This is the dataset for ["Reinforcement learning and Bayesian inference provide complementary models for the unique advantage of adolescents in stochastic reversal" (Eckstein, Master, Dahl, Wilbrecht, Collins, 2022) in *Developmental Cognitive Neuroscience*][1] **Paper abstract** During adolescence, youth venture out, explore the wider world, and are challenged to learn how to navigate novel and uncertain environments. We investigated how performance changes across adolescent development in a stochastic, volatile reversal-learning task that uniquely taxes the balance of persistence and flexibility. In a sample of 291 participants aged 8–30, we found that in the mid-teen years, adolescents outperformed both younger and older participants. We developed two independent cognitive models, based on Reinforcement learning (RL) and Bayesian inference (BI). The RL parameter for learning from negative outcomes and the BI parameters specifying participants’ mental models were closest to optimal in mid-teen adolescents, suggesting a central role in adolescent cognitive processing. By contrast, persistence and noise parameters improved monotonically with age. We distilled the insights of RL and BI using principal component analysis and found that three shared components interacted to form the adolescent performance peak: adult-like behavioral quality, child-like time scales, and developmentally-unique processing of positive feedback. This research highlights adolescence as a neurodevelopmental window that can create performance advantages in volatile and uncertain environments. It also shows how detailed insights can be gleaned by using cognitive models in new ways. **Dataset Description** This repo contains one csv file per participant. Section 4.1 in our openly accessible [DCN paper][2] details the sample composition (children, adolescents, adults) and exclusion criteria. The data files are organized in the *long format*, i.e., one row per trial. The following columns are present in each csv file: * RT: Response times in miliseconds * selected_box: 0 when participant selected the left box; 1 for the right box * reward: 0 if trial was not rewarded (no coin won); 1 if trial was rewarded (coin won) * correct_box: which box was the "correct" one on this trial, i.e., had a probability of 0.75 of returning a reward * sID: participant ID * TrialID: Trial ID * rewardversion: 4 different pre-randomized reward pay-off tables were used across participants. This is the ID of the pay-off table used * ACC: accuracy of the trial, i.e., whether the "correct" box was chosen (independent of whether a reward followed or not) * switch_trial: whether the correct box switched sides on this trial (boolean); switching occurred when a pre-determined, pseudo-random number of rewards had been obtained for the previous side * block: incremented by 1 each time the correct box switched sides * trialsinceswitch: number of trials before / after a switch trial We also obtained saliva samples, height, weight, pubertal development, parental education, and socio-economic status for (most of) our sample. This information is provided at the [overarching project repo][3] in the file EcksteinEtAl_eLife2022_Demographics.csv. **Contact** For any questions or remarks, please get in touch with maria.eckstein@berkeley.edu and/or annecollins@berkeley.edu! [1]: https://www.sciencedirect.com/science/article/pii/S1878929322000494 [2]: https://www.sciencedirect.com/science/article/pii/S1878929322000494#sec4 [3]: https://osf.io/h4qr6/