Model fitting analyses were conducted using Mx (Neale 2004), and the fit of all models was assessed using the minus twice the log-likelihood (−2LL) fit statistic. The model we applied to our data was the model described by Purcell (Purcell 2002), often referred to as a gene-environment interaction model (see Figure 1). In this model, the phenotypic variance can be partitioned into the latent, or unmeasured, genetic (a), shared environment (c), and non-shared environment (e) effects, and also the interaction between each of these parameters (βX, βY, and βZ respectively) with a moderating variable (i.e. religiosity). The model also examines the moderating effect of religiosity on the means (βM). Any gene-environment correlation between religiosity and problem alcohol use (i.e. any genetic influences shared in common by the two phenotypes) are included in the means model, and are not explicitly tested in the model. To test the significance of each of the interaction parameters, we dropped each one in turn and compared the fit of the nested models (those in which the parameter was dropped) with the fit of the model
