Introducing all two-way interactions markedly improved model fit (G 2 = 2.42, df = 1, p = 0.119), and the likelihood-ratio comparison against the main-effects model was highly significant (ΔG 2 = 18.37, Δdf = 3, p = 0.00037).
Daytime sleepiness was not significantly associated with injuries among females ( p = 0.148), although the effect size suggested a possible trend (OR = 1.47, 95% CI: 0.86–2.51; φ = 0.10) ( Table 3 ).
Log-linear modeling supported this observation by showing that two-way dependencies significantly improved model fit, although the three-way multiplicative interaction did not reach statistical significance.