Loading Events

« All Events

CANCELLED: Safe, Anytime-Valid Inference (SAVI)

May 25 - May 29

Summary

A large fraction of published research in top journals in applied sciences such as medicine and psychology has been claimed as irreproducable. In light of this 'replicability crisis', traditional methods for hypothesis testing, most notably those based on p-values, have come under intense scrutiny. One central problem is the following: if our test result is promising but non-conclusive (say, p = 0.07) we cannot simply decide to gather a few more data points. While this practice is ubiquitous in science, it invalidates p-values and error guarantees and makes the results of standard meta-analyses very hard to interpret. This issue is not unique for p-values: other approaches, such as replacing testing by estimation with confidence intervals, suffer from similar optional continuation problems. Over the last few years several distinct but closely related solutions have been proposed, such as anytime-valid confidence sequences, anytime-valid p-values, and safe tests.

Remarkably, all these approaches can be understood in terms of (sequential) gambling. One formulates a gambling strategy under which one would not expect to gain any money if the null hypothesis were true. If for the given data one would have won a large amount of money in this game, this provides evidence against the null hypothesis. The test statistic in traditional statistics gets replaced by the gambling strategy; the p-value gets replaced by the (virtual) amount of money gained. In more mathematical terms, evidence against the null and confidence sets are derived in terms of non-negative supermartingales. While this idea in essence goes back to Wald’s sequential testing of the 1950's and its extensions by Robbins and co in the early 1960's and Lai in the 1970's, it never really caught on because it used to be applicable only to very simple statistical models and testing scenarios.

However, recent work shows that this idea is essentially universally applicable – one can design supermartingales for large classes of tests, both parametric and non-parametric, and many estimation problems, yielding anytime P-values using non-asymptotic versions of the law of the iterated logarithm. A variation of the idea, the S-value, can even be applied to completely arbitrary tests - at the price of only allowing for a weaker form of optional stopping. All these techniques have both a gambling and a Bayes factor interpretation. Thus, these directions are able to somewhat unite Bayesian and frequentest ways of thinking; with the explicit ability to use prior knowledge, with frequentest error control and confidence bounds, but often using Bayesian techniques.

This workshop aims to get together two groups of researchers --- those who have been developing the mathematical, probabilistic and statistical foundations of this area, and practitioners who have studied and written about the reproducibility crisis in the sciences.

This conference supports the Welcoming Environment Statement of the Association for Women in Mathematics (AWM).

Sponsors

     

 

    

Organisers

Peter Grünwald CWI, Amsterdam
Aaditya Ramdas Carnegie Mellon University

 

Speakers

Akshay Balsubramani Stanford University Genetics, Computer Science
Leonhard Held University of Zurich Biostatistics, Reproducibility
Chris Jennison University of Bath Statistics
Wouter Koolen CWI, Amsterdam Computer Science (ML)
Daniel Lakens TU Eindhoven Human-Technology Interaction, Reproducibility
Theis Lange University of Copenhagen Biostatistics, Reproducibility
Alexander Ly University of Amsterdam and CWI Psychology, Computer Science
Alan Malek Optimizely Statistics
Deborah Mayo Virginia Tech Philosophy
Luigi Pace University of Udine Economics, Statistics
Leonid Pekelis Opendoor Statistics
Don van Ravenzwaaij Groningen University  Psychologym Statistics
Christian Robert Université Paris-Dauphine + Warwick University Statistics
Alessandra Salvan University of Padova Statistics
Glenn Shafer Rutgers University Foundations of Probability and Statistics
Mark Simmonds University of York Review and Dissemination
Anne Lyngholm Sørensen University of Copenhagen  Statistics
Vladimir Vovk Royal Holloway London Computer Science (ML), Statistics
Eric-Jan Wagenmakers University of Amsterdam Psychology
Ruodu Wang University of Waterloo Statistics, Actuarial Science
Larry Wasserman Carnegie Mellon University Statistics, ML

Programme

Will follow soon

Abstracts

Will follow soon

Registration

Link to online registration form: REGISTRATION
Registration is free, but compulsory for all speakers and participants.

Please note: A possible cancellation of your participation needs to be done by email (koorn@eurandom.tue.nl) before Monday  18 May. When you don’t cancel your participation and you do not show up at the event, we are unfortunately obliged to charge you 100 euro no-show fee (invoice will be sent to you).

Practical Information

Link to travel and accommodation information

Details

Start:
May 25
End:
May 29

Venue

MF 11-12 (4th floor MetaForum Building, TU/e)