This cross-sectional study was conducted using the Testable Minds (www.testable.org) platform, a subject pool for psychology experiments with a participant verification system (50). Testing and recruitment platforms such as that used in the present study can provide researchers access to hitherto less accessible study populations such as non-treatment seeking cannabis users (51). Testable Minds employs verification checksat sign up and for each study participation to minimise the issues (e.g., multiple accounts, “bot” responses) encountered by other such participant pool platforms (52). Participation in this study was restricted to “verified minds”; to qualify as such, participants are required to submit an official photo ID and take a live webcam photo of their face. The photo ID is used to manually verify the first and last names, date of birth, sex, and country of residence. It was reported that, as of June 2020, ~15,000 members are registered in the Testable Minds participant pool, mean age 34 years (SD = 11.5) and 51.7% male. In terms of location, 42% are from USA, 19% from UK, 18% from EU countries, 2% from Canada and 2% from Australia and New Zealand (50).

Testable Minds members were invited to participate in a study with both questionnaire and PC-based task elements. They were informed that the study was designed to examine the relationship between personality and lifestyle factors and attentional function. The inclusion criteria for the study were (i) individuals aged between 18 and 55 years old and (ii) from a predominantly English-speaking location. Participants were excluded if they (i) had a history of neurological disease or brain injury and/or (ii) had been formally diagnosed with a psychiatric illness. Ethical approval was obtained from the Social Research Ethics Committee of University College Cork. All participants provided informed consent and were compensated £5 for their time.

Participants were first asked to complete a brief demographics survey, which included questions about age, sex, nationality, and education level. They were then directed to complete the following: Cannabis Experience Questionnaire modified version [CEQmv; (16)]; Schizotypal Personality Questionnaire [SPQ; (53)]; ASI (54).

The CEQmv was administered to collect information on history of cannabis use (including information such as age at first use, duration of use, frequency of use, type), as well as other recreational drugs including tobacco and alcohol (16). In the present analysis, measures of interest were: lifetime cannabis consumption (“ever vs. never”); current cannabis consumption (defined as frequent use of cannabis consumption during the previous 12 months; “current” vs. “other” comparison); frequency of use which referred to the most recent period of cannabis consumption (a six-level categorical response item: never, only once or twice, a few times each year, a few times each month, more than once a week, every day) (15).

Aberrant salience status was measured using the 29-item ASI instrument (54). This self-report measure aims to assess subjective experiences of aberrant salience attribution (1) and yields a score of 0–29; a higher score corresponds to a greater degree of aberrant salience, derived from yes/no responses to 29 statements. The ASI provides a total score calculated as number of positive responses and has the following five subscales: Increased significance–reflecting heightened attribution of salience to stimuli (e.g., item 10: “Do you ever feel the need to make sense of seemingly random situations or occurrences?”); Senses sharpening–this is also consistent with aberrant salience (e.g., item 17: “Has your sense of taste ever seemed more acute?”); Impending understanding–referring to an individual's feeling of perceived significance associated with a psychotic episode (e.g., item 6: “Do you sometimes feel like it is important for you to figure something out, but you're not sure what it is?”); Heightened emotionality–signifying greater anxiety during psychosis, particularly in the early phases of a psychotic episode when the individual attempts to understand the increased importance of stimuli (e.g., item 8: “Do you ever have difficulty telling if you are thrilled, frightened, pained, or anxious?”); Heightened cognition–reflective of the experience of being part of something not apparent to others (e.g., item 25: “Do you ever perceive an overwhelming significance to things that are usually not significant to you?”).

