Pooled marginal slicing approach via SIR\({}_\alpha\) with discrete covariates (Q964647)

A sliced inverse regression (SIR) technique is considered in the case of multivariate responses and the presence of some categorical predictor which stratifies the analyzed population. Two pooled marginal slicing estimators are proposed for the effective dimension reduction space, utilizing mixed SIR-I and SIR-II approaches. They can be used for homoscedastic and heteroscedastic errors. Asymptotic normality of the estimates is established. Results of simulations are presented.