Estimation and results Directory UMM :Data Elmu:jurnal:E:Economics of Education Review:Vol19.Issue1.Feb1999:

52 M.J. Hilmer Economics of Education Review 19 2000 47–61 Table 2 Mean values for explanatory variables Direct Attendees All University University Transfer University Transfer Community College Transfers Up Down Transfer Individual Characteristics: Black 0.057 0.047 0.049 0.045 0.037 Hispanic 0.038 0.051 0.037 0.064 0.053 Other Race 0.018 0.015 0.020 0.010 0.043 Math Test 0.159 0.148 0.163 0.133 0.122 0.29 0.27 0.29 0.25 0.24 Reading Test 0.154 0.145 0.162 0.130 0.102 0.28 0.27 0.29 0.25 0.21 Family Income 5.100 5.344 5.169 5.522 5.040 1.58 1.49 1.62 1.33 1.71 HS Grades 3.352 3.249 3.250 3.248 3.052 0.54 0.58 0.62 0.55 0.57 College Performance: College GPA 2.990 3.032 3.082 2.984 2.948 0.51 0.53 0.50 0.55 0.54 College Major: Business 0.253 0.193 0.146 0.238 0.271 Engineering 0.204 0.233 0.340 0.130 0.167 Science 0.159 0.112 0.075 0.148 0.133 Social Science 0.223 0.229 0.167 0.288 0.244 Educ. and Letters 0.092 0.148 0.199 0.100 0.059 Other Major 0.069 0.085 0.073 0.096 0.126 Labor Market Experience: Employed in Major 0.235 0.262 0.262 0.261 0.249 Business 0.145 0.111 0.077 0.144 0.103 Engineering 0.109 0.149 0.226 0.074 0.081 Science 0.015 0.000 0.000 0.000 0.036 Social Science 0.019 0.044 0.029 0.059 0.038 Educ. and Letters 0.012 0.061 0.104 0.190 0.000 Other Major 0.044 0.048 0.053 0.039 0.072 Postgrad Degree 0.017 0.019 0.034 0.005 0.001 Postgrad Attendee 0.096 0.111 0.086 0.135 0.138 Full-time Employee 0.826 0.845 0.884 0.808 0.669 Institutional Characteristics: Research I 0.239 0.243 0.264 0.222 0.221 Doctoral Program 0.120 0.108 0.060 0.155 0.049 Enrollment 13,354.36 13,665.01 13,979.18 13,346.02 16,532.34 10,904.08 10,671.58 11,029.80 10,336.37 12,402.34 Number of 551 155 77 78 83 Observations Notes: Standard deviations are in parentheses. Observations with missing values are not included in the calculation of those variables. Data are weighted using Panelwt4.

3. Estimation and results

The purpose of the empirical work presented below is to examine whether the return to university quality dif- fers for transfer students and direct attendees. This analy- sis starts by estimating a simple form of the post-gradu- ation wage function for different subsamples of students. As with previous studies, this simple wage function only controls for the quality of university from which a stud- ent graduates. The simple wage function to be estimated can be written as: W i 5 aQ G i 1 dX i 1 e i 1 where W i is the log 1992 hourly wage of student i, Q G i is the quality of university from which student i gradu- ates, X i is the vector of explanatory variables described above, and e i is a normally distributed error term. Para- meters to be estimated are a and d. 53 M.J. Hilmer Economics of Education Review 19 2000 47–61 It is important to consider how the graduation quality measure enters Eq. 1. Entering quality as a continuous variable, as in previous studies, constrains the return to an incremental increase in quality to be equal across the entire range of university qualities. This may not provide a full picture of the return to university quality, however, as it is entirely possible that the return to incremental quality differs depending on an institution’s quality level. In particular, the return to quality might be expected to be greater for students who succeed in graduating from high quality, highly competitive, elite institutions than for students who graduate from lower quality, moder- ately selective institutions. By entering quality as a con- tinuous variable, previous studies have failed to consider this possibility. To fully explore potential differences in the return to quality, graduation quality is entered both as a continuous variable and as a series of dummy variables representing different ranges in graduation quality. This latter specification is desirable as it allows the return to quality to vary across different ranges of institutional quality, whereas the former specification constrains the return to be constant across all ranges of institutional quality. Table 3 presents the results of estimating Eq. 1 by OLS for students in the sample. It should be noted that estimating Eq. 1 by OLS yields potentially biased esti- mates of a and d because post-graduation wages are only observed for those students who graduate from a univer- sity and not for the population as a whole. This potential self-selection bias can be corrected using the two-stage methodology of Lee 1983. 