Categorical Data Analysis Using The SAS® System, 2nd Edition

8.6. A Note on Diagnostics

While goodness-of-fit statistics can tell you how well a particular model fits the data, they tell you little about the lack of fit, or where a particular model fails to fit the data. Measures called regression diagnostics have long been useful tools to assess lack of fit for linear regression models, and in the 1980s researchers proposed similar measures for the analysis of binary data. In particular, work by Pregibon (1981) provided the theoretical basis of extending diagnostics used in linear regression to logistic regression. Both Hosmer and Lemeshow (1989) and Collett (1991) include lengthy discussions on model-checking for logistic regression models; Collett includes many references for recent work in this area. ...