Model Directory UMM :Data Elmu:jurnal:E:Economics of Education Review:Vol19.Issue3.Jun2000:

283 J. Monks Economics of Education Review 19 2000 279–289 Table 3 Summary measures Variable Mean Standard deviation Minimum Maximum Experience weeks52 7.79 3.34 0.38 18.46 Tenure weeks52 3.04 2.82 0.02 16.50 Male 0.51 0.50 0.00 1.00 White 0.80 0.40 0.00 1.00 Armed Forces Qualification Test 1.75 0.59 0.03 3.09 Public institution 0.63 0.48 0.00 1.00 Masters, doctoral or research university 0.76 0.43 0.00 1.00 Specialized institutions 0.01 0.08 0.00 1.00 Non or less competitive 0.27 0.44 0.00 1.00 Competitive 0.45 0.50 0.00 1.00 Very competitive 0.21 0.41 0.00 1.00 Highly or most competitive 0.06 0.24 0.00 1.00 1979 net family income in 10K 2.55 1.46 0.00 7.50 1979 net family income missing 0.22 0.42 0.00 1.00 Log hourly wage 2.18 0.47 1.08 6.08 Year 88.48 3.17 79.00 93.00 Number of person-year observations 4977 because of differences in academic ability. This problem is minimized by calculating the ratio of each person’s test score to the average test score for his or her age. Finally, the 1979 net family income is used as a control for an individual’s ability to pay. 4 Table 3 provides means, standard deviations, minimums and maximums for all of the variables for the entire sample.