The SPQ is a 72-item index of schizotypy, based on DSM-IIIR criteria for schizotypal personality disorder, that can be measured in the general population. The SPQ comprises nine subscales that form three dimensions: (a) Cognitive-perceptual (including the four subscales “ideas of reference”, “magical thinking”, “unusual perceptual experiences” and “paranoid ideation”). An example of this subscale is the item “Have you often mistaken objects or shadows for people, or noises for voices?”; (b) Interpersonal (related to negative symptoms, including the four subscales “social anxiety”, “no close friends”, “constricted affect” and “paranoid ideation”). An example is the item “I have little interest in getting to know other people”; and (c) Disorganised (including the two subscales “eccentric behaviour” and “odd speech”). An example is item “I sometimes forget what I am trying to say” (55). It has been suggested that conceptual and structural differences exist between the various schizotypy scales (e.g., O-LIFE, SPQ) with respect to the measurement of the disorganisation dimension (56). The SPQ uses a dichotomous response format. A total SPQ score is based on the sum of all nine subscale scores. This scale has high reliability and validity (53).

The present blocking procedure was based on Oades' KB task as described by Moran et al. (36, 37, 40), and minimally adapted for delivery via the Testable interface (https://www.testable.org). Participants are instructed that there is a hungry mouse trying to find his cheese in a “house”. On a typical trial, one of six sets of tri- or bi- coloured rectangular bars appears on the screen for 1 s. There are six sets of colours [i.e., set 1 = grey (colour 1), green (colour 2); set 2 = pink (colour 1), brown (colour 2); set 3 = yellow (colour 1), turquoise (colour 2), orange (colour 3); set 4 = red (colour 1), green (colour 2), blue (colour 3); set 5 = purple (colour 1), blue (colour 2), red (colour 3) and set 6 = yellow (colour 1), green (colour 2), pink (colour 3)]. Each set of colours corresponds to a particular location in the “house” which are numbered 1–8. Participants are instructed that (1) there are two main rooms in the house, each with four possible hiding places (1–4 and 5–8) and (2) the mouse will always be in the opposite room for the one where the cheese is hidden. Participants identify the location of the cheese by pressing the corresponding number keys 1–8 on a QWERTY keyboard. If the answer is correct, a piece of cheese appears on the location chosen by the participant and “Correct” appears in the middle of the house plan. If an incorrect choice is made, “Incorrect” is displayed in the middle of the screen. If there is a failure to respond within the response interval period then “No response detected” is displayed.

At the beginning of each fixed length 6-s trial, mouse and colours were superimposed on the house template (the template remained on the screen throughout the experiment, except for breaks between sessions). Moreover, a timer is displayed at the top right of the screen to let participants know how much time they have to respond and to engage them to answer as quickly as possible. At a time of 1 s into the trial, the colours disappear. Subjects were instructed to press one out of eight keyboard keys (four on each hand) as quickly and accurately as possible in a 5-s window to answer. Participants needed to position their left hand in order to have the little finger correspond to number 1 running through to number 4 for the index finger and position their right hand in order to have the index finger correspond to number 5 running through to number 8 for the little finger. Once a key press was detected, feedback information was superimposed on the house template for 2-s and then the next trial followed immediately.

The KB task is run in two conceptual phases and they both take approximately 10 min to complete: (1) Overshadowing (OS) with a test phase and (2) Blocking (BL) with a test phase. In OS there were 2 training phases: (1) 44 trials of bi-coloured rectangular bars were presented (one half of each set 1 and 2) and then (2) 36 trials of tri-coloured bars were shown (one half of each set 3 and 4). Test trials involved 12 trials of colour 1 and 12 trials of colour 3 that are a segment of the tri-coloured sets. Test trials probed how much learning had accrued to individual elements of the tri-colour blocks. In BL there were also two training phases including 44 trials with a bi-colour bar (one half of each set 5 and 6) followed by 36 trial where an additional colour (colour 3) was added to create a tri-colour bar. These sessions were followed by 24 test trials probing how much learning had accrued to individual elements (colours 1 and 3) of the tricolour blocks. As training with the two colour bar fully predicts the spatial location, blocking should be demonstrated as slower or no learning about the third added colour.

