observe the impacts of IC status on change in accident rates.20 Additionally, measurement error in the dependent variables along with the categorization of responses may have limited the findings. The reviewers also commented that measurement bias may have been a factor because some of the safety and health data were based on recall (most from archival data but recall on remaining). An additional concern was the fact that the two dependent variables appear to have been ordinal, yet the results of an ordinary least squares regression were provided. The authors should have considered a modeling technique more appropriate for ordinal data.
Torp et al. (2000) (described in the intermediate outcomes section above) also provided evidence on musculoskeletal symptoms (including pain), since the questionnaire distributed to motor vehicle repair garage employees contained seven questions related to musculoskeletal symptoms experienced in the last 30 days. Six were derived from the Subjective Health Complaints Questionnaire and one question on knee pain was added. The severity of pain was rated on a four-point scale. One question on sick leave asked respondents whether they had been away from work in the last 30 days (dichotomous response).
There were two measures of IC implementation, as discussed previously. Findings were presented from a number of multiple regression analyses with each of the final outcome variables as the dependent variable and including each IC measure along with control variables for company size and unionization. For the model involving musculoskeletal symptoms, coefficients for managers’ rated IC status were significant (0.077, p<0.05). These results indicated a negative relationship between IC status and musculoskeletal symptoms (employees in garages in which the manager rated IC status as higher reported fewer symptoms). Neither coefficient for the two measures of IC implementation in their respective models for sick leave were statistically significant.
The evidence for final outcomes from Torp et al. (2000) is unreliable because of the possibility of selection bias, response bias, and most importantly, because of the limitations of their study design. The authors point out that conclusions about causality are not possible because of the cross sectional design. Although the investigators made efforts to select the most appropriate garages for the study, the selection was non-random as 130 had managers enrolled in an OHS course and 200 others were selected by other means.21 Response rates for the respective groups of garages were not provided, raising further concern over possible section bias.
20 Additionally, the reviewers felt that the finding of a significant decrease in absenteeism may have been an artifact because it was found over such a short period of observation and up to only 6 months after the legislation.
21 130 of the garages in the non training group were matched to the firms participating in the course and another 70 were randomly selected.
Effectiveness of Occupational Health & Safety Management Systems: A Systematic Review