To determine the set of latent classes separately for each outcome from day 1 to day 26 of the diaries, we used Mplus Version 4.21 to employ a growth mixture mode with fixed and random intercepts and slopes specified separately for each latent trajectory class (Muthén and Shedden, 1999). For each outcome and class, the quadratic effects were not statistically significant. Because the distribution of the scores for each outcome was not normally distributed, tending to be both skewed and to cluster around whole numbers, we recoded each of the scores into four levels. Alcohol craving was recoded as follows: (1) equal to zero; (2) greater than 0 but less or equal to 1; (3) greater than 1 and less than or equal to 2; and (4) greater than 2. Positive affect was recoded as: (1) less than or equal to 2; (2) greater than 2 and less than or equal to 3; (3) greater than 3 but less than or equal to 4; and (4) greater than 4. Negative affect was recoded to: (1) equal to 1; (2) greater
