140
Case 1 results, analysis and discussions The proportion of hazards identified (HR index) across work-periods are shown in Figure 4 and the analysis results are presented in Table 2. Model I and II were estimated using conventional regression. Model I was estimated by regressing the HR index on the three predictor variables TD, and SC) as indicated in Table 2 and Model II was estimated similarly by regressing the HR index on only the level-change dummy variable (D. Following the estimation of
the parameters of both models, the model comparison test (see Equation 2) was used to select the preferred mathematical model. Comparing the obtained F value (F
obt
= 1.659) with the critical value (F
critical
= 3.682) using an alpha level of 0.05 and the degree of freedom (
df = 2,12), model II was selected as the preferred model. Also model Ir) explained slightly more of the variability in the measured data than model Ir, further supporting our choice. For Crews 2 and 3, the same approach was followed and model II was determined to be appropriate in each case. Therefore,
according to the results, each crew demonstrated only a level-change improvement. The Levene’s test for the homogeneity of the error variance and the Anderson-Darling test for the normality of errors,
in each case, yielded a p-value above 0.05. Hence, it was reasonable to accept homoscedasticity of error variance and the normality for errors. The Durbin-Watson test statistics revealed no evidence of autocorrelation, implying the adequacy of the selected mathematical model presented in Table 1. Thus, additional parameters that account for autocorrelation was unnecessary for further analysis.
