Statistical hypothesis: Beta coefficients will be equal to zero
Test statistics: (1st Regression – t statistic for Bill 70) Manufacturing = -4.1, Retail = -0.6; (2nd Regression – t statistics for 1980, 1981 and 1982+) Manufacturing = -2.6, -3.0, -5.0, Retail = -0.6, -0.3, -0.5-
Cost of Intervention Economic Outcomes
Did the Design Lack Statistical Power?
No. Since statistical significance was achieved in the manufacturing sector regression analyses, it is clear there was enough statistical power. As for the retail sector, if we use the rule of ten observations per predictor variable, we can see we have enough power. Specifically, five predictor variables are used in the first retail sector regression analysis and eight in the second analysis. Therefore, we would need 10*5 observations for the first analysis (n=50) and 10*8 for the second (n=80). Since there are 201 retail firms included in the analysis, the study was sufficiently powered to detect an association if one existed.
Were Any Harms of the Intervention Identified? No. IWH Reviewers’ Comments:
The study does have a number of limitations that should be raised. The authors suggested that a number of other variables may have had an impact on OHS performance and may have biased the findings; one major confounder would be a change in workers’ compensation administration and they mention that over the study period there was a shift to the New Experimental Experience Rating (NEER) system in Ontario but that only one rate group (plastics) entered the program. The reviewers also identified some weaknesses in the statistical analysis (a Poisson regression would have been a better approach - or the normality of the distribution of the injury/illness frequencies should have been provided to justify the use of multiple regression), potential for selection bias, as well as the possibility of the effect of confounding variables.
Effectiveness of Occupational Health & Safety Management Systems: A Systematic Review