fit the data better than a 2-class solution (Table 2). Furthermore, a 2-class solution without binge drinking fit the data significantly better than a 2- or 3-class model that included binge drinking (Table 2). A 4-class solution identified a class with only 4 individuals, which was too small for further analyses and did not fit as well as a 3-class model. The factor mixture model with two classes and a dimensional latent factor did not fit the data as well as the LCA with two classes (Table 2). Therefore, we concluded that the 3 classes were distinct and pursued the 3-class LCA solution of 10 criteria without binge drinking in further analyses because models without binge drinking were significantly better than any model that included binge drinking. The probability of being assigned to class 1 given a class 1 profile was 0.92, for class 2 it was 0.90 and for class 3, 0.91. The probability of being assigned to class 1 given a class 3 profile or vice versa was 0, indicating that perfect classification occurred between classes 1 and 3 with no overlap whatsoever. The probability of error between class 1 and 2 was 0.08 and between class 2
