How to find outliers in proc logistic?

[MikeTurner]

In ordinary regressions, we can use 2*sd or students' residual to detect and remove outliers, but in logistic regression, how to find outliers?

thanks

[SteveDenham]

If you are using PROC LOGISTIC, check out the syntax for the OUTPUT statement.  It will give you many of the parameters you are looking for--studentized residuals, leverage, dfbetas, etc.

Steve Denham

[StatDave · Accepted solution]

The standardized deviance residuals and the likelihood residuals (available in PROC LOGISTIC in SAS 9.3, or in PROC GENMOD in earlier releases) have distributions reasonably close to standard normal.  So, values greater than 2 or less than -2 might be considered suspect.