Variable importance measures based on discriminant analysis and multivariate analysis of variance are useful for identifying variables that discriminate between two groups in multivariate group designs. Variable importance measures are developed based on trimmed and Winsorized estimators for describing group differences in multivariate non-normal populations.
Sajobi, T. T., & Lix, L. M. (2017). Robust Measures of Variable Importance for Multivariate Group Designs. Journal of Modern Applied Statistical Methods, 16(2), 99-123. doi: 10.22237/jmasm/1509494760