The Confounding of Measured and Unmeasured Variables

Abstract

Given the impossibility of measuring and including more than a few of the variables that ideally belong in a theoretical model, we know that biases are introduced when omitted variables are confounded with measured variables. In particular, professional or ideological biases may affect the labels attached to interrelated sets of variables. In sociology we tend to confound "status effects" with many possible nonstatus effects of objective characteristics such as education, occupation, or income. It is shown that, without the introduction of simplifying assumptions, such status and nonstatus effects cannot be disentangled. The case of "status" and "power" effects is also discussed.