The Effect of Contextual and Personal Factors on the Use of Probabilistic Recommenders in E-Markets

Abstract

E-markets of the future may extend to real-time, on-line auctions of, for example, real estate or antiques. To operate successfully in real-time e-markets, knowledge-based recommenders may become indispensable. Real-time emarket trading decisions typically take place in dynamic and uncertain environments in which high stakes, time pressure, multiple players, and feedback loops are key characteristics. A 2x2x2 design was used in which risk, time pressure and availability of a probabilistic recommender were manipulated. The effect of the recommender on the confidence and performance of three groups of decision-makers (novices, journeymen and expert) was examined. Finally, deductive logic ability was assessed. 84 participants engaged in a laboratory-based computer simulation of a customized version of Blackjack that approximated a real-world e-market decision environment. Recommender use differed with expertise level. Novices were significantly more confident when the recommender was available and used its advice to avoid decision errors. The recommender did not alter the confidence of journeymen and experts but they used it to avoid errors as contextual factors became more complex. There were significant group differences in recommender use to improve strategic outcomes: recommender use enhanced experts' performance, damaged journeymen's and had little effect on novices'. In addition, deductive logic ability was significantly related to fewer decision errors. These results suggest that, to be effective in dynamic e-market environments, knowledge-based recommenders should be tailored to individual differences, particularly decision-makers' expertise and reasoning ability.