Bank of Canada, other federal sources as well as unpublished reports. Pooled time series regression analyses, based on the three years of annual data across the 19 manufacturing sectors, were used to develop six different specifications of the authors’ theoretical equation explaining variation in TFP growth. All of the specifications had satisfactory explanatory power (R-squared values from 0.54 to 0.77) and were relatively consistent. 22
The specification with all the variables included was used to explain the impact of each on the TFP growth. Three of the five OHS variables were significant in the final model, with Protective (rate of protective reassignments) having a negative coefficient thus a negative impact on TFP growth. Infraction (rate of penalties imposed) and Prevent (percent of firms that have adopted prevention programs) both were significant and positive and the implied contribution to productivity growth was estimated at 0.007 (the Prevent variable itself was calculated by the reviewers to make an estimated TFP growth contribution of 0.006).23 The Protective variable had a large impact on TFP growth (-0.019) and taken together the overall impact of the OHS variables was -0.012 which was larger than the impact found for OHS regulation in the United States (-0.003).
The findings of Dufour et al. (1998) provide insights into the impacts of important aspects of the Quebec regulation and CSST activities on the rate of productivity growth over the 3 year study period. The authors suggested that prevention programs and penalties may have reduced workplace injury incidence, leading to reductions in both direct and indirect costs related to accidents thus enhancing productivity growth. The finding of a statistically significant positive impact of prevention programs (the variable was significant at the 5 per cent level in all the specifications) on productivity growth has application to this review’s research questions. The model was well developed theoretically and the statistical analyses were appropriate and comprehensive.
Although the findings for the prevention programs were consistent and the model robust, the nature of the study design raises concerns about the strength of the evidence. The time series design, conducted with data at the industry level and with no control group, may not account for the possible effects of other factors affecting the study outcome. Although numerous potential confounding variables were included in the model there remained the possibility of a common underlying factor (e.g., management competency) that could be associated with both a) the more rapid development and report to CSST of a prevention program over time in response to a legislative change and b) productivity growth over time. Additionally, the measure of the prevalence of prevention programs in the
22 The reviewers questioned the accuracy of R-squared values of some of the reported models (model 3, with three significant variables removed, had a higher R-squared value as compared to model 5 with them included).
Annual rate of growth was 0.0032 over the period of the study.
Institute for Work & Health