Provincial aggregate employment data

replacement restrictions potentially reduce capital investment so that the wage effect is ambiguous. Extending this reasoning, strike replacement legislation has an ambiguous predicted effect on employment depending on the legislation’s effect on capital and wages and the relationship between capital, wages and employment. Thus, two concluding points are in order. One, nearly all of the claims about replacement policies and employment pertain to the threat of using replacements, not the actual usage. Two, there is considerable debate over whether strike replacement legislation affects employment. What happens in practice is an empirical question.

3. Provincial aggregate employment data

The aggregate data used in this study were constructed primarily from monthly and annual Statistics Canada sources. Appendix A contains details on sources and variable construction. The resulting data set contains 3480 observations consisting of a monthly time series of data covering January 1966 to December 1994 for each Ž of the 10 provinces. Following the minimum wage literature Card, 1992; Neu- . mark and Wascher, 1992 , the dependent variable is the provincial employment- to-population ratio which is the ratio of a province’s number of people employed Ž . Ž . aged 15 and older to the province’s population aged 15 and older . The average employment-to-population ratio for 1966–1994, weighted by provincial popula- tion, is 57.921. As described above, Quebec enacted legislation banning the use of strike replacements in February 1978 while Ontario and British Columbia enacted similar laws in January 1993. To consider what happened to the employment-to- population ratio, Table 1 illustrates that 18 months before the Quebec law became effective, Quebec’s employment-to-population ratio was 55.755. Eighteen months after the law, Quebec’s employment-to-population ratio had increased by 1.755 to 57.510. To assess the impact of the law, however, one needs to know the counterfactual of what would have happened in the absence of such a law. To this end, Table 1 illustrates that in this same period, the average employment-to-popu- lation ratio, weighted by provincial employment, in the other nine provinces increased by 2.233. If the only determinant of employment-to-population ratios that differs across the 10 provinces is a change in provincial strike replacement legislation, then the employment-to-population ratio change of the other nine provinces can be used to infer what Quebec’s change would have been without a new strike replacement law. This yields the difference-in-difference estimate of the effect of Quebec’s strike replacement ban of 1.755–2.233 s y0.478, or a 0.478 reduction in the employment-to-population ratio. The lower half of Table 1 repeats this difference-in-difference estimation strategy for the January 1993 Ontario and Table 1 Difference-in-difference estimates of the effect of strike replacement bans on the employment-to-popu- Ž . lation ratio. Source: Statistics Canada various issues The Labour Force, Ottawa, Ontario, Catalogue CS71-001 a Average provincial employment-to-population ratio Ž . Ž . 1 2 Quebec’s February 1978 strike replacement ban Quebec All others August 1979 57.510 63.140 August 1976 55.755 60.907 Difference 1.755 2.233 Difference-in-difference y0.478 Ontario and British Columbia’s January 1993 Strike Replacement Ban Ontario and British Columbia All others July 1994 61.340 59.477 July 1991 63.067 60.357 Difference y1.727 y0.880 Difference-in-difference y0.847 a Weighted by provincial population. British Columbia strike replacement bans. The result is a 0.847 reduction in the employment-to-population ratio. The validity of these difference-in-difference estimates, however, relies on the assumption that all of the factors affecting provincial employment except strike replacement legislation are changing in the same way across provinces. Thus, differencing the treatment and control group employment-to-population ratio changes removes the impact of the other determinants. If this assumption is not satisfied, the difference-in-difference estimate will be confounded by other factors. Consequently, a multivariate regression estimation strategy is undertaken and the results reported in Table 2. In an attempt to control for observable differences in provincial employment, a variety of additional variables were merged to the provincial employment-to-popu- lation ratio data. These variables are listed, with their sample means and standard deviations, in Table 2 and described in more detail