ABSTRACT
Of course many explanatory variables encountered in statistical practice are categorical rather than continuous in nature. Mismeasurement arises for such a variable when the actual and recorded categories for subjects can differ. Fundamentally this misclassification differs from measurement error as discussed in Chapter 2, as now the surrogate variable cannot be expressed as a sum of the true variable plus a noise variable. Rather, one must characterize the mismeasurement in terms of classification probabilities, i.e., given the true classification, how likely is a correct classification. This chapter focusses primarily on the impact of misclassification in binary explanatory variables, with some discussion of polychotomous explanatory variables at the end of the chapter. Schemes to ameliorate the impact of misclassification are not considered until Chapter 5. Thus, as with Chapter 2, this chapter’s role is to instill qualitative understanding of when unchecked mismeasurement can produce very misleading results. While the respective literatures on misclassification of categorical explanatory variables and measurement error of continuous explanatory variables have few points of contact, this chapter demonstrates considerable similarities between the respective impacts of both these kinds of mismeasurement.
