Analysis of recreational psychedelic substance use experiences classified by substance

Outcome variable assumption checks revealed problems with the normality assumption on negative emotion, anxiety, and sadness words and problems with homoscedasticity on positive emotion, negative emotion, anxiety, sadness words, OAV scale semantic similarity, and Hood’s M scale semantic similarity scores. As a consequence, we used ggstatsplot’s robust option for positive emotion words, OAV scale similarity index, and Hood’s M scale similarity index and the nonparametric option for negative emotion, anxiety, and sadness words and interpreted the results appropriately. Due to normality and homoscedasticity violations on age and weight, respectively, we also used appropriate alternative tests for these variables.

Figure S3 displays the trimmed-means ANOVA results and significant pairwise comparisons for the affective process words outcome. There was a significant report group effect, F_{trimmed-means} (5503.16) = 75.66, p < 0.001 and ξ² = 0.45, CI 95% (0.40, 0.51). Bonferroni-corrected post hoc tests showed that MDMA reports were associated with the highest proportion of affective process words (M = 5.26), significantly higher than in psilocybin mushroom (M = 4.28, p < 0.001), LSD (M = 4.26, p < 0.001), Ayahuasca and DMT (M = 3.71, p < 0.001), and ketamine reports (M = 3.71, p < 0.001). Moreover, antidepressant reports featured a significantly higher proportion of affective process words than Ayahuasca and DMT (p < 0.001), ketamine (p < 0.001), LSD (p < 0.001), and psilocybin mushroom reports (p < 0.001). Finally, Ayahuasca and DMT and ketamine reports featured a significantly lower proportion of affective process words than LSD and psilocybin mushroom reports, respectively (all p < 0.001).

Figure S4 displays the Kruskal–Wallis test results and significant pairwise comparisons for the sadness words outcome. There was a significant report group effect, χ² (5) = 79.44, p < 0.001 and ε²_ordinal = 0.03, CI 95% (0.02, 0.04). Post hoc comparisons showed that Ayahuasca and DMT reports (Median = 0.23) featured a lower median proportion of sadness words than LSD (Median = 0.26, p = 0.029), psilocybin mushroom (Median = 0.28, p = 0.001), MDMA (Median = 0.30, p = 0.001), and antidepressant reports (Median = 0.57, p < 0.001). Antidepressant reports featured a higher median proportion of sadness words than each of the other report groups (all p < 0.001).

Figure S5 displays the Kruskal–Wallis test results and significant pairwise comparisons for the anxiety words outcome. There was a significant report group effect, χ² (5) = 21.70, p = 0.001 and ε²_ordinal = 0.01, CI 95% (0.00, 0.02). Post hoc comparisons showed that ketamine reports (Median = 0.27) featured a lower median proportion of anxiety words than psilocybin mushroom (Median = 0.42, p = 0.002), LSD (Median = 0.41, p = 0.010), and antidepressant reports (Median = 0.46, p = 0.006).

The present study analyzed anonymous online experience reports linked to the use of various psychedelic drugs. The exploratory analyses revealed clear differences between report groups, which amounted to very small (anxiety words), small (sadness words), medium (affective process, negative emotion, positive emotion, and cognitive process words; analytical thinking index), and large (OAV scale and Hood’s M scale semantic similarity) effect sizes (Field 2013; Wilcox & Tian 2011). These effect sizes should be kept in mind as the different language profiles of the different report groups will be discussed step-by-step below.

