What Predicts Beneficial Outcomes in Psychedelic Use? A Quantitative Content Analysis of Psychedelic Health Outcomes

Psychedelics encompass serotonergic tryptamines and phenethylamines and, more recently in research contexts, non-serotonergic substances such as ketamine and MDMA. Earlier work has reported potential therapeutic effects across clinical and non-clinical settings, including reductions in substance use problems, improvements in mental health and well-being, and lifestyle changes. Despite growing interest, mechanisms that might underlie enduring benefits remain contested. Two prominent explanatory frameworks are metaphysical belief theory, which emphasises adoption of comforting supernatural or non‑naturalist beliefs after mystical experiences, and predictive self‑binding theory, which emphasises the generation of new psychological insights following perturbations of self‑representations (for example via ego dissolution). Both accounts invoke overlapping subjective states, making empirical discrimination difficult with many designs used to date. Acevedo and colleagues set out to adjudicate between these competing models by analysing unsolicited, naturalistic psychedelic experience reports. The study aimed to test whether altered metaphysical beliefs or psychological insight better predict enduring beneficial outcomes, and whether psychological insight mediates relations between predictors (ego dissolution and therapeutic intent) and outcomes. The analysis used a quantitative content‑coding approach applied to reports drawn from the Erowid.org experience vaults, focusing on non‑clinical, self‑reported effects and longer‑term health and behavioural outcomes.

The researchers conducted a quantitative content analysis of user‑submitted psychedelic experience reports on Erowid.org. Sampling was restricted to the site’s “psychedelics” category (tryptamines and phenethylamines) and limited to reports that described psychedelic use without co‑ingestion of other drugs. From sub‑categories representing positive effects (Health Benefits, Therapeutic Outcomes, Mystical Experiences) and negative effects (Health Problems, Bad Trips, Nightmares and Trainwrecks) they randomly sampled reports to yield an equal number across sub‑categories. A pilot of five reports per sub‑category trained coders and was excluded; the final analytic sample comprised 240 reports (initially 270 were collected, 30 used for pilot training). A power analysis indicated that 215 reports would detect a moderate effect with goodness‑of‑fit tests, so 240 was considered sufficient. Three independent coders applied an a priori codebook derived from established psychometric instruments to create researcher‑generated measures. Key coded variables were therapeutic intent (binary: present/absent), ego dissolution (Likert 1–5 informed by the Ego Dissolution Inventory), altered metaphysical belief (Likert 1–5 informed by the Non‑Physicalist factor of a Metaphysical Belief Questionnaire, counted only when beliefs were explicitly attributed to the psychedelic experience), psychological insight (Likert 1–5 informed by the Psychological Insight Questionnaire), and enduring beneficial and detrimental outcomes (Likert 1–5 based on changes in substance use, lifestyle, and mental health domains). Word count was recorded as a control. Interrater reliability was reported as good to excellent for quantitative variables (intraclass correlation coefficients) and excellent for the categorical therapeutic intent variable (Fleiss’ kappa); coders resolved the few discrepancies by majority rule. For analysis, the investigators built a path model in IBM SPSS Amos reflecting the competing theoretical pathways: predictors were therapeutic intent and ego dissolution; mediators were psychological insight and altered metaphysical belief; outcomes were beneficial and detrimental outcomes. Bentler‑Weeks procedures were used to identify the model. Two variables showed non‑normal distributions (altered metaphysical belief and detrimental outcomes), so bootstrapping with 2000 samples was performed. Multicollinearity diagnostics indicated no issues. The model allowed correlations between the mediator error terms due to conceptual overlap. The authors also tested model fit after trimming nonsignificant paths and examined whether inclusion of word count altered results.

Descriptive statistics showed that 50 reports indicated therapeutic intent (21%), while 190 did not (79%). Report lengths averaged 1,049 words (range 119–4,705). Mean scores (and ranges) were: psychological insight 1.66 (SD = 0.84; range 1–4.67), ego dissolution 1.66 (SD = 0.89; range 1–5), altered metaphysical belief 1.18 (SD = 0.46; range 0.67–4.33), beneficial outcomes 1.74 (SD = 0.92; range 1–5), and detrimental outcomes 1.31 (SD = 0.67; range 1–4.67). Path analysis results supported the predictive self‑binding account. Therapeutic intent positively predicted psychological insight (standardised β = 0.36, p < 0.001), and ego dissolution also positively predicted psychological insight (β = 0.15, p = 0.002). Ego dissolution modestly predicted altered metaphysical belief (β = 0.13, p = 0.050). Psychological insight was the only mediator that uniquely predicted outcomes: it positively predicted beneficial outcomes (β = 0.66, p < 0.001) and negatively predicted detrimental outcomes (β = –0.23, p < 0.001). Altered metaphysical belief did not significantly predict either beneficial or detrimental outcomes. None of the predictors (therapeutic intent, ego dissolution) had significant direct effects on outcomes once mediators were included, indicating full mediation by psychological insight. Bootstrapped indirect‑effect tests corroborated mediation: therapeutic intent had a significant indirect effect on beneficial outcomes via psychological insight (β = 0.49, BCa 95% CI [0.20, 0.79]) and on detrimental outcomes (β = –0.18, BCa 95% CI [–0.45, –0.12]). Ego dissolution had a significant indirect effect on beneficial outcomes through psychological insight (β = 0.16, BCa 95% CI [0.06, 0.28]). After trimming nonsignificant paths the final model showed excellent fit: χ2 = 10.14, df = 10, p > 0.05; χ2/df = 1.01; CFI = 1.00; TLI = 1.00; RMSEA = 0.01 (90% CI 0.00–0.10). Including word count in the model did not change results.

Acevedo and colleagues interpret the findings as clear support for predictive self‑binding theory over metaphysical belief theory in this sample of unsolicited, non‑clinical psychedelic reports. Psychological insight emerged as the central predictor of enduring beneficial outcomes and as the sole negative predictor of detrimental outcomes; it fully mediated the relations between therapeutic intent and ego dissolution and subsequent beneficial outcomes. Altered metaphysical belief, by contrast, did not predict beneficial or detrimental outcomes, suggesting that adoption of supernatural beliefs is unlikely to be the primary mechanism behind reported long‑term improvements among recreational users in this dataset. The authors situate these results alongside prior studies that have linked psychological insight to favourable outcomes and note that mystical features such as ego dissolution can still play a role insofar as they facilitate insight. They caution that insights reported by users need not be factually accurate and that the same loosening of prior beliefs that enables therapeutic insights may also permit false insights. The discussion links the centrality of altered self‑representations to psychotherapeutic approaches and highlights congruence with therapies that foster psychological flexibility and value‑based insight, such as Acceptance and Commitment Therapy; the authors also suggest that preparatory work to set therapeutic intent might increase the likelihood of insight during psychedelic experiences. Key limitations acknowledged include the correlational, non‑causal design; the anecdotal, retrospective and self‑selected nature of Erowid reports, which introduces risks of recall bias, dishonest reporting and selection bias; and omission of potentially relevant variables such as psychedelic type and dose, setting features, and prior experience. The investigators further note that their measures were researcher‑generated from existing scales and should not be treated as equivalent to validated psychometric instruments. Nonetheless, they argue that analysing unsolicited reports mitigates some concerns about demand characteristics and complements experimental and survey research. The authors conclude that results are tentative but suggestive: in recreational contexts, therapeutic intent, ego dissolution and—critically—psychological insight are key predictors associated with enduring beneficial outcomes, and psychological insight mediates the effects of the other two variables. They recommend replication and further research using other methods before translating findings into clinical practice or policy.