- This event has passed.
Webinar: Biostatistics and Bioinformatics Seminar – ‘Akaike information criterion, how confident are we of the selected model?’
January 13, 12:00 pm to 1:00 pm
Presented by: Jason Liao, PhD, Professor, Division of Biostatistics and Bioinformatics
About the presentation: Akaike information criterion (AIC) and many of its refined forms are widely used in model comparison and selection. The considerable sampling variation in them, however, is seldom considered as part of the inference. As such, when one model is selected over the other, we are not sure of the strength of the support. The paper fills this gap by introducing formal statistical tests for H_0: two candidate models approximate the underlying true distribution equally well vs. H_1: one model approximates better than the other. We show that a constrained bootstrap test is effective and superior. The resulting p-value allows us to make more informative model selection. The paper also serves as a succinct introduction to the ideas behind AIC.
To join by phone, dial: 929-205-6099
To join via Zoom, visit https://pshealth.zoom.us/j/94860982030
Meeting ID is 948 6098 2030 and the passcode is 320367