Ayahuasca and DMT reports contained relatively little affective language, notably on negative emotion and sadness words. They also featured relatively few cognitive process words and high scores on the analytical thinking variable. Given that some aspects of the Ayahuasca experience (nausea, vomiting, diarrhea, bad taste of the beverage; Hamill et al. 2019) would be expected to feel quite unpleasant, it is somewhat surprising that this was not reflected in the reports via a greater proportion of negative emotional language. This counterintuitive finding could be explained with Ayahuasca reports making up less than a fifth of the reports in the Ayahuasca and DMT group. Indeed, when we analyzed the proportion of negative emotion words for Ayahuasca and DMT reports separately, we found that DMT reports featured significantly a lower median proportion of negative emotion words than the other report groups, whereas Ayahuasca reports featured a higher median proportion of negative emotion words (albeit not significantly different from the other psychedelic report groups; see Fig. S6). Hence, despite being a key component of the psychedelic effects of Ayahuasca, DMT may not be responsible for Ayahuasca’s negative side effects. Ayahuasca and DMT reports also featured the greatest semantic similarity with the referential psychedelic and mystical experience scales. This might indicate that on average, Ayahuasca and DMT reports described the most psychedelic and mystical experiences. The latter would be consistent with the findings of Griffiths et al. (2019), who found that complete mystical experiences were most frequent, and mystical experience questionnaire scores the highest in a group of users who reported having had a God encounter after consuming DMT, relative to having consumed psilocybin, LSD, or no drug (Fig. S7 shows that the finding remains consistent even when examining DMT reports separately).

Ketamine reports were associated with relatively little affective language, particularly on positive emotion and anxiety words. As controlled research has observed facial affective flattening on ketamine (Pomarol-Clotet et al. 2006), one could speculate whether the language and facial expressions associated with ketamine’s effects both reflect an underlying emotional blunting. Although one review lists emotional blunting as a potential side effect of ketamine (Caddy et al. 2015), more research is needed to test this idea. Ketamine reports also featured relatively many cognitive process words. This might be because report writers noticed changes in their cognitive function, which they then discussed in their reports. The direction of these changes could be positive or negative. For example, affective flattening could lead to less interference of negative emotion with cognitive processes (Scheidegger et al. 2016), thereby increasing cognitive processing and reports thereof. Contrarily, it might also be that ketamine produced cognitive impairments (Morgan et al. 2004), and this in turn led to the increased cognitive process words observed in the current study. As could generally be the case for this study’s results, an alternative explanation for the cognitive process words finding could be that the difference was not due to the drug’s effects themselves but rather due to some unknown third factor like selection bias. For example, individuals who choose to use ketamine might habitually use less cognitive process words than individuals who choose to use other drugs. Given that ketamine dosages in these reports were likely greater than the low-dose ketamine and esketamine therapeutic use in current clinical practice (e.g., Bartoli et al. 2017), these results may not apply to clinical settings.

LSD reports featured the highest proportion of negative emotion words among the psychedelic drug report groups, although only the difference with Ayahuasca and DMT reports reached statistical significance, and the proportion was nearly half that of antidepressant reports. The same pattern was observed on sadness words and on anxiety words; LSD reports were significantly different from ketamine only. LSD reports were on neither of the extreme ends of the cognitive processes and analytical thinking variables. LSD reports were associated with fewer cognitive process words than ketamine MDMA, and antidepressant reports, lower analytical thinking scores than Ayahuasca and DMT reports, and higher analytical thinking scores than MDMA reports. The finding of a high negative emotion word proportion in LSD reports might have been due to writers’ direct experience of negative emotions, or an analytical focus on their own, or others’ negative emotions in other situations. The latter explanation would be consistent with experimental findings of enhanced emotional empathy under the influence of LSD (Dolder et al. 2016). Interestingly, Dolder and colleagues also found that LSD reduced the recognition of negative emotion, which would support the idea that the observed effects in this study may have been due to a conceptual-analytical focus on, rather than an increased experience of, negative emotion.

