We then compared the full moderation model with all parameters freely estimated (−2 log likelihood = 3738.92, AIC = 996.92, df=1371), and a moderation model with the five significant moderation paths (−2 log likelihood = 3739.78, AIC = 995.78, df=1372), to the no-moderation model. When compared to the no-moderation model, both moderation models resulted in significantly better fit. The full moderation model resulted in a slightly better −2 log likelihood value, while the five parameter moderation model had a slightly better AIC value. Thus, given the almost equal fit of the two models, we chose the full moderation for its ability to provide the fullest picture of the change in the etiology of Internalizing as a function of the level of Marital Quality. Of note, in the final model, the confidence intervals around five of the six moderation parameters included zero, and were thus not statistically significant in the p-value sense (see Table 4). Nevertheless, as a set, these parameters are important to include in the model, because the full moderation model fit better than a model with no moderation (No Moderation AIC=1007.34, whereas Moderation AIC = 996.92).
