topic Re: Regression Question: Best way to make predictions outside the training dataset? in Statistical Procedures

Regression Question: Best way to make predictions outside the training dataset?

Statistics91 — Sun, 24 Mar 2013 06:24:16 GMT

Hello everyone,

I'm trying to learn the best way to get both predicted values and confidence intervals for observations in a test dataset when you are using a model based on a separate training dataset. Better yet, what if I'm just working with one dataset, but some of the observations are missing the dependent variable. How can I get the predicted value/confidence interval for these assuming I have all of the predictor variables present?

In my coursework, I've learned to use Proc Score which does give you the predicted values, but it doesn't get you the confidence intervals. Is there a way to get this in SAS?

Thanks in advance to anyone that can help!

Re: Regression Question: Best way to make predictions outside the training dataset?

PGStats — Sun, 24 Mar 2013 11:50:39 GMT

Look at the OUTPUT statement, it exists for most if not all regression procedures. In proc reg for example, the statistics PREDICTED= LCL=, LCLM=, UCL= and UCLM= will provide the predicted values and confidence intervals (for the mean and for individual observations) for all observations in your dataset, including those with a missing dependent variable value.

Re: Regression Question: Best way to make predictions outside the training dataset?

Statistics91 — Sun, 24 Mar 2013 15:26:15 GMT

Thank you so much!

I've always tried that, but for two years, I didn't realize that you need a period to denote "missing dependent variable" get an estimate for it. I feel a lot better about my regression skills in SAS now. Thanks again for responding!