Abstract Recent research on multi-attribute decision-making has challenged the view that in open-view conditions, under time pressure, humans mostly rely on simplified strategies that only examine part of the choice information, as in Take the Best ( TTB ) or the priority heuristics. Here we examine and test a probabilistic extension of TTB which preserves the central heuristic idea that each decision is made based on a single attribute but selects this attribute probabilistically (rather than deterministically as in TTB ) and maintains choice accuracy at levels found in human data. We show that this single probabilistic attribute ( SPA ) model produces choice patterns similar to the compensatory (normative) weighted-average ( WAV ) model, and we computationally compare the SPA model with a similar model called gTTB (Bergert & Nosofsky, Journal of Experimental Psychology: Learning, Memory, and Cognition, 33:107, 2007), showing that SPA provides better fit for 3 attributes to choice data (and about equal fit for 4 and 5 attributes). We then show that the heuristic (SPA/gTTB) and compensatory (WAV) models can be distinguished based on decision times, by contrasting high vs. low choice-polarization trials. To arbitrate between the SPA and the normative model, we collected data on a speeded multi-attribute decision task with 3, 4 and 5 numerical attributes, in a main and a replication experiment (total N = 117 participants). Our data shows significant individual differences in decision strategy. While about 30% of the participants appear to deploy a TTB strategy, the majority (70%) show choices that are consistent with either the SPA or the WAV models. Contrary to Bergert and Nosofsky (Bergert & Nosofsky, Journal of Experimental Psychology: Learning, Memory, and Cognition, 33:107, 2007), we found that the examination of decision-time provided strong evidence against the SPA model and supported instead the normative weighted-average account: when presented with choice information in free view most participants were able to carry out fast (mean-RT < 1.5 s) and compensatory decisions that attend to (and weight) all choice attributes.
Atun et al. (Fri,) studied this question.