What can be done when a test is destructive? How do we characterize measurement error? How can we determine if a test method is adequate for a given product or application? How can we check for bias?

All of the techniques for assessing the quality of a measurement system require us to make multiple measurements of the same thing. This allows us to isolate the test-retest error from the product variation so that we can estimate and characterize measurement error. However, when the act of obtaining a measurement also destroys the sample being measured, it becomes impossible to test the same thing twice. So what do we do?

In those cases where we can obtain paired samples that are thought to be reasonably similar, we test the paired samples with our destructive measurement system, and use the difference between the values obtained as an approximation to the test-retest error. While this difference will also contain some amount of product variation, hopefully it will be small. By doing this repeatedly and averaging the results, it is possible to obtain an upper bound on the test-retest error for the measurement system. Our first two examples will illustrate this approach.

 …