Consumer sequential search: Not enough or too much?

W e study sequential search behavior in a generalized "secretary problem" in which a single object is to be selected from a set of n alternatives. Alternatives are inspected in a random order, one at a time, and only the rank order of the current alternative relative to the ones that have already been observed can be ascertained. At each period, the consumer may either accept the current alternative, continue to search and pay a fixed cost, or recall an alternative that has already been inspected. A recalled alternative is assumed to be available with a known probability. The consumer's goal is to select the overall best alternative from the fixed set. We describe the results of an experiment designed to test the optimal model and compare it to a behavioral decision model that incorporates local patterns of the observed sequence. Both set size and search cost are manipulated experimentally in a 2x2 factorial design. Our results show that cost and set size affect the amount of search in the predicted direction. However, in the two no-cost conditions subjects search too little in comparison to the optimal model, whereas in the two cost conditions they search too much. The behavioral decision rule that we propose provides a possible account for the observed pattern of the behavioral regularities