A1003
Title: On a fast and consistent test for equality of means for high-dimensional data
Authors: Sayan Das - Washington University in St. Louis (United States) [presenting]
Debraj Das - Indian Institute of Technology Bombay (India)
Subhajit Dutta - IIT Kanpur (India)
Abstract: A two-sample test is proposed for high-dimensional means using a logistic regression framework. The crux of the method relies on identifying significant feature variables via LASSO (using the logit link) that can capture the population mean. The proposed test is based on the regression coefficients of these selected features, effectively reducing dimensionality. Theoretically, it is shown that the proposed test is consistent against a wide range of alternatives. In the ultra-high-dimensional regime, the test leverages the sparse structures of the means, making it computationally more efficient compared to other methods in the existing literature. Additionally, the method is extended to test for equality of means across multiple groups. Finite sample studies corroborate the theoretical findings and reveal the superiority of the tests compared to several existing tests.
