98 dummy variable (D) alone as indicated in the previous section. Then, the model comparison test statistics as shown in Equation 2 was used to select the most representative model. The obtained value of F (F
obt
= 12.9) was compared with the critical value of F (F
critical
= 3.682) using a liberal alpha level of 0.05 and the degree of freedom (
df = 2,12). This comparison test clearly indicated that model I explained much more variation (r) than model II (r, implying that model I is more appropriate. For crews 2 and 3, the same methodology was followed and Model II was determined as being the apt model for each case. This implies that for crew 1, the results indicate both a level change and slope change in hazard recognition and communication performance.
On the other hand, crew 2 and 3 indicated only a level change in hazard recognition and communication performance The Levene’s test for homoscedasticity of error variance and the Anderson-Darling test for normality of errors yielded a p-value above an alpha of 0.05 for each case. This indicated that the assumption of equality of variance and normality for error variance were reasonable. The
Durbin-Watson test statistics and the Huitema-McKean test for independence of errors revealed no evidence of autocorrelation implying the adequacy of the model as presented in Table 3 and that additional parameters for autocorrelation were unnecessary.
