On matrix-variate regression analysis

Three-way data arise in different application domains when multiple responses are measured at different time points or locations. A new regression model for analyzing three-way data is proposed. By assuming the matrix normal distribution for the error term, we will show that the proposed model represents the natural generalization of multiple and multivariate regression analysis. Inferential properties of the model estimators are derived. The model fit is illustrated on a real application. (C) 2012 Elsevier Inc. All rights reserved.