A KB score was calculated from the mean difference in latency to respond to the first and third colour bars in the two conditions OS and BL: reaction times (RT) are calculated in milliseconds (ms). Calculation of KB as a function of an overshadowing condition removes any potential confounding influence of overshadowing or stimulus presentation on reaction times. This blocking score was calculated for each of the pairs of test trials (“Trial 1”–“Trial 12”) and the latency difference was averaged across pairs of trials (“KB score”).

Normality of data was examined using Shapiro-Wilk tests. As total and individual KB trial scores were normally distributed (after data cleaning), independent sample t-tests were used to examine the effects of current or lifetime cannabis use (dichotomous variables in both cases) on total KB score. A one-way analysis of variance (ANOVA) subsequently examined the effect of frequency of cannabis use on total KB score. For individual KB trial scores, a linear mixed-effects model was used with current cannabis use as a fixed effect. Mann–Whitney U-tests were used to assess differences in SPQ and ASI scores based on current cannabis use (as these scales were all non-normal and positively skewed, each p < 0.001). Linear regression analyses were used to examine whether SPQ and ASI (total and sub-scale scores) predicted total KB score. Finally, correlation analyses were performed between SPQ, ASI, and individual KB trial data using Kendall's rank correlation.

A parallel analysis strategy was applied to the data, in which both frequentist and Bayesian methods were used. This approach was chosen due to some critical null findings of the data, wherein Bayesian methods can discern whether a p > 0.05 is the result of a true null effect or insensitivity of the data to detect an effect. For interpreting Bayes Factors, they are the ratio evidence for the alternative hypothesis relative to the null hypothesis (BF₁₀). For example, a BF₁₀ of 5 means there is five times more evidence for the alternate hypothesis relative to the null. This can be converted into evidence for the null by dividing 1 by the BF₁₀ (now the BF₀₁). Common cut-off criteria for BFs are as follows: values between 3 and 0.333 can be considered to indicate lack of sensitivity to detect effects (requiring more data), a BF 3 > | <0.333 represents moderate evidence for the alternate and null, respectively, 10 > | <0.1 strong evidence, 30 > | <0.033 very strong evidence, and 100 > | <0.01 decisive evidence (57).

All data were collated and transferred into R Studio (R Studio Team, 2015), which was used to complete the linear mixed-effects models using the lmertest (58) and emmeans (59) packages. Graphs and figures were created in R Studio using ggplot2 (60) and plotly (61). All Bayesian Analyses were conducted in JASP (JASP Team, 2020; https://jasp-stats.org/).

Of the 310 participants that attempted the experiment, three were removed due to not completing the KB task. Data cleaning procedures were applied to the remaining 307 participants across total KB score and the 12 individual KB trials. Initially, 230 data points were removed resulting from participants responding slower than 3 s, which were assumed to be lapses in attention. Seven of the 13 KB variables were non-normally distributed according to Shapiro-Wilk tests [all p < 0.02 with False Discovery Rate (FDR) correction]. Outliers were removed according to the median +/– 2.5 (the Median Absolute Deviation), resulting in 53 data points being removed across the 13 KB variables (0.013 data points per variable). In total, 7.091% of the data (283 data points) for KB trials were removed. After outlier removal, all variables were normally distributed (all p > 0.152, FDR corrected). Linear mixed effects models were used for KB trial scores, which are appropriate for repeated measures data with incomplete data.

Table 1 presents the sociodemographic and cannabis use characteristics of the study sample. Further drug use patterns reported in the present sample are presented in Supplementary Table 1.

Male participants were more likely to have “ever” used cannabis (χ2 = 10.72, p = 0.03), but no sex differences were observed for current cannabis use (χ2 = 0.97, p > 0.05) or cannabis use frequency (χ2 = 16.43, p > 0.05). Family history of mental illness was associated with increased likelihood of current cannabis use (χ2 = 17.54, p < 0.001), but no association was observed for lifetime cannabis use (χ2 = 5.23, p > 0.05) or frequency of use (χ2 = 16.53, p > 0.05). Educational attainment level was not associated with any cannabis use measure (all p > 0.05).

