Multi-attribute sequential search

Abstract

This article describes empirical and theoretical results from two multi-attribute sequential search tasks. In both tasks, the DM sequentially encounters options described by two attributes and must pay to learn the values of the attributes. In the continuous version of the task the DM learns the precise numerical value of an attribute when she pays to view it. In the threshold version the DM learns only whether the value of an attribute is above or below a threshold that she sets herself. Results from the continuous condition reveal that DMs tended to terminate their searches too early relative to the optimal policy. The pattern reversed in the threshold condition: DMs searched for too long. Maximum likelihood comparisons of two different stochastic decision models showed that DMs under both information conditions performed in ways consistent with the optimal policies. Those offered continuous-valued attribute information did not, however, spontaneously degrade this information into binary (acceptable/unacceptable) form, despite the theoretical finding that satisficing can be a very effective and efficient search strategy.