The effects of dichotomization on the diagnostic accuracy of hypothesis tests: How avoiding dichotomization is in a researcher’s self-interest

Although researchers have valued dichotomization because it can simplify conducting and presenting analyses, statisticians have criticized the general use of this procedure because it reduces power and can lead to invalid results. Despite this, researchers continue to regularly dichotomize continuous measures, as evidenced by the presence of median splits and other forms of artificial categorization in the published literature. This suggests that researchers do not find the currently presented arguments against dichotomization compelling enough to avoid using the procedure. In response, this article provides a new, more pragmatic examination of the effects of dichotomization focused on how this practice can impede the pursuit of researchers’ professional and personal goals. Specifically, we present an examination of the effects of dichotomization phrased in terms of the sensitivity, specificity, positive predictive value, and negative predictive value of hypothesis tests, showing that dichotomization makes it more likely researchers will publish false results and reduces their ability to detect true findings in their data.