The authors provide a systematic treatment of the major problems involved in using regression analysis. They clearly and concisely discuss the consequences of violating the assumptions of the regression model, procedures for detecting violations, and strategies for dealing with these problems.

Where an assumption of unidirectionality in causal effects is unrealistic, 'recursive' models cannot be used, and more complex 'nonrecursive' models are necessary. Unfortunately, many nonrecursive models (unlike recursive models) are 'unidentified', which makes meaningful parameter estimation impossible. Even when they are identified, it would be inappropriate to use OLS regression techniques (appropriate for recursive models) for the purpose of estimation. The concept of identification, and the factors that lead to it are explained; and various tests for determination are provided. Illustrations from a variety of social science disciplines are used throughout the book.