Default Bayesian Model Selection of Constrained Multivariate Normal Linear Models

Abstract: The multivariate normal linear model is one of the most widely employed models for statistical inference in applied research. Special cases include (multivariate) t testing, (M)AN(C)OVA, (multivariate) multiple regression, and repeated measures analysis. Statistical procedures for model selection where the models may have equality and order constraints on the model parameters of interest are limited however. This paper presents a default Bayes factor for this model selection problem. The default Bayes factor is based on generalized fractional Bayes methodology where different fractions are used for different observations and where the default prior is centered on the boundary of the constrained space under investigation. First, the method is fully automatic and therefore can be applied when prior information is weak or completely unavailable. Second, using group specific fractions, the same amount of information is used from each group resulting in a minimally informative default prior having a matrix Cauchy distribution, resulting in a consistent default Bayes factor. Third, numerical computation can be done using parallelization which makes it computationally cheap. Fourth, the evidence can be updated in a relatively simple manner when observing new data. Fifth, the selection criterion can be applied relatively straightforwardly in the presence of missing data that are missing at random. Applications for the social and behavioral sciences are used for illustration.