Predicting Risk-Sensitivity in Humans and Lower Animals: Risk as Variance or Coefficient of Variation

Abstract

In this paper, we examine the determinants of risk-sensitivity exhibited by humans and other animals. Our dependent measure is the proportion of respondents who choose a sure option over a risky option with equal expected value. We present a meta-analysis of human risk-preference data and compare it to the results of a similar meta-analysis of animal data by Shafir (2000). Both sets of data show that the coefficient of variation (CV), a relative measure of risk per unit of return, significantly predicts choices across a broad range of decision situations. In those situations where the CV can be compared to outcome variance, a more traditional (absolute) measure of risk, the CV outperforms variance as a predictor of risk sensitivity. This is especially true when decision makers (humans, or animals foraging for food) acquire information about choice outcomes and their variability experientially and over time, as demonstrated in an experiment in which we attempted to put students into a risky learning and decision making situation comparable to the experiential information acquisition in risky foraging choice tasks in animal experiments.