HIERARCHICAL LINEAR-MODELS FOR MULTIVARIATE OUTCOMES

In this article, we develop a class of two-stage models to accommodate three common characteristics of behavioral data. First, behavior is i nvariably multivariate in its conceptualization and communication. Sep arate univariate analyses of related outcome variables are fraught wit h potential interpretive blind spots for the researcher. This practice also suffers, from an inferential standpoint, because it fails to tak e advantage of any redundant information in the outcomes. Second, stud ies of behavior, especially in experimental research, employ smaller s amples. This situation raises Issues of robustness of inference with r espect to outlying individuals. Third, the outcome variable may have o bservations missing because of accidents or by design. The model permi ts the estimation of the full spectrum of plausible measurement error structures while using all the available information. Maximum likeliho od estimates are obtained for various members of a multivariate hierar chical linear model (MHLM), and, in the context of several illustrativ e examples, these estimates march closely the results from a Bayesian approach to the normal-normal MHLM and to the normal-t MHLM.