Measurement

Scientists from numerous disciplines frequently make sense of the world by using yardsticks that they hope will show how their study participants are performing, what they are thinking, and how they interact with others. Numbers are faithfully recorded, spun through various forms of software, and prepared for publication. All this is fine if the yardsticks themselves are true—all the time, in every single place they are used, regardless of who is doing the actual recording of the numbers, and regardless of the circumstances in which the numbers are obtained. But what if the yardsticks themselves are shaky?

In epidemiologic analyses based strictly on counting, a few units in dispute here or there may seem rather unlikely to significantly change the overall interpretation of the data set. However, even a single reclassification of a case from one cell to another, for instance, can force a confidence interval to bracket 1.0 where it otherwise might not do so, or a statistical test to just miss threshold. Although it might not appear as an issue when the data are highly differentiable, the need for quality measurement is fundamentally inescapable. Equally important, whole sections of the field of epidemiology have long ago been unbound from the simple exercise of counting, working instead in arenas in which measurements take the form of scores, scales, and other assessments. In such settings, the challenges to designing and conducting strong and reproducible studies are magnified.

The domain of psychometrics gives criteria concerning the quality of a measurement. Although “psychometrics” has been a label narrowly applied to a particular specialized branch of mathematical and statistical thinking within educational research, this entry uses the term in a broader sense. It explores a handful of concepts that are crucial to all measurements and considers recent examples in the epidemiologic literature that show the importance of such considerations.