Using Mechanisms Built in a Design Class to Test Methods for Decision Under Uncertainty

An approach for testing methods for decision-making under uncertainty has been proposed by the authors and their coworkers. The testing approach uses data for which the generating mechanism and probability distribution is unknown. The approach simulates a very large number of decisions and outcomes of uncertain events to test methods. The study presented in this paper follows the above approach to test methods for decision under uncertainty experimentally on a decision problem involving two decision makers, an inventor and a financier who want to develop and market a new device. Real-life data for simulating the outcomes of the project was collected using 133 slider-crank mechanisms that undergraduate students constructed. The mechanisms were constructed and measured to simulate the entire risky venture of developing and marketing a new device or product. The data obtained was used to simulate thousands of decisions of the inventor and the financier on a computer, using different methods for decision under uncertainty; standard probability, imprecise probability and Bayesian probability. These methods were then judged on the basis of the expected utilities that they produced when used by the two decision makers and also on their sensitivities to changes in the amount of available information or the risk attitudes of the decision makers. In general, the imprecise probability method was found insensitive to the changes in the problem parameters. Bayesian probability method, although sensitive to prior models of uncertainty, produced the highest utilities in general and it was also found insensitive to the information level of a decision maker, when a good prior was chosen.