Case Number: 7-112-003, Year Published: 2012
HBS Number: KEL637
Economics, Market Research, Statistical Methods
This is the third in a series of lecture notes which, if tied together into a textbook, might be entitled “Practical Regression.” The purpose of the notes is to supplement the theoretical content of most statistics texts with practical advice based on nearly three decades of experience of the author, combined with over one hundred years of experience of colleagues who have offered guidance. As the title “Practical Regression” suggests, these notes are a guide to performing regression in practice. This technical note explains how to choose predictor variables to include in regression. The note begins by explaining the many virtues of parsimony. Sometimes analysts include predictors simply because they are in the available data. Including such "junk" predictors increases the chances of obtaining confusing or misleading results. The note also explores multicollinearity, a favorite topic in some statistics classes that is rarely a problem in real world empirical work. The note concludes by explaining how to work with groups of related variables and describes how to implement the partial F test for joint significance.
Students will learn the following: -How to identify appropriate variable to include in a regression -The dangers of including endogenous variables in a regression -The problem of multicollinearity, how to diagnosis it, and why it is rarely a problem in practice -How to test for the significance of a group of related variables using a partial F test and how to determine which variables, if any, to include in the model
Return to Search Results