Hello, I am modeling weekly drinking data during twelve weeks of treatment using the two-part model described by Olsen & Schafer (2001). At the first week, only 32.7% of the sample reported no drinking in the week prior. However, by week 5, 67.7% reported no drinking, and by the final week, 86.3% reported no drinking. Due to these changes during treatment, the continuous variables near the end of treatment have many missing values, as those people fall in the non-drinking category. The covariance coverage falls below .05 when any of the continuous variables past week 6 are included in the model. Are there any techniques to complement the two-part model that can account for the missingness due to participants being categorized as non-drinking?
Missingness is accounted for in the 2-part model using the usual ML-MAR approach. Nothing more needs to be done. The coverage threshold of 0.10 used in Mplus sometimes has to be lowered for 2-part applications.