BookmarkSubscribeRSS Feed
EPM22
Calcite | Level 5

Good morning, 

I am using NHATS data with complex sampling design and weights. Does anyone have SAS code to plot the ROC and compute c-statistic with confidence intervals that properly consider the complex sampling design?  Could you share code if you have done this before.  Thanks 

1 REPLY 1
StatDave
SAS Super FREQ

The ROC curve area and its confidence interval resulting from any binary-response model or classifier can be computed using the ROC statement in PROC LOGISTIC. Several examples are shown in this note. As can be seen in the note, the computations depend only on the predicted and actual classifications from the model/classifier. The method used is a nonparametric method based on U statistic theory as discussed in "Receiver Operating Characteristic Curves" in the Details section of the PROC LOGISTIC documentation. As such, I believe that if you use the predicted classifications from a proper analysis of your survey data, as could be done using PROC SURVEYLOGISTIC, then the area and confidence interval can be obtained using the ROC statement in PROC LOGISTIC. You can further investigate by seeing the DeLong et al. paper cited in the above documentation section which details the method including the variance computation.

hackathon24-white-horiz.png

Join the 2025 SAS Hackathon!

Calling all data scientists and open-source enthusiasts! Want to solve real problems that impact your company or the world? Register to hack by August 31st!

Register Now

What is ANOVA?

ANOVA, or Analysis Of Variance, is used to compare the averages or means of two or more populations to better understand how they differ. Watch this tutorial for more.

Find more tutorials on the SAS Users YouTube channel.

Discussion stats
  • 1 reply
  • 393 views
  • 2 likes
  • 2 in conversation