11 For the current analysis, however, the selectivity corrections are not statistically significant and do not significantly affect the coefficient estimates. Consequently, the results in Table 3 are the uncorrected OLS estimates. The first column of Table 3 attempts to replicate pre- vious studies by entering university quality as a continu- ous variable and estimating Eq. 1 for the full sample of college graduates. Previous studies have estimated the return to graduation quality to be between three and seven percent for a 100 SAT point increase in quality Rumberger Thomas 1993; James et al. 1989; Mueller 1988; Wise 1975; Wales 1973. The five and one-half percent wage premium estimated here is consistent with those previous results. Thus, the return to quality for students in this sample does not appear to differ systematically from that of students in samples used in previous studies. 11 This methodology consists of estimating a four-way multi- nomial logit non-attendance, community college attendance, university dropout, and university graduate and using those results to calculate selectivity correction terms that are included as additional regressors in Eq. 1. The results of this model gives nearly identical results to those presented here. To explore potential cross-quality differences in the return to graduation quality, the second column of Table 3 includes a set of dummy variables representing the quality range in which the student’s graduation univer- sity falls. The omitted range is 500 to 800 SAT points. Thus, the coefficient estimates represent the difference between the return to quality for students graduating from universities in a particular quality range and stu- dents graduating from universities in the 500 to 800 SAT point range. For example, the results indicate that stu- dents who graduate from universities that are between 1,200 and 1,400 SAT points earn roughly thirty-eight percent more than those who graduate from universities that are between 500 and 800 SAT points. This is likely the effect that James et al. 1978 had in mind when they estimated a significant positive return of roughly ten percent for private colleges in the Eastern United States. However, the findings presented here may paint a clearer picture as some Eastern private colleges are of lower quality and some non-Eastern public universities are of higher quality. Comparing across quality ranges suggests that the return to incremental quality does differ with university quality. As expected, students who graduate from the highest quality universities realize the largest return to incremental quality. The return to a given increase in quality for students graduating from universities that exceed 1,100 SAT points are more than twice as large as those for students in the remaining ranges. Further, the quality premium is nearly one-third larger for qual- ities above 1,200 SAT points than for qualities between 1,100 and 1,200 SAT points. Somewhat surprisingly, the return to quality is not as large for universities between 1,000 and 1,100 SAT points as for universities between 900 and 1,000. Looking at the remaining estimates in columns 1 and 2, college performance and labor market experiences appear to be much more important to a student’s future than individual characteristics. As might be expected, being employed full-time and having more work experi- ence both have significant positive effects on a student’s future earnings. Engineering majors who are employed as engineers earn a substantial wage premium over those who are not. Likewise, students who receive degrees in social science and education and letters earn significantly less upon graduation. 