An Experimental Investigation of the Role of Errors for Explaining Violations of Expected Utility

One possible conclusion from recent experimental research on decision making under risk is that observed behaviour can be reasonable accommodated by expected utility plus an error term. This conclusion implies that the violation rate of expected utility should decrease if errors are excluded. The present paper presents an experiment which investigates this implication. Indeed, the results show that the exclusion of errors leads to a significant reduction of the violation rate for most of the subjects and most of the choice problems under risk. However, it turns out that for decision problems under ambiguity the exclusion of errors in contrast increases the violation rate significantly. In this sense the Ellsberg paradox can be regarded as a more serious challenge of expected utility than the Allais paradox. More general, while expected utility plus error term may be regarded as a reasonable representation for choice under risk this does not seem to be true for ambiguous choice problems.