For each of the three daily diary measures, the LMR test of the number of classes for a given latent class growth curve mixture and the stability of classes shows that the three class models provided additional information beyond that of the two class models. The improvement provided by a four-class model was not significant for alcohol craving and positive affect. For negative affect, the four-class model significantly improved the fit over the three-class model based on the LMR test, but the size of the two of the four classes was small. Specifically, the middle class in the three-class model class split into two middle level negative affect classes with 41 and 11 subjects respectively in the four-class model. Because of these small sample sizes, we based the results on the three-class model.
