[Statlist] Reminder: ETH Young Data Science Researcher Seminar Zurich, Virtual Seminar by Barrett Thomas (EANSAE, Paris), 22.05.2020

[Maurer Letizia]

Dear all

We are happy to announce the 'Young Data Science Researcher Seminar Zurich', a weekly virtual seminar where young Statisticians and Data Scientists have a platform to present their work. Our first speaker this Friday at 3pm CET will be Thomas Berrett from ENSAE Paris. More information about the upcoming talks can be found at: https://math.ethz.ch/sfs/news-and-events/young-data-science.html

"The conditional permutation test for independence while controlling for confounders"  
by Thomas Berrett, ENSAE, Paris

Time: Friday, 22.05.2020 at 15.00 h
Place: Zoom at https://ethz.zoom.us/j/92367940258

Abstract: In this talk I will discuss the problem of testing conditional independence. In standard independence testing problems, given a test statistic measuring the strength of dependence, a simple and practical approach to find critical values and calibrate our tests is to permute the data uniformly at random and recalculate the test statistic in order to simulate its behaviour under the null hypothesis of independence. Unfortunately, this is no longer effective when testing conditional independence, and may result in misleading conclusions. We propose a general new method, the conditional permutation test, for testing the conditional independence of variables X and Y given a potentially high-dimensional random vector Z that may contain confounding factors. The proposed test permutes entries of X non-uniformly, so as to respect the existing dependence between X and Z and thus account for the presence of these confounders. Like the conditional randomization test of Candès et al., our test is useful in the `Model-X’ framework, in which an approximation to the distribution of X|Z is available—while Candès et al.’s test uses this estimate to draw new X values, for our test we use this approximation to design an appropriate non-uniform distribution on permutations of the X values already seen in the true data. We provide an efficient Markov Chain Monte Carlo sampler for the implementation of our method, and establish bounds on the Type I error in terms of the error in the approximation of the conditional distribution of X|Z, finding that, for the worst case test statistic, the inflation in Type I error of the conditional permutation test is no larger than that of the conditional randomization test. We validate these theoretical results with experiments on simulated data and on the Capital Bikeshare data set. This talk is based on joint work with Yi Wang, Rina Foygel Barber and Richard Samworth. 

We hope that you are able to join us and we are looking forward to the talk.

Best wishes,

M. Löffler, A. Taeb, Y. Chen