An empirical test of the role of value certainty in decision making

Abstract

Most contemporary models of value-based decisions are built on value estimates that are typically self-reported by the decision maker. Such models have been successful in accounting for choice accuracy and response time, and more recently choice confidence. The fundamental driver of such models is choice difficulty, which is almost always defined as the absolute value difference between the subjective value ratings of the options in a choice set. Yet a decision maker is not necessarily able to provide a value estimate with the same degree of certainty for each option that he encounters. We propose that choice difficulty is determined not only by absolute value distance of choice options, but also by their value certainty. In this study, we first demonstrate the reliability of the concept of an option-specific value certainty using three different experimental measures. We then demonstrate the influence that value certainty has on choice, including accuracy (consistency), confidence, response time, and choice-induced preference change (i.e., the degree to which value estimates change from pre-to post-choice evaluation). We conclude with a suggestion of how popular contemporary models of choice (e.g., race model, drift-diffusion model) could be improved by including option-specific value certainty as one of their inputs.

Coherence Shifts in Attribute Evaluations

“…Tasks that neither focus on value nor require comparison of options (see Experiments 4-5) should lead to lower (but still detectable) degrees of refinement. Such a pattern would be consistent with previous work showing that merely repeating isolated evaluations of options causes the evaluations to become more precise (Lee & Coricelli, 2020).…”

“…We then assessed the relationship between choice difficulty measured by the difference in initial overall value ratings between the options, dV1, where the subscript distinguishes the initial ratings (1) from the final ratings (2). As in previous studies (Lee & Coricelli, 2020;Lee & Daunizeau, 2020a, 2020b, dV1 had a reliable negative relationship with SoA and RT, and a reliable positive relationship with choice consistency and choice confidence. We used GLM to regress dV1 on overall SoA, on RT, on consistency, and on confidence, separately.…”

“…The experiments were constructed using the online experimental platform Gorilla (gorilla.sc). The experimental stimuli were drawn from a set of 200 digital images used in a previous study (Lee & Coricelli, 2020), each representing a distinct snack food item. For Experiment 1, we used a subset of 100 stimuli, identical for all participants.…”

“…For this task, 50 pairs of stimuli were displayed on the screen, one pair at a time, in a sequence randomized across participants. The pairings of items for each choice trial were created so as make the choices relatively difficult, as assessed by small differences in value ratings between the two items in a choice pair in a previous study (Lee & Coricelli, 2020). Each individual item occurred in a single choice pair.…”

Trading Mental Effort for Confidence in the Metacognitive Control of Value-Based Decision-Making

“…Here, we also ran, for each participant, a multiple linear regression of response times against |ΔVR 0 | Figure 6 below shows the quantitative model predictions and summarizes the corresponding empirical data. (Lee and Daunizeau, 2020;Lee and Coricelli, 2020). In addition to expected changes in value ratings, the MCD model predicts that the precision of value representations should increase after the decision has been made (cf.…”