Modeling Strategy Switches in Multi-attribute Decision Making
- PDF / 825,147 Bytes
- 16 Pages / 595.224 x 790.955 pts Page_size
- 81 Downloads / 179 Views
ORIGINAL PAPER
Modeling Strategy Switches in Multi-attribute Decision Making Michael D. Lee1
· Kevin A. Gluck2
Accepted: 11 September 2020 © Society for Mathematical Psychology 2020
Abstract We develop and demonstrate a method for inferring changes in strategy use, applicable to decision making in multi-attribute choice. The method is an extension of one developed by Lee, Gluck, and Walsh (Decision 6:335–368, 2019) and continues to rely on a Bayesian approach for inferring strategy switches based on spike-and-slab priors. The extensions improve the existing method in two ways. The first is by using a hierarchical approach to make inferences about the underlying propensity to switch strategies simultaneously at both the individual and group levels. The second is by making inferences about the probability different strategies are used, including the transition probabilities between strategies when switches are made. We demonstrate the method by applying it to data sets from five previous experiments, involving a range of experimental designs and sets of strategies of interest. We conclude by discussing the potential of the method to contribute to addressing basic questions in human decision making involving the nature of adaptation, learning, and self-regulation. Keywords Decision making · Strategies and heuristics · Bayesian methods · Change-point detection · Spike-and-slab priors
Introduction The idea that people can adapt the decision strategies they use is not a new one. Research studying how people search for information then decide between alternatives has often focused on what sorts of information environments, task demands, or individual differences lead to the use of one strategy over another (Bergert and Nosofsky 2007; Bobadilla-Suarez and Love 2018; Br¨oder 2000; Br¨oder and Schiffer 2006; Hilbig 2008; Lee and Cummins 2004; Mata et al. 2007; Newell et al. 2003; Newell and Shanks 2003; Newell and Lee 2011). The different strategies considered have included non-compensatory one-reason decision mechanisms like take-the-best, relatively simple exhaustive strategies that tally the features in favor of each alternative, and more complicated strategies that weight and combine the evidence provided by the features (Gigerenzer
Michael D. Lee
[email protected] 1
Department of Cognitive Sciences, University of California, Irvine, CA, USA
2
711 Human Performance Wing, Air Force Research Laboratory, Wright-Patterson AFB, Dayton, OH, USA
and Goldstein 1996; Gigerenzer et al. 1999). This line of work has tended, however, to focus on whether and why strategy use differs across individuals or experimental conditions. As a consequence, it has typically assumed that the same person uses the same strategy throughout an experimental condition. More fine-grained conceptions of changes in strategy use are provided by models of how people learn or adapt over a sequence of decisions. The most influential of these models is strategy selection learning (SSL; Rieskamp and Otto 2006), which is based on reinforcement learning.
Data Loading...
