Isaiah Lankham, University of California Office of the President, Oakland, CA;
Matthew Slaughter, Kaiser Permanente Center for Health Research, Portland, OR
Validation is essential for assessing a predictive models performance with respect to optimism or overfitting. While traditional sample-splitting techniques (like cross-validation) require us to divide our data between model building and model assessment, bootstrap validation enables us to use the full sample for both. Theres a simple method for efficiently calculating bootstrap-corrected measures of predictive model performance in SAS. While several SAS procedures have options for automatic cross-validation, bootstrap validation requires a more manual process. See examples that focus on logistic regression using the LOGISTIC procedure in SAS/STAT(r), with additional discussion of how these techniques can be extended to other procedures and statistical models.
The running example can be downloaded in Jupyter Notebook and .sas file format from https://github.com/saspy-bffs/sgf-2021-bootstrap-validation.
Watch A Framework for Simple and Efficient Bootstrap Validation in SAS, with Examples as presented by the authors on the SAS Users YouTube channel.
READ the PAPER
Please see attached, or visit https://github.com/saspy-bffs/sgf-2021-bootstrap-validation