Janet Van Niekerk, from KAUST, will present the Department of Statistical Science seminar with a talk entitled, "What do we know about priors and contraction to simpler models?"

Abstract: Generalized models are being proposed almost daily to accommodate various challenges posed by real data. Often, specific parameter value choices causes the model to reduce to the simpler model, but how often do we estimate this to be the case?

How can we formulate priors that contract (not shrink) toward a simpler model with interpretable properties and assurance of never doing "the wrong thing"?

Many questions with few solid answers.

In this talk I will illustrate the need for this type of thinking using the Weibull model in a survival context and also the skewed-probit regression model for binary data.

The mathematical methodology underpinning these formulations is briefly explained and these estimation frameworks are available in the INLA R-package.

Join Zoom Meeting


Meeting ID: 983 3776 5461

Passcode: 2l858q