Multiple Linear Regression: Variable Selection

When there are multiple variables to select from that may be of potential importance in a regression model, we may have to select some smaller subset due to a smaller sample size (e.g., smaller n than p) or to try and achieve a more parsimonious model (i.e., fewer variables may be easier to interpret and disseminate). In this lecture we highlight some variable selection strategies, as well as describe the algorithms for automatic procedures for forward, backward, and stepwise selection. These should not be used without careful consideration and evaluation of the final model (both for statistical and scientific/clinical/biological sense). As an extra topic, the concept of model averaging is introduced that can combine the estimates across multiple regression models.

Table of Contents:

00:00 - Intro Song
00:16 - Welcome
00:48 - Variable Selection
05:48 - Automatic Procedures (Caution!)
07:19 - Forward Selection
08:27 - Backward Selection
09:17 - Stepwise Selection
11:12 - Example in R
15:36 - FYI - Model Averaging