Practical Guide to Logistic Regression
Ulrike Grömping · Journal of Statistical Software · 2016
Practical Guide to Logistic Regression" introduces the analysis of binary response data for working analysts and researchers who have an understanding of basic statistical principles and linear regression analysis -this is its intention as declared in the preface.The book is structured as follows: a brief section (Chapter 1, 12 pages) on statistical models (focusing on basic ideas of logistic regression, the Bernoulli distribution, and ML estimation) is followed by a detailed introduction (Chapter 2, 36 pages) to the logistic model with a single predictor, which starts with one binary predictor, presents all formal tools in this context, and moves on to a multi-category and a continuous predictor.Chapter 3 (22 pages) discusses logistic regression with multiple predictors, covering interpretation of coefficients, model fit statistics, information criteria (e.g., AIC, BIC), adjustment of standard errors (by scaling, through the sandwich estimator or via the bootstrap), and common expressions used in risk modeling, like "risk factor", "confounder" or "effect modifier".Chapter 4 ("Testing and Fitting a Logistic Model", 36 pages) covers goodness-of-fit tests, diagnostic plots, numeric instabilities resulting from (almost) perfect separation, exact logistic regression, and -perhaps somewhat out of place -analysis of data given as a table of counts.The next chapter, "Grouped Logistic Regression" (Chapter 5, 20 pages), would have been a more natural place for the analysis of data given as frequency tables; it starts with an introduction to the binomial distribution with more than one trial and moves on to overdispersion and beta binomial regression (for which the book's author is a renowned expert).The last chapter covers "Bayesian Logistic Regression" (Chapter 6, 24 pages).A reference section and an index complete the book.The book presents many worked examples, and the choice of interesting data sets all of which are available to the reader is one of its greatest assets.Data availability makes it easy for readers to reproduce the examples from the book, and example code is available for R, SAS and Stata: R code is incorporated into the book chapters, and the end of each chapter gives SAS and Stata code.The data sets can be downloaded from the book author's website in all three formats, and it is very convenient for R users that the package LOGIT (on CRAN) contains all data sets and several functions that the book author provides.The SAS code for the examples can also be found on the book author's website, while I have not been able to