To test our hypotheses, we estimated a series of conditional multilevel models. We fitted each model to all M=10 data sets with imputations of missing data and combined the parameter estimates and standard errors using SAS PROC MIANALYZE, which implements procedures developed by Rubin (1987). To test the effects of time-varying (i.e., within-person) versus proximal and distal (i.e., between-person) effects, we followed Raudenbush and Bryk (2002, p. 134– 141; see also Curran & Bauer, 2009). Specifically, we added person-mean centered time-varying covariates for mothers’ and fathers’ alcohol-related symptoms as repeated measures and the report of these symptoms averaged over repeated assessments as the proximal effect. We also added interactions between each of these predictors with study to test for differences in findings based on membership in MLS versus AFDP.