12 Finally, it is interesting to discuss 12 The estimated returns to college major are similar to those in Eide and Grogger 1995. They differ slightly, however, from those in James et al. 1978. While the magnitudes differ some- what, the signs, for the most part, are the same. The one excep- tion is social science majors for whom they estimate an insig- nificant positive effect. There are two possible explanations for these discrepancies. First, the major groups used here are the same as those in Eide and Grogger but different from those in James. Second, this study includes variables that represent 54 M.J. Hilmer Economics of Education Review 19 2000 47–61 the set of institutional characteristics. As with James et al. 1978 student enrollment and whether the institution is classified as a doctoral granting institution both have insignificant effects on a student’s future earnings. A school’s research status is seen to have a significant negative effect on future earnings. Specifically, holding enrollment constant, students who graduate from Research I institutions earn roughly two percent less than those who do not. In the current study, however, the esti- mate is statistically significant while in James et al. it is not. This may result from the fact that Research I insti- tutions concentrate more of their resources on research than on teaching and student services. Or perhaps, stu- dents who graduate from Research I institutions simply choose different types of jobs. For example, such stu- dents may be more likely to obtain jobs with a training component that initially offer lower wages but promise a higher post-training earnings path. The last six columns in Table 3 repeat the analysis separately for the subsets of direct attendees, university transfers, and community college transfers. The results suggest that a student’s educational path choice does sig- nificantly affect the return to graduation quality. Specifically, when entered as a continuous variable the significant positive return to university quality is only observed for university transfers. The return to a 100 SAT point increase in quality is significant and roughly fourteen percent for university transfers, while it is insig- nificant for both direct attendees and community college transfers. This suggests that it is important to control for a student’s educational path when estimating the return to university quality. Indeed, a Chow test Greene 1993 rejects the hypothesis that the estimated coef- ficients are equal for each of the three educational paths. 13 Turning to the quality range estimates, the significant positive effect appears to exist only for university and community college transfers who graduate from the highest quality universities. Those students observe a return to incremental quality that is more than 100 per- cent greater than that observed by students graduating from the lowest quality universities. This should not be that surprising. Transfer students in the highest gradu- ation quality range must have started at lower quality institutions before transferring to and graduating from whether the student was employed in the major while the others do not. 13 The Chow test follows an F-distribution. In this case the test statistic is 3.53, and the table value for 5 percent signifi- cance is 1.46. A potential shortcoming of the Chow test is that it is based on the assumption of equal variances in both regression equations. However, a Wald test Greene 1993 that allows for unequal variances also rejects the null hypothesis of equal coefficients. their high quality institutions. 