A SAS/IML Macro for Goodness-of-Fit Testing in Logistic Regression Models with Sparse Data

The logistic regression model has become the standard analyzing tool for binary responses in a variety of disciplines. Methods for assessing goodness-of-fit, however, are less developed and this is especially pronounced in calculating goodness-of-fit tests with sparse data, when the standard tests (deviance and Pearson test) behave unsatisfactorily. In our paper we show two solutions to the problem that are implemented in the LOGISTIC procedure in SAS software, and introduce five additional testing procedures from the statistical literature. By means of a simulation study we show that these additional tests are valid instruments for assessing goodness-offit in logistic regression models, even with sparse data. Finally, we present the SAS/IML macro %GOFLOGIT which allows calculation of the introduced tests and illustrate the macro with an example from occupational epidemiology on hand eczema in hairdressers.