Instrument Relevance Identifi cation

Equation 3 that are augmented with province- specifi c linear trends. 28 If parental in- come and government resources spent on education increased faster in oil- producing provinces than in other provinces and are the main factors that reduced the proportion of young men neither enrolled in school nor employed, estimates of ␤ 1 should drop substantially when moving from Equation 3 to alternative versions that include these trends. We fi nd no such evidence: Most of the estimated impact of real wages on young men’s likelihood of being neither in school nor employed generally remains in the aggregate when we use these alternative versions of Equation 3. Whether we aug- ment Equation 3 with province- specifi c trends or not, we generally detect no statisti- cally signifi cant impact of real wages on young men’s likelihood of being both en- rolled in school and employed at the aggregate level. Furthermore, our 2SLS point estimates of ␤ 1 in models of young men’s propensity to combine school and work are negative in the aggregate, even though they are potentially biased upward. Hence, our main results regarding young men’s probability of being neither enrolled in school nor employed or being both enrolled and employed do not appear to be driven by the omission of parental income or government spending on education.

B. Instrument Relevance

Tables 1 to 3 show that the instrument selected is strongly correlated with log after- tax real wages. For all models and samples considered in these tables, the Kleibergen- Paap Wald F- statistic for OIL aert ranges from 10.0 to 73.0. As expected, increased oil prices are positively correlated with real wages, with coeffi cients for OIL aert generally close to 2.0. This suggests that a doubling of oil prices from their 2002 level would increase by roughly ten points the log after- tax real wages of young men for whom the prob- ability of working in the oil industry was equal to 5 percent during the 1997–2000 period. 29