MDMA reports were characterized by relatively affective language. This was consistent with other natural language processing-based research (Bedi et al. 2014), and in part due to a high proportion of positive emotion words, consistent with findings from previous LIWC-based experimental research (Bedi et al. 2014; Wardle & de Wit 2014). MDMA reports also featured a relatively high proportion of negative emotion words, which is inconsistent with the results reported by Wardle and de Wit (2014), who found no differences between MDMA and placebo on negative emotion word frequency. In other research, there were mixed results as MDMA was associated not only with increased anxiety but also with decreased recognition of specific negative emotions (sadness and anger) in photos of human faces (Kirkpatrick et al. 2014). There also were a relatively high proportion of cognitive process words and the lowest score on the analytical thinking variable. The cognitive process words finding is interesting, because previous research has demonstrated cognitive decrements due to MDMA (Parrott & Lasky 1998). Thus, it could be that writers were aware of their cognitive difficulties, and writing about this in the reports increased the frequency of cognitive process words. The analytical thinking variable finding indicates that MDMA reports were associated with language rich in personal pronouns, impersonal pronouns, auxiliary verbs, conjunctions, adverbs, and negations and low in articles and prepositions (Pennebaker et al. 2014). This represents a more dynamic and personal, as opposed to analytical and impersonal narration style; potentially due to intensified emotional states on MDMA. Lastly, of all the psychedelic substance report groups, MDMA reports featured the least semantic similarity with Hood’s mystical experience scale (comparable to antidepressant reports), which seems consistent with previous research results (Lyvers & Meester 2012).

Psilocybin mushroom reports were not on the extreme ends of any outcome variable. Notably, they featured less affective language than MDMA (also on positive emotion words) and antidepressant reports (due to negative emotion and sadness words, as the reverse was true for positive emotion words) but more than Ayahuasca and DMT (also on negative emotion and sadness words) and ketamine reports (also on positive emotion and anxiety words). There also was a lower proportion of cognitive process words than in MDMA, ketamine, and antidepressant reports and a higher analytical thinking score than in MDMA and antidepressant reports. There was more semantic similarity to Hood’s mysticism scale relative to MDMA and antidepressant reports but less similarity relative to Ayahuasca and DMT and LSD reports, as well as less similarity to the OAV scale relative to Ayahuasca and DMT reports. One could compare these results to those of a survey study of individuals who had a God encounter (Griffiths et al. 2019), which found that psilocybin-induced God encounters reportedly featured less strong and less frequent complete mystical experiences than DMT-induced encounters (whereas there was no difference to LSD and Ayahuasca experiences; Griffiths et al. 2019).

Antidepressant reports, which were included as a non-psychedelic substance comparison regarding Erowid users’ language, featured a distinct profile. They featured a relatively high proportion of affective language (only outscored by MDMA reports), but in contrast to MDMA reports, this was due to a significantly higher proportion of negative emotion and sadness words relative to the other report groups, the proportion of positive emotion words being lower than that in the other reports. This finding might be naturally explained by antidepressant users’ higher prevalence of depression and associated negative emotional biases. Antidepressant reports also featured the highest proportion of cognitive process words, which one could attribute to cognitive characteristics of depression like increased rumination. On analytical thinking, antidepressant reports featured lower scores than Ayahuasca and DMT and psilocybin mushroom reports, but did not differ from the remaining psychedelic drugs (and scored higher than MDMA reports), indicating a notable difference between psychedelic substances on this dimension. Antidepressant reports also featured less semantic similarity with the psychedelic and mystical experience scales than the psychedelic drug report groups.

Though no prior study has compared the present combination of psychedelic substances on psycholinguistic variables like this study did, previous research used similar outcomes or analytic methods. Studies that examined the effects of psychedelic drugs on the psychological states or processes underlying the psycholinguistic outcomes of this study reported emotional effects of psychedelic drugs, such as an emotionally intensifying profile on MDMA (Agurto et al. 2020; Bedi et al. 2014), decreased negative emotions like sadness on ketamine (Hasler et al. 2020), increased 5D-ASC blissful state and increased emotional excitability with a bias toward positive emotion on LSD (Carhart-Harris et al. 2016a; Liechti et al. 2017; Schmid et al. 2015), and emotional excitability on psilocybin (Hasler et al. 2004). Furthermore, some research also observed cognitive disorganization on LSD (Carhart-Harris et al. 2016b). However, few studies compared psychedelic drugs against each other. Holze et al. (2020) and Liechti et al. (2017) compared LSD against MDMA on the Mystical Experience Questionnaire and found a seemingly contradictory result to the present study, where LSD was associated with higher “deeply felt positive mood” and “positive mood” subscale scores than MDMA. Furthermore, LSD was associated with higher 5D-ASC blissful state scores (Holze et al. 2020). Zamberlan et al. (2018) compared many different psychedelic drugs with latent semantic analysis of Erowid experience reports but only correlated the semantic similarity (to other report groups) with the molecular structural similarity between drugs to infer something about the effects that molecular differences might have on psycholinguistic outcomes. These results emphasize that ideally, experimental research should compare different psychedelic drugs directly on outcomes derived from both natural language processing methods and validated questionnaires to (a) determine differences between psychedelic drugs and (b) the relationship between linguistic indicators of psychological variables and self-report measures of those same variables.