Before investigating predictors of KB performance, we first assessed whether KB was exhibited in the sample overall. To do so, the average RT of the 12 individual KB trials were compared against a reference value of 0 ms (i.e., no difference between overshadowing and blocking trials). Multiple one sample t-tests (with FDR correction) suggested that Trial 2 (p = 0.004), Trial 5 (p = 0.004), Trial 7 (p = 0.018), and Trial 9 (p = 0.004) all significantly differed from 0. The remaining trials did not significantly differ from 0 ms (all p > 0.11).

Analyses were conducted to replicate our previous findings between cannabis use and both schizotypy and aberrant salience (15) and to inform upcoming analyses of potential moderating variables. The results of parallel analysis Mann–Whitney U tests found that those who currently use cannabis had significantly higher levels of disorganised schizotypy (interpolated median = 6.000) compared to those who do not currently use cannabis (interpolated median = 3.967) at a small effect size (p = 0.01, r_{rank−biserial} = 0.196 [0.048, 0.335]). However, the BF reported only anecdotal evidence for the null hypothesis and thus suggested that more data is needed (BF₁₀ = 0.595). The groups did not differ in their scores on the Cognitive-perceptual (Cannabis = 8.813, No Cannabis = 6.731, p = 0.13, r_{rank−biserial} = 0.117 [−0.033, 0.262], BF₁₀ = 0.234), Interpersonal (Cannabis = 13.125, No Cannabis = 12.900, p = 0.82, r_{rank−biserial} = −0.017 [−0.166, 0.133], BF₁₀ = 0.143), or total SPQ scales (Cannabis = 24.750, No Cannabis = 20.333, p = 0.20, r_{rank−biserial} = 0.099 [−0.051, 0.245], BF₁₀ = 0.242). Moreover, the Bayesian analyses suggested that there was moderate evidence for the null hypothesis for these latter three effects (all BF₁₀ < 0.3). Similarly, those who currently use cannabis demonstrated significantly higher scores for total ASI (11.000) relative to those who do not (8.375) at a small effect size (p = 0.004, r_{rank−biserial} = 0.221 [0.073, 0.359]), but again the BF was insensitive (BF₁₀ = 1.934). When analysing the five ASI subscales, the Increased Significance (Cannabis = 3.700, No Cannabis = 2.444, p = 0.01, r_{rank−biserial} = 0.202 [0.054, 0.342, BF₁₀ = 0.954), Senses Sharpening (Cannabis = 1.955, No Cannabis = 1.000, p < 0.001, r_{rank−biserial} = 0.263 [0.117, 0.397], BF₁₀ = 3.383), Impending Understanding (Cannabis = 1.900, No Cannabis = 1.316, p = 0.02, r_{rank−biserial} = 0.171 [0.021, 0.313], BF₁₀ = 0.606), and Heightened Emotionality (Cannabis = 3.033, No Cannabis = 2.311, p = 0.03, r_{rank−biserial} = 0.169 [0.019, 0.311], BF₁₀ = 0.575) sub-scales were all significantly higher in current cannabis users. However, most of the BF calculations suggested that more data were needed. The Heightened Cognition scale however was non-significant (Cannabis = 1.063, No Cannabis = 0.871, p = 0.25, r_{rank−biserial} = 0.085 [−0.066, 0.232], BF₁₀ = 0.219), with the BF suggesting moderate evidence to accept the null hypothesis.

Subjects that do not currently use cannabis (n = 231) had a lower total KB score (Mean [M]= −17.90 ms, Standard Deviation [SD] = 328.81) than those that do currently smoke cannabis (n = 74, M = −0.69, SD = 339.15 ms), but this difference did not attain statistical significance and produced a small effect size with large confidence intervals (mean difference = −17.21 ms, t (303) = −0.389, p = 0.70, d = −0.052 [−0.31, 0.21]; Figure 1). Homogeneity of variance was not violated (p = 0.81). The Bayesian adaptation suggested there was 6.39 –fold more (“moderate”) evidence for the null hypothesis than the alternate hypothesis (Cauchy prior scale = 0.71, BF₁₀ = 0.16). As a robustness check, a sequential analysis was performed at different prior settings to assess if these results were stable across participants and priors. This analysis suggested the evidence was insensitive (3 > BF <0.333) until approximately the 120 th participant, after which the evidence for the null increased steadily to moderate strength.

