We addressed the issue of missing data in our time-invariant and time-varying covariates through multiple imputation for each of our three sub-samples (i.e., mother-, father-, and adolescent-report) independently (Schafer, 1997a). For these analyses, we used the externalizing Item Response Theory scores from 2,801 assessments of 1,026 participants in the mother-report sample, 2,448 assessments of 938 participants in the father-report sample and 1,349 assessments of 454 participants in the adolescent-report sample. We used SAS PROC MI (SAS, 1999) to impute missing data in the time-invariant covariates and the R package PAN (Schafer, 1997b) for imputation of the time-varying covariates. Specifically, we first created 10 data sets for which the missing data in the time-invariant covariates were imputed, and for each we proceeded to impute the missing time-varying covariate values using PAN. Following standard recommendations in the multiple imputation literature (Rubin 1996), we included all predictors in both imputation models and independent as well as dependent variables in the PAN model.
