Fixed Effects Regression Models » (New Edition)

Paul D. Allison is Professor of Sociology at the University of Pennsylvania, where he teaches advanced graduate courses on event history analysis, categorical data analysis, and structural equation models with latent variables. He is the author of seven books and more than 50 journal articles. Every summer he teaches 5-day workshops on survival analysis and logistic regression analysis that draw about 100 researchers from around the U.S. A former Guggenheim Fellow, Allison received the 2001 Lazarsfeld Award for distinguished contributions to sociological methodology.

This book demonstrates how to estimate and interpret fixed-effects models in a variety of different modeling contexts: linear models, logistic models, Poisson models, Cox regression models, and structural equation models. Both advantages and disadvantages of fixed-effects models will be considered, along with detailed comparisons with random-effects models. Written at a level appropriate for anyone who has taken a year of statistics, the book is appropriate as a supplement for graduate courses in regression or linear regression as well as an aid to researchers who have repeated measures or cross-sectional data.

About the Author vii

Series Editor's Introduction ix

1 Introduction 1

2 Linear Fixed Effects Models: Basics 6

The Two-Period Case 7

Extending the Difference Score Method for the Two-Period Case 10

A First-Difference Method for Three or More Periods per Individual 12

Dummy Variable Method for Two or More Periods per Individual 14

Interactions With Time in the Fixed Effects Method 19

Comparison With Random Effects Models 21

A Hybrid Method 23

Summary 26

3 Fixed Effects Logistic Models 28

The Two-Period Case 28

Three or More Periods 32

Interactions With Time 37

A Hybrid Method 39

Methods for More Than Two Categories on the Response Variable 42

Summary 47

4 Fixed Effects Models for Count Data 49

Poisson Models for Count Data With Two Periods per Individual 49

Poisson Models for Data With More Than Two Periods per Individual 54

Fixed Effects Negative Binomial Models for Count Data 61

A Hybrid Approach 65

Summary 68

5 Fixed Effects Models for Events History Data 70

Cox Regression 71

Cox Regression With Fixed Effects 73

Some Caveats 77

The Hybrid Method for Cox Regression 79

Fixed Effects Event History Methods for Nonrepeated Events 79

Summary 85

6 Structural Equation Models With Fixed Effects 87

Random Effects as a Latent Variable Model 87

Fixed Effects as a Latent Variable Model 91

A Compromise Between Fixed Effects and Random Effects 92

Reciprocal Effects With Lagged Predictors 93

Summary 97

Appendix 1 Stata Programs for Examples in Chapters 2 to 5 99

Appendix 2 Mplus Programs for Examples in Chapter 6 108

References 113

Author Index 116

Subject Index 118