Regression results Directory UMM :Data Elmu:jurnal:E:Economics of Education Review:Vol18.Issue1.Feb1999:

73 T.C. Buchmueller et al. Economics of Education Review 14 1999 65–77 Table 3 COGEE sample publications by employment sector AcademicPh.D. Academicother Non-Academic Percentage publishing 1 or 84.3 70.0 63.6 more articles Total publications 4.3 4.3 2.1 2.6 2.3 3.0 Percentage publishing 1 or 67.6 45.0 34.8 more in top 50 journals Total top 50 publications 2.3 2.7 0.8 1.2 0.8 1.4 N 108 60 66 Note: Standard deviations in parentheses. roughly 57 percent of their graduates in the AcademicPh.D. sector, whereas Tier 2 programs place 45 percent. Tiers 4 and 5 have the lowest placement rate in the AcademicPh.D. sector 15 percent and the high- est placement rate in the Non-Academic sector 38 percent. The connections between publications and employ- ment are described in Table 3 and in Fig. 3. Ph.D. econ- omists in the AcademicPh.D. sector publish more than their counterparts in the other two sectors. For all four measures reported in Table 3, differences between the AcademicPh.D. sector and the other two sectors com- bined are statistically significant at the P 5 0.01 level. The small differences in publication rates between the Non-Academic and AcademicOther sectors are not stat- istically significant at conventional levels. Similarly, Fig. Fig. 3. Average publications by employment sector — all publications — for 1977–78 cohort COGEE sample. 3 shows that economists in the AcademicPh.D. sector publish successively more each year through year six, whereas the patterns for the other two sectors are more erratic. Given these patterns and our small sample size, we define just two sectors of employment — the AcademicPh.D. sector and the other two sectors com- bined — in the remainder of our analysis.

