Revisiting the evidence for cardinal treatment of ordinal variables

Abstract

Well-being (life satisfaction or happiness) is a latent variable that is impossible to observe directly. Moreover, it does not have a unit of measurement. Hence, survey questionnaires usually ask people to rate their well-being in different domains. The common practice of comparing well-being by means of averages or linear regressions ignores the fact that well-being is an ordinal variable. Since data is ordinal, monotonic increasing transformations are permissible. We illustrate the sensitivity of empirical studies to monotonic transformations using examples that relate to well-known empirical papers, and provide two theoretical conditions that enable us to rank ordinal variables. In our examples, monotonic increasing transformations can in fact reverse the conclusion reached.