The assumption that experts can make approximately optimal subjective and unaided decisions underlies some parameter estimation methods and decision support approaches. We developed experimental testing software and evaluated the performance of student subjects in this kind of decision making for an operational business planning problem having three decision variables and two performance criteria. We computed a measure of error between an observed trial and an optimal one called decisional regret. Some theory-based assumptions about such measures are that they have either a gamma or Weibull distribution with a certain bound on the shape parameter, and also exhibit learning in repeated trials. We found strong support for the distributional assumption and learning properties, with weaker support for the shape parameter assumption. Overall, our results are supportive of the basic assumption.
