In measurement science, “bias” refers to the systematic error component of the measurement system. Unlike other types of measurement error that are randomly distributed, a bias predictably shifts a measurement in the same direction.
For instance, I recently facilitated a “round robin” measurement correlation study with two other companies, where we compared the outputs of our hardness testers using the same set of test samples. While preparing for that study, I realized that one of our hardness testers, on average, tested 1.1 Rockwell B points higher than the reference sample. It wasn’t testing exactly 1.1 points over, but instead ranged from .8 to 1.4 points over across a series of tests. In other words, we had two error components: a bias of +1.1 points and a random error of +/- .3 points. To compensate for this bias, we shifted down the output reading of the tester by 1.1 points, leaving only the random error component in our tester’s output values.[Read more…]