Assessing and adjusting for dependent observations in group treatment research using multilevel models. |
| |
Authors: | Tasca, Giorgio A. Illing, Vanessa Ogrodniczuk, John S. Joyce, Anthony S. |
| |
Abstract: | Group treatment data are nested by design, that is, clients nested in groups. Dependence associated with the nesting of group intervention data can inflate Type I error rates, which poses unique challenges to group treatment researchers. This article evaluates the extent and variability of dependence in data taken from 3 previously published randomized clinical trials of group psychotherapy. Three methods of assessing dependence by calculating intraclass correlation coefficients (ρ) were examined. Results showed great variability in ρs across studies, across methods of calculating ρ, and across outcome variables. The distribution of ρs suggested that the amount of dependence in the data was moderate. Two methods of addressing dependence in grouped treatment data through multilevel modeling were used. These methods resulted in minimally compromised statistical power compared with results from uncorrected data. These 2 methods may allow researchers to reliably assess their group treatments. Group intervention researchers are encouraged to consider their assumptions and conceptualizations of treatment change, and to choose corresponding methods of assessing for and addressing ρ. (PsycINFO Database Record (c) 2010 APA, all rights reserved) |
| |
Keywords: | multilevel modeling hierarchical linear modeling group treatment group effects dependent observations group treatment research group psychotherapy |
|
|