Scientists often build statistical models to explain a phenomenon or to make predictions. In the course of doing so, scientists usually consider several-to-many candidate models prior to selecting a 'winning' model. A number of methods have been advanced for selecting an optimal model. Because of their complete accounting of uncertainty, Bayesian models have become widely used in the last two decades. In this project we will restrict our attention to Bayesian statistical models. Many Bayesian model selection criteria have been presented in the literature. The mere fact that so many criteria have been proposed and are in useis an indication that none have been found to be optimal in all cases. I will examine the performance of a number of the more commonly used criteria. In this project, model 'goodness' will be equated with predictive ability.
|Effective start/end date||9/14/15 → 8/31/18|
- National Institute of Food and Agriculture (National Institute of Food and Agriculture (NIFA))