Competing theories of multialternative, multiattribute preferential choice.

Abstract

In accounting for phenomena present in preferential choice experiments, modern models assume a wide array of different mechanisms such as lateral inhibition, leakage, loss aversion, and saliency. These mechanisms create interesting predictions for the dynamics of the deliberation process as well as the aggregate behavior of preferential choice in a variety of contexts. However, the models that embody these different mechanisms are rarely subjected to rigorous quantitative tests of suitability by way of model fitting and evaluation. Recently, complex, stochastic models have been cast aside in favor of simpler approximations, which may or may not capture the data as well. In this article, we use a recently developed method to fit the four extant models of context effects to data from two experiments: one involving consumer goods stimuli, and another involving perceptual stimuli. Our third study investigates the relative merits of the mechanisms currently assumed by the extant models of context effects by testing every possible configuration of mechanism within one overarching model. Across all tasks, our results emphasize the importance of several mechanisms such as lateral inhibition, loss aversion, and pairwise attribute differences, as these mechanisms contribute positively to model performance. Together, our results highlight the notion that mathematical tractability, while certainly a convenient feature of any model, should neither be the primary impetus for model development nor the promoting or demotion of specific model mechanisms. Instead, model fit, balanced with model complexity, should be the greatest burden to bear for any theoretical account of empirical phenomena. (PsycINFO Database Record

Robust Modeling in Cognitive Science

“…Thus, simply relying on a checklist of effects misses important aspects of the psychological phenomena being explored, and mischaracterizes the behavior for which an explanation is sought. A better approach is to evaluate detailed cognitive models of the decision processes involved, testing the accuracy of their predictions about the individual decisions that individual people make on these tasks (Evans, Holmes, & Trueblood, 2018;Turner, Schley, Muller, & Tsetsos, 2017).…”

A map of decoy influence in human multi-alternative choice

“…Previous models have sought to explain all three decoy effects in a single computational framework (10). To date, a unified account has proved elusive.…”

“…the price and quality of a product, or the likeability and competence of a political candidate), the introduction of an irrelevant distracter item to the choice set leads to rich and stereotyped biases in decision-making. A major research goal in psychology and economics has been to identify a simple and elegant computational principle that can explain the biases provoked by an irrelevant "decoy" stimulus (10).…”

“…This has led to a proliferation of empirical results and a plethora of computational explanations, including models that rely on loss aversion (11), pairwise normalisation (12), attentional weighting (13)(14)(15)(16), associative biases (14), power-law transformation of attribute values (14), sampling from memory (17,18), or various other forms of reference-dependent computation (19)(20)(21)(22). However, there has been a notable lack of consensus about the computational principles that give rise to decoy effects (10). There are a number of potential reasons for this, but here we focus on one limitation of past studies: most have tested for decoy effects by selecting fixed attribute values for !…”

Dissociable mechanisms govern when and how strongly reward attributes affect decisions

“…Lastly, we note that an attribute-specific, independent race model is not a parsimonious explanation of our data. In an independent race model, the first competitor to cross the threshold determines how the decision is made 91 .…”