What Did Unions Do? An Analysis of Canadian Strike Data, 1901-14

Michael Huberman, Trent University
Denise Young, University of Alberta

1. Motivation and Theory

Gavin Wright (1987) observed that economic history forms a bridge between labor economics and labor history. The gap between the disciplines is most evident in their different approaches in understanding the determinants of strikes and strike dimensions. Using a new data source on individual strikes in Canada, the aim of this paper is to bring together recent contributions by labor economists and historians in an analysis of the determinants of strikes durations between 1901 and 1914.

The new information economics maintains that strikes serve the important function of generating information. In its most popular variant workers go on strike to get information about the profitability of firms; firms are more willing to settle in periods of expansion, and as a result strike durations are countercyclical. Empirical tests of this model for Canada, Harrison and Stewart (1989) is the best example, have been restricted to the post 1945 period. But how robust are these models? It would be expected that in the early period of union consolidation, as collective bargaining became regularized, environmental variability would be reduced. More information would have been available to both parties on wage offers and demands, and there would have been alternatives to strikes. It would be expected, therefore, that as unionization grew strike frequencies would be reduced and they would be limited increasingly to periods of cyclical change when environmental variability was greatest. Similarly, durations would become increasingly countercyclical over time.

For more than a decade, a new generation of Canadian labor historians has been exploring many of these issues. Much of this work has been published in Labour/Le Travail, and summaries of this literature are found in Palmer (1992) and Morton (1988). Although there is some agonizing over finding trends or explaining Canadian 'exceptionalism,' this new research situates strikes in their historical context: their location, sector, and political and social background. Kealey (1989, p. 228) wrote: "Each national working class experience must be studied historically and understood in light of contrasting experience, not held up against a reified model, which never existed."

This literature has eschewed quantitative methods and questions. An exception is the work of Kealey himself (1989, Cruikshank and Kealey 1987) who, in a project for the Canadian Historical Atlas has revised Canadian strike data and, using the familiar framework of Shorter and Tilly (1974), has produced statistics on strike dimensions from 1891 to 1961. In his own description of the data until 1930, Kealey emphasized the strike waves of 1899-1903, 1912-1913, and 1917-20, as well as the political, regional and sectoral features of these mobilizations. Kealey also observed, however, the cyclical character of strikes before 1930, and as for strike durations he found that they declined throughout the period, with the exception of the 1920s. Kealey relies exclusively on descriptive statistics. Multivariate analysis offers a more rigorous method of separating cyclical effects and strike waves and their determinants.

It is precisely here that the historical end economics literature intersect. The economics literature needs to incorporate the historical context in evaluating their models, while the historian needs theory, however reified, to assess the relative importance of mobilization and cyclical strikes.

2. Data

The data for this paper is based on the Strikes and Lockouts File of the Department of Labour in Canada. In response to the growing number of work stoppages in the late 19th and early 20th centuries, the Department began collecting detailed information on strikes: length of dispute, size of work force involved, cause of dispute, how the strike was settled, in whose favor, and so on. The reports did not include some important information, notably on the actual wage settled upon at the end of the dispute. From these accounts we have been able to assemble a data set of 1129 strikes. This does not represent the entire population of Canadian strikes in the period, but the time series of the strikes from the files corresponds closely with the aggregate time series reported by the Department of Labour. As for the geographic and sectoral distribution of strikes, it is probably the case that the reports to the Department may underestimate strikes in Quebec and in the eastern and western regions, but it is unclear how seriously this bias affects the analysis that follows.

The basic descriptive statistics of the sample are reported in the first column of Table 1. For indicator variables, like the location of the strike, the means are just the proportion of the sample with the given attribute. Some features stand out in comparison to the post 1945 period. Harrison and Stewart (1989) reported that for their sample of contact data between 1946 and 1983 the mean and median duration were 35.51 and 20 days; in the early period the corresponding figures were 22.7 and 7 days. Another difference is that after 1945 the median duration of wage-issue strikes was greater than non-wage strikes (20 and 9 days) but that this pattern was reversed in the earlier period. The median wage strike lasted 6 days, while the median duration of a strike about a union issue (e.g. recognition) was 12 days before 1914. In the longer version of the paper we also compare survival rates in the two periods.

3. Method and Results

A hazard model is used to model strike durations, taking into account firm and worker characteristics, policy initiatives, macroeconomic indicators, a strike wave indicator, and other relevant variables. In such a model the focus is on the likelihood that an ongoing strike will end in the next short interval. There are several possible functional forms for hazard functions and we use the log-logistic form. The coefficients on the independent variables in the hazard model are then the partial derivatives of (the expected value of) log duration with respect to the corresponding independent variable. If duration is negatively related to an independent variable, the settlement rate is positively related to that variable.

