User's Guide to Ratio Variables

Abstract

Some sociologists intuit a necessary relation between ratio variables with common components, and they warn that observed relations between such variables are likely to be artifactual. In rebuttal, other sociologists argue that the use of ratio variables is justified when the ratios themselves, rather than their components, are of Both arguments are based on mistaken ideas about estimating causal effects in nonexperimental research. To realize unbiased estimates, social scientists often must controlfor a dominant confounding variable Z (e.g., population size). One control procedure is to divide the other variables by Z, i.e., to use ratio variables. Contrary to the artifactuality argument, this control procedure does not introduce bias; but contrary to the argument, the control procedure need not be restricted to instances where the ratios themselves are of theoretical interest. Conventional wisdom notwithstanding, theoretical interest is not a reliable guide for determining whether or not to use ratio variables.