|Friday, February 24|
|CS02 Beyond the Basics: Advanced Modeling Methods||
Fri, Feb 24, 9:15 AM - 10:45 AM
River Terrace 3
Don't Be Silly; Do It Bayesian (303357)*Perceval Sondag, Arlenda
Keywords: Bayesian, modeling, complexity
Over the past few years, Bayesian statistics have gained in popularity in the Pharmaceutical industry. A simple proof would be the number of Bayesian-related talks and posters at statistical conferences. Despite this growing enthusiasm among the new generation of statisticians, some concerns are still perceptible about using Bayesian for the day to day analysis. Often, scientists and/or statisticians will prefer not to use the Bayesian statistics because they want to “keep things simple”. That is, Bayesian statistics are considered too complex. However, while the learning curve might be different, it is often the complete opposite. Indeed, while it takes a little longer to understand the fitting and interpretation of a linear model in Bayesian statistics compared to the simple Least Square method, once this is done, dealing with moderate to highly complex models doesn’t require that much more effort. This talks walks through examples of problems of increasing complexity (from inference on a normal iid sample to nonlinear mixed effect models) and suggests putting them into three categories. The simplest problems for which Bayesian statistics aren’t useful (except for educational purpose); the case-by-case scenarios for which the best choice is a matter of context and/or personal preference; and the moderate to complex problems, for which frequentist solutions are approximations based on complex formulae, while Bayesian modeling allows to provide simple and accurate solution.