How useful are perception- and experienced-based measures in predicting actual food choice? Evidence from an in-store field experiment using a multi-response approach

Multi-response approaches have gained popularity in product-based research for obtaining deeper and more generalizable product insights related to product performance. Through a natural in-store field experiment, this study examined how perception-based and experience-based measures, within a multi-response approach, serve to predict actual product choice. We investigated consumers’ evaluation and acceptance of a novel biofortified orange-fleshed sweetpotato (OFSP) puree bread with a set of 141 (Treatment A: OFSP bread) and 204 (Treatment B: other bread) randomly selected bread buyers in Nairobi, Kenya. Non-parametric machine learning methods were used for partitioning and predictive purposes. They included unbiased conditional inference trees and theory-based model-based recursive partitioning. After identifying an individual specific reference product, we first examined how Just-About-Right scaling for sensory evaluation related to nutritional beliefs, liking, and economic valuation. The latter measure was based on an incentive-compatible and consequential Becker-DeGroot-Marshak approach. The results revealed structural differences in the way that the evaluation of sensory attributes are predictive of the perception-based measures. Liking evaluation was strongly predictive of actual choice, while the economic valuation was only weakly so. Notably, neither nutritional beliefs as a food quality characteristics nor any of the dimensions related to product conceptualizations by CATA counts, including that of emotions, were predicative of the actual choice. These results suggest a further need to develop and integrate measures related to circumstantially derived state of wanting to better predict actual product choice in a natural environment