in Appendix A. These variables Notes to Table 2: a Ž . Sample means sample standard deviations in parentheses weighted by provincial population. b Dependent variable: provincial employment-to-population ratio. Each model also contains an intercept and is weighted by provincial population. The standard errors in columns 2–4 are robust to arbitrary forms of heteroskedasticity. U Ž . Statistically significant at the 0.05 level two-tailed test . Tests for province, time trend, year and month effects are joint F-tests. J.W. Budd r Labour Economics 7 2000 225 – 247 233 Table 2 Ž . Regression analysis of the provincial employment-to-population ratio, 1966–1994 standard errors in parentheses . Source: see text a b Variable Sample statistics Regression models Ž . Ž . Ž . Ž . 1 2 3 4 Ž . Provincial employment-to-population ratio 57.921 5.286 Dependent variable Dependent variable Dependent variable U U U Ž . Ž . Ž . Ž . Strike replacement ban 0.211 0.408 y1.276 0.158 y0.976 0.170 y0.467 0.155 U U U Ž . Ž . Ž . Ž . Ž . Reinstatement rights and no replacement ban 0.327 0.469 0.552 0.099 0.691 0.108 0.331 0.010 U Ž Ž . Ž . Ž . Ž . Professional strikebreaker ban and no replacement 0.079 0.270 y0.684 0.155 y0.252 0.170 y0.015 0.152 . ban or reinstatement rights U U U Ž . Ž . Ž . Ž . Ž . Real provincial gross domestic product growth 3.105 3.967 0.027 0.011 0.031 0.009 0.021 0.008 U U U Ž . Ž . Ž . Ž . Ž . Real provincial consumer expenditures growth 3.126 2.630 0.323 0.019 0.288 0.019 0.242 0.017 U U Ž . Ž . Ž . Ž . Ž . Real provincial corporate profits growth 3.130 18.267 y0.005 0.019 y0.004 0.002 y0.003 0.002 Ž . Ž . Ž . Ž . Ž . Provincial real new automobile sales growth y4.115 16.913 y0.002 0.002 y0.003 0.002 y0.002 0.001 Ž . U Ž . U Ž . U Ž . National consumer price index percent change 5.750 3.235 0.182 0.027 0.169 0.025 0.149 0.022 U U U Ž . Ž . Ž . Ž . Provincial union membership percent change 2.742 3.807 0.060 0.007 0.078 0.006 0.072 0.006 U U U Ž . Ž . Ž . Ž . Provincial union density 29.767 5.102 y0.118 0.010 y0.385 0.016 y0.416 0.017 U U Ž . Ž . Ž . Ž . Mandatory strike vote 0.481 0.500 0.030 0.126 y1.096 0.170 y0.376 0.146 Ž . Ž . Ž . Ž . Dues checkoff 0.522 0.500 y0.147 0.086 0.084 0.092 0.104 0.091 U U U Ž . Ž . Ž . Ž . Ž . Cooling-off period days 7.370 6.497 y0.065 0.012 y0.109 0.013 y0.068 0.012 U Ž . Ž . Liberal provincial premier 0.266 0.442 – – 1.043 0.083 U Ž . Ž . New Democratic Party provincial premier 0.133 0.340 – – y0.452 0.082 U Ž . Ž . Parti Quebecois provincial premier 0.090 0.287 – – y0.363 0.116 ´ ´ U Ž . Ž . Social Credit provincial premier 0.096 0.295 – – y1.099 0.131 U Ž . Ž . Union Nationale provincial premier 0.030 0.171 – – 1.117 0.217 U U U Province effects – Yes Yes Yes Province-specific time trends – No Yes U Yes U U U U Year effects – Yes Yes Yes U U U Month effects – Yes Yes Yes 2 Adjusted R – 0.962 0.970 0.975 Sample size 3480 3480 3480 3480 are intended to capture the economic, political and labor policy climate of each province for a given point in time and were selected based on the previous literature, ex ante expectations and data availability. As indicators for the strength of the provincial economy, the data set includes the real growth rates of provincial Ž . gross domestic product GDP , consumer expenditures, corporate profits and the dollar value of new automobile sales. Other explanatory variables included to represent economic and labor market conditions are the national inflation rate, the percent change in provincial union membership and the provincial union density rate. 5 As proxies for the political climate, dichotomous variables were constructed Ž indicating the ruling party in the provincial legislature Conservative is the omitted . category . Since other public policies are sometimes enacted at the same time as strike replacement legislation, three variables indicating the presence of a manda- tory strike vote, dues checkoff and required cooling-off period are also included. Of particular interest is the impact of strike replacement legislation. Thus, three indicator variables were created to indicate the strongest statutory strike replace- Ž . ment policy in effect in each month in each province see table in Appendix A . Ž . The first dummy variable indicates if there was a strong ban on using strike replacements. This variable includes the restrictive bans of British Columbia, Ontario and Quebec, but not the less restrictive bans on permanent strike replace- ments only. The second variable equals one if reinstatement rights for strikers was Ž . the most restrictive policy in