The Influence of Trade-off Difficulty Caused by Preference Elicitation Methods on User Acceptance of Recommendation Agents Across Loss and Gain Conditions.
Information Systems Research, 2011, Volume 22,
Issue 4, Page 867-884.
Prior studies on product recommendation agents (RAs) have been based on the effort-accuracy perspective in which the amount of effort required to make a decision and the accuracy of such decisions are two dominant antecedents of user acceptance of RAs. The current study extends the effort-accuracy perspective by considering trade-off difficulty, a type of negative emotion that arises when attainment of one's goals is blocked by the attainment of other goals; consequently, one must make trade-offs among the conflicting goals. Many product purchase choices for which RAs are used require users to make trade-offs among conflicting product attributes. A key feature of RAs, the preference elicitation method (PEM), often compels users to make explicit trade-offs. We examine whether an RA's PEM generates trade-off difficulty, which, in turn, affects users' evaluations (i.e., perceived amount of effort and perceived accuracy of recommendations) and the resultant acceptance of the RA. Trade-off difficulty influences users' evaluations of an RA via perceived control over execution of the RA PEM. In addition, the decision context in which users employ a PEM moderates the degree to which that PEM generates trade-off difficulty. Specifically, a PEM generates a greater degree of trade-off difficulty in a choice context that leads to a loss than in a choice context that leads to a gain. Consequently, users exert more effort to cope with trade-off difficulty in a loss condition. Because users voluntarily spend more effort, the negative influence of perceived effort on users' acceptance of an RA-which is supported in prior studies-decreases in a loss condition. A laboratory experiment was conducted using two between-subject factors: two RAs, one that employed a trade-off-compelling PEM and the other a trade-off-hiding PEM, and two decision contexts, one of which was a loss condition and the other a gain condition. The results supported all of the hypotheses.