CMStatistics 2017: Start Registration
View Submission - CMStatistics
B1830
Title: New tests for generalized linear models based on lasso Authors:  Sylvain Sardy - University of Geneva (Switzerland) [presenting]
Abstract: We propose new statistics based on glm-lasso to test a null-hypothesis. We show the statistics is asymptotically pivotal under $H_0$. The new test behaves better than the classical deviance test (better level and power) when $P<N$ and can still be applied when $P\geq N$ (more covariates then observations).