Models of decreasing complexity were fitted. First, the model of simple regression of all main effects and all combinations of two- and three-way interactions was fitted. Subsequent models removed nonsignificant parameters as measured by p values >0.05, until no significant increase in deviance resulted. The number of paternal ASP symptoms was not found to be a significant predictor of CD nor to improve model fit as either a main or interaction term and was excluded from the model. Models were compared using goodness-of-fit and parsimony, assessed by the deviance and Akaike’s Information Criterion (AIC) respectively. Lower values of AIC indicate more parsimonious models and lower values of deviance imply improved model fit (Neale & Maes, 2002).
