Two for the Price of One: Info-Gap Robustness of the 1-Test Algorithm

Analysts in many domains must choose a design, a strategy, or an intervention without being able to test all relevant alternatives. We consider a situation in which one of two alternatives must be chosen, while only one alternative can be tested prior to decision. The probability of success from blind choice is 1/2. The probability of success if the distribution of the system attributes is known is 3/4. The 1-test algorithm assures probability greater than 1/2 of choosing the better system based on a single test, even without knowing the probability distribution of the system attributes. If the distribution is poorly known, then info-gap theory can robustify the 1-test algorithm. Using the info-gap robustness function we show that robust-satisficing algorithms may differ from the nominally optimal algorithm when the attribute distribution is uncertain.