Consumer Learning of New Binary Attribute Importance Accounting for Priors, Bias, and Order Effects

• Published in: Marketing science (Providence, R.I.) R.I.), 2012-07, Vol.31 (4), p.549-566
• Main Authors: Chylinski, Mathew B., Roberts, John H., Hardie, Bruce G. S.
• Format: Article
• Language: English
• Subjects: Analysis; Associative learning; Attribution theory; Bias; Consumer advertising; Consumer behavior; Consumer behaviour; Consumer education; Consumer psychology; consumer utility; Consumers; Economic theory; Experimental methods; Laboratories; Learning; Marketing; Modeling; Noise; Objectives; Observational learning; Parametric models; preference dynamics; Preferences; Probability; Studies; Utility functions; Utility models
• ISSN: 0732-2399; 1526-548X
• DOI: 10.1287/mksc.1120.0719

This paper develops and calibrates a simple yet comprehensive set of models for the evolution of binary attribute importance weights, based on a cue-goal association framework. We argue that the utility a consumer ascribes to an attribute comes from its association with the achievement of a goal. We investigate how associations may be represented and then track back the relationship of these associations to the utility function. We explain why we believe this to be an important problem before providing an overview of the extensive literature on learning models. This literature identifies key phenomena and provides a foundation for our modeling of binary attribute importance learning, which can test for three departures from "rational" learning-bias, existence of priors, and the unequal weighting of sample observations (order effects). We apply our models in a laboratory setting under a number of different relationship strengths, and we find that, in our application, consumers' learning about attribute-goal associations exhibits bias and the effects of prior beliefs when the sample realizations occur with and without noise, and order effects when the sample realizations occur with noise. We provide an example of how our models can be extended to learning about more than one attribute.