Descriptive results Directory UMM :Data Elmu:jurnal:E:Economics of Education Review:Vol20.Issue2.2001:

152 B.J. Surette Economics of Education Review 20 2001 151–163 colleges to which transfers are accepted. Rouse 1995 examines the effect of starting at a two-year college on total years of college completed and the probability of completing a bachelor’s degree. The decision to transfer is not examined. Finally, Grubb 1989a examined indirectly the decision to transfer from a two-year to a four-year college using aggregate state-level data. His results suggest that women are less likely to transfer, though data limitations prevent him from directly exam- ining that phenomenon. This paper models attendance at both two-year and four-year colleges, as well as the decision to transfer from a two-year to a four-year college, using a 12-year panel from the National Longitudinal Survey of Youth NLSY. These data allow the direct examination of the decision to transfer and provides an understanding of which factors play a major role in that decision. The main conclusion from this paper is that, despite the large number of economic and demographic variables used to control for differences in college attendance decisions, women are less likely than men to transfer.

2. Descriptive results

The data for this study are drawn from the 1979 through 1990 waves of the NLSY. To ensure that all post-secondary education decisions during this period are observed, individuals who turned 18 prior to 1979 are excluded. 1 Only high school graduates are retained. The final data set contains 2413 women and 2349 men. For this analysis, successive waves of data are retained for each individual. 2 The final data contain 20,432 female and 18,832 male person-years. Mean characteristics for 1 Individuals with extreme values for the variables of inter- est, or those with obviously inconsistent data, were dropped. The military sample and the poor white samples are also excluded because membership in these groups is based at least in part on individual and family decisions that may be related to the outcomes examined in this paper. The model retains the over-samples of black and Hispanic individuals, which are sim- ply additional members of these demographic groups, drawn randomly. 2 Prior to the first missed interview missing data were imputed—primarily by examining individuals’ responses to rel- evant survey questions over several years. For cases in which this did not provide the necessary information, predicted values from multivariate models or group means replaced missing values. Only a small fraction of any one variable was imputed based on predictions or group means. Collectively, imputation allows the retention of a great deal of non-imputed information that would otherwise be discarded. this sample are presented in Table 1. Definitions of the variables are provided in Appendix A. The college attendance and accumulated credit vari- ables discussed here were derived from FICE 3 codes and self-reported credit histories provided by each college student in the Geocode version of the NLSY. FICE codes were matched to US Department of Education hereafter “Education Department” data on the highest degree offered at each school. Two-year colleges are defined as those offering an associate’s degree as their highest degree. Schools offering bachelor’s or graduate degrees are defined to be four-year colleges. 4 The use of FICE codes facilitates a consistent definition of college type for each person; misreporting of college type is reduced or eliminated. College attendance is defined in this study by whether or not an individual enrolled in and completed credits at each type of college in each calendar year. 5 Attendance rates at two-year and four-year colleges are presented by age in Table 2. The table shows that women are more likely than men to attend two-year college during the prime college-going ages 18–22. 6 It is well known that since the 1960s, women’s overall college attendance rates have increased dramatically and have now over- taken men’s college attendance rates. 7 These data indi- cate that the attendance gender gap favoring women reflects higher two-year attendance. This paper defines a “transfer” as any student who attends four-year college concurrent with or subsequent 3 Federal Interagency Committee on Education FICE codes are unique identifiers for each accredited college or university in the United States. 4 Schools that offer only a vocational certificate are not treated as colleges in this paper. Rather, credits earned at vocational institutes are included in the analysis as part of the formal vocational training programs. 5 It would have been interesting to identify individuals who attended college but did not complete any credits. Owing to the complexity of tracking individuals and college types over a 12- year panel, it was simply not feasible with these data to do so. Such individuals are treated as having never attended. 6 The 1996 Digest of education statistics United States Department of Education, 1996 reports college enrollment rates for the early 1980s higher than those reported in Table 2. Two factors may explain this difference. First, my data include the NLSY’s black and Hispanic over-samples; these groups have lower college attendance rates than the population as a whole. Excluding them raises attendance rates by a few percent- age points. A second cause is definitional. Students in my data must complete credits to be counted as having attended college. The Education Department defines attendance by enrollment status as of October 1. Students who drop out after October 1, but before completing any credits, are not counted in my defi- nition of attendance, but are in the Education Department’s definition. 7 1996 Digest of education statistics p. 182, Table 175. 