Common Statistical Methods for Clinical Research with SAS Examples, Third Edition, 3rd Edition

Appendix I. Commonly Used Parameterizations in Logistic Regression Models

• I.1 Introduction 509

• I.2 The 'Reference' Parameterization 509

• I.3 The 'Effect' Parameterization 511

• I.4 More Than Two Nominal Levels 512

• I.5 Other Parameterizations 515

I.1 Introduction

There are several ways in which a statistical model may be written. In a logistic regression model (Chapter 20), you might include numeric covariates (either continuous or ordinal) and nominal categorical factors. Generally, you will have one parameter (β_i) associated with each numeric covariate (X_i). For a nominal covariate (or any effect used in a CLASS statement in SAS), the number of parameters depends on your model specification or 'parameterization' as well as the number of levels of the ...