Source of Decline in School Enrollment

categories, seven education levels, and eight regions; and b an alternative defi nition, which interacts two age groups with the seven education and eight region catego- ries Table 4. While estimates from grouping estimators are generally smaller in abso- lute value than those from 2SLS, they confi rm that increased real wages are associated with rising employment rates and a reduced likelihood of young men being neither in school nor employed. 35 UEVE estimates also confi rm that increased real wages reduce school enrollment: Of the eight wage parameter estimates four of which are shown in Table 4 associated with the two group defi nitions and the four percentiles considered, six are statistically signifi cant at the 5 percent level. Second, our main fi nding is robust to functional form. As Appendix Table A2 shows, moving from a level- log wage model to a log- log wage model when using EWALD and our basic group defi nition yields fairly similar wage elasticities of employment and the likelihood of being neither in school nor employed, both for all young men and for young men with a high school diploma or more education. For these two samples, wage elasticities of school enrollment increase in absolute value when moving to log- log specifi cations, thereby strengthening our fi nding that reductions in school enrollment are a second channel through which relatively more- educated young men increase their labor market participation. 36 Third, considering all young men or those with a high school diploma or more education, both weighted and unweighted 2SLS estimates confi rm that the growth in young men’s labor market participation that follows rising wages comes from a drop in school enrollment and the re- entry into the labor market of some who were neither in school nor employed rather than an increase in the proportion combining school and work Appendix Table A3. Unweighted 2SLS estimates also confi rm that young men with no high school diploma do not appear to reduce school enrollment following improved wage offers. For this subsample, unweighted wage parameter estimates for various outcomes are generally insignifi cant at conventional levels.

