 10626 −20000 −30000 29185 0 .2 .4 .6 .8 1 Propensity Score UT−Austin Other four−year public

Overlap of Propensity Scores Panel B: TAMU 5000 3312 3221 2558 2361 1055 811 103 0 69 900 −5000 3062 6398 −15000 −10000 10061 13081 14298 0 .2 .4 .6 Propensity Score TAMU Other Four−year Public

Overlap of Propensity Scores Panel C: Community College 5000 3345 3121 2437 1717 1797 1254 959 477 0 176 30 565 1456 −20000−15000−10000 −5000 3250 7181 15690 19525 0 .2 .4 .6 .8 1 Propensity Score Community College Other Four−year Public

UT-Austin & Non-flagship Earnings Distributions Panel A: UT−Austin 2 UT Austin−Observed Other 4 Year−Counterfactual Other 4 Year−Observed 1 Log Earnings Residual 0 −1 −2 0 10 20 30 40 50 60 70 80 90 100 Percentile

QTT Estimates of UT-Austin Graduation on Earnings Mean: 0.115 (0.007) Panel A: UT−Austin .4 .3 Log Earnings Residual .2 .1 0 −.1 −.2 −.3 0 10 20 30 40 50 60 70 80 90 100 Percentile

Texas A&M & Non-flagship Earnings Distributions Panel B: Texas A&M 2 TAMU−Observed Other 4 Year−Counterfactual Other 4 Year−Observed 1 Log Earnings Residual 0 −1 −2 0 10 20 30 40 50 60 70 80 90 100 Percentile

QTT Estimates of Texas A&M Graduation on Earnings Mean: 0.212 (0.006) Panel B: Texas A&M .4 .3 Log Earnings Residual .2 .1 0 −.1 −.2 −.3 0 10 20 30 40 50 60 70 80 90 100 Percentile

Two Year & Non-flagship Earnings Distributions Panel C: Community Colleges 2 Community College−Observed Other 4 Year−Counterfactual Other 4 Year−Observed 1 Log Earnings Residual 0 −1 −2 0 10 20 30 40 50 60 70 80 90 100 Percentile

QTT Estimates of Two Year Graduation on Earnings Mean: -0.100 (0.005) Panel C: Community Colleges .4 .3 Log Earnings Residual .2 .1 0 −.1 −.2 −.3 0 10 20 30 40 50 60 70 80 90 100 Percentile

Robustness Checks: UT-Austin .4 Log Earnings Residual .2 0 Baseline 1996−1998 HS Fixed Effects Attendees Background Sample Background Sample − w/controls −.2 0 20 40 60 80 100 Percentile

Robustness Checks: Texas A&M .4 .3 Log Earnings Residual .2 .1 Baseline 1996−1998 HS Fixed Effects Attendees Background Sample Background Sample − w/controls 0 0 20 40 60 80 100 Percentile

Robustness Checks: Community College .1 0 Log Earnings Residual −.1 −.2 Baseline 1996−1998 HS Fixed Effects Attendees −.3 0 20 40 60 80 100 Percentile

Conclusion Mean estimates do a poor job of characterizing the returns to college quality for most students. There are large differences in the distribution of the quantile treatment effects across school types. Why? This paper highlights the need to look beyond mean treatment effects, particularly when thinking about how students make higher education investment decisions.