Logistic regression residuals

[nathanb_1993]

Hi,

I have a dataset which has been created with both predicted and actual values from a logistic regression model (the only other information in the dataset are the original inputs into the model). 

The dataset was already created so no PROC LOGISTIC / GLM has been used.

I know all of the individual parameters in the model and their estimates, however I need to do some analysis on the residuals which are not included in the dataset.

Probably a longshot, but does anyone know how I could compute the residuals from these logistic regression model estimates and actuals?

Thanks!

Nathan

[SAS_Rob]

I assume when you say you have the predicted values these are not predicted probabilities but some kind of classification based on the predicted probability.  If it is actually the predicted probability then you should be able to compute the residual in a data step.

If my assumption is correct then you could put the parameter estimates into a data set that is in the correct format for use as input into Proc LOGISTIC with the INEST= option and MAXITER=0.  You could then use the OUTPUT statement to recreate the predicted probabilities/residuals. 

You could also use one of the methods here as well http://support.sas.com/kb/33/307.html

[Reeza]

Residuals are actuals - estimate. 

[nathanb_1993]

Hi Reeza,

Agreed, though due to this being a logistic regression, rather than simple liner, can it be computed in a manner as simple as this?

I actually think my statistical knowledge is getting all a bit confused!

[Reeza]

I believe so, but if you're really, really concerned, test it. Run one of the demo codes for logistic regressions and see if the residuals are the same calculating it both ways.

[StatDave]

You say you have the actual response values and all the inputs, which suggests you can simply refit the model in PROC LOGISTIC and use the OUTPUT statement to generate the regression diagnostics of interest.