Analyzing Models With Multiple‐outcome Variables *

Multivariate designs have been useful in testing models with more than one output variable. This paper explains the use of the step‐down F statistic, a technique that allows a more refined analysis than does the traditional multivariate or univariate F test by examining sequential interactions among dependent variables in a multiple‐response set. An illustration is presented within the context of a laboratory experiment designed to examine a manipulation effect on a three‐variable causal chain. The example illustrates the misleading interpretation that could result without the information provided by the step‐down statistic.