An Introduction to Categorical Data Analysis

This new edition of Alan Agresti's basic account of applied categorical data analysis retains the spirit of the 1996 original. Changes in emphasis and order reflect developments in statistical methodology and in computing software. A major addition is that two new chapters cover models for clustered correlated categorical data. Agresti, an expert in the field, provides lucid explanations for users wanting to learn how to apply relevant methods without becoming bogged down in technical detail. Derivations and proofs are avoided, the emphasis being on the formulation and use of models that are appropriate to a range of situations. Those who want detailed coverage of the theory should refer to Agresti (2002). Modern statistical software is essential for the laborious computations that are required for the estimation and testing that are associated with sophisticated models. References are given to relevant software including SAS and R, as well as to more specialized packages and to author-maintained Web sites. The book is intended as an introductory text-book or as a reference guide for analysts dealing with categorical data. It fulfils the latter role admirably and would be an excellent text-book for an extensive module on categorical data analysis. However, instructors would need to tailor the coverage and to select material carefully for shorter courses. Presentation may need to be adapted if students were familiar with, for example, generalized linear models and logistic regression before starting the course. Key features of the book include detailed comments on the appropriateness of particular models, and the differences between marginal- and conditional-distribution-based inferences are illustrated by interesting real data examples. The relationship between approaches that are based on logistic regression and on log-linear models and the differences between them are highlighted, as are the relative merits of alternative analyses of the same data. This requires much cross-referencing. Sophisticated modelling makes complicated notations unavoidable, but these are all carefully explained. A minor disappointment is a sparsity of references that might be useful especially to research workers in other disciplines who perform categorical data analyses. There are a few in-text references to works that are mentioned in the short bibliography, reduced from 39 entries in the first edition to 26 in this one. In several places a reference to more detailed discussion of a topic would be helpful. For example, on page 48 the use of unconditional tests as an alternative to conditional tests in a 2×2 table with binomial counts when some marginal totals are not fixed a priori is mentioned and rightly described as ‘beyond the scope of this text’. Nevertheless, some readers may want to pursue the topic further and references, for example, to some of Barnard's pioneer papers on this approach and to the general discussion in Yates (1984) would not have gone amiss.