effect i.e., temporary replacements are not restricted . In this coding scheme, permanent strike replacement bans are considered reinstate- ment rights, not general replacement bans. The third variable indicates province- months in which professional strikebreakers are banned and the other two types of strike replacement legislation are not present. Each of these variables is con- structed from the provincial statutes using the effective dates listed above. As illustrated in column 1 of Table 2, 21.1 of the weighted monthly observations Ž include a strike replacement prohibition, 32.7 include reinstatement rights and . no ban on temporary replacements , and 7.9 are covered by a professional strikebreaker prohibition only. Since the general replacement bans of Quebec, Ontario and British Columbia clearly place the greatest restrictions on employers’ abilities to use replacement workers, this variable is predicted to have the strongest effect on employment. A professional strikebreaker ban is a weak, potentially obsolete, restriction and is predicted to have minimal employment consequences. Conceptually, the effect of the reinstatement rights variable on employment is somewhere between the other two policies. However, in practice there has been some confusion about provincial 5 The reduced form regressions do not include wage or earnings variables due to potential endogeneity problems. However, including the real provincial average weekly earnings and its growth rate, various lagged values of the variables or instrumenting for average weekly earnings using lagged values does not alter the results. Ž permanent strike replacement policies. For example, Spriggs as quoted in U.S. . Congress, 1991, p. 324 , claims that ‘‘all Canadian provinces, through legislation or jurisprudence, require the reinstatement of striking employees to their jobs at Ž . the conclusion of a work stoppage.’’ Martinello and Meng 1992, p. 178 claim that ‘‘all jurisdictions prohibit the hiring of permanent replacements for striking workers.’’ These claims are incorrect, but they may cloud the interpretation of the results if such confusion is widespread. 6 Column 2 of Table 2 presents the results of regressing the provincial employ- ment-to-population ratio on the independent variables described above, except the political variables, plus year effects, month effects and province effects weighted by provincial employment. 7 The model explains more than 96 of the variance in the monthly employment-to-population ratio. As one would probably expect, real GDP growth and real consumer expenditures growth are both positively related to employment. The coefficients of interest, however, are the first three in column 2. The presence of a strike replacement ban is estimated to reduce the employment-to- population ratio by 1.276, on average ceteris paribus. Moreover, this coefficient is statistically significant at all conventional levels of significance with an absolute Ž t-statistic greater than 8. In December 1992, the population of Ontario aged 15 . and older was 7,926,000 so each one point reduction in the employment-to-popu- lation ratio translates to a reduction in employment of approximately 80,000 jobs. While the 295,000 estimate of Ontario jobs lost produced by Ernst and Young Ž . Council of Ontario Construction Associations, 1992b seems exaggerated, the estimate in column 2 implies significant job loss from enacting a replacement ban. The professional strikebreaker estimate is also statistically significant and Ž . negative, albeit not as large in absolute value as the replacement ban coefficient which is what one would expect since a professional strikebreaker ban is not as restrictive. More puzzling is the reinstatement rights coefficient which is statisti- cally significant and positive. To investigate the robustness of these results, columns 3 and 4 of Table 2 report the results of two alternative specifications. First, while the regression in column 2 controls for province-specific and year-specific fixed effects, ideally one would like to be able to control for province-specific effects that vary over time. For example, while the year effects control for aggregate demographic changes, they do not account for province-specific demographic changes. However, strike re- 6 Ž . Provincial labor law specifies that striking workers remain employees legally while engaged in a legal strike, but only the federal jurisdiction has been unambiguous in ruling that this legal status renders permanent replacements illegal. 