Abstract: Cook and Setodji (2003) introduced the notion of model-free reduced-rank in multivariate regression. However, they only focused on continuous predictors. In this article, we propose an extension of model-free multivariate reduced-rank regression to incorporate a mixture of continuous and categorical predictors. A test for reduced-rank is proposed that requires no parametric model specification. Simulations and a data analysis are provided to demonstrate its effectiveness.
Key words and phrases: Central mean subspace, dimension reduction, multivariate reduced-rank regression.