A SAS MACRO FOR RESIDUAL DEVIANCE OF ORDINAL REGRESSION-ANALYSIS

In this paper, a SAS macro is described for calculating the likelihood of the 'saturated' model in the analysis of ordinal regression. The o utcome variable is multinomial on an ordinal scale, while the explanat ory variables can be nominal or ordinal. Several ordinal regression mo dels may be fitted to the data. One method of testing for the goodness of fit of these regression models is by comparing the residual devian ce with the chi(2) distribution. In SAS, PROC LOGISTIC may be used to fit this type of data with proportional odds model. Unfortunately, the residual deviance is not available from the output. Our SAS macro wil l supplement the SAS output so that the residual deviance test may be carried out. The data from an ongoing HIV study is used as an illustra tion.