7 The unweighted regression results are very similar. For example, the replacement ban coefficient Ž . standard error in the unweighted regression analogous to the weighted regression reported in column Ž . 2 of Table 2 is y1.428 0.190 . placement legislation coefficients cannot be estimated in a model with province Ž . and year effect interactions. Thus, following Levine et al. 1996 , the regression in column 3 adds province-specific time trends. Second, column 4 reports the results of adding a set of five indicator variables Ž denoting the ruling party of the provincial government Conservative is the . omitted category to the regression model. The rationale for the inclusion of these Ž regressors is that if political parties affect provincial employment e.g., via . regulation or taxes and if strike replacement legislation is correlated with various political parties being in control, then the strike replacement coefficients in columns 2 and 3 suffer from omitted variable bias. In fact, the three strike replacement bans were enacted by New Democratic Party or Parti Quebecois ´ ´ governments and the estimates in column 4 imply that political parties are correlated with provincial employment. Focusing on the column 4 results, the strike replacement ban coefficient is Ž . significantly smaller in absolute value relative to columns 2 and 3, but is statistically significant at conventional levels of significance with a p-value of 0.003. The professional strikebreaker coefficient is very small and imprecisely estimated. Since the strike replacement bans are very restrictive whereas the professional strikebreaker ban is not, these results are consistent with the nature of the laws. In contrast, the reinstatement rights point estimate is positive and statistically significant — a surprising result which will be addressed after Table 3. For this natural experiment methodology to yield reliable inferences, however, the policy changes need to be exogenous. If, for example, violent strikes caused a provincial government to restrict the use of replacements and caused business to reduce investment and employment, then the legislation is not exogenous. Lacking suitable instruments, Table 3 reports the results of regressing the employment-to- population ratio on the control variables from Table 2 plus three dummy variables for each strike replacement policy. The first dummy variable for each policy captures what happened to the employment-to-population ratio in the 2-year period prior to the legislation. The other two dummy variables divide each policy into two periods: the first 2 years and after the first 2 years. The first row of Table 3 illustrates that the provincial employment-to-popula- tion ratio during the 2 years leading up to a new strike replacement policy is significantly different than what is predicted to occur with no policy changes, which suggests that policy changes may not be exogenous. However, the negative coefficients in the first row imply that the estimates in Table 2 are biased towards zero, not away from zero. The second two rows are intended to estimate short run and long run effects of each policy change. The strike replacement ban’s negative relationship with aggregate employment outcomes is consistent in both time periods whereas for the other two policies, the estimate is only statistically significant in either the short run or the long run. Since the replacement ban is the most restrictive policy, it was Table 3 Time-varying effects of strike replacement policies on the provincial employment-to-population ratio, Ž . 1966–1994 standard errors in parentheses . Source: see text a Time period Regression coefficients Strike Reinstatement Professional replacement rights strikebreaker ban ban Ž . Ž . Ž . 1 2 3 U U U 1 to 24 months before policy y0.564 y0.361 y1.110 Ž . Ž . Ž . effective date 0.131 0.102 0.107 U U Less than 2 years after y0.685 0.110 y1.142 Ž . Ž . Ž . policy effective date 0.183 0.120 0.254 U U More than 2 years after y1.236 0.451 0.482 Ž . Ž . Ž . policy effective date 0.214 0.120 0.207 Control variables from Yes column 4 of Table 2 U Province effects Yes U Province-specific time Yes trends U Year effects Yes U Month effects Yes 2 Adjusted R 0.976 Sample size 3480 a Dependent variable: provincial employment-to-population ratio. The standard errors are robust to arbitrary forms of heteroskedasticity. U Ž . Statistically significant at the 0.05 level two-tailed test . predicted that its effect would be the strongest. The results are consistent with this prediction, but the results for the other two policies are more puzzling. Taken literally, the