A parallel accumulator model accounts for decision randomness when deciding on risky prospects with different expected value

In decision-making situations individuals rarely have complete information available to select the best option and often show decisional randomness, i.e. given the same amount of knowledge individuals choose different options at different times.  Dysfunctional processes resulting in altered decisional randomness can be considered a target process for psychiatric disorders, yet these processes remain poorly understood.  Advances in computational modeling of decision-making offer a potential explanation for decisional randomness by positing that decisions are implemented in the brain through accumulation of noisy evidence, causing a generally less preferred option to be chosen at times by chance.  One such model, the linear ballistic accumulator (LBA), assumes that individuals accumulate information for each option independently over time and that the first option to reach a threshold will be selected.  To investigate the mechanisms of decisional randomness, we applied the LBA to a decision-making task in which risk and expected value (EV) were explicitly signaled prior to making a choice, and estimated separate drift rates for each of the four task stimuli (representing high and low EV and high and low risk).  We then used the fitted LBA parameters to predict subject response rates on held-out trials for each of the 6 possible stimulus pairs.  We found that choices predicted by LBA were correlated with actual choices across subjects for all stimulus pairs.  Taken together, these findings suggest that sequential sampling models can account for decisional randomness on an explicit probabilistic task, which may have implications for understanding decision-making in healthy individuals and in psychiatric populations.
Methods
Participants

Forty-four college students (age: 19.34 ± 2.28 years; 26 females and 18 males) participated in this study. Subjects were recruited from San Diego State University through an online system as part of Psychology 101 class during the spring of 2013. They were contacted and scheduled for an experimental session during winter quarter 2013. The study was approved by the Human Research Protections Program at San Diego State University. All participants provided written informed consent, and were compensated $25 and 2.5 course credits for completing the study. Prior to completing the experimental task, participants completed the State-Trait Anxiety Inventory, Trait scale (STAI-T) to assess trait anxiety.

Decision-Making Task

Subjects completed a decision-making task consisting of 4 blocks with 24 trials per block. On each trial, subjects were asked to choose one of two gambles. They were asked to imagine they were choosing between two different random drawings with different numbers of chips worth 0, 20, or 40 points (subjects played for virtual points rather than real money). For each of the two options, subjects were shown the number of each type of chip (out of a total of 100). Subjects were thereby shown the value and probability of each possible outcome. There were four different stimuli (i.e. four different drawings) used in the task with different risk and expected value (EV) profiles: high risk/high EV, high risk/low EV, low risk/high EV, and low risk/low EV. High risk gambles had variance 384 (with larger numbers of 0-point and 40-point chips), while low risk gambles had variance 96 (with larger number of 20-point chips). High EV gambles had EV of 24 points, while low EV gambles had EV of 20 points. With 4 different stimuli, there were 6 different possible pairs of stimuli, each of which was encountered 16 times during the course of the task (subjects were never asked to choose between two identical stimuli). After making a selection, subjects were shown the outcome of their choice (either 0, 20, or 40 points, which was determined by a random number generator in accordance with the stated probabilities). Half of blocks were “Counterfactual Feedback” blocks, in which subjects were shown what they would have received if they had made the opposite choice. The other half were “No Counterfactual Feedback” blocks, in which subjects were not shown the outcome of the opposite choice. The order of “Counterfactual Feedback” and “No Counterfactual Feedback” blocks was counterbalanced between subjects. After receiving feedback, subjects were asked to indicate their level of satisfaction with their choice on a visual analog scale.