B. Source of Decline in School Enrollment

In Table 5, we investigate the source of the reduction in school enrollment found in Tables 1 and 2. To do so, we show the estimated impact of real wages on full- time uni- versity enrollment for potential male university enrollees. 37 The fi rst and second pan- els of Table 5 present results from all estimators OLS, 2SLS, EWALD, and UEVE using the basic group defi nition and the alternative group defi nition, respectively. 35. In seven cases out of eight defined by four percentiles and the two group definitions, most of the esti- mated EWALD impact of real wages on young men’s likelihood of being neither in school nor employed remains when adding province- specific trends to Equation 3. This suggests that the estimates shown in Table 4 are not driven solely by the omission of time- varying factors such as parental income and government spending on education. 36. Wage elasticities of school enrollment also increase, in absolute value, for young men with no high school diploma. However, whether we use real wages or relative wages, these elasticities are not statistically significant at conventional levels for most percentiles considered. 37. We exclude from potential university enrollees young men who have no high school diploma, who have a trades certificate or diploma, or who have a bachelor’s degree or above. The latter restriction is imposed to enable the construction of reliable relative wages: LFS sample sizes for young employed men with a Master’s degree or a PhD are fairly small, thereby precluding the construction of a reliable measure of relative wages for young male bachelor degree holders. Table 5 Real Wages and Full- Time University Enrollment – Potential Male University Enrollees Percentile 15th 1 25th 2 35th 3 45th 4 I. Basic group defi nition four age groups, seven education levels, and eight regions OLS –0.161 –0.100 –0.037 0.032 0.014 0.014 0.015 0.015 2SLS –0.169 –0.176 –0.197 –0.208 0.113 0.123 0.139 0.150 EWALD –0.134 –0.117 –0.112 –0.066 0.046 0.045 0.046 0.043 UEVE –0.132 –0.118 –0.118 –0.077† 0.040 0.040 0.041 0.042 Kleibergen-Paap Wald F statistic 2SLS 50.8 36.8 29.4 26.9 Sample size 106,321 Number of clusters groups 84 Number of grouped observations 621 Average sample size per grouped observation 171.2 II. Alternative group defi nition two age groups, seven education levels, and eight regions OLS –0.166 –0.107 –0.045 0.028 0.019 0.019 0.020 0.020 2SLS –0.174 –0.182† –0.208† –0.211† 0.089 0.096 0.111 0.112 EWALD –0.172 –0.160 –0.150 –0.149 0.048 0.048 0.050 0.052 UEVE –0.169 –0.156 –0.145 –0.146 0.042 0.041 0.043 0.043 Kleibergen-Paap Wald F statistic 2SLS 26.1 21.1 17.6 14.7 Sample size 108,265 Number of clusters groups 47 Number of grouped observations 354 Average sample size per grouped observation 305.8 Full-time university enrollment in 2001 0.218 Source: Authors’ calculations from Labour Force Survey. Notes: The sample consists of unmarried men aged 17–24 with no children and who have a high school diploma, some postsecondary education, a college diploma, a CEGEP diploma, or a university certifi cate below bachelor’s degree. The numbers show the estimated impact of log after- tax real wages on the probability of attending uni- versity on a full- time basis. All models include group fi xed effects, year effects, and all other regressors used in Tables 1–3. Standard errors clustered at the ageeducationregion level are between parentheses for OLS, 2SLS, and EWALD. †: p- value 0.10; : p- value 0.05; : p- value 0.01; : p- value 0.001. The fi rst observation is that under the basic group defi nition, 2SLS estimates indi- cate that a ten- point increase in log after- tax real wages reduces full- time university enrollment by roughly two percentage points, from a baseline enrollment rate of 22 percent. However, the wage parameters are estimated imprecisely. Using the alterna- tive group defi nition yields similar but more precisely estimated 2SLS parameters that are statistically signifi cant at the 6 percent level. 38 Using grouping estimators with the alternative group defi nition—under which sample size per cell averages 305.8 obser- vations—yields parameters that are estimated even more precisely. These parameters suggest that a ten- point increase in log after- tax real wages reduces full- time uni- versity enrollment by about 1.5 percentage point. In addition, re- estimating EWALD and UEVE with log after- tax relative wages under the alternative group defi nition indicates that a ten- point increase in log relative wages lowers full- time university enrollment by between 1.0 and 1.3 percentage point. 39 Together, these results confi rm that declines in full- time university enrollment underlie part of the drop in school enrollment shown in Tables 1 and 2. 40 Could this drop in school enrollment also come from less- educated young men leav- ing school in response to increased wages? While 2SLS results detect no statistically signifi cant reduction in school enrollment for young men with no high school diploma, these results are potentially biased upward due to the omission of parental income and government spending on education. Since our instrumental variable is weakly cor- related with wages when province- specifi c trends are added to Equation 2, we cannot assess how the 2SLS wage parameters of the school enrollment equation change when adding these trends. For this reason, we pursue a different strategy. To allow for the possibility that parental income and government spending on education might have risen faster in oil- producing provinces than in other provinces, we estimate an alterna- tive version of Equation 3 that includes province- specifi c trends. We then compare the resulting estimates to those obtained from Equation 3. We do so using our basic group defi nition. Whether we use EWALD or UEVE, real wages or relative wages, and include province- specifi c trends or not in Equation 3, we fail to detect a negative and statistically signifi cant school enrollment response to increased wages for young men with no high school diploma results available in Table 3.1 in the online appendix. Combined with the possibility that selective interprovincial migration might lead to a spurious drop in school enrollment among this group in oil- producing provinces 41 , this fi nding provides additional evidence that the reduction in school enrollment found in the aggregate does not originate from young men with no high school diploma. 38. Under this alternative group definition, we also get fairly similar 2SLS estimates that are statistically significant at the 8 percent level when we impute wages of nonemployed male potential university enrollees using the 55th, 65th, 75th, or 85th percentile of cell- specific wage distributions. Results are available on Table 5.1 in the web appendix that can be found at http:jhr.uwpress.org. 39. All underlying parameters are statistically significant at the 5 percent level. Results are available on Table 5.2 in the web appendix. 40. In contrast, regression analyses of full- time enrollment in community colleges, junior colleges, or CEGEPs yield little evidence that increased real or relative wages reduce full- time attendance in these postsecondary institutions. 41. This would happen if nonemployed young men who have no high school diploma and have completed their schooling moved to oil- producing provinces in search of high- wage jobs. If so, oil- producing provinces would register a drop in school enrollment among young men with no high school diploma even if school enrollment were inelastic with respect to wages.

C. Selective Migration