The analysis was repeated for those who stated that they had tried cannabis at least once (but may or may not currently use cannabis i.e., lifetime cannabis use). Those who had never used cannabis (n = 125) had a lower mean KB Score (M = −21.61 ms, SD = 315.83 ms) than those that had smoked cannabis (n = 179, M = −9.95 ms, SD = 342.02), but this difference did not attain statistical significance and produced a very small effect size with large confidence intervals (mean difference = −11.66, t(302) = −0.302, p = 0.76, d = −0.035[−0.26, 0.193]). Homogeneity of variance was again not violated (p = 0.28). The Bayesian adaptation suggested there was 7.467 more evidence for the null hypothesis (Cauchy prior scale = 0.707, BF₁₀ = 0.13). The sequential analysis found evidence for the null increased steadily and was consistent across prior setting.

The analysis was repeated with cannabis use frequency of the most recent period of cannabis consumption as the grouping variable. This led to the removal of participants reporting they do not currently use cannabis, leaving 74 complete cases. These data were plotted in Figure 2, which appeared to indicate a more consistent increase in KB score with increased cannabis use frequency. The following ANOVA met the assumptions of homogeneity of variance (p > 0.51) and normally distributed residuals. The results of the frequentist analysis returned a non-significant effect of cannabis use frequency at a medium effect size [F(4, 68) = 1.45, p = 0.23, ηpartial2 = 0.08]. The Bayesian ANOVA found the most probable model was the null model (BF₁₀ = 2.74) rather than the proposed model containing cannabis use frequency (BF₁₀ = 0.37), with the R² of the model being small (R² = 0.06 [0.007, 0.17]). However, as the BF₁₀ was insensitive, more data is needed before conclusions can be drawn.

A follow-up analysis was conducted between current cannabis use and individual KB trial data (trials 1 through 12) by applying a linear mixed-effects model to the data. As we also planned to assess interactions with interpersonal (i.e., negative) schizotypy, two participants needed to be removed who failed to respond to the questionnaire (n = 305). To assess whether cannabis use affected trial RT, increasingly complex nested models were compared. At each comparison, a significant χ² difference test suggests the more complex model was a better fit to the data. Akaike Information Criterion (AIC) values are also reported, with lower values indicating a more parsimonious model. Due to the aim of this analysis being model comparison, restricted maximum likelihood (REML) was not used. In all following analyses, inspection of qqplots suggested residuals were approximately normal. The first model contained no fixed effects (independent variables) and only one random effect of participant (random intercept model). This random intercept model was the null model for future comparisons (AIC: 55593), which also allows estimation of the proportion of variance in trial RTs explained by participant's different baseline RTs (i.e., their personal intercepts). The analysis found that the random effect of participant explained 0.337% of the total variance, suggesting little individual differences in baseline RT. Next, this model was compared to a second model also containing trial type as a fixed effect (“main effect”). This second model was a significantly better fit to the data than the null model [χ²(11) = 58.2, p < 0.001, AIC: 55557] and explained an additional 1.96% of variance (totalling 2.297%). This suggested the extent of KB differed between trials. A further model was proposed that investigated a potential interaction between cannabis use and trial type. This model was not significantly better than the previous model [χ²(12) = 13.52, p = 0.33, AIC:55568] and explained an additional 0.42% of the variance (total 2.717%). Using the ANOVA function from the base R stats package, the fixed effect of trial type remained significant in this model [F(11, 3093) = 4.78, p < 0.001], while both the fixed effect of cannabis use [F(1, 3356) = 0.003, p = 0.96] and the interaction of current cannabis use and trial type were non-significant [F(11, 3094) = 1.149, p = 0.32]. As it was predicted that interpersonal (negative) schizotypy would predict KB score, we assessed a final model adding interpersonal schizotypy and its interaction with both cannabis use and trial type (three-way interaction). This model was not a significantly better fit to the data relative to the previous model [χ²(24) = 21.54, p = 0.61, AIC: 55594, R² = 3.310]. Overall, the analysis suggests that RTs differ by trial type but not by cannabis use, interpersonal schizotypy, nor the interaction of these variables.

