Seminar, Hosted by International Seminar on Selective Inference
Co-sufficient weighting: leveraging sparsity for more powerful variable selection in the linear model
Harvard University
Abstract
For controlled variable selection in the Gaussian linear model, Souhardya Sengupta shows that informative, p-value-independent weights for weighted multiple testing can be derived entirely from sufficient statistics and a belief in sparsity, using only the data and no external quantitative side information. Using these co-sufficient weights in weighted versions of Holm's and the Benjamini-Hochberg procedures gives substantially higher power than the unweighted procedures when the true coefficient vector is sparse, while retaining Holm's exact validity and the highly robust empirical validity of Benjamini-Hochberg regardless of sparsity. The methods are compared with their unweighted analogues and other competitors in a wide range of simulations and in an application to HIV drug resistance.