6. Regression results

6.1. Employment sector probit results Before turning to the publication regressions, we examine the correlates of placement into the AcademicPh.D. sector. The relationship between gradu- 74 T.C. Buchmueller et al. Economics of Education Review 14 1999 65–77 ate program reputation and job placement evident in Table 3 resurfaces in the probit results reported in Table 4. Students in Tiers 4 and 5 are significantly less likely than students in higher-ranked programs to obtain jobs in the AcademicPh.D. sector. Differences among Tiers 1, 2, and 3 are not statistically significant. One puzzling finding is that the average number of faculty publications has a negative estimated effect on placement in the AcademicPh.D. sector. The coefficients on the ability proxies may represent an effect of observed quality in the presence of queuing for jobs in this sector. Having gone to a college with a high average SAT score significantly increases the prob- ability of sorting into the AcademicPh.D. sector. Because of measurement error this variable is imputed for non-U.S. B.A.s this coefficient is likely biased toward zero. The finding of a positive and significant at the P 5 0.10 level coefficient on MATHDIS is compat- ible with both the conventional wisdom that graduate departments have a preference for hiring more math- ematically-oriented job candidates and the dissatisfaction expressed by small college departments and non-aca- demic employers about the highly mathematical orien- tation of recent Ph.D.s Hansen, 1991, p. 1085. The coefficient on SCIENCE is also positive, but not signifi- cant. The coefficients on the indicators of early research experience reveal an interesting pattern. Submitting a paper for publication while a graduate student Table 4 Employment sector probit results Predictor variable Coefficient Constant 20.942 1.778 TIER1 1.318 0.489 TIER2 0.858 0.395 TIER3 1.300 0.491 AGE 20.043 0.042 MALE 0.310 0.307 KIDS 0.114 0.280 SCIENCE 0.262 0.241 MATHDIS 0.149 0.081 RA 0.090 0.197 FACPUB 20.332 0.157 STUDENTS 20.077 0.126 COLSAT 0.002 0.001 TTPROP 0.010 0.010 SUBMIT 0.381 0.227 PRETOP 0.973 0.350 PREPUB 20.633 0.202 N 234 Notes: Dependent variable ACPHD 5 1 if first job is in AcademicPh.D. sector. Standard errors in parentheses. Model also includes eight dummy variables for field of dissertation. significant at P 5 0.05 level; significant at P 5 0.10 level. SUBMIT is positively related to selection into the AcademicPh.D. sector. The effect of prior publications in a top 50 journal — measured by the sum of the coef- ficients on PRETOP and PREPUB — is also positive, though not statistically significant. The negative coef- ficient on PREPUB implies that individuals with prior publications in lesser ranked journals are more likely to take jobs in the Non-Academic or AcademicOther sec- tors, all else equal. The explanation for this latter result may have to do with the nature of their research. To the extent that these articles address a non-academic audi- ence, the authors may prefer to continue to do such research in a non-academic setting. Conditional on other attributes, we find no significant effect of gender on employment sector selection. McMillen and Singell 1994 report a similar finding. 6.2. Publications regressions The results from several publications regressions are reported in Table 5. Separate equations are estimated for all publications columns 1 and 2 and those in top 50 journals columns 3 and 4. The specifications in col- umns 1 and 3 do not use employment sector as a predic- tor variable; the other two columns include an indicator variable for individuals whose first job was in the AcademicPh.D. sector. Consider first the estimates for the models without the job placement variable. Conditional on other factors, graduates of programs in Tiers 1 and 2 i.e. top 10 programs are predicted to have 1.4 to 1.9 more publi- cations and 0.8 to 1.4 more top 50 publications than their counterparts from Tiers 4 and 5. Our finding that graduates of programs with highly productive faculties as indicated by both tier and faculty publications are more productive themselves replicates that of Hogan 1981. Research exposure gained as a research assistant RA is positively associated with post-Ph.D. publications. Ceteris paribus, former RAs are predicted to have 0.77 more articles 0.63 in top journals than individuals who were not RAs. Submitting a paper for publication while still in graduate school is also positively related to publi- cations after leaving graduate school. The coefficients on the two measures of prior publi- cations have the same pattern as in the job placement equation: the coefficient on PRETOP is positive and slightly larger in magnitude than that on PREPUB, which is negative. Their sum is not statistically different from zero using a two-tailed test. This pattern implies that individuals with prior publications in other than the top 50 journals have lower than average publication rates after leaving graduate school. Again, this result may reflect the choice of a career where one’s research is directed primarily to a non-academic audience. Individuals who take longer to meet graduate program 75 T.C. Buchmueller et al. Economics of Education Review 14 1999 65–77 Table 5 Publication regression results All publications Publications in top 50 journals Predictor variable 1 2 3 4 Constant 3.876 3.314 4.096 3.187 20.040 2.093 0.193 1.973 ACPHD — 1.627 0.504 — 1.132 0.307 TIER1 1.742 1.079 0.997 1.100 1.402 0.601 0.884 0.586 TIER2 1.865 0.884 1.358 0.920 1.209 0.522 0.856 0.516 TIER3 1.371 0.962 0.744 0.949 0.755 0.552 0.317 0.528 AGE 20.068 0.094 20.064 0.091 20.024 0.051 20.022 0.049 MALE 20.105 0.673 20.199 0.690 0.605 0.360 0.540 0.346 KIDS 20.430 0.291 20.399 0.279 20.203 0.162 20.181 0.156 SCIENCE 20.425 0.520 20.518 0.499 20.255 0.329 20.319 0.313 MATHDIS 20.216 0.118 20.274 0.124 20.020 0.068 20.060 0.071 RA 0.768 0.450 0.709 0.439 0.623 0.299 0.583 0.284 FACPUB 0.678 0.378 0.821 0.381 0.299 0.235 0.398 0.232 STUDENTS 0.013 0.008 20.011 0.008 20.009 0.005 20.007 0.005 COLSAT 0.001 0.002 0.000 0.001 0.001 0.001 0.000 0.001 TTPROP 20.004 0.002 20.004 0.002 20.002 0.001 -0.002 0.001 SUBMIT 1.261 0.547 1.059 0.544 0.762 0.314 0.622 0.303 PRETOP 2.123 1.279 1.669 1.206 1.320 0.716 1.003 0.680 PREPUB 21.417 0.574 21.156 0.550 20.733 0.350 20.551 0.331 R 2 0.255 0.292 0.233 0.285 Notes: Sample size 5 234. General-heteroscedasticity-corrected standard errors in parentheses. Each model also includes eight indi- cator variables for dissertation field. Significant at P 5 0.05 level; significant at P 5 0.10 level. requirements, as measured by the time from entry to approved proposal, tend to be less productive after com- pleting graduate school. The coefficient on TTPROP implies that an additional year spent completing a pro- posal corresponds to 0.4 fewer articles. This result is interesting in light of the concern expressed by the COGEE panel about observed increases in the average time-to-degree among economics Ph.D.s Krueger et al., 1991. Since we consider TTPROP as a proxy for indi- vidual aptitude, we do not attach causal significance to this result. That is, it does not imply that graduate depart- ments can make their graduates more productive by rush- ing their students through the program. However, this result does have important implications for both graduate faculty and employers, as it suggests that the speed at which students progress through a program is one predic- tor of their future publications success. When we add the initial job placement variable columns 2 and 4, we find that economists employed at research universities are predicted to have 1.6 more publications 1.1 more in top journals than are other economists. As noted, this result may reflect sector-spe- cific differences in research resources, constraints, and incentives, as well as the sorting of individuals with strong publishing prospects into the AcademicPh.D. sec- tor. 13 Conditional on job placement and other variables, 13 To the extent that there is a quality–quantity tradeoff in the production of articles and the returns to article quality are the estimated coefficients on TIER1 and TIER2 are now smaller and no longer significantly different from zero.

7. Implications and conclusions