Table 3.

Structural equation models fit to empirical data and key indirect paths tested^∗

Model	χ2	df	χ2/df	p	CFI	RMSEA	AIC	Interruptions → WCFS action → Number of near accidents β (CI 90)
Independence model	350.64	55	6.38	0.00	0	0.23	394.64	—
Saturated model	0	0	0	—	1.00	0.00	154.00	—
Measurement model	48.18	36	1.34	0.08	0.96	0.06	108.18	—
Hypothesized indirect path model	48.62	37	1.31	0.10	0.96	0.05	106.62	0.11 (0.01–0.23)
Only interruptions by persons included	76.25	70	1.09	0.28	0.98	0.03	146.25	0.07 (−0.01 to 0.19)
Only interruptions by malfunction included	34.12	37	0.92	0.60	1.00	0.04	92.12	0.12 (0.004–0.28)
Only interruptions by blockings included	75.39	58	1.30	0.06	0.95	0.00	141.39	0.11 (0.001–0.23)
Alternative accident-prone person model	48.21	37	1.30	0.10	0.96	0.05	106.21	WCFS action → Interruptions → Number of near-accidents: −0.01 (−0.103 to 0.08)

AIC, Akaike information criterion, which should be as low as possible. A nonsignificant χ² and CFI higher than 0.90 in the indirect path model reflect an acceptable fit between the model and the data [26]. The comparably low Akaike information criterion attests to the parsimonious informative modeling in the hypothesized indirect path model; CFI, comparative fit index; CI, confidence interval; df, degrees of freedom; p, probability of the discrepancy to differ from zero (should be nonsignificant in a good model); p value of minimum discrepancy divided by its degrees of freedom, which should be nonsignificant; RMSEA, root-mean-square error of approximation, a measure of fit that takes into account the population moments rather than sample moments, RMSEA ≤ 0.05 can be considered a good fit; values between 0.05 and 0.08 indicate an adequate fit [26]; WCFS, Workplace Cognitive Failure Scale; χ², indicates the minimum discrepancy between empirical covariance structures and those implied by the model; χ²/df, minimum discrepancy divided by its degrees of freedom, as an indicator of fit.

^∗The models are as follows: (1) independence model: no associations between study variables were assumed; (2) saturated model: assumes all variables were interrelated—estimates best possible fit of model variables and empirical data; (3) measurement model: all latent variables were specified and assumed to be nondirectionally interrelated; (4) hypothesized indirect path model: model as shown in Fig. 1; and (5) alternative accident-prone person model: trait model, conscientiousness predicts WCFS action and compliance with safety regulations, and WCFS action predicts interruptions that directly link to near-accidents.