14 It may be that the unob- served characteristics that enable such students to move from low quality institutions to some of the most elite institutions in the United States are characteristics that are valued in the labor market. If so, one would expect to see a large estimated return. That the returns to quality are generally lower for direct attendees than transfer stu- dents might suggest that employers place greater empha- sis on observed educational experiences for students who transfer and greater emphasis on other characteristics for students who do not. Indeed, it appears that employers place more emphasis on individual characteristics for direct attendees. There are other interesting differences between direct attendees and transfer students. Whereas the return to college major is generally positive for direct attendees, it is generally negative for university transfers, and gen- erally less significant for community college transfers. Direct attendees who are business and engineering majors receive significant and positive wage premiums, while university transfers in each major receive signifi- cant and negative wage premiums. Likewise, the return to post college experience and institutional character- istics differ according to educational path. Being employed full-time and having more work experience are more important for direct attendees than university transfers. The characteristics of the institution from which a student graduates have the most significant effect on the earnings of university transfers. As the results in Table 3 indicate, the return to gradu- ation quality differs systematically for each of the three educational paths. To get a better idea of the causes of this difference, the following analysis concentrates on transfer students. The major difference between transfer students and direct attendees is that transfer students spend some fraction of their career at different quality institutions, while direct attendees spend their entire career at the same institution. Thus, in examining the return to university quality for transfer students, it is important to also control for the quality of other insti- tutions attended and the length of time spent at other institutions. The wage function to be estimated for trans- fer students can thus be written as: W i 5 a 1 Q G i 1 a 2 Q G i 1 a 3 T i 1 dX i 1 e i 2 where W i , Q G i , and X i are defined as before, Q 1 i is the quality of university last attended by student i before 14 Among university transfers who graduated from univer- sities with qualities between 1,200 and 1,400 SAT points the average increase in quality was 239 SAT points. The lowest initial quality for those students was 826 SAT points with a corresponding quality increase of 474 SAT points. The highest initial quality was 1,360 SAT points with a corresponding quality decrease of 20 SAT points. 55 M.J. Hilmer Economics of Education Review 19 2000 47–61 Table 3 Wage regressions without educational path controls All College Graduates Direct Attendees University Transfers Community College Transfer Graduation Quality100 0.0553 – 0.0194 – 0.1367 – 0.0079 – 0.0165 – 0.0192 – 0.0380 – 0.0718 – Quality Dummies 800–900 – 0.1261 – 0.1110 – 0.1296 – 0.4480 – 0.1000 – 0.1201 – 0.2391 – 0.3430 900–1,000 – 0.1866 – 0.2070 – 0.0238 – 0.3807 – 0.0990 – 0.1183 – 0.2372 – 0.3229 1,000–1,100 – 0.1530 – 0.1292 – 0.2192 – 2 0.0171 – 0.1039 – 0.1257 – 0.2445 – 0.3334 1,100–1,200 – 0.2816 – 0.1979 – 0.4207 – 0.2136 – 0.1157 – 0.1390 – 0.2739 – 0.3839 1,200–1,400 – 0.3812 – 0.1611 – 1.0461 – 0.8904 – 0.1225 – 0.1439 – 0.3139 – 0.4531 Individual Characteristics: Black 0.1114 0.1156 0.1344 0.1584 0.3185 0.2486 2 0.2577 2 0.0600 0.0753 0.0768 0.0845 0.0868 0.1911 0.1896 0.3218 0.3157 Hispanic 0.1013 0.0952 0.1962 0.1926 0.0980 0.0969 0.0147 0.0745 0.0809 0.0812 0.1015 0.1020 0.1758 0.1780 0.2552 0.2449 Other Race 0.1717 0.1524 0.2238 0.2283 2 0.0487 2 0.1273 0.0944 0.1351 0.1155 0.1163 0.1431 0.1439 0.3007 0.2956 0.2860 0.2711 Math Test 0.0376 0.0173 2 0.0112 2 0.0732 2 2.7848 2 1.0867 2 0.6727 2 0.5500 0.2022 0.2023 0.2341 0.2354 6.6454 6.8031 0.4748 0.4527 Reading Test 0.0839 0.0965 0.1078 0.1603 2.8082 1.0838 0.9437 0.6277 0.2059 0.2060 0.2361 0.2368 6.6367 6.7944 0.5785 0.5565 Family Income 0.0197 0.0216 0.0300 0.0354 2 0.0215 2 0.0210 0.1121 0.1140 0.0112 0.0114 0.0134 0.0137 0.0307 0.0303 0.0465 0.0452 HS Grades 0.0084 0.0107 0.0665 0.0737 2 0.0766 2 0.0462 0.0209 0.1419 0.0324 0.0325 0.0405 0.0409 0.0792 0.0793 0.1401 0.1526 College Performance: College GPA 0.0547 0.0502 0.0452 0.0366 0.0517 0.0843 0.0696 0.0387 0.0344 