Mystical and Affective Features of Naturalistic Psychedelic Experiences: Dataset and Supplementary Materials

Project Overview This repository contains the data, supplementary tables, and prompts for a study investigating the subjective effects of psychedelics using web-scraped trip reports (N = 7,303) from Shroomery.org. The project replicates established dictionary-based text analysis methods and extends them using large language models to extract multidimensional psychometric indicators of mystical experiences and sentiment.Preprint: Will be avaliable on Research Square (pending, in preparation). --- Abstract: Background: First-person narrative accounts of altered states of consciousness shared online offer a naturalistic complement to clinical research on psychedelic substances. However, traditional text-analysis approaches based on static word dictionaries may have limited ability to capture the contextual and phenomenological meaning of psychedelic experiences. The present study replicated a prior dictionary-based analysis of mystical and affective language in a large corpus of naturalistic psychedelic experience reports and extended it using large language models (LLMs) to rate specific dimensions of mystical-type experience. Methods: A total of 7303 English-language psychedelic experience reports were collected from an online forum focused around psilocybin-containing mushrooms. Reports were scored using a custom word-count dictionary targeting mystical-type language and a rule-based sentiment tool, replicating a prior published methodology. Subsequently, a large language model rated each report on four dimensions corresponding to established components of mystical experience (unity, noetic insight, sacredness, time transcendence) and overall sentiment, guided by a structured natural-language prompt. Associations with self-reported experience intensity were assessed using chi-square tests, Kruskal-Wallis tests with post-hoc pairwise comparisons, and Spearman correlations, with Bonferroni correction for multiple comparisons. Reliability of the language-model procedure was assessed across eight different models on a subsample of texts from the corpus using Krippendorff’s alpha. Results: The dictionary-based replication reproduced the previously reported U-shaped distribution of mystical language across self-reported experience intensity, alongside a decline in positive sentiment at higher intensity levels. In contrast, LLM-derived ratings of unity, noetic insight, time transcendence, and sacredness increased monotonically with intensity. LLM-derived sentiment converged with two independent sentiment measures, while ratings across eight LLMs showed substantial inter-model agreement. Conclusions: The divergence between lexical and LLM-based measures suggests that keyword frequency and contextual ratings capture partially distinct aspects of mystical experience. Whereas dictionary-based analysis appears useful for detecting the presence of mystical vocabulary, its normalized word-frequency score may be less suited to estimating the intensity of the construct of interest. LLM-based contextual scoring therefore represents a potentially useful exploratory complement to established text-analysis methods, although validation against human coders remains necessary.