topic Re: Interpretation of calibration plot in Statistical Procedures

Interpretation of calibration plot

edasdfasdfasdfa — Sun, 05 May 2019 15:52:30 GMT

Hey guys,

I have attached an image of my calibration plot.

My interpretation: When probability is about 0.6 and greater, predicted probabilities are close to or on target as it relates to the actual outcome, but below that probability they tend to be inaccurate in terms of predicting event?

Re: Interpretation of calibration plot

PGStats — Sun, 05 May 2019 21:08:38 GMT

You should be asking WPS related questions to the WPS community.

Re: Interpretation of calibration plot

Ksharp — Mon, 06 May 2019 14:22:00 GMT

Calling @Rick_SAS . Rick have written several blogs about it .

Re: Interpretation of calibration plot

Rick_SAS — Mon, 06 May 2019 14:34:21 GMT

That looks like a decile calibration plot. You can read about the decile calibration plot and why many researchers do not recommend their use.

I've written other articles about the loess-smoothed calibration plot:

- How to create if you are running SAS/STAT prior to 15.1

- How to create a calibration plot directly in PROC LOGISTIC in SAS?STAT 15.1 or later.

The first two links include a discussion of how to interpret the plots and references to the literature.

Regarding your interpretation: yes, your interpretation seems correct. For low values of the predicted probabilities, the model seems inaccurate. I'd add that the very high predicted probabilities seem to have some issues as well.