0.0350 0.0411 0.0416 0.0880 0.0884 0.1111 0.1113 College Major: Business 0.0406 0.0379 0.2112 0.2251 2 0.4258 20.3608 0.2532 0.0212 0.0759 0.0759 0.0925 0.0926 0.1792 0.1772 0.2650 0.2865 Engineering 0.0793 0.0718 0.2666 0.2673 2 0.4019 20.3567 0.4092 0.1692 0.0796 0.0796 0.0959 0.0959 0.1882 0.1875 0.2831 0.2869 Science 2 0.0759 2 0.0745 0.0473 0.0622 2 0.2886 2 0.2557 2 0.3384 2 0.2559 0.0733 0.0735 0.0891 0.0899 0.1629 0.1639 0.2597 0.2561 Social Science 2 0.1404 20.1434 0.0632 0.0747 2 0.6461 20.5135 20.1773 2 0.3813 0.0686 0.0686 0.0852 0.0855 0.1556 0.1576 0.2484 0.2502 Education letters 2 0.1389 2 0.1435 0.0037 0.0135 2 0.5041 20.3958 0.4760 0.6265 0.0816 0.0818 0.0986 0.0996 0.1792 0.1786 0.3122 0.3045 Labor Market Experience: Postgrad Degree 2 0.0343 2 0.0257 2 0.0840 2 0.0890 0.1109 0.2283 0.3136 0.8204 0.1338 0.1334 0.1545 0.1541 0.2945 0.2930 1.8761 1.7861 Postgrad Attendee 0.0969 0.0830 0.1294 0.1184 2 0.0841 2 0.1701 0.3572 0.3652 0.0569 0.0570 0.0688 0.0690 0.1266 0.1260 0.1981 0.1947 Fulltime Employee 0.2667 0.2671 0.2832 0.2783 0.2071 0.2313 0.5905 0.5504 0.0426 0.0427 0.0516 0.0517 0.1087 0.1094 0.1507 0.1539 Work Experience 0.0337 0.0335 0.0489 0.0485 0.0136 0.0225 2 0.0669 2 0.0905 0.0096 0.0096 0.0111 0.0111 0.0244 0.0247 0.0519 0.0518 Continued 56 M.J. Hilmer Economics of Education Review 19 2000 47–61 Table 3 Continued All College Graduates Direct Attendees University Transfers Community College Transfer Employed in Major: Business 0.0486 0.0526 0.0451 0.0362 2 0.0295 2 0.0236 2 0.5523 20.4330 0.0658 0.0658 0.0766 0.0769 0.1731 0.1740 0.2610 0.2815 Engineering 0.2616 0.2658 0.2568 0.2617 0.3167 0.3138 2 0.0061 0.0652 0.0731 0.0731 0.0853 0.0852 0.1757 0.1750 0.3359 0.3290 Science 0.0143 0.0127 2 0.0423 2 0.0300 – – 0.4600 0.5042 0.1424 0.1426 0.1632 0.1639 – – 0.3324 0.3184 Social Science 2 0.0078 2 0.0104 0.0031 2 0.0149 0.2698 0.1100 0.0243 0.3030 0.1056 0.1061 0.1459 0.1470 0.2012 0.2009 0.3068 0.3124 Educ. Letters 2 0.1054 2 0.1109 2 0.0945 2 0.1104 2 0.3243 2 0.3232 – – 0.1248 0.1247 0.1834 0.1831 0.2023 0.1990 – – Institutional Characteristics: Log Enrol Research I 2 0.0159 20.0147 20.0065 2 0.0038 2 0.0337 20.0377 20.0317 2 0.0394 0.0054 0.0054 0.0063 0.0064 0.0136 0.0136 0.0188 0.1886 Doctoral Program 0.0434 0.0305 0.0640 0.0565 2 0.2230 2 0.2133 0.4456 0.1057 0.0560 0.0574 0.0645 0.0664 0.1412 0.1419 0.3093 0.3126 Log Enrollment 2 0.0139 2 0.0065 2 0.0647 20.0736 0.1039 0.1488 0.0563 0.1449 0.0203 0.0215 0.0242 0.0254 0.0530 0.0544 0.0859 0.0883 R-square 0.2301 0.2348 0.2849 0.2920 0.4007 0.4463 0.5098 0.5961 Number of Observations 794 794 551 551 155 155 88 88 Dependent variable is log hourly wage. Standard errors in parentheses. See text for individual characteristics controlled for in each equation. Regressions also include dummy variables to indicate missing values for some variables. , significant at 0.05 and 0.10 levels. Data are weighted using Panelwt4. transferring, and T i the fraction of total schooling spent at institutions other than the one from which the student graduates. Parameters to be estimated are a 1 , a 2 , a 3 , and d . Again, the quality measures will be entered as both continuous variables and series of dummy variables. Unfortunately, because community colleges usually have open-door policies, they do not require students to take the SAT test in order to gain admission. Therefore, the initial quality term is unobservable for community col- lege transfers. The length of time spent at previous insti- tutions is observable for all students, however. Tables 4 presents the results of estimating Eq. 2 for university and community college transfer students. The potential for self-selection bias exists, but once again the selectivity corrections are statistically insignificant and the uncorrected results are reported. The first column presents estimates with initial and graduation quality entered as continuous variables. The second column enters quality range dummies for graduation quality while the third adds quality range dummies for initial quality. According to column 1, adding the full set of edu- cational