Is multivariate analysis of PET data more revealing than the univariate approach? Evidence from a study of episodic memory retrieval.

In a functional imaging study of cued paired associate retrieval, in which the strength of association between pair members was systematically varied, we predicted increased right frontal activity as a function of weakening semantic linkage. An initial univariate analysis found the opposite effect, with greater right frontal activity during recall of strongly linked paired associates. This unexpected result led us to perform a multivariate analysis of covariance (MANCOVA), an approach which proved more informative. This analysis showed that the most significant source of task-related variance was accounted for by a nonlinear relationship not predicted by the prior hypothesis and not revealed by the standard univariate approach. This application of the MANCOVA supports the assertion that multivariate analysis can provide an important adjunct to univariate approaches like statistical parametric mapping (SPM). New perspectives engendered by the MANCOVA still allow for statistical inference but are not constrained by explicit hypotheses about specific task-dependent effects.