This study was limited in several ways. First, the design and analyses of the study did not permit causal inference. The fact that there was no control over the administration, dosage, and quality of psychedelic substances, nor over the time at which the experience was recalled and reported, shows that these results should only be interpreted as indirect indicators of potential substance group effects. For example, it is not clear whether the use of the examined substances produced the differences in language that the analyses detected or whether this was due to a third variable predisposing individuals toward recreational use of a specific substance that also influenced the linguistic outcomes. We did, however, check demographic variables (age, weight, and gender) and found significant differences between substances. For example, Ayahuasca and DMT, psilocybin, and antidepressant reports were associated with older age than LSD and MDMA reports. Ayahuasca and DMT reports were associated with heavier report authors than LSD and MDMA reports, the latter also involving significantly lighter authors than psilocybin and antidepressant reports. Females were underrepresented in Ayahuasca and DMT, ketamine, and psilocybin reports and overrepresented in MDMA reports. Some of these effects might be interrelated, and some of them might also be related to the findings on the linguistic outcomes. For example, the lower weight of MDMA report writers might be due to the relative overrepresentation of women in this group, which might in turn explain part of the more emotional language in this group (Newman et al. 2008). Rigid experimental research balancing the proportions of men and women per substance could provide insights into these relationships, although such research should be very difficult to conduct with the combination of substances examined here. Future research could also monitor other potential sources of bias, such as cultural background or socioeconomic status. Another limitation concerns the antidepressant report group, which is not an ideal control condition, as would be a placebo group in a hypothetical controlled experimental version of this study. Given the constraints of the study design, though, the antidepressant reports did provide useful information about the language of Erowid authors who did not use psychedelic substances.

Given the absence of experimental data regarding the psycholinguistic effects of different psychedelic drugs, this study provided valuable insights. In light of its findings and limitations, future research could build on this study by conducting a methodologically robust experimental trial examining the substances reviewed here, and a potential control or placebo substance, using free speech transcripts or written texts produced during the experience. Another limitation concerns the dearth of demographic information about the sample, as more data could help explain who uses psychedelic drugs and with what motivation. The available descriptive information (i.e., age, gender, weight) provided some insight (e.g., women, lighter, and younger people being relatively overrepresented in MDMA reports; Ayahuasca and DMT report writers being relatively old; Ayahuasca and DMT, ketamine, and psilocybin report groups being relatively male), but more variables might be needed to show that differences between substances persist when controlling for individual differences. Future research could thus examine demographic variables not available here (e.g., educational/social status, cultural background, substance preference).

This study examined differences in linguistic markers of psychological function between different psychedelic drug experience reports. These differences indicated certain psychological effect profiles, although more controlled studies are necessary to more closely examine these profiles and replicate the current results. The study indicates that some drugs may have an emotionally intensifying profile accompanied by many cognitive process words and dynamic, personal language (MDMA), whereas others are characterized by reductions of emotional language, relatively few cognitive process words, and increases in analytical thinking-associated language (Ayahuasca and DMT), and again others show a mixed effect profile (ketamine, LSD, psilocybin mushrooms). With this article, we hope to contribute to the development of psychedelic-assisted psychotherapy or psychiatric treatments for psychiatric problems, for example, by optimizing the selection of appropriate substances based on patient needs.