LOGISTIC Procedure

The LOGISTIC procedure includes several new MODEL statement options that provide additional control over the model-fitting process. The ABSFCONV= option specifies the absolute function convergence criterion, the FCONV= option specifies the relative function convergence criterion, the GCONV= option specifies the relative gradient convergence criterion, and the XCONV= option specifies the relative parameter convergence criterion. The RIDGING= option specifies the technique used to improve the log-likelihood function when its value is less than that of the previous step.

PROC LOGISTIC now supports the PREDPROBS= option in the OUTPUT statement. This option requests individual, cumulative, or cross validated predicted probabilities. The LACKFIT option now enables you to specify a number n to be subtracted from the number of partitions to give the correct degrees of freedom for the Hosmer and Lemeshow test.

[IMAGE] The LOGISTIC procedure supports the CLASS statement and the specification of model effects similar to the GLM procedure. You can specify the type of parameterization to use, such as effect coding and reference coding, the ordering of the classification variables, and the reference level. Such specifications can be done globally or for individual variables. See the information on the CLASS statement in the SAS/STAT User's Guide for more detail.

Chapter Contents
Previous
Next
Top of Page