path controls does not affect the estimate for the continuous graduation quality variable. A 100 point increase in graduation quality is still expected to increase future earnings by roughly fourteen percent. The esti- mated coefficients for the graduation quality dummies are similar to those above, albeit somewhat smaller in magnitude, with the notable exception of the 900–1,000 SAT range. The estimated coefficient for that range actu- ally switches signs. As discussed below, this may be caused by the fact that most transfer students who gradu- ate from universities in that range will have transferred down in quality. The return to initial quality merits some discussion. Regardless whether graduation quality is entered as a continuous variable or as a series of dummy variables, the return to initial quality is negative and significant. The coefficient estimate can be interpreted as follows. Holding everything else constant, for two university transfers who graduate from the same quality university, the one who initially attends a university that is 100 SAT points higher in quality earns roughly eight and one-half percent less upon graduation. As it is more likely that the student who initially attends the higher quality uni- versity is decreasing quality when transferring, this result suggests that firms place a negative value on transferring down in quality. Further support for this argument is given by the pre-transfer tenure term. The length of time spent at initial institutions has a significant negative effect on a transfer student’s post-graduation earnings. In other words, the longer a transfer student spends at his initial institution the less he is paid upon graduation, suggesting that employers look negatively on students 57 M.J. Hilmer Economics of Education Review 19 2000 47–61 Table 4 Wage regressions with educational path controls University Transfers Community College Transfers Graduation Quality100 0.1423 – – 0.0162 – 0.0386 – – 0.0737 – Graduation Quality Dummies 800–900 – 0.0311 0.0394 – 0.5580 – 0.2296 0.2369 – 0.3531 900–1,000 – 2 0.1056 2 0.0032 – 0.4534 – 0.2311 0.2405 – 0.3239 1,000–1,100 – 0.0799 0.0998 – 0.0574 – 0.2392 0.2408 – 0.3346 1,100–1,200 – 0.2842 0.3491 – 0.2510 – 0.2687 0.2727 – 0.3794 1,200–1,400 – 1.0716 1.0613 – 1.0276 – 0.3005 0.3028 – 0.4537 Transfer Quality100 2 0.0891 2 0.0957 – – – 0.0393 0.0377 – – – Transfer Quality Dummies 800– 900 – – 2 0.1717 – – – – 0.2218 – – 900–1,000 – – 2 0.3212 – – – – 0.2115 – – 1,000–1,100 – – 2 0.3865 – – – – 0.2271 – – 1,100–1,200 – – 2 0.3212 – – – – 0.2491 – – 1,200–1,400 – – 2 0.7922 – – – – 0.2819 – – Pre-Transfer 2 0.1117 2 0.4557 2 0.4389 2 0.4459 2 0.4700 0.1902 0.1924 0.1996 0.4215 0.3877 Individual Characteristics: Black 0.2145 0.1843 0.2057 2 0.3680 2 0.1839 0.1923 0.1866 0.1948 0.3344 0.3203 Hispanic 0.0964 0.1041 0.1460 2 0.0926 2 0.0657 0.1713 0.1700 0.1794 0.2696 0.2529 Other Race 2 0.1426 2 0.2147 2 0.1521 0.0689 0.1105 0.2945 0.2829 0.2847 0.2876 0.2676 Math Test 0.2708 2.6132 5.8543 2 0.5150 2 0.4068 6.5660 6.5970 6.8182 0.5056 0.4712 Reading Test 0.2625 2 2.6402 2 5.9367 0.8320 0.5494 6.5611 6.5915 6.8193 0.6125 0.5715 Family Income 0.0090 0.0284 0.0322 0.1030 0.1082 0.0336 0.0330 0.0327 0.0480 0.0461 HS Grades 2 0.0583 2 0.0391 2 0.0469 2 0.0111 0.1014 0.0817 0.0794 0.0799 0.1430 0.1520 Continued. who transfer late in their careers. The remaining coef- ficient estimates indicate that controlling for a student’s educational path also increases the significance of the college major choice and institutional characteristic vari- ables. The final two columns of Table 4 presents results for community college transfers. This analysis is clearly inferior to that above due to the lack of the initial quality measure. Nonetheless, it is interesting to examine these results. When entered as a continuous variable, gradu- ation quality is again not significant for these students. When entered as a series of dummy variables, the return to graduation quality is significant and positive only for the highest quality range. Consistent with university transfers there is a large, negative