Regression analysis was used to examine whether the total and subscale totals of the ASI and subscales of the SPQ were predictors of total KB Score. Consistent with the previous analyses, BFs were also calculated for regression coefficients. Each of the following regression models passed the assumptions of no autocorrelation, homoscedasticity, normal distribution of residuals, lack of outliers/influential points, and no multi-collinearity. The first set of linear regressions entered total ASI (simple linear regression) or its five subscales (multiple regression) as predictors of KB Score. Total ASI score did not predict KB Score (β = 0.037[-0.08, 0.15], p = 0.53) and explained little variance (R² = 0.001), with the BF supporting the null (BF₁₀ = 0.15). As ASI scores were increased in those that currently use cannabis, we investigated a potential interaction effect between ASI and cannabis use in an additional regression. There was trend level evidence that this model was a better fit to the data [F(2, 298) = 2.533, p = 0.08], which likely came from the significant interaction effect between ASI and cannabis use (B = 12.589 [1.505, 23.674], p = 0.026). ASI score remained a non-significant predictor (B = −2.220 [-8.202, 3.761], p = 0.466) and current cannabis use returned trend (B = −131.524 [−287.637, 24.590], p = 0.098). To discern whether this effect was driven by a total ASI or an individual subscale, the analysis was repeated using the ASI subscales as predictors. As calculating the interaction between all ASI subscales and cannabis use status would be unnecessarily complex, a binary logistic regression with cannabis use status as the outcome variable and the ASI subscales as predictors was implemented. This was used to determine whether the association between cannabis use and ASI subscales was general (1 > scale) or specific (only one scale) and thus which interactions to calculate. The binary logistic regression suggested that only Senses Sharpening was a significant predictor of cannabis use status (OR = 1.413, p = 0.008), whereas the remaining three scales were not (all p > 0.279). Consequently, only the interaction effect between Senses Sharpening and cannabis use was included. The regression analysis (Table 2) found that the only ASI subscale that predicted KB score was Senses Sharpening (B = −46.923 [-85.9, −7.91], p = 0.019) and all other subscales returned non-significant (all p > 0.20). Current cannabis use did not significantly predict KB score (p = 0.20), however, there was a significant interaction effect between cannabis use and Senses Sharpening (B = −61.146 [5.388, 116.9], p = 0.032). To illustrate this interaction effect, post-estimation of marginal effects was conducted. As can be seen from Figure 3, in those who do not currently use cannabis Senses Sharpening predicted decreased KB score, whereas Senses Sharpening appeared unrelated to KB score in people who do currently use cannabis.

Next, the three SPQ dimension scores were added as predictors of KB score in a separate multiple linear regression (Table 3). The model was significantly greater than the null model of intercept alone (F(3, 299) = 2.861, p = 0.04), explaining 2.8% of the total variance (adjusted R² = 1.8%). This effect derived primarily from the disorganised scale, in which for every additional item endorsed on the 16 item scale, participant's KB reaction time increased by 17.645 ms (β = 0.237 [0.077, 0.397], p =0.004, BF₁₀ = 2.474). However, the BF suggests more data are needed to clarify the robustness of this effect. In terms of partial R² values, the Disorganised dimension also accounted for the vast majority of the model's total R² (Rpartial2 = 2.755%). Of the remaining dimensions, the Cognitive-perceptual (β = −0.076 [−0.244, 0.093], p = 0.38, BF₁₀ = 0.358) and Interpersonal dimensions (β = −0.096 [−0.259, 0.068], p = 0.25, BF₁₀ = 0.447) both returned non-significant and insensitive. As disorganised schizotypy was increased in current cannabis users, a further model was tested including an interaction effect between current cannabis use and disorganised schizotypy. However, this model was not a significantly better fit to the data (F(2, 297) = 0.416, p = 0.66), the interaction effect returned non-significant (B = 8.422 [−10.7, 27.6], p = 0.39), and the model explained minimal additional variance (R² = 3.062%). This suggested the model in Table 3 was the most parsimonious.