Table 1 presents some of our results. (The longer version of the paper includes OLS results with log duration as the dependent variable.) The last column in the Table reports the change in the predicted median duration due to (1) the change of the respective dummy variable from 0 to 1; or (2) the change of the non-dummy variable from its mean value to mean +1. For example, the addition of one hundred extra workers to a strike, adds about two days to a strike. Our more important findings are:

* Bigger (i.e. more strikers) strikes, strikes involving more than one firm, violent strikes and strikes of skilled and machine workers lasted longer.

* The number of strikers (whether lagged or current) - an indicator of a strike wave - was positive and significant when the entire sample was used. In a smaller sample which distinguishes between the type of union involved (Canadian or international), the strike wave variable is insignificant.

* Disputes including women were not different from those involving men only.

* Union led strikes were longer, but their was no difference whether the strike was led by an international or Canadian union.

* Durations of strikes was not significantly different across the country. The labor history literature suggests otherwise. Moreover, labor historians have observed that sectors like textiles where the intrusion of new managerial techniques were greatest, incurred lengthy disputes. The results do not confirm this. The literature also suggests that the introduction of legislation in 1907 which introduced mandatory conciliation in certain sectors, had the result of lengthening disputes that did occur. The dummy for 1907 onwards is insignificant.

* The real output (per capita) variables indicate that the response of durations to output (measured in this case as deviations from trend in levels) depended on the strike issue. Wage and work issue strikes were countercyclical. The relevant coefficients are the sum of the GNP and interaction variables. The absolute value of the t-statistics for both work and wage issue strikes are both greater than two. Union issue strikes, however, were insensitive to the business cycle. Other ways of measuring output changes were tried, including an indicator for good and bad years, but the results did not change greatly.

4. Remarks

The period 1901-14 were years of transition in Canadian industrial relations. Although there is some support for both the strike wave hypothesis and information model, perhaps it is best to see the two approaches as complementary. Union organizing drives were a feature of strike waves that did not necessarily correspond to the business cycle before 1914. (The t-test value for the combination of the GNP variable and the UnionxGNP is 0.77.) This result confirms similar findings using aggregate data on union membership. As unions consolidated they began to serve the important function of acquiring information for their members about the profitability of firms; the duration of union led disputes about wages and working conditions became increasingly longer and countercyclical. In light of our introductory observations, models of strikes need to be considered in their historical settings.



Variable		Mean	     Coefficient	           T-Ratio        ∆Duration

Constant				 30204		 2.15
Strikers	183.81	 0.0003		 3.76		 0.002
Firms involved	  0.35	 0.5768		 5.27		 4.786
Females		  0.082	-0.1576		-0.73	 	-1.147
Union involved	  0.81	 0.7046		 4.55		 4.534
Skill			    0.63	 0.2219		 1.97		 1.675
Violence		    0.03	 0.7058		 3.45		 7.691
Legal (1907)				-0.1208		-0.48		-0.958
Number of strikes
	in year				  0.0042		 1.96		-0.033
Year					-31.715		-2.15		-0.201
Yearxyear				  0.0036		 2.16		-0.202
Output effects:
	GNP				 -0.0205		-0.64		-0.158
	UnionxGNP			 -0.0009		-0.05		-0.007
	IsunionxGNP			 -0.0211		 0.68		 0.166
	IsworkxGNP			 -0.0347		-1.00		-0.265
	IswagexGNP			 -0.0347		 1.23		-0.264
	IsmultxGNP			 -0.0161		-0.57		-0.124
Location effects:
	West		    0.10	 -0.3028		-0.60		-2.332
	East		    0.22	   1.1152		 0.64		 0.939
	Central		    0.67
Strike issues:
	Multiple issue            0.25	  0.7483		 3.74		 6.733
	Wage issue	    0.67	  0.4511		 2.28		 3.740
	Union issue	    0.20	  0.4590		 2.14		 4.209
	   Conditions     	    0.20	  0.4485		 1.68		 4.301
	Misc. issue	    0.20
Industry effects:
	Building		    0.06	-0.7763		-2.53		-4.470
	General		    0.26	-0.7566		-2.70		-5.123
	Apparel		    0.11	-0.3053		-0.09		-0.234
	Food&Tobacco	    0.05	 0.0882		 0.25	   	 0.713
	Machine		    0.02	 0.7667		 1.80		 8.849
	Metals		    0.09	-0.3297		-0.11		-0.253
	Mining          	    0.07	 0.1011		 0.34		 0.819
	Shoes		    0.03	 0.0327		 0.09		 0.257

Table 1 continued

	Transport		    0.13	-0.0871		-2.94		-5.005
	Wood products	    0.04	-0.5194		-1.15		-3.187
	Services		    0.04	-0.6508		-1.83		-3.805
	Misc. mfg.	    0.05	-0.2694		-0.76		-1.853
	Other		    0.13

1 Maximim likelihood estimation. Log of likelihood function = -2689.24. N= 698.

2 Indicates that 8.00% of strikers were women.