Computational Mechanisms of Attribute Translations

Attribute translations, a choice architecture intervention technique aiming to promote behavior change by translating decision-relevant information into more comprehensible and meaningful units for laypersons (e.g., translating nutritional information into a rating scale), have been extensively adopted by policymakers in recent years. However, little is known about the computational mechanisms that underlie their effects on behavior. To address this gap, we investigate the role of visual attention in inducing behavior change by modeling the dynamic interplay of information acquisition and evidence accumulation. For our analyses, we used a pre-existing data set from an online process tracing study in which participants completed a multi-attribute value-based decision-making task. Participants performed the task twice, with and without an additional translation of the items' energy and water consumption. Our modeling results suggest that behaviorally less effective attribute translations (in the form of numeric information) only affected participants' response caution. In contrast, behaviorally effective attribute translations (in the form of a qualitative rating) impacted participants' preference formation by shifting the attribute weights in favor of the translated attribute as well as by decreasing the bias on the attended option. These findings add critical insights to the ongoing debate on the degree of agency that is promoted by choice architecture interventions as they imply that attribute translations may promote more considerate, preference-aligned decisions rather than simply provide pre-decisional shortcuts.