Probability of Selecting the "Correct" Model and Determination of Sample Size in Regression
Abstract
**Please note that the full text is embargoed** ABSTRACT: Selection of variables in multiple linear regression is a common problem in model building. Let the "correct" model be the model which includes all variables that influence the dependent variable and excludes
all others. This paper derives the probability of selecting the correct model for the sequential deletion procedure. A concept of least favorable selection
is given. Based on this concept, the sample size is determined for
given probability of selecting the correct model. The procedure is illustrated by considering the polynomial regression.