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

68 T.C. Buchmueller et al. Economics of Education Review 14 1999 65–77 ships among graduate training, job placement, and publi- cation productivity.

4. Data

4.1. The COGEE survey Our principal source of data is the COGEE survey of Ph.D. economists receiving their degrees in 1977–78 and 1982–83. The COGEE sample is not a random sample of all Ph.D.s within the two cohorts. Rather, it follows a stratified sampling procedure based on separating graduate programs into “quality tiers” according to department rankings presented in Jones et al. 1982. The first tier consists of the top six programs. Tiers 2 and 3 consist of programs ranked 7 to 15 and 16 to 30, respect- ively. Tier 4 is defined as programs ranked from 31 to 50, and Tier 5 includes the remaining 44 Ph.D. programs listed in Jones et al. With the exception of one that refused to participate, all programs in Tiers 1 and 2 are in the sample frame, as are seven of the 15 Tier 3 programs, and 14 of the 61 programs in Tiers 4 and 5. Each graduate program in the COGEE sample frame supplied names and addresses of graduates receiving their Ph.D.s in 1977–78 and 1982–83. Using these address lists, surveys were sent to all graduates residing in the United States. The response rate was roughly 60 percent. We have reason to believe that the sample resulting from this process differs from the population in several ways. First, and most obviously, our sample is weighted more heavily toward graduates of top programs. 4 Even within tiers, there are two other likely sources of bias. First, Ph.D.s not residing in the U.S. were not sampled. Second, Ph.D.s who are most involved in the profession may have been more likely to respond to the survey. Thus, our sample should have a higher publication rate than the population of all Ph.D.s. A comparison of publi- cations by our sample and by the population of Ph.D.s in these cohorts, presented in Section 5, confirms this belief. This bias is likely to vary across sub-groups. In particular, the bias will be larger for groups with rela- tively low participation in the profession — e.g. individ- uals in non-academic jobs — than for groups with rela- tively high participation — e.g. those in academic jobs. 4 Tier 1 graduates represent 34 of our COGEE sample, but only 20 of the population. The corresponding percentages for Tier 2 are 42 and 21. Relative to the population, Tiers 3, 4 and 5 are under-represented in our sample. The COGEE results reported in Hansen, 1990, 1991; and Krueger et al., 1991 were weighted to account for the sampling scheme. Table 1 presents definitions and summary statistics for the variables used in our analysis. The variables can be divided into five categories: research publications, post- training employment setting, graduate department characteristics, early research experience, and individ- ual characteristics. 4.2. Description of variables 4.2.1. Research publications Names of the Ph.D. economists in our sample were matched with bibliographic entries from the JEL’s on- line Economic Literature Index ELI Ekwurzel and Saf- fran, 1985. We use two different measures of research output. The first measure counts all publications, be they journal articles, conference reports, or chapters in books; the second counts only articles in journals ranking among the top 50 Leibowitz and Palmer, 1984. 5 Neither measure differentiates among publications with multiple authors. 6 Our regression analysis of the COGEE sample data focuses on articles published within the first six years after leaving graduate school. Implicit in this taxonomy is our assumption that “the clock starts” when an econ- omist begins his first job rather than when he receives the Ph.D. Six years is chosen as the interval length to correspond with the typical tenure schedule. 7 In addition, the names of all 1977–78 and 1982–83 Ph.D. recipients published in the JEL were also matched with the ELI data. Since we lack survey data on the full population we cannot use these data to investigate the impact of graduate training on research publications. However, the population publications data help to pro- vide a more complete picture of economists’ early pub- lishing activity and allow us to better understand how our sample differs from the population. 5 Leibowitz and Palmer present several different rankings. The one we use ranks journals on the basis of “impact-adjusted” citations number of citations weighted by the quality of the citing journal for articles published between 1975–1979. 6 The second measure makes some distinction based on the perceived quality or selectivity of the publication outlet, but substantial heterogeneity remains. An alternative approach is to use the Leibowitz and Palmer rating scale to construct a “qual- ity-adjusted” output measure, as did Sauer 1988. Preliminary analyses using such a measure yielded results which were quali- tatively similar to those reported below. Given this similarity, we prefer total publications because the interpretation is more straightforward. 7 Long 1978 argues that studies which find research output to be a powerful determinant of job placement are often flawed, because they do not limit measured research to work done prior to obtaining the job. As such, these analyses, such as Hansen et al. 1978, confound the effect of research output on job placement with that of job placement on subsequent output. Our analysis should not be subject to this criticism. 