A company is a product testing a new medical device. Each day the device is tested on
Question:
A company is a product testing a new medical device. Each day the device is tested on 25 subjects. A QC statistician is hired to calibrate the device so it accurately detects high or low levels of a particular chemical substance. The statistician fits a ridge-type logistic regression model to the first 4 days' worth of data and uses leave-one-out cross-validation to select lambda. He then applies leave-one-out CV again to the first 4 days' worth of data with the selected lambda to estimate the misclassification error. He is delighted to find it is 2% and tells the company that he will be able to predict high and low levels on data from day 5 with 98% accuracy.
What's wrong with this procedure? How can you design a different procedure for an unbiased measure of misclassification error?
Managing Supply Chain and Operations An Integrative Approach
ISBN: 978-0132832403
1st edition
Authors: Thomas Foster, Scott E. Sampson, Cynthia Wallin, Scott W Webb