1st November 2005 — Aleks Jakulin: Variable interactions

Aleks Jakulin will give a guest talk.

Variable Interactions: Analysis, Testing and Modelling

While there are many definitions of an interaction, we will present one that can be defined regardless of what is the underlying probabilistic model. With this approach, we can ask questions such as “How much does variable A tell us about variable B?”, or “How much does variable A tell us about variable B, once we control for variable C?” Such questions turn out to be highly revealing when people attempt to understand the data.

Of course, the answers to the above questions are often uncertain themselves. We will compare Bayesian and frequentist approaches to hypothesis testing, and find out that they are like apples and oranges. I will end up describing how frequentists integrate their models out, and provide recommendations about what priors should Bayesians use to compete with frequentists.

The talk will be based on the materials available at http://kt.ijs.si/aleks/int/.