Efficient coding accounts for faster and more accurate choices on high-valued items

Abstract

One of the ultimate goals of decision-making research is to identify cognitive processes underlying choice behavior similar to the ones we make in our lives as consumers. One common observation in the choice of the goods is that consumers tend to select a set of more valuable goods and choose the one within the set. This is in line with a commonly accepted human decision-making process where the brain represents a value of options first and then compares them to choose one. However, it has not been fully understood how subject value-based choice behavior varies with the value level (i.e., high or low). In this study, we aim to investigate subject-value based choice behavior with various value levels in terms of the speed and accuracy of choice at different levels of choice difficulty (i.e., smaller or larger value difference). We also aim to account for an underlying value computation process by comparing two alternatives of mechanisms – efficient coding and attention ¬– in the framework of sequential sampling models (SSMs) of binary choice. Here, we chose snacks as a target goods. To categorize snacks into different subjective value levels, we first collected a set of snacks that participants had tried before (familiarity check), and evaluated how much they want to eat each of the collected snacks by measuring ratings from -10 (not at all) to 10 (very much). This rating task was repeated three times with a random order of stimulus presentation and the mean rating values of each snack were used for classifying snacks into three value levels (i.e., low, medium, and high). The value level was determined such that the number of snacks in ??? was maximized while each snack could belong to only one value level or none. Choice pairs were created using two snacks in the same value level with four levels of value difference. Accordingly, there were 12 conditions of choice pairs with maximum of 30 trials for each condition. Behavior results showed that the larger the value difference was, the faster the choice was made with higher accuracy. Increases in speed and accuracy as a function of the value difference were steeper when choosing between high-valued items than low-valued ones. Based on the behavioral results, we fitted to three SSMs to the behavioral data: 1) a typical SSM only concerning choice difficulty (tSSM); 2) an SSM employing efficient coding for high-valued items (eSSM); and 3) an SSM employing attention to high-valued items (aSSM). From the model analysis, eSSM was found to be the only model that could describe the interaction between the value level and the value difference on choice behavior. Our findings highlight the role of efficient and precise valuation in the choice of daily goods and help to understand subjective value-based decision-making processing.