Charlie Brown and Statistics: An Exchange

To the Editor.— In demonstrating the usefulness of simulation methods for comparing interrater reliability indexes, Spitznagel and Helzer have provided a valuable new perspective on the matter. Their argument that Yule's Y is a better means of quantitatively expressing interrater agreement on psychiatric diagnosis, however, is not compelling. First, their results and conclusions are predicated on the assumption that "the errors of one diagnostician are independent of those of the other." This is a highly unrealistic assumption when made in connection with psychiatric diagnosis, one of its consequences being that it rules out the possibility of a disorder in which different patients display different levels of symptom intensity, leading to differences in their likelihood of receiving a positive diagnosis by any diagnostician. In short, if some patients' conditions are easier to diagnose than others, the "independent errors" assumption does not hold. Schizophrenia and major depression, to name only two, are disorders that therefore