Evidence Accumulates for Individual Attributes during Value-Based Decisions

When choosing between different options, we tend to consider specific attribute qualities rather than deliberating over some general sense of the options overall values. The importance of each attribute together with its quality will determine our preference rankings over the available alternatives. Here, we test the hypothesis that the most prominent class of model for simple decisions - sequential sampling or evidence accumulation to bound - can be bolstered by explicitly incorporating variables related to individual attributes in addition to the standard usage of overall value estimates. We examine six datasets in which participants evaluated snack foods both in terms of overall value and individual attributes, then chose between pairs of the same snacks, and show that only models that explicitly incorporate information about the individual attributes are able to reproduce fundamental patterns in the choice data, such as the influence of attribute disparity on decisions, and such models provide quantitatively better fits to the choice outcomes, response times, and confidence ratings compared to models based on overall value alone. Our results provide important evidence that incorporating attribute-level information into computational models helps us to better understand the cognitive processes involved in value-based decision- making.