Revealing the Cognitive Computational Mechanisms underlying Social Reward Sensitivity Using a Human Expression Probabilistic Reward Task

This dataset is derived from a behavioral study on the "social version of the Probabilistic Reward Task (social PRT)". It uses real facial expressions (smiling/neutral/sad) as target stimuli. Social feedback consists of female smiling expressions, thumbs-up gestures, and encouraging text, with cumulative rewards displayed in real-time during the task. Participants completed an "emotion two-choice judgment" and were instructed to respond as quickly as possible while ensuring accuracy. The experiment was divided into 3 blocks, with 100 trials per block (300 trials in total), and the ratio of target emotions to neutral stimuli was 1:1. Under correct response conditions, a 3:1 asymmetric reinforcement was used, while no feedback was provided for incorrect responses, and the specific reward probabilities were not explicitly disclosed to participants. The study included Experiment 1 (four between-subject "emotion-feedback" combination conditions) and Experiment 2 (focusing on two high-reward scenarios).The data generation and processing workflow is as follows: The experiment was programmed and presented using E-Prime (E-Prime/E-Studio). Raw trial-level data was automatically recorded by E-Prime (including stimulus presentation and event timestamps, key-press responses, correctness/incorrectness, feedback provision, reaction time, etc.) and exported as text logs and experiment quality reports for subsequent analysis. Reproducible data cleaning and quality control were then performed in Excel: first, trials with raw reaction times (RT) less than 150 ms or greater than 2500 ms were excluded; then, outliers of ln(RT) were removed based on the mean ± 3SD. At the participant level, if the number of non-outlier trials in any block was less than 80 after outlier removal, or if the reward ratio of "rewarded high-reward trials / rewarded low-reward trials" was less than 1.5, or if the number of rewarded high-reward trials was less than 14 and rewarded low-reward trials was less than 2, the participant was deemed to have failed quality control and was excluded entirely. After the above screening, Experiment 2 ultimately retained 83 participants in the "smiling-high reward" group and 46 in the "sad-high reward" group for modeling and statistical analysis (N=129). Some participants, although completing the experimental tasks, were not included in subsequent collation and analysis due to incomplete corresponding questionnaires or excessively high lie-detection scores in the questionnaires indicating unreliable responses.The raw data folder "Raw E-Prime Data" is used to store the original outputs of E-Prime, with three types of files of the same name generated from each task run stored separately by experiment and condition: (1) The ".edat2" file is the original database file of E-Prime, containing complete trial-level event flows, variable assignments, key-press responses, correctness, feedback presentation, and time information, serving as the most original and comprehensive record; (2) The ".txt" file of the same name is a plain text log exported from E-Prime, facilitating batch parsing and indicator calculation using Python/R; (3) The "-ExperimentAdvisorReport.html" file of the same name is an automatically generated operational quality report by Experiment Advisor, which can be opened in a browser and used to quickly check the overall task operation, potential anomalies, and basic distribution characteristics. The above raw files correspond one-to-one with "Processed Data.xlsx".The processed spreadsheet file "Processed Data.xlsx" is used to summarize participant-level indicators and (partial) intermediate calculation results that can directly enter statistical/modeling analyses. This file contains 6 worksheets: First, "Experiment 1_Basic Information, Questionnaire Scores, and Calculated Indicators Included in the Paper" with 142 records (each row = 1 participant). The main column labels are: participant ID, age (unit: years), whether in the first menstrual cycle (1=yes, 2=no), total RCSAS score (scale score, unit: points), group (group codes for the four emotion-feedback combinations: 1–4), and values of four core behavioral indicators across three blocks (block1–block3): discriminability log d, response bias log b, composite score CS, and composite score bias CSbias (these indicators are the main dependent variables of the paper, reflecting reward learning/bias and the comprehensive characteristics of speed-accuracy; specific calculations correspond to formulas 1–4 in the paper). Second to fifth are the supplementary tables of Experiment 1 ("Smiling High Reward", "Smiling Low Reward", "Sad High Reward", "Sad Low Reward"), corresponding to the participant subsamples under the four conditions of Experiment 1 (with 34, 36, 36, and 36 records respectively). They are used to store performance statistics split by block and intermediate variables required for indicator calculation (e.g., various response counts, reaction times and their logarithmic transformations, log d/log b/CS/CSbias etc. Sixth, "Experiment 2_Basic Information, Questionnaire Scores, and Calculated Indicators Included in the Paper" contains 129 records (each row = 1 participant who passed quality control). In addition to demographic data and RCSAS, it includes four core parameters output by HDDM: drift rate v, boundary height a, starting point bias z, and non-decision time t. Among them, v, a, and t are organized by "reward level (high/low reward stimuli) × block (block1–block3)", and z is organized by block (t is usually in seconds, z is a proportional measure with a value range of 0–1, and the other parameters are continuous variables under the model scale). All statistical analyses were mainly completed in SPSS 22.0 and R 4.5.1; the HDDM modeling for Experiment 2 was performed in the dockerHDDM environment by launching Jupyter Notebook through an image to complete cleaning, sampling, diagnosis, and result export, and routine diagnosis of MCMC convergence was conducted using R-hat, ESS, etc.