A Matrix-variate Regression Model with Canonical States: an application to elderly Danish Twins

In many situations we observe a set of variables in different states (e.g. times, replicates, locations) and the interest can be to regress the matrix-variate observed data on a set of covariates. We define a novel matrix-variate regression model characterized by canonical components with the aim of analyzing the effect of covariates in describing the variability within and between the different states. Despite the seeming complexity, inference can be easily performed through maximum likelihood. We derive the inferential properties of the model estimators and a general approach for hypothesis testing. Finally, the proposed method is applied to data coming from the Longitudinal Study of Aging Danish Twins (LSADT), so to investigate the causes of variation in cognitive functioning.