These practices have other inherent limitations. First, classical and modern test theory emphasize that observed measures (i.e., items), whether binary, categorical or otherwise, contain measurement error (Crocker and Algina 1986) and are thus imperfect “indicators” of the underlying construct of interest. Indicators that contain large measurement errors are less reliable limiting the statistical power to detect significant effects (Crocker and Algina 1986). Second, binary items are frequently used to represent the presence or absence of trauma without providing information about the severity of the trauma experienced. Third, binary items treat the trauma as a unidimensional construct. However, various forms of trauma exist which, although interrelated, may each lie on its own severity continuum. Fourth, when multiple items are added to form a sum score (which often is then dichotomized), the items are equally weighted in that sum. This is equivalent to assuming that each item measures the construct equally well despite the possibility that certain items might comprise more measurement error than others.