effect associated with spending additional time at initial institutions. However, in this case the estimated effect lacks significance. This may be due to the relatively small sample of community 58 M.J. Hilmer Economics of Education Review 19 2000 47–61 Table 4 Continued University Transfers Community College Transfers College Performance: College GPA 0.0992 0.1634 0.1478 0.0939 0.0732 0.0881 0.0873 0.0874 0.1131 0.1114 College Major: Business 2 0.4515 2 0.3201 2 0.3117 0.2190 2 0.0425 0.1786 0.1746 0.1756 0.2683 0.2845 Engineering 2 0.5047 2 0.4004 2 0.3876 0.3986 0.1890 0.1925 0.1891 0.1948 0.2922 0.2915 Science 2 0.2969 2 0.2536 2 0.2019 2 0.3474 2 0.2803 0.1623 0.1595 0.1649 0.2608 0.2539 Social Science 2 0.6554 2 0.5016 2 0.4351 2 0.2379 2 0.4781 0.1519 0.1508 0.1562 0.2553 0.2531 Education Letters 2 0.4644 2 0.3087 2 0.2616 0.4200 0.5933 0.1784 0.1745 0.1814 0.3213 0.3058 Labor Market Experience: Postgrad Degree 0.0303 0.1389 0.1423 0.5260 1.1201 0.2880 0.2807 0.2860 1.8898 1.7683 Postgrad Attendee 2 0.1247 2 0.2269 2 0.2266 0.3161 0.3118 0.1240 0.1210 0.1213 0.2017 0.1942 Fulltime Employee 0.1372 0.1794 0.1791 0.6249 0.5796 0.1099 0.1081 0.1092 0.1547 0.1542 Work Experience 0.0142 0.0324 0.0274 2 0.0534 2 0.0692 0.0238 0.0239 0.0244 0.0556 0.0532 Employed in Major: Business 2 0.0393 2 0.0867 2 0.0135 2 0.5293 2 0.3884 0.1717 0.1698 0.1767 0.2665 0.2793 Engineering 0.3928 0.4047 0.4692 2 0.0502 2 0.0823 0.1763 0.1714 0.1756 0.3634 0.3583 Science – – – 0.4787 0.5503 – – – 0.3341 0.3154 Social Science 0.2369 0.0534 0.0718 2 0.0404 0.1878 0.1969 0.1930 0.1986 0.3156 0.3169 Educ. Letters 2 0.3687 2 0.2436 2 0.2153 – – 0.2114 0.2064 0.2129 – – Institutional Characteristics: Log Enrol Research I 2 0.0320 2 0.0339 2 0.0372 2 0.0266 2 0.0301 0.0134 0.0130 0.0132 0.0215 0.0200 Doctoral Program 2 0.2850 2 0.2565 2 0.2777 0.5111 0.2120 0.1435 0.1406 0.1454 0.3273 0.3203 Log Enrollment 0.1004 0.1381 0.1320 0.0464 0.1235 0.0522 0.0519 0.0542 0.0913 0.0905 R-square 0.4524 0.5158 0.5365 0.5239 0.6222 Number of Observations 155 155 77 88 88 Dependent variable is log hourly wage. Standard errors in parentheses. See text for individual characteristics controlled for in each equation. Regressions also include dummy variables to indicate missing values for some variables. , significant at 0.05 and 0.10 levels. Data are weighted using Panelwt4. college transfers. The remaining estimates do not differ significantly from the estimates presented in Table 3. A final question to be addressed is why students choose to transfer and if they do choose to transfer what factors affect their decision to increase or decrease qual- ity. Presumably, students choose to transfer if they are dissatisfied with their initial institutions. Dissatisfaction may arise for a variety of reasons. Students at large insti- tutions may be uncomfortable with large class sizes and may choose to transfer to smaller institutions. Students who initially leave home may want to move closer to family and friends. Still others may decide that their initial institution does not offer or is not strong in their desired field of study. The most likely reason that a stud- 59 M.J. Hilmer Economics of Education Review 19 2000 47–61 ent transfers, however, is a mismatching of the student’s ability andor motivation and the quality of his or her initial institution. Students who perform above their ability in high school or who come from high-income families may be accepted to high quality institutions at which they do not belong. Likewise, low-income stu- dents or students who perform poorly in high school and look bad “on paper” may not be accepted to universities that are high enough quality. Such students will be mis- matched with their initial institutions and may choose to transfer to institutions that fit them better. The summary statistics in Tables 1 and 2 suggest that this is the case for students in the sample. Transfer stu- dents who decrease quality come from higher income families, have lower standardized test scores and perform worse in college than both transfer students who increase quality and