Predicted probabilities for each observation can be saved in a data set by using the P= or PREDPROBS= option in the OUTPUT statement. Typically, the PREDPROBS= option is preferred for these reasons:

• A predicted response level is also produced (and saved in variable _INTO_)
• Predicted probabilities that are associated with each response level are produced
• For responses with more than two levels, all information for an input observation is output to one observation rather than to several observations as is done by the P= option

For more information about the PREDPROBS= option, see "Details of the PREDPROBS= Option" in the Syntax: OUTPUT Statement section of the PROC LOGISTIC documentation.

Upper and lower confidence limits can also be added to the output data set by specifying the U= and L= options in the OUTPUT statement. Predicted probabilities and their confidence limits are computed as shown in "Linear Predictor, Predicted Probability, and Confidence Limits" in the Details section of the PROC LOGISTIC documentation.

For examples and more discussion, see the Allison reference.