69 T.C. Buchmueller et al. Economics of Education Review 14 1999 65–77 Table 1 COGEE data: variable definitions and summary statistics Variable name Description Mean standard deviation Research output ALLPUB Total publications in first 6 years after 3.15 3.69 graduate school TOP50J Publications in top 50 economics journals, 1.47 2.17 first 6 years after graduate school Post-training employment ACPHD First job was in AcademicPh.D. Sector 0.462 0.500 0,1 ACOTH First job was in AcademicOther Sector 0.256 0.436 0,1 NONAC First job was in Non-Academic Sector 0.282 0.450 0,1 Graduate department characteristics TIER1 Graduate of Tier 1 program 0,1 0.342 0.475 TIER2 Graduate of Tier 2 program 0,1 0.415 0.494 TIER3 Graduate of Tier 3 program 0,1 0.098 0.298 FACPUB Average faculty publications 2.99 0.77 STUDENTS Number of graduate students in program 3.27 0.88 Early research experience PRETOP Number of publications in top 50 0.090 0.341 economics journals prior to completing graduate school PREPUB Number of publications prior to 0.376 0.582 completing graduate school SUBMIT Submitted a paper for publication while in 0.282 0.451 graduate school RA Worked as a research assistant while in 0.356 0.481 graduate school 0,1 Individual characteristics COLSAT Mean SAT score of undergraduate Alma 1129 151 Mater SCIENCE BS or MS in mathematics or science 0,1 0.226 0.419 MATHDIS Level of mathematics used in dissertation 2.68 1.27 TTPROP Time years taken to complete approved 3.33 1.39 dissertation proposal MALE Male 0,1 0.889 0.315 AGE Age at time of Ph.D. 28.8 2.810 KIDS Had children at time of Ph.D. 0,1 0.222 0.416 Notes: Number of observations 5 238. Source:Jones et al. 1982. 4.2.2. Post-training employment characteristics The employment settings in which economists work can be characterized in several ways. The first and most obvious cut is to distinguish academic from non-aca- demic jobs. We further divide the academic sector into two categories: jobs at universities granting Ph.D.s in economics are defined to be in the “AcademicPh.D.” sector, while those in other universities and in exclus- ively undergraduate institutions are defined as being in the “AcademicOther” sector. While some heterogeneity remains in each group, we do not attempt a finer categor- ization because of our small sample size and limited data on job characteristics. 4.2.3. Graduate department characteristics We measure departmental reputation using the quality tiers described above. The average number of publi- cations by the department’s faculty is used as an additional measure of program quality. A third graduate program characteristic is the size of the program, meas- ured by the number of students. 8 8 The source of both of the department characteristics is Jones et al. 1982. 70 T.C. Buchmueller et al. Economics of Education Review 14 1999 65–77 4.2.4. Early research experience If considerable practice in conducting research facili- tates future research performance, then students who gain early research experience should be more likely to produce acceptable dissertations within reasonable time limits and to become more successful researchers after receiving their Ph.D.s. The COGEE survey data provide several indicators of such experience. One important source of early research experience is work as a research assistant. As apprenticeships of sorts, these opportunities may give students an improved understanding of the research process and, in some cases, will directly lead to collaborative work. We also expect individuals who submitted papers for publication prior to completing their Ph.D.s would be more productive after doing so. The variable SUBMIT equals 1 if the individual submitted a paper for publi- cation while in graduate school. The JELELI data allow us to construct two additional measures of pre-Ph.D. research experience. The variable PREPUB represents the number of publications prior to leaving graduate school. PRETOP is the number of publications in top 50 economics journals during that same time period. 9 4.2.5. Individual characteristics To estimate the effect of early research experience on post-Ph.D. publications, it seems desirable to condition on individual aptitude. If research assistantships are given to the best students in a department, then without adequate controls for ability, we would overstate the true impact of the experience gained in such positions. Simi- larly, we expect that individuals who submit papers for publications while in graduate school will have a greater than average interest in and aptitude for research. The COGEE survey provides information on where each respondent received his or her baccalaureate degree. We construct a variable equaling the average SAT score from the individual’s undergraduate Alma Mater. 10 Two additional variables are included to assess the plausibility of the common perception that mathematical ability is highly correlated with success in economics graduate programs and in the profession more generally. 9 Note that a positive value for either of these publication measures does not imply that SUBMIT equals 1, because these articles may have been written prior to entering graduate school. Similarly, SUBMIT 5 1 does not imply a positive value for either PREPUB or PRETOP, because not all articles submitted as a graduate student will be published prior to leaving graduate school, if at all. 10 The survey asked respondents to report their graduate rec- ord examination scores, but a high rate of non-response for this question keeps us from using this information. The source of the SAT data is Barron’s Guide to Colleges 1990. For non- U.S. B.A.s we impute this variable using the average for the individual’s graduate program. SCIENCE is a dichotomous variable equal to one for all persons with a B.S. or M.S. in mathematics, science, or engineering, and equal to 0 otherwise. The variable MATHDIS is a self-report using a five point scale of the level of mathematics used in the dissertation Hansen, 1991, pp. 1073–1075. A time line reported by COGEE survey respondents allows us to observe the time it took each individual to progress through the graduate program. The variable TTPROP measures the length of time from entering the program until the completion of an approved dissertation proposal. We choose this measure over time-to-degree because the latter is not independent of an individual’s experience on the job market or his first job. We expect individuals who take longer to get to the dissertation stage will tend to be less productive after receiving their degrees. Some demographic variables are also included in the analysis — age at time of degree, the presence of depen- dent children at time of degree, and gender. 11 We also have data on dissertation field as categorized according to the JEL classification scheme.

5. Descriptive analysis