BSB123 Lecture Notes - Lecture 10: Linear Regression, Stepwise Regression, Multicollinearity
Document Summary
Categorical variable is not measured on quantitative scale. We cannot use the original categories (e. g. male, female) in regression. A dummy variable assumes only two values: 0 or 1. We use one less dummy than the number of categories for any categorical variable. 1 implies the case is in a particular category and 0 not. We can throw in all iv"s in the regression model. We can also successively add or delete iv"s to determine the best model stepwise regression. However model building using stepwise is controversial. In this unit, we focus on interpretation of slope coefficients and illustration of a problem in regression known as multi-collinearity. To obtain the model that has the most explanatory power next slide. We successfully add one iv at a time until no remaining iv/s make a significant contribution. The iv with the highest correlation with the dv will usually be added first. Begin with all iv"s in the model.