Student Seminar Series - July 7, 2006
University of Minnesota
School of Statistics
College of Liberal Arts

Inference and Model Selection


Yingwen Dong


Friday, July 7, 2006
10:00 AM, 170 Ford Hall
Minneapolis, East Bank Campus

Refreshments at 9:30 AM
300 Ford Hall


Abstract


Model selection is often employed in many scientific studies and has a significant effect on inference of estimated parameters. Common practice is to use a selected model to estimate parameters of interest ignoring uncertainty introduced by the process of model selection. It is well known that this could yield overoptimistic inferences, resulting in misleading conclusions. In this proposal, we propose an inferential tool given a selected model, together with an estimated probability of selecting the model, which measures the uncertainty of model selection in contrast to simply applying the standard inferential tools to the estimators. The probability can be estimated by using a data perturbation technique. We show that the proposed technique yields approximately unbiased estimation through theoretical analysis and simulations. We also examine the proposed method in the context of hypothesis testing, and demonstrate through simulation that it performs well in the situation when the commonly used method breaks down. The results illustrate that the proposed method is attractive.