Correlates of Hallucinatory Experiences in the General Population: An International Multisite Replication Study

Hallucinatory experiences can occur in both clinical and nonclinical groups. However, in previous studies of the general population, investigations of the cognitive mechanisms underlying hallucinatory experiences have yielded inconsistent results. We ran a large-scale preregistered multisite study, in which general-population participants (N = 1,394 across 11 data-collection sites and online) completed assessments of hallucinatory experiences, a measure of adverse childhood experiences, and four tasks: source memory, dichotic listening, backward digit span, and auditory signal detection. We found that hallucinatory experiences were associated with a higher false-alarm rate on the signal detection task and a greater number of reported adverse childhood experiences but not with any of the other cognitive measures employed. These findings are an important step in improving reproducibility in hallucinations research and suggest that the replicability of some findings regarding cognition in clinical samples needs to be investigated.


S1. Further details on the headphone check
For online data collection, participants were required to pass a short headphone check task before gaining access to the main task platform. The task is designed such that it can only be successfully completed if the participant is wearing headphones. The participant completes 6 trials, each comprising of 3 sequentially presented pure tones (200Hz, duration = 1000ms), with one tone presented at a lower volume than the others, and with one of the tones presented 180° out of phase. The task is designed so that when played through loudspeakers, the tones are indistinguishable, yet when played through headphones, the quietest tone is easily detected. The participant is required to select which of the tones is the quietest, and 'passes' the task if they correctly choose the quietest tone on at least 5/6 trials (which would occur at a rate of approximately 1.8% by chance). As part of the check, participants were also presented with two short bursts of noise to each ear, and asked to ensure that the headphones were positioned the correct way round (for the dichotic listening task). No data was gathered from participants who did not proceed past the headphone check.

S2. Further details on the attention checks
The first question, within the demographics section, stated: "The following question is an attentional check. Please do not answer the sports question -just click the 'continue' button, below, to move on to the next part of the study. Which sport do you like?". Participants were provided with a list of sports to choose from, but only passed the attention check if they did not answer. The second question, embedded within the CAPS questionnaire, stated "While watching TV, have you ever had a fatal heart attack?", with participants failing by selecting 'yes'. The third question, placed at the end of the study, stated "The questions below are an attention check. Please select "Other" and type "Memory" into the text box. Once you have done that, click "continue" to complete the study! What was this study about?", followed by a number of options, as well as the option to type into a free-text box. Participants failed this check if they did not correctly select "Other" and type into the text-box as instructed.

S3. Exclusions based on pre-registered criteria
CAPS -A further 4 participants were excluded based on a high number of missed responses to the CAPS scale (> 33%), or due to unfeasibly fast response to questions on the CAPS (< 1.5s per item).
Source memory -13 participants were excluded from the source memory data, meeting one or more of the pre-registered exclusion criteria: due to 100% task accuracy (n = 1), scoring at or below 33.3% accuracy (n = 9), or scoring at or below 50% on old-new accuracy (n = 5).
Dichotic listening -126 participants were excluded from the dichotic listening task performance, based on Bless et al. (2015), due to scoring < 50% accuracy on homonymous trials in any of the three task conditions (n = 83) and/or scoring a laterality index of 100% to one ear (n = 44).
Digit span -32 participants were excluded from the backwards digit span data, scoring a mean span ≤ 3 (n = 26) or ≥ 12 (n = 6). One further participant was excluded because data from the digit span task did not save due to a technical error Auditory signal detection -11 participants were excluded from the auditory signal detection task data, scoring a ′ of ≤ 0 (indicating at or below chance performance) (n = 10), and/or a hit rate of ≤ 10% (n = 3).
Adverse childhood experiences -data on adverse childhood experiences was excluded if participants completed ≤ 66% of the questionnaire (n = 19), or spent < 1.5s per item (n = 14). In total, 22 participants' data was excluded from the ACE.

S4. Source memory -further analysis
Exploratory (non-preregistered analysis) showed that there was a very small correlation between old-new accuracy and CAPS score (rs = 0.09, 95% CI = [0.04, 0.14], p <0.001), suggesting that increased reports of hallucinations were weakly associated with a higher performance on recognition memory. However, this association was not found when the LSHS-E was used as a measure of hallucinations (rs = 0.01, 95% CI = [-0.04, 0.06], p = 0.704). One further analysis strategy has been to adjust performance for guessing by subtracting the number of new words misrecalled as heard from the number of imagined words recalled as heard (Woodward & Menon, 2011). This adjusted variable did not correlate with CAPS score (rs = .017, p = .537).

S5. Dichotic listening -task performance analysis
In the non-forced attention condition, as expected an overall right ear advantage was evident; that is, participants were more likely to correctly identify the syllable presented to their right ear (M = 13.06, 95% CI = [12.85, 13.26

S6. Comparison of lab and online data
The below table provides means for important variables, split into lab and online data.
To test whether there was any divergence between findings from data collected in the lab and data collected online, we repeated the analyses for H1-5 for lab data only and online data only, with no difference in results: there was no association between CAPS and dichotic listening, digit span, or source memory performance, but there was an association between CAPS score and adverse childhood experiences, in both sets of data. There was also no difference between the number of exclusions due to failed attention checks in the lab data (3.86%) as compared to the online data (4.73%), ( 2 = 0.48, p = .488).