reinstatement rights estimates imply that banning permanent strike replacements will not have adverse employment consequences — and may even increase employment. However, this effect is not very robust to alternative specifications. For example, in regressions similar to those reported in Table 2, restricting the sample period to 1966 to 1980, 1985 or 1992 yields a significantly negative strike replacement ban as in Table 2, but negative and imprecisely estimated coefficients for the reinstatement rights variable. Excluding British Columbia or allowing the province effects to be different before and after 1980 also have the same effect. Thus, it appears that the reinstatement rights estimates are being driven by very narrow geographical and time factors. Moreover, as noted above there appears to be some confusion as to the extent that permanent strike replacements are legal in Canada. 8 Lastly, as illustrated in 8 An additional complication is that the aggregate provincial data include federal workers covered by federal jurisprudence providing reinstatement rights. J.W. Budd r Labour Economics 7 2000 225 – 247 238 Table 4 Ž . Regression analysis of private sector unionized bargaining unit employment growth, 1966–1993 standard errors in parentheses . Source: see text a b Variable Sample statistics Regression models Ž . Ž . Ž . Ž . Ž . 1 2 3 4 5 Ž . Bargaining unit employment 0.235 14.079 Dependent variable Dependent variable Dependent variable Dependent variable Ž . growth annualized U U U U Ž . Ž . Ž . Ž . Ž . Strike replacement ban 0.173 0.378 y8.304 2.475 y8.707 3.344 y12.980 4.021 y12.217 3.463 Ž . Ž . Ž . Ž . Ž . Reinstatement rights 0.402 0.490 y2.481 1.895 y1.922 2.050 y1.945 2.455 y2.801 2.170 Ž . and no replacement ban Ž . Ž . Ž . Ž . Ž . Professional strikebreaker 0.133 0.340 y1.281 2.522 y3.662 3.766 y2.030 3.791 y1.499 2.699 Ž ban and no replacement . ban or reinstatement rights Ž . Ž . Ž . Ž . Ž . Ž . Contract duration months 27.018 8.629 y0.162 0.117 y0.168 0.119 y0.166 0.114 y0.217 0.123 Ž . Ž . Ž . Ž . Ž . Provincial industry 1.068 5.644 0.056 0.147 0.045 0.147 0.035 0.146 0.021 0.157 Ž . employment growth U U Ž . Ž . Ž . Ž . Ž . Real provincial gross 3.590 3.482 y0.505 0.274 y0.477 0.268 y0.548 0.264 y0.544 0.258 Ž . domestic product growth U U U U Ž . Ž . Ž . Ž . Ž . Real provincial consumer 3.595 2.384 1.364 0.564 1.394 0.570 1.392 0.544 1.442 0.515 Ž . expenditures growth Ž . Ž . Ž . Ž . Ž . Real provincial corporate 3.848 17.245 y0.076 0.048 y0.076 0.049 y0.063 0.051 y0.067 0.052 Ž . profits growth Ž . Ž . Ž . Ž . Ž . Provincial real new y4.389 15.753 y0.033 0.027 y0.033 0.027 y0.030 0.026 y0.030 0.026 Ž . automobile sales growth Ž . Ž . Ž . Ž . Ž . National consumer price 6.303 2.986 y0.432 0.548 y0.390 0.547 y0.308 0.607 y0.115 0.621 index percent change Ž . Ž . Ž . Ž . Ž . Provincial unemployment 7.792 2.947 0.458 0.389 0.450 0.402 0.299 0.397 0.423 0.440 rate J.W. Budd r Labour Economics 7 2000 225 – 247 239 Ž . U Ž . U Ž . U Ž . U Ž . Mandatory strike vote 0.409 0.492 7.797 2.052 7.214 2.842 5.657 2.747 7.224 2.527 Ž . Ž . Ž . Ž . Ž . Dues checkoff 0.520 0.500 0.551 1.405 0.926 1.786 1.016 1.802 1.034 1.697 Ž . Ž . Ž . Ž . Ž . Ž . Cooling-off period days 7.448 6.474 0.461 0.277 0.489 0.301 0.422 0.298 0.510 0.281 Ž . Ž . Ž . Liberal provincial premier 0.248 0.432 – – y1.998 2.173 y1.580 1.860 U Ž . Ž . Ž . New Democratic Party 0.079 0.269 – – y1.859 2.010 y3.164 1.600 provincial premier Ž . Ž . Ž . Parti Quebecois provincial 0.112 0.315 – – 3.775 4.763 3.905 4.952 ´ ´ premier Ž . Ž . Ž . Social Credit provincial 0.155 0.362 – – y1.792 2.173 y2.672 2.984 premier Ž . Ž . Ž . Union Nationale provincial 0.035 0.183 – – 1.274 4.010 y0.201 3.511 premier Province effects – Yes Yes Yes No Province-specific time trends – No Yes Yes No U U U U Year effects – Yes Yes Yes Yes Industry effects – Yes U Yes U Yes U No Bargaining unit effects – No No No Yes 2 Adjusted R – 0.084 0.083 0.085 0.017 Sample size 3629 3629 3629 3629 3629 a Ž . Sample means sample standard deviations in parentheses weighted by bargaining unit employment. b Dependent variable: bargaining unit employment annualized percent change over the life of the contract. Each model also contains an intercept and is weighted by bargaining unit employment. The standard errors in columns 2–5 are robust to arbitrary forms of heteroskedasticity. U Ž . Statistically significant at the 0.05 level two-tailed test . Tests for province, time trend, year, industry and bargaining unit effects are joint F-tests. 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