Extended Data Fig. 2: Cumulative response time distributions for sDDM, tDDM and bDDM.

Cumulative distributions for response times by participant type, choice outcome and data source. Participants estimated to consider taste or health first are plotted in the top and bottom rows, respectively. Response times for choices in favour of (91) less healthy but more tasty (LH_MT), (2) more healthy but less tasty (MH_LT), or (3) both more healthy and more tasty (MH_MT) outcomes are shown in columns 1-3, respectively. Choices in favour of the option rated as less healthy and less tasty were rarely made (less than 5% of trials) and are omitted for clarity. Responses generated by human participants are shown in green lines. Responses generated by simulated agents using the best-fitting sDDM, tDDM, and bDDM parameters are shown in orange, purple, and magenta lines respectively. All three models can recreate the RT patterns in the empirical data equally well when choice outcomes align with the attribute participants consider first. However, the sDDM and bDDM both generate response times that are too fast relative to the empirical data when participants that consider taste first ultimately choose in favour of a more healthy, but less tasty option (row 1, column 2) or if participants that consider health first ultimately choose in favour of a less healthy, but more tasty option (row 2, column 1). In contrast, the tDDM is able to reproduce the observed response time distributions in these cases well.