In one study, researchers used catastrophe models to examine proximal and distal predictors of post-treatment drinking among individuals with alcohol use disorders [31]. Catastrophe models accounted for more than double the amount of variance in drinking than that predicted by linear models. Similar results have been found using the much larger Project MATCH dataset [33]. Two additional recent analyses of the MATCH dataset showed that nonlinear approaches can detect processes that may go unobserved in the context of linear models. Witkiewitz and colleagues [34,82] used catastrophe modeling and latent growth mixture modeling to re-assess two of the matching hypotheses that were not supported in the original study--that individuals low in baseline self-efficacy would respond more favorably to cognitive-behavioral therapy (CBT) than motivational enhancement therapy (MET) and that individuals low in baseline motivation would respond more favorably to MET than CBT [18]. In the first study [34], catastrophe models provided the best fit to the data, and latent growth analyses confirmed the predicted interaction: frequent drinkers with low initial self-efficacy had better outcomes in CBT than in MET, while those high