direct attendees. This suggests that students who choose to transfer down in quality are those whose family wealth allowed them to initially attend higher quality universities than their ability merited. Conse- quently, they are unable to compete as well with their fellow students and receive worse grades before choos- ing to transfer. Transfer students who increase quality, on the other hand, performed better on standardized tests but received lower high school and higher grades than direct attendees. This suggests that such students were initially forced to attend lower quality institutions due to their relatively poor high school records. Once at those institutions, however, they performed well relative to their classmates and were able to transfer to higher qual- ity institutions. The factors affecting a student’s decision to transfer to a different institution or persist at his or her initial institution can be examined more closely by estimating the following equation: L i 5 bZ i 1 y i 3 where L i is a dummy variable equal to one if the student i transferred and zero if he or she did not, Z i is a vector of characteristics affecting the decision to transfer, and y i is a normally distributed error term. Parameters to be estimated are b. As Eq. 3 describes a discrete choice problem and the error term is assumed to be normally distributed, it is appropriate to use probit analysis to estimate the para- meters. Table 5 presents the estimated marginal effects for the student’s transfer decision. The entries should be interpreted as the effect that changes in the independent variables have on the probability of choosing to transfer relative to choosing not to transfer, holding all else con- stant. The first two columns compare all transfer students to direct attendees while the final four columns compare students who transfer up and students who transfer down, respectively, to direct attendees. The results in Table 5 tend to confirm that students primarily choose to transfer due to an initial mismatching between students and institutions. The only variables that consistently have a significant effect on the decision to transfer are high school and college grades and the qual- ity of university initially attended. High school grades have a negative effect on the decision to transfer to a higher quality institution while college grades have a positive effect. This suggests that students who transfer up are indeed those who perform below their capabilities in high school and are forced into lower quality insti- tutions at which they are able to excel and improve their academic record to the point that they can gain admission to higher quality institutions. The fact that initial quality has a negative effect on the decision to transfer up strengthens this interpretation by suggesting that students who transfer up initially attend lower quality universities. The results for the decision to transfer down in quality are somewhat less clear. As would be expected, initial quality has a positive effect on the decision to transfer down, suggesting that students who do transfer down initially attend high quality universities. This would be consistent with the story that such students are over- matched at their initial institutions. However, high school grades have a negative effect and college grades have an insignificant effect on this decision, whereas under the overmatching story they would be expected to have positive and negative effects, respectively. It is not readily apparent why this should be so. It is interesting to briefly discuss the variables that do not significantly affect the decision to transfer. It does not appear that a student’s ethnicity systematically affects his or her decision. A student’s college major does is also not a significant determinant, suggesting that students in one particular major are no more likely to transfer than students in another major. Finally, enrollment does not have a significant effect suggesting that students at larger institutions are no more or less likely to transfer than those at smaller universities.

4. Conclusions