Neural correlates of the behavioral differences between descriptive and experiential choice: An examination combining computational modeling with fMRI

Recent findings indicate that individuals overweight small probability events when given full descriptive information about available options and the likelihood of their outcomes (descriptive choice) but underweight them when the same information must be learned through repeated choice and feedback (experiential choice). The major hypothesis of this thesis is that feedback is the crucial arbiter of this difference. Furthermore, it is hypothesized that this behavioral disparity is the result of differential recruitment of neural systems. To test the first hypothesis, two groups of individuals received full descriptive information about available options and made repeated choices on two option sets. The only dissimilarity between the groups was that one received feedback after every trial and the other received none. The results supported the hypothesis: the reception of feedback engendered the relative underweighting of small probability events when present and the overweighting of small probability events when absent. Then I fit decision field theory, a computational model of choice, to each individual's data. The model fitting procedure supported the above conclusion but also revealed that feedback drove individuals towards objective probability weighting. To test the second hypothesis, I conducted the same behavioral task, this time while participants underwent a functional magnetic resonance imaging (fMRI) scanning session. Here, individuals separately encountered both feedback and no feedback conditions. Again, behavior depended on the reception of feedback; furthermore, after statistically removing the effects of the outcome phase, a difference in blood oxygenation level dependent (BOLD) activation - contingent on the reception of feedback - was observed in both the posterior cingulate and anterior cingulate (ACC) during the decision phase. The pattern of posterior cingulate activity corresponded with preference for a risky option in the first study. Additionally, I extended the aforementioned computational model of choice to incorporate learning in order to better clarify the neural computations during the choice task. The model was fit to the behavioral data and a time series signal was then extracted - representing a probability prediction error - and correlated with the BOLD activity. This procedure revealed that ACC and posterior parietal activity correlated with the extracted data, suggesting that these regions signal the occurrence of an unexpected event.