ABSTRACT
Multivariate multilevel regression models are multilevel regression models that contain more than one response variable. As such, they are comparable to classical multivariate analysis of variance (MANOVA) models, where we also have several outcome measures. The reason for using a multivariate model is usually that the researchers have decided to use multiple measurements of one underlying construct, to achieve a better construct validity. A classic example is in medical research when diseases manifest themselves in a syndrome that leads to a pattern of related effects (Sammel, Lin, & Ryan, 1999). By using several outcome measures, researchers can obtain a better and more complete description of what is affected by changes in the predictor variables. Tabachnick and Fidell (2007) mention several advantages of using a multivariate approach instead of carrying out a series of univariate analyses. One advantage of multivariate analysis is that carrying out a series of univariate statistical tests inflates the type I error rate, which is controlled better in a multivariate analysis. A second advantage of multivariate analysis is that it often has more power. On each individual response measure, the differences may be small and insignificant, but for the total set of response measures, the joint effect may produce a significant effect (Stevens, 2009; Tabachnick & Fidell, 2007). However, the disadvantage of multivariate models is that they are more complicated, and that their interpretation is more ambiguous.