153 B.J. Surette Economics of Education Review 20 2001 151–163 Table 1 Descriptive characteristic of the sample, by gender a Women Men 1 All HS 2 Ever attended 3 All HS 4 Ever attended graduates b 2-yr college c graduates b 2-yr college c Demographics Currently married 0.338 0.166 0.229 0.062 Have young children under 5 0.349 0.148 0.151 0.027 AFQT score 0.483 0.643 0.501 0.757 Black random sample 0.079 0.076 0.063 0.064 Black over-sample 0.229 0.221 0.241 0.206 Hispanic random sample 0.043 0.041 0.039 0.041 Hispanic over-sample 0.127 0.161 0.126 0.168 Urban location 0.793 0.826 0.768 0.794 Age 22.6 26.5 22.6 26.4 Costs and benefits of attendance Return to college 0.474 0.596 0.472 0.586 HS graduate wage in logs 1.923 1.954 1.926 1.971 Two-year tuition US1000s 0.859 0.864 Four-year tuition US1000s 1.715 1.724 Tuition differential 3.430 3.673 Unemployment rate 8.042 9.065 7.979 9.104 Human capital 2-yr credits divided by 30 0.259 1.158 0.221 1.168 4-yr credits divided by 30 0.734 0.752 LM experience years 1.775 0.382 1.991 0.403 Vocational training 0.146 0.512 0.144 0.564 Gap in schooling n.a. 0.851 n.a. 0.858 Vocational field n.a. 0.151 n.a. 0.129 Academic field n.a. 0.038 n.a. 0.032 Number of observations 20,432 764 18,832 626 a Definitions of variables are provided in Appendix A. b Includes multiple observations of individuals observed in more than one of the 12 years examined. c Includes individuals only in the last year they are observed during the 12-year panel. Variables that can change with time are set to their value at age 20 and are designated with an asterisk for use in the transfer model. to two-year attendance. 8 In other words, if a student attends two-year college at age 19 and attends a four- year college at age 20, she is considered to be a transfer from age 20 onward for descriptive purposes. Table 2 shows cumulative transfer rates by age and demonstrates that men are about 5 percentage points more likely to 8 Concurrent spells need to be included as transfers because I cannot separately identify spring, summer, or fall attendance for a particular calendar year. This is due to high rates of miss- ing data for enrollment dates, especially the month of enrollment. transfer than women. 9 This difference is statistically sig- nificant. 10 These figures demonstrate two potentially contradic- tory facts about women’s and men’s college attendance patterns: women are more likely than men to attend two- 9 Grubb 1989b reports similar aggregate transfer rates based on the NLS-72 and the High School and Beyond surveys. 10 Table 2 shows slight declines with age in cumulative trans- fer rates. Attrition from the sample accounts for these declines; older individuals need to be in the panel longer. This suggests attrition may be non-random. We address this potential problem by conducting the analysis on individuals in the last year they are observed, not at any particular age. Moreover, as the declines are more severe for men than for women it seems unlikely that attrition explains women’s lower transfer rates. 154 B.J. Surette Economics of Education Review 20 2001 151–163 Table 2 College attendance rates among high school graduates Age Two-year attendance rates Four-year attendance rates Cumulative transfer probability Women Men Women Men Women Men 18 14.0 12.1 29.3 28.7 19 17.8 14.2 28.9 28.3 8.6 7.6 20 17.5 14.4 29.1 28.1 20.8 21.3 21 12.2 10.7 26.5 27.3 29.1 31.8 22 8.8 6.4 23.4 25.7 33.1 39.1 23 4.7 5.2 14.0 17.7 33.1 37.7 24 3.2 3.7 7.0 9.7 30.1 36.8 25 3.8 2.8 5.0 5.2 29.6 34.3 P-values: two-tailed test 0.007 a 0.768 a 0.012 b of no gender difference Source: Author’s calculations from the 1979–1990 NLSY. a P-value for two- and four-year attendance calculated over ages 18 through 23. b P-value for cumulative transfer probability calculated for individuals older than 20 in the last year they are observed in the data. year college but are less likely to transfer. This suggests that women and men may have different reasons for attending college, and the same reasons that cause women to attend two-year college at higher rates than men may also explain why they transfer less often. Two of the most obvious potential explanations for these pat- terns are differing domestic responsibilities and occu- pational ambitions. The model presented in the next sec- tion uses a human capital framework to sort through these and several other plausible explanations for the transfer rate difference. Table 3 shows that the lower transfer probabilities observed for women may affect overall education levels. It presents degree completion rates for the oldest cohorts in my sample age 21 to 29 in the last year they are observed. Among all high school graduates row 1, women have a higher probability of completing an Table 3 Degree completion rates among HS graduates aged 21 + years a Associate’s degree Bachelor’s degree b Either degree b Women Men P-value c Women Men P-value c Women Men P-value c Obs HS graduates 10.7 8.5 0.011 18.9 19.0 0.914 27.7 25.3 0.064 4471 2-yr students 27.1 25.0 0.424 14.8 19.6 0.929 37.7 38.4 0.797 1208 4-yr students 16.1 15.1 0.569 49.5 50.7 0.622 60.9 60.0 0.702 1656 Transfers 48.8 44.0 0.246 36.2 41.1 0.223 71.3 68.4 0.451 575 Source: Author’s calculations from the 1979–90 NLSY. a Includes high school graduates who never attended college. Only uses observation from the last year that an individual is observed in the data. b Includes individuals who attended both 2-year and 4-year college. c P-values are for two-tailed test of no gender difference in means. associate’s degree, but have about the same probability as men of completing a bachelor’s degree. Among those who attended a two-year college at some point row 2, men have a higher probability of completing a bachelor’s degree. These differences are statistically significant at the 5 percent level. Among those who actually transfer row 4 the difference bachelor’s degree completion rates is not statistically significant—though this may be due to the relatively small number of observations in this cell. The degree completion figures paint an ambiguous picture of the overall effects of two-year college attend- ance and transferring on women’s and men’s educational attainment. Whether two-year college attendance affects overall education levels, and whether there is a gender difference in these effects, has been the subject of several previous studies with inconclusive results Grubb, 1989a; Rouse, 1995. This study’s objective is more 155 B.J. Surette Economics of Education Review 20 2001 151–163 modest: it documents the fact that women differ from men in their college attendance and transfer decisions and attempts to discover what factors can and cannot explain these differences.

3. A bit of economic theory