? ? Driven by both clinical and cost considerations, the vast majority, of treatment for substance-abusing patients in community-based programs is delivered in a group therapy milieu. Moreover, the most common approach to sustain groups in community programs is the use of """"""""rolling"""""""" admission, whereby patients enter groups at various times, while other members leave (due to graduation or, more likely, dropout). This process of continual change in group structure over time (and in a nonrandom fashion) creates unique analytic challenges to investigators who wish to account for and understand the dependencies created by rolling admission into therapy groups and for the proper modeling of treatment effects in this context. This thorny analytic problem has had a stifling effect on group therapy research in drug abuse treatment. Many investigators avoid the analytic challenges of modeling rolling admissions by (a) ignoring dependencies created by rolling treatment groups (leading to biased parameter estimation and/or standard errors), (b) designing studies that center on individual-based interventions (as opposed to groups), or (c) designing studies that restrict enrollment after a certain period (i.e., """"""""closed"""""""" groups). The major costs of each of these three approaches are incorrect inferences concerning treatment effects (i.e., Type I and/or Type II errors) (under condition [a]) and a disconnect between how treatment efficacy trials are conducted and how treatment takes place in community settings (under conditions [b] and [c]). Most investigators have avoided group therapy research altogether because there has not been a framework to properly address these challenges, particularly in grant and article submissions. The purpose of this exploratory/developmental R21 project is to (a) demonstrate that statistical developments in the area of multiple membership modeling and non-ignorable missing data (NIMD) modeling may actually provide theoretically sound frameworks for the analysis of rolling group data, (b) demonstrate the consequences associated with current approaches to rolling groups (as compared to NIMD methods) through real data examples and simulation modeling and (c) provide substance abuse treatment researchers with clear guidelines on how to analyze such data, along with requisite programming code from popular statistical packages (e.g., SAS, STATA, MPIus) to carry out these analyses. ? ? ?
