Curated activation-event data derived from the emotion-coding subset of the Stanford Suppes Brain Lab Psychotherapy EEG Dataset

This dataset contains curated CSV tables derived from the public Emotion_codings.zip archive in the Stanford Suppes Brain Lab Psychotherapy EEG Dataset. The source archive contains 51 Matlab emotion-coding files representing 23 couple-therapy sessions. Each session was coded independently by at least two human coders, with five sessions coded by three coders. Each source row identifies the timestamp of a salient emotional event and contains, for the female and male partners, intensity values for five emotion categories: joy/happiness, sadness, fear/anxiety/tension, anger, and other. A value of zero indicates that a category was not coded; when all five categories for a partner are zero, that partner is treated in the curated data as inactive/neutral at that coded event. The purpose of the curation is not to test a single research hypothesis, but to make this unusual public set of naturalistic, independently coded psychotherapy interactions accessible for secondary analysis and to support investigation of emotional activation, dyadic interaction, and annotation reliability. The package preserves the original emotion categories and intensity values while adding derived activation/inactivation variables, dyadic activation states, partner- and session-level summaries, threshold-based activation-event runs, and one-to-one coder-event matches within 1500 ms. The data show that coder agreement can be examined at several distinct levels rather than treated as a single reliability question: whether coders selected approximately the same moments as emotionally salient, whether they agreed that a partner was emotionally active at matched moments, and whether they assigned the same specific emotion when both identified activation. A high-confidence subset identifies events at which at least two coders selected nearby timestamps, identified the same partner as active, and assigned the same source emotion category. The threshold-based files group nearby active coded events using 750, 1000, 1500, and 2000 ms contiguity rules. These should be interpreted as sensitivity analyses of event clustering, not as estimates of the duration of emotional expression, because the public archive does not include the underlying transcripts, word offsets, audio, or video. The package is intended for reuse in psychotherapy-process research, affective computing, dyadic interaction analysis, behavioral annotation reliability, and computational analysis of expressed emotion. It does not redistribute the original Matlab files, EEG recordings, audio, video, or transcripts.