3. Model

This section focuses on the underlying economic model and methodology used to estimate the earnings differentials across institutional characteristics con- ditional upon individual traits and labor market experi- ences. Properly controlling for relevant individual attri- butes and a number of college characteristics enhances the understanding of variation in earnings across specific college characteristics. The process underlying the human capital investment decision to enroll in a parti- cular college is one which relates wages to human capital such as experience, tenure, and individual and insti- tutional characteristics. Institutional characteristics may influence earnings conditional on individual characteristics for a number of reasons. First, institutions may facilitate the accumu- 4 For those individuals who either did not take the ASVAB battery of tests used in the construction of the AFQT score or had missing 1979 net family income, a dummy variable indicat- ing that the variable was not reported was included among the regressors and the individual was assigned the average value for that variable from the sample used in this study. lation of human capital at different rates. For example, if there are peer effects, then attending a selective insti- tution where one is surrounded by bright students may increase human capital accumulation. Similarly, if the instructional quality is better at private institutions, then graduates of private colleges and universities may have greater human capital. Classroom dynamics and overall curricular design may also be important in the production function of human capital. If this is true, then graduates of different types of institutions may have accumulated different levels of human capital. A second reason why institutional characteristics may influence earnings separ- ately from individual characteristics is that employers may identify institutional attributes as a signal of ability. This may especially be the case if institutional character- istics are more visible than individual ability measures. The human capital and signaling explanations for the potential importance of institutional characteristics in determining earnings are not mutually exclusive. It may be that graduates of certain types of institutions earn more because of both human capital and signaling. In either case, individuals attempting to maximize the net present value of lifetime wealth would attempt to enroll in those institutions whose graduates earn a premium. Further compounding the college choice decision is that most institutions are selective to some degree in their admissions processes. So not only do individuals choose institutions based on future earnings and costs, but institutions choose individuals based on individual characteristics, such as academic ability, and in some cases ability to pay. As a result, an individual’s academic ability and financial resources are primary determinants in the college matriculation process. 284 J. Monks Economics of Education Review 19 2000 279–289 A reduced-form equation relating the log of wages at time t, for individual i, who attended institution j, as a function of individual and institutional characteristics is: ln W ijt 5 X 0it b 1 1 X 1i b 2 1 Q j b 3 1 d i 1 e ijt 1 where ln W ijt is the log of hourly wages; X 0it are individ- ual time varying labor market experiences; X i 1 are non- time-varying individual characteristics which influence earnings; Q j are college characteristics; d i is a normally distributed individual specific error component; and e ijt is a normally distributed random error. Estimation of Eq. 1 will result in biased estimates of the returns to insti- tutional characteristics because it does not account for the selection process by individuals and institutions in the enrollment process. Individuals are clearly not randomly allocated to dif- ferent institutions and institutional types. They are both chosen by the institution and choose the institution them- selves. Accounting for this selection process is at best difficult and problematic. Attempting to correct for selec- tion using various estimation techniques relies on assumptions which may act to exacerbate the problem. For example, Brewer et al. use multinomial logit to esti- mate the institutional type chosen. This approach becomes complicated as the number of institutional characteristics are increased, and it also assumes the independence of irrelevant alternatives, which is not likely to be the case in choosing which college to attend. Additionally, the instrumental variable estimation approach followed by Behrman et al. 1996a works well when there are few variables to instrument, but when the number of college characteristics increases this approach too becomes problematic. Because I include a number of institutional character- istics among the regressors, I am unable to use either a multinomial logit correction or instrumental variables. As a result, I attempt to control for the college selection process in the least restrictive means possible by includ- ing among the regressors individual attributes which influence the enrollment process. Because academic ability and ability to pay are primary determinants in the college selection process, I include among the regressors of Eq. 1 one’s AFQT score and 1979 net family income. In as much as AFQT score and family income are controls which may not fully capture the endogeneity of the institutional characteristics, the resulting estimates of the coefficients on the institutional characteristics may be biased. One should not interpret the coefficients on the institutional characteristics as a true return to these characteristics in the sense of what an individual ran- domly assigned to this institution could expect to receive. Instead these coefficients reflect the average earnings of graduates from institutions of certain types, conditional upon their observable individual character- istics. The log of hourly wages is first regressed against experience, experience squared, tenure, tenure squared, a male dummy variable and a white dummy variable as a benchmark for the incorporation of individual academic ability measures and college characteristics. This speci- fication is then expanded to include the additional col- lege selectivity, control, classification dummies and AFQT score, and then the respondent’s 1979 net family income. All specifications are estimated using gen- eralized least squares to control for heteroskedasticity introduced to the disturbance structure from the individ- ual specific error component. Additionally, a number of previous analyses have found significant differences across demographic groups in the returns to education Bok Bowen, 1998; Cohn Addison, 1998; Cooper Cohn, 1997; Behrman et al., 1996a; Loury Garman, 1995; Card Krueger, 1992. I allow for varying coefficients across demographic groups by performing separate regression analyses by race and gender. Most other studies of the returns to institutional attributes ignore possible differences across demographic groups in the impact of college character- istics on labor market outcomes. Because labor market and higher education opportunities may differ across these groups, the returns to these opportunities may dif- fer as well. If the accumulation of human capital varies across certain groups within an institution, perhaps due to peer effects or the classroom dynamics of race and gender, then the earnings of graduates of certain insti- tutional types may vary across these groups. Bok and Bowen 1998 found that black students at selective institutions “under-performed” relative to what would be predicted based on their standardized test scores. 5 There- fore standardized test scores may not accurately control for systematic differences across groups of students in academic performance while in college, and the returns to institutional characteristics may vary across demo- graphic groups as a result. It may also be the case that race and gender may interact with institutional character- istics in forming a signal to employers of ability. Additionally, Kolpin and Singell 1997 show that in the presence of affirmative action, individuals from a pre- ferred group may receive different rates of return to their individual and institutional characteristics. While a complete test of these hypotheses is beyond the scope of this study, I allow for possible variation in the returns to individual and institutional characteristics by estimating separate regressions for males and females, and whites and non-whites. Chow tests are performed to test for the existence of significant variation in the coefficients across gender and racial groups. 5 Bok and Bowen 1998 use the Scholastic Aptitude Test SAT as their standardized test. 285 J. Monks Economics of Education Review 19 2000 279–289 Table 4 Returns to college characteristics. Dependent variable: log hourly wages Specification 1 Specification 2 Specification 3 Intercept 1.533 1.295 1.242 0.035 0.064 0.066 Experience 0.077 0.071 0.071 0.006 0.006 0.006 Experience squared10 2 0.019 2 0.017 2 0.017 0.004 0.004 0.004 Tenure 0.037 0.035 0.035 0.005 0.005 0.005 Tenure squared10 2 0.025 2 0.023 2 0.023 0.005 0.005 0.005 Male 0.155 0.128 0.131 0.026 0.022 0.022 White 0.035 2 0.049 2 0.069 0.029 0.029 0.030 Armed Forces Qualification Test 0.111 0.107 0.022 0.022 Public institution 2 0.045 2 0.033 0.027 0.027 Masters, doctoral or research 0.137 0.126 0.031 0.032 Specialized institution 0.189 0.169 0.089 0.089 Non or less competitive 2 0.047 2 0.042 0.028 0.028 Very competitive 0.081 0.079 0.030 0.030 Highly or most competitive 0.151 0.131 0.049 0.050 Net family income 0.030 0.008 Adjusted R-squared 0.86 0.89 0.89 Number of observations 4977 4977 4977 Notes : 1 Standard errors are in parentheses. 2 Specifications 2 and 3 include dummy variables for industry and occupation, and missing AFQT and net family income, not shown. 3 Significant: at the 1 level; at the 5 level; at the 10 level.

4. Results