609 – Methodological Advances in Model Selection
A Connection Between Discrepancy Function Estimation and the P-Value
Andrew Neath
Southern Illinois University Edwardsville
Joseph Cavanaugh
The University of Iowa
Benjamin Riedle
The University of Iowa
Consider the common statistical problem of using data to decide between a null model and a more general alternative. Within the significance testing framework, one will decide in favor of the alternative (reject the null hypothesis) only when the p-value is sufficiently small. Within the discrepancy function / model selection framework, the decision is based on which model is deemed to provide the most accurate depiction of the underlying data generating mechanism. In this talk, we establish a connection between the frameworks. We will show how the probability of selecting the null model under a discrepancy function framework can be estimated by the p-value. Furthermore, we will discuss the implications of imposing significance testing principles on a discrepancy function based model selection problem.