Disaggregated unionized employment data

the table in Appendix A, the professional strikebreaker ban variable is constructed from only a single province. Therefore, it is prudent to refrain from strong conclusions regarding these two variables until other data sources are used.

4. Disaggregated unionized employment data

The disaggregated data for this study consist of 3629 Canadian private sector collective bargaining agreements. Human Resources Development Canada main- tains a major wage settlements database of collective agreements covering 500 or more workers. This database includes bargaining unit employment as one of the pieces of information collected. 9 By matching successive contracts, the annual- ized percent change in bargaining unit employment over the life of the collective bargaining agreement can be calculated. After merging control variables similar to those used in the aggregate analysis and deleting observations with missing data, Ž 3629 private sector collective bargaining agreements remain see Appendix A for . details . The resulting panel data set spans 1966 to 1993. The data contain 482 union-establishment bargaining pairs each of which have at least four contracts in the data set. The sample means and standard deviations of the variables used in the disaggregated employment analysis are presented in column 1 of Table 4. Most of the variables are the same as in the aggregate analysis and only the differences will be highlighted here. Since all of the observations are union contracts, the two union membership variables are dropped from the analysis. As additional indica- tors of the economic environment, the provincial 1-digit industry employment growth rate and the provincial unemployment rate are used. Finally, a set of industry effects is added to the regressions. 10 The results of regressing the bargaining unit annualized employment growth rate on the independent variables, weighted by bargaining unit employment, are reported in column 2 of Table 4. 11 In sharp contrast to the employment-to-popula- 9 This Human Resources Development Canada database also serves as the basis for the data sets Ž . Ž . Ž . used by Gunderson and Melino 1990 , Gunderson et al. 1989 , Budd 1996 and Cramton et al. Ž . 1999 to analyze the effect of strike replacement and labor policy legislation on strike activity and wages. 10 As in the aggregate regressions, the bargaining unit regressions do not include wage or earnings variables due to potential endogeneity problems. However, including the annualized percent change over the life of the contract in the real base contract wage, real provincial average weekly earnings and its growth rate, various lagged values of the variables or instrumenting for the base wage growth rate using lagged values does not alter the results. 11 The unweighted regression results are similar with the primary difference being that the replace- Ž . ment ban coefficients are smaller in absolute value in the unweighted regressions, e.g., an unweighted estimated of y6.327 compared to a weighted estimate of y8.304 in the specification of column 2 of Table 4. tion ratio models, the disaggregated employment growth rate models have low predictive ability. To wit, the adjusted R 2 from the regression reported in column 2 is only 0.084. The R 2 values are, however, very similar to those reported in Ž . Ž . Abowd and Lemieux 1991, Table 13.4 and Long 1993, Tables 3 and 4 in their analyses of unionized employment growth in Canada. Notwithstanding the poor overall regression model performance, the estimate in column 2 of the strike replacement ban coefficient is negative and statistically Ž . significant at the 1 level p-value s 0.001 . The reinstatement rights and professional strikebreaker ban coefficients are negative, but both are imprecisely estimated. The hypothesis that the true coefficient for each of these two strike replacement policies is zero cannot be rejected at conventional levels of signifi- cance. To investigate the sensitivity of these results to alternative specifications, columns 3–5 report the regression results for three different specifications. Col- umn 3 adds province-specific time trends as in column 3 of Table 2. Column 4 adds indicator variables for the provincial government ruling party as was done in column 4 of Table 2. To control for unobservable time-constant bargaining unit heterogeneity, column 5 adds a set of bargaining unit fixed effects. As was the case for the aggregate analysis, the pattern of results for the strike replacement legislation coefficients across these various specifications is fairly Table 5 Time-varying effects of strike replacement policies on private sector unionized bargaining unit Ž . employment growth, 1966–1993 standard errors in parentheses . Source: see text a Time period Regression coefficients Strike replacement ban Reinstatement rights Professional strikebreaker ban Ž . Ž . Ž . 1 2 3 Ž . Ž . Ž . 1 to 24 months before 1.160 3.672 7.873 4.348 3.678 2.840 policy effective date U Ž . Ž . Ž . Less than 2 years after y12.967 5.878 0.751 2.170 y2.919 5.778 policy effective date U Ž . Ž . Ž . More than 2 years after y8.739 3.833 0.874 2.353 0.866 2.866 policy effective date Control variables from Yes column 5 of Table 4 U Year effects Yes Bargaining unit effects Yes 2 Adjusted R 0.021 Sample size 3629 a Dependent variable: bargaining unit employment annualized percent change over the life of the contract. The standard errors are robust to arbitrary forms of heteroskedasticity. U Ž . Statistically significant at the 0.05 level two-tailed test . stable. The reinstatement rights and professional strikebreaker ban coefficients remain negative, imprecisely estimated and statistically insignificant. The esti- mated strike replacement ban coefficient is consistently negative and statistically Ž . significant with the largest p-values equalling 0.009 column 3 . Moreover, the point estimates for the strike replacement ban are not modest. For example, relative to the mean growth rate of 0.235 per year in the full sample, the estimate in column 2 predicts that bargaining units in a province with a strike replacement ban will be shrinking by 8.304 per year. The point estimates in the Ž . other specifications are even larger in absolute value . Table 5 presents the results of allowing the strike replacement policy effects to vary over time. In contrast to the aggregate case, one cannot reject the null hypothesis that unionized employment growth rates deviate from the no-policy mean during the 2 years leading up to a policy change, suggesting that endogenous policy changes might not be affecting the results. In terms of short run vs. long run policy implications, the estimates for reinstatement rights and professional strike- breaker bans are statistically insignificant in both time periods. The strike replace- ment ban effect gets smaller 2 years after the policy effective date, but is still quite large. Note that because the Ontario and British Columbia bans were enacted towards the end of the sample time period, the long run strike replacement variable relies solely on Quebec’s experience.

5. Conclusion