Naturalistic multiattribute choice

Sudeep Bhatia,Neil Stewart

doi:10.1016/j.cognition.2018.05.025

Sudeep Bhatia, Neil Stewart

Open Access

https://doi.org/10.1016/j.cognition.2018.05.025

Copy DOI

Abstract

We study how people evaluate and aggregate the attributes of naturalistic choice objects, such as movies and food items. Our approach applies theories of object representation in semantic memory research to large-scale crowd-sourced data, to recover multiattribute representations for common choice objects. We then use standard choice experiments to test the predictive power of various decision rules for weighting and aggregating these multiattribute representations. Our experiments yield three novel conclusions: 1. Existing multiattribute decision rules, applied to object representations trained on crowd-sourced data, predict participant choice behavior with a high degree of accuracy; 2. Contrary to prior work on multiattribute choice, weighted additive decision rules outperform heuristic rules in out-of-sample predictions; and 3. The best performing decision rules utilize rich object representations with a large number of underlying attributes. Our results have important implications for the study of multiattribute choice.

Highlights

Most choices that people make on a day-to-day basis can be seen as involving objects defined on two or more attribute dimensions
This was followed by equal weights heuristic (EW), weighted pros heuristic (WP), tallying heuristic (TAL), fast and frugal tree (FFT), and lexicographic heuristic (LEX)
The food dishes were obtained from www.AllRecipes.com, and there were a total of 100 unique food dishes used in the study

Summary

Introduction

Most choices that people make on a day-to-day basis can be seen as involving objects defined on two or more attribute dimensions These choices involve trading off the relative values of the component attributes, so as to select the object whose attributes are, overall, the most desirable (Keeney & Raiffa, 1993). Some consumer decisions do involve the evaluation of a small set of explicitly presented and quantified attributes, many other common decisions – involving, for example, movies to watch or food items to eat – do not The objects in these common decisions may be listed using only their names (without any attribute information), but the underlying attribute structure is typically very rich and complex (e.g. Fig. 1b). Decision makers do often have knowledge about these objects and their underlying attributes, but this knowledge is represented in the decision makers’ minds after having been learnt through prior experience with the choice domain

Objectives

Methods

Results

Conclusion