Log-Linear Models and Logistic Regression (Springer Texts in Statistics)

Rating:
Summary: graduate level book on categorical data
Review: Ron Christensen is an excellent writer who has written many intermediate to advanced level texts on linear models, multivariate analysis, time series and categorical data analysis. They all provide a good mix of theory and applications and cover topics often neglected in the standard text books including Bayesian approaches to inference. This book is no exception.

Christensen's emphasis is on log linear models much in line with the earlier text by Bishop, Fienberg and Holland (the text he learned from). However Christensen attempts to create a balance by not being quite as advanced as Bishop, Fienberg and Holland or Haberman or Santner and Duffy but not as introductory as Fienberg or Everitt. Also Christensen's text is much more current than many of these texts that were published in the late 1970 through the 1980s.

The second edition has added more material on logistic regression and logistic regression has even been added to the title. The other major change in the text is the addition of chapter 13 on Bayesian binomial regression. Chapter 12 on likelihood theory for categorical data is taken from Christensen's linear models book.

Like Hosmer and Lemeshow this book includes many examples and illustrates the use of various available software packages. It differs in that it covers more theory and emphasizes log-linear models whereas Hosmer and Lemeshow deal strictly with logistic regression.

Many people favor Agresti and view his text as the bible on categorical data analysis. Christensen's book os a worthy competitor. A unique feature is the inclusion of graphical models, a topic rarely covered except in specialized texts such as Edwards. Christensen also covers the Bradley-Terry model for paired comparisons, a very useful model for ranking players or teams in sports, bridge or chess tournaments. He also shows how the modeling fits into the generalized linear model framework. He also addresses the improtant distinction between fixed and random zeros in contingency tables.

The similarities and differences between categorical and continuous data are made and issues of variable selection in regression modeling are addressed.

Rating:
Summary: graduate level book on categorical data
Review: Ron Christensen is an excellent writer who has written many intermediate to advanced level texts on linear models, multivariate analysis, time series and categorical data analysis. They all provide a good mix of theory and applications and cover topics often neglected in the standard text books including Bayesian approaches to inference. This book is no exception.

Christensen's emphasis is on log linear models much in line with the earlier text by Bishop, Fienberg and Holland (the text he learned from). However Christensen attempts to create a balance by not being quite as advanced as Bishop, Fienberg and Holland or Haberman or Santner and Duffy but not as introductory as Fienberg or Everitt. Also Christensen's text is much more current than many of these texts that were published in the late 1970 through the 1980s.

The second edition has added more material on logistic regression and logistic regression has even been added to the title. The other major change in the text is the addition of chapter 13 on Bayesian binomial regression. Chapter 12 on likelihood theory for categorical data is taken from Christensen's linear models book.

Like Hosmer and Lemeshow this book includes many examples and illustrates the use of various available software packages. It differs in that it covers more theory and emphasizes log-linear models whereas Hosmer and Lemeshow deal strictly with logistic regression.

Many people favor Agresti and view his text as the bible on categorical data analysis. Christensen's book os a worthy competitor. A unique feature is the inclusion of graphical models, a topic rarely covered except in specialized texts such as Edwards. Christensen also covers the Bradley-Terry model for paired comparisons, a very useful model for ranking players or teams in sports, bridge or chess tournaments. He also shows how the modeling fits into the generalized linear model framework. He also addresses the improtant distinction between fixed and random zeros in contingency tables.

The similarities and differences between categorical and continuous data are made and issues of variable selection in regression modeling are addressed.