The final analysis added the SPQ scales, ASI scales, and KB trial scores into a Bayesian correlation matrix using two-tailed Kendall's Tau. Correlations were used to reduce the complexity of the large number of comparisons, Kendall's Tau was chosen due to the positive skew of the SPQ and ASI scales, and the Bayesian approach was applied as BFs do not need correction for multiple comparisons (60). A beta distribution prior was used where all correlation coefficient values were equally likely (a stretched prior width of 1 and robustness check at 0.5). As can be seen from the correlation matrix in Supplementary Table 2A, total SPQ (r_Kendall = 0.032, BF₁₀ = 0.106) and both its Cognitive-perceptual (r_Kendall = 0.016, BF₁₀ = 0.082) and Interpersonal (r_Kendall = 0.007, BF₁₀ = 0.076) dimensions remained unrelated to KB scores, with moderate to strong levels of evidence. The previous significant prediction of KB Score by Disorganised SPQ scores was insensitive in the current correlations (r_Kendall = 0.099, BF₁₀ = 2.064). When assessing individual trial correlations, the source of this effect was unclear, although the largest (but insensitive) correlation belonged to its relationship with Trial 6 (r_Kendall = 0.106, BF₁₀ = 1.974). For the SPQ subscales, Odd speech was associated with both Trial 6 (r_Kendall = 0.139, BF₁₀ = 31.084) and overall KB Score (r_Kendall = 0.119, BF₁₀ = 8.959). These associations with Odd Speech likely explain the link between the Disorganised dimension and KB Score in the previous regressions, as well as strengthen the idea that the link between Disorganised and KB Score is through Trial 6. Total ASI levels were largely unrelated to trial scores, with the null hypothesis supported for 11 of the 13 variables. The ASI subscales followed a similar pattern, with 82% of BFs accepting the null and 16.9% being insensitive, although one correlation between Heightened Emotion and Trial 8 had moderate evidence (r_Kendall = 0.115, BF₁₀ = 4.339).

Finally, we repeated this correlation matrix in current cannabis users (Supplementary Table 2B). This was conducted to assess whether the previous associations in Supplementary Table 2A remain stable between those who do and do not use cannabis. Briefly, the correlation between disorganised schizotypy and KB score remained insensitive, but descriptively increased in effect size (r_Kendall = 0.150, BF₁₀ = 0.874). The association between Odd Speech and Trial 6 remained a similar effect size but became insensitive, which most likely reflects the reduction in sample size (n = 74) rather than moderation. A new association was reported between the Heightened Cognition subscale of the ASI and KB score (r_Kendall = 0.210, BF₁₀ = 4.605), which primarily came from the association to Trial 6 (r_Kendall = 0.297, BF₁₀ = 66.703). Although the remaining subscale were not found to be related to overall KB score (r_Kendall = 0.080–0.152, BF₁₀ = 0.205–0.904), associations were found between Trial 6 and both Increased Significance (r_Kendall = 0.277, BF₁₀ = 29.806) and Senses Sharpening (r_Kendall = 0.231, BF₁₀ = 6.056), with Impending Understanding (r_Kendall = 0.168, BF₁₀ = 1.115) and Heightened Emotion (r_Kendall = 0.181, BF₁₀ = 1.506) returning insensitive. These correlations corroborate the both differences in SPQ and ASI scores between cannabis users and non-users (replication analyses section) and the significant interaction effect reported between current cannabis use and ASI when predicting KB score.