A1602
Title: Resampling-based approaches for nonparametric MANOVA in the presence of missing data
Authors: Lubna Amro - TU Dortmund University (Germany) [presenting]
Markus Pauly - Technical University of Dortmund (Germany)
Abstract: Multivariate analysis of variance (MANOVA) is widely used across various fields to examine multivariate endpoints. Traditional MANOVA methods require complete data and assume multivariate normality and homogeneous covariance matrices, but these assumptions often do not hold. Missing data can complicate these issues, potentially leading to inflated type-I error rates or low statistical power. To address this, resampling-based methods are introduced that handle missing data without the need for imputation or the exclusion of observations. The approach, which integrates resampling with quadratic form-type test statistics, is asymptotically valid and, accommodates heteroscedastic designs and allows for singular covariance matrices. Extensive simulations demonstrate that our methods effectively control type-I error rates and perform well across various distributional scenarios under missing completely at random (MCAR) and missing at random (MAR) mechanisms. Additionally, the methods are applied to a real data example to illustrate their practical applicability.