C. Identifi cation

In Figure 1, we group the data on log after- tax real wages and our instrumental vari- able in several ways to assess which dimensions—age, education, region—provide identifi cation of wage impacts. Figures 1a, 1b, and 1c show that movements in oil prices are strongly positively correlated with changes in young men’s real wages when the data is grouped by ageeducationregion, educationregion or ageregion. Figure 1d shows that this correlation is not as strong and that movements in our in- strumental variable are more limited when the data is grouped by ageeducation. This 28. The same procedure cannot be used for Equation 2 since our instrumental variable becomes a weak in- strument when province- specific trends are added to this equation. This is expected since the identification strategy used relies partly on cross- provincial variation in wage growth and youth outcomes. 29. The magnitude of the implied wage increase is plausible. As oil prices doubled from 2001–2002 to 2007–2008, average log real wages of less- educated young men increased roughly 15 points faster in Alberta and Saskatchewan than in the non- oil- producing provinces. As mentioned above, about 6 to 8 percent of less- educated young men were employed in the oil industry in Alberta and Saskatchewan during the 1997– 2000 period. Table 2 Real Wages and Outcomes of Young Men with a High School Diploma or More Education Percentile 15th 1 25th 2 35th 3 45th 4 A Being employed I. OLS 0.871 0.662 0.428 0.179 0.037 0.032 0.029 0.021 II. 2SLS 0.273 0.287 0.319 0.342 0.093 0.102 0.117 0.131 B Being enrolled in school I. OLS –0.218 –0.146 –0.066 0.017 0.025 0.028 0.031 0.030 II. 2SLS –0.329 –0.345 –0.384 –0.411 0.137 0.149 0.170 0.187 C Being neither enrolled in school nor employed I. OLS –0.278 –0.214 –0.139 –0.056 0.018 0.015 0.011 0.007 II. 2SLS –0.091 –0.095 –0.106 –0.114 0.039 0.031 0.046 0.049 D Being enrolled in school and employed I. OLS 0.374 0.301 0.223 0.140 0.051 0.046 0.038 0.031 II. 2SLS –0.146† –0.154† –0.171† –0.183† 0.076 0.081 0.091 0.099 First-stage regressions Dependent variable is log after-tax real hourly wages OIL_aert 2.50 2.38 2.14 2.00 0.35 0.39 0.38 0.39 Kleibergen-Paap Wald F statistic 2SLS 50.3 37.2 31.0 26.1 Sample size 119,750 Number of clusters 116 Mean of young men’s outcomes in 2001 Being employed 0.625 Being enrolled in school 0.472 Being neither in school nor employed 0.110 Being enrolled in school and employed 0.207 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, a trades certifi cate or diploma, or more education. The numbers show the estimated impact of log after- tax real wages on the probability of being employed, being enrolled in school, being neither enrolled in school nor employed, and being both enrolled in school and employed. Separate regressions are run based on various percentiles used for im- puting the wages of nonemployed men. All regressions include group fi xed effects, year effects, month indicators, a renter indicator, a CMA CA indicator, the unemployment rate, and the rate of involuntary part- time employment defi ned at the ageeducationregion level, as well as province- specifi c log real minimum wages, log average real tuition fees for a bachelor’s degree, and levels of Social Assistance income potentially available to single individu- als. Standard errors clustered at the ageeducationregion level are between parentheses. †: p- value 0.10; : p- value 0.05; : p- value 0.01; : p- value 0.001. Table 3 Real Wages and Outcomes of Young Men with No High School Diploma Percentile 15th 1 25th 2 35th 3 45th 4 A Being employed I. OLS 1.271 0.938 0.626 0.244 0.089 0.059 0.060 0.070 II. 2SLS 0.531 0.645 0.755 0.870 0.195 0.244 0.302 0.373 B Being enrolled in school I. OLS –0.095 –0.021 0.040 0.106 0.036 0.048 0.057 0.068 II. 2SLS –0.058 –0.071 –0.083 –0.096 0.131 0.162 0.193 0.224 C Being neither enrolled in school nor employed I. OLS –0.454 –0.353 –0.238 –0.094 0.054 0.044 0.032 0.022 II. 2SLS –0.226 –0.275 –0.321† –0.371† 0.112 0.139 0.172 0.204 D Being enrolled in school and employed I. OLS 0.723 0.564 0.429 0.257 0.156 0.125 0.108 0.076 II. 2SLS 0.246† 0.299† 0.350† 0.404† 0.145 0.168 0.185 0.221 First-stage regressions Dependent variable is log after-tax real hourly wages OIL_aert 2.01 1.66 1.42 1.23 0.45 0.42 0.43 0.39 Kleibergen-Paap Wald F statistic 2SLS 20.5 15.6 10.7 10.0 Sample size 50,882 Number of clusters 48 Mean of young men’s outcomes in 2001 Being employed 0.497 Being enrolled in school 0.624 Being neither in school nor employed 0.136 Being enrolled in school and employed 0.258 Source: Authors’ calculations from Labour Force Survey. Notes: The sample consists of unmarried men aged 17–24 with no children and with no high school diploma. The numbers show the estimated impact of log after- tax real wages on the probability of being employed, being enrolled in school, being neither enrolled in school nor employed, and being both enrolled in school and employed. Separate regressions are run based on various percentiles used for imputing the wages of nonemployed men. All regressions include group fi xed effects, year effects, month indicators, a renter indicator, a CMA CA indicator, the unemploy- ment rate, and the rate of involuntary part- time employment defi ned at the ageeducationregion level, as well as province- specifi c log real minimum wages, log average real tuition fees for a bachelor’s degree, and levels of Social Assistance income potentially available to single individuals. Standard errors clustered at the ageeducationregion level are between parentheses. †: p- value 0.10; : p- value 0.05; : p- value 0.01; : p- value 0.001. Panel A: Age-education-region-year Cells Panel C: Age-region-year Cells Panel B: Education-region-year Cells Panel D: Age-education-year Cells –. 4 –. 2 .2 .4 –.1 –.05 .05 .1 .15 –. 4 –. 2 .2 0. 4 –.1 –.05 .05 .1 .15 –. 4 –. 2 .2 0. 4 –.1 –.05 .05 .1 .15 –. 4 –. 2 .2 0. 4 –.1 –.05 .05 .1 .15 dm_w15 = – 0.000 + 2.706 dm_oil dm_w15 = – 0.000 + 3.126 dm_oil [0.000] [0.289] [0.000] [0.349] dm_w15 = 0.000 + 2.801 dm_oil dm_w15 = 0.000 + 2.211 dm_oil [0.001] [0.561] [0.000] [0.645] Figure 1 Log After- tax Real Wages and Oil Prices Notes: Deviations over time of log after- tax real wages from group- specifi c means on the Y- axis are plotted against oil prices demeaned on the X- axis. Log after- tax real wages from wage imputations for nonemployed young men based on the 15th percentile. A linear fi t is plotted and the resulting equation is shown under each panel. Standard errors from these weighted regressions are clustered at the group level and are between brackets. p 0.001; p 0.01; p 0.05; † p 0.10. suggests that most of the identifi cation comes from cross- regional variation in young men’s wage growth and in changes in our instrumental variable. 30 Likewise, Figures 2a– 2d indicate that movements in young men’s employment rates and in our instrumental variable are more strongly correlated when the data is grouped at least by region Figures 2a–2c than when it is not Figure 2d. Together, Figures 1 and 2 suggest that identifi cation of the impact of wages on young men’s employment originates mainly from cross- regional variation in young men’s employment move- ments, wage growth, and exposure to rising oil prices. Graphical analysis of other outcomes also suggests a predominant identifi cation role for cross- regional variation in young men’s wage growth and in changes in our instrumental variable.

IV. Results