A problem can occur in the calculation of raw residuals when missing values are present in the data, causing these residuals to be incorrect. The conditions for this problem to occur are the following:

• binary data with y=(0,1)
• OUTPUT statement
• output statistics requesting the raw residual without blups
• the random effect is missing
• no other information is missing that would prevent the computation of the predicted value (without blup).

This combination of conditions should be fairly rare.

When an observation is not used in estimating the model, the response value for that observation will be recoded to match the appropriate 0/1 response as predicted for that observation. If that encoding results in a change of the response value, then the corresponding residual will be incorrect. This problem also occurs when invalid frequencies are present for observations in the input dataset.

To circumvent the problem, use the EVENT='1' response variable option (or any other response variable option) that maps the response value to the same underlying value for the binary response.

Select the Hot Fix tab in this note to access the hot fix for this issue.

Operating System and Release Information

A fix for SAS 9.2 (TS1M0) for this issue is available at:

A fix for SAS 9.2 (TS1M0) with Asian Language Support (DBCS) for this issue is available at: