Using Virtual Reality to Test for Telepathy: A Proof-of-Concept Study

Vernon, D., Sandford, T., & Moyo, E. (2020). Using Virtual Reality to Test for Telepathy: A Proof-of-Concept Study. Journal of Scientific Exploration, 34(4), 683–702. https://doi.org/10.31275/20201833

AI Assessment

A telepathy feasibility study whose pre-specified primary test was null; a post-hoc reanalysis reached nominal significance, and the authors themselves call the result more suggestive than conclusive. Eleven pairs took turns as sender and receiver, with the sender immersed in a 360-degree virtual-reality clip while the receiver, partially isolated, ranked five images per trial. The primary direct-hit rate (25% against a 20% baseline) did not differ from chance. A secondary, explicitly post-hoc analysis of receivers’ top-two choices reached 52% against a 40% baseline, which the authors flag as possibly a Type I error in an underpowered study.

Provenance

Source. Peer-reviewed open-access article (CC-BY-NC): Journal of Scientific Exploration, 34(4), pp. 683–702, 2020. Submitted 25 April 2020; accepted 20 June 2020; published 15 December 2020. School of Psychology, Politics, and Sociology, Canterbury Christ Church University, UK.

Study type. Within-participants cross-over telepathy experiment: a virtual-reality-immersed sender and a partially isolated receiver, with a five-alternative forced-choice ranking and a 20% chance baseline.

Funding. None stated; an academic study at Canterbury Christ Church University. The authors thank Richard Weatherall and Mark Titus for help with the VR equipment.

Data availability. None stated; no open-data repository is referenced.

Source basis. Figures confirmed against the published JSE article (pp. 683–702), Results section and Tables 1 and 2.

What the paper reports

Eleven pairs of participants (22 people, each acting as both sender and receiver) were tested in a cross-over design. On each trial the sender was immersed in a 360-degree VR clip of a positive, high-arousal activity while the receiver ranked five candidate images. The authors predicted, a priori, that VR immersion would push the receiver’s direct-hit rate above the 20% chance baseline. It did not: across trials the mean direct-hit rate was 25%, not significantly above chance (t(20) = 0.960, p = .17, one-tailed; d = 0.20).¹

A secondary analysis the authors describe as post hoc compared receivers’ top-two choices against a 40% baseline and reached 52% (t(19) = 2.259, p = .018, Bonferroni-corrected; d = 0.50). Of the belief and relationship measures, only the psi subscale of the Revised Paranormal Belief Scale correlated with hit rate (r = .454, p = .04). The authors conclude the result is “more suggestive than conclusive.

How it was run

• Within-participants cross-over: all 22 participants served as both sender and receiver, counterbalanced. The 11 pairs (14 female, 8 male; aged 19 to 55, mean 28.73) were opportunity-sampled; 2 were spousal couples and 9 self-identified as friends or colleagues.
• The sender wore an Oculus Rift headset and viewed five 30-second 360-degree clips of positive, high-arousal activities (downhill skiing, racecar driving, skydiving), each separated by a 45-second relaxing beach clip, and was told to imagine the receiver was present.
• The receiver sat in an adjoining room, partially isolated with headphones playing pink noise, viewing a synchronized slideshow, and ranked the five images of each set from 1 (most likely the target) to 5 (least likely).
• A trial was a hit if the receiver ranked the target image “1”; each receiver completed five trials; chance was one in five, or 20%.
• Targets were drawn from a pool of 50 positive-valence, high-arousal images (45 from the International Affective Picture System), arranged in 10 matched sets of five; image order, target selection, and clip order were randomized using random.org.
• Belief and closeness were measured with the Revised Paranormal Belief Scale (RPBS) and two relationship measures (the Inclusion of the Other in the Self scale and a 100 mm intensity line).

Results, as reported

The pre-specified primary outcome is the direct-hit rate against the 20% baseline, which was non-significant. The top-two analysis against a 40% baseline is labelled post hoc by the authors; one participant was excluded from the primary analysis for assigning a rank of 1 to more than one image, and one further participant was excluded from the post-hoc analysis for ranking only first choices.

Eleven-dimension audit

Pre-registration

A single directional prediction (VR immersion of the sender would push the receiver’s hit rate above chance) was stated in advance, and the design (11 pairs, five trials each, 20% baseline) was fixed before data collection. As a 2020 journal article there is no formal pre-registration. Importantly, the top-two reanalysis is explicitly labelled post hoc by the authors, which keeps the confirmatory and exploratory tests distinct.

Randomization

Image order within each set, target selection, and the order of the VR clips were randomized using random.org, and sender/receiver assignment was randomized and counterbalanced in the cross-over design. Randomization is adequate and described.

Sensory leakage

Sender and receiver were in separate adjoining rooms with no direct channel, and scoring was an objective forced-choice rank, so there is no obvious leakage path for the target’s identity. Isolation is weaker than in ganzfeld work, however: the receiver was only partially isolated (headphones and pink noise, but eyes open viewing a screen), which the authors raise as a possible source of noise rather than leakage.

Blinding

Target identification was a closed forced-choice rank against a fixed target with image positions randomized, so scoring required no subjective judgement and no separate judge to blind. The experimenter seated with the receiver started the matched slideshow on cue but did not score or influence the rankings.

Optional stopping

The sample was fixed at 11 pairs and five trials each, and there is no indication of data-dependent stopping. The study is explicitly a small feasibility, or proof-of-concept, exercise.

Outcome measure

The pre-specified primary measure was the direct hit (target ranked 1) against 20%. The authors themselves criticise this measure as possibly insensitive and propose a top-half/bottom-half re-coding for future work; the 52% figure comes from a post-hoc top-two measure against a 40% baseline.

Effect size

Small for the primary test (d = 0.20, non-significant) and medium for the post-hoc test (d = 0.50). The authors’ own G*Power analysis shows that detecting a d = 0.20 effect at power .8 would require 156 participants (80 pairs), so this study of 11 pairs was substantially underpowered.

Multiple comparisons

The post-hoc top-two test carried a Bonferroni correction (α/2 = .025) and remained significant. The belief analysis, however, ran correlations against all seven RPBS subscales, and only the psi subscale reached p = .04, which would not survive correction for seven tests, alongside two further non-significant relationship correlations. The belief result should be read as exploratory.

Internal replication

The cross-over design has every participant serve as both sender and receiver, but this is a single small series rather than a repeated experiment, and the primary and post-hoc analyses draw on the same data.

External replication

The one prior virtual-reality telepathy study (Murray et al., 2007) found no effect,² and the wider telepathy literature is contested, with null ganzfeld replications (Milton and Wiseman, 1999)³ set against positive meta-analytic and experimental claims (for example Bem and Honorton, 1994).⁴

Transparency

The methods, apparatus, exclusions (two participants removed for invalid rankings, with reasons given), the Bonferroni step, and a power analysis are all reported, and the paper is open access (CC-BY-NC). There is no open-data repository.

The adversarial record

• The result that matters is the primary null: the pre-specified direct-hit test did not differ from chance (25% versus 20%, p = .17). The 52% figure is a post-hoc top-two reanalysis, and the authors explicitly warn it “could simply be a Type I error.”
• The belief correlation (RPBS psi subscale, p = .04) is one of seven subscale tests and would not survive correction; the authors note it might instead reflect low-belief participants performing below chance.
• Reading against the study: it is a small, self-described underpowered feasibility study (11 pairs); the receiver was only partially isolated; and while the confirmatory-versus-post-hoc framing in the body is honest, the abstract states the post-hoc 52% figure, which a hurried reader could mistake for the headline result.
• The authors themselves catalogue the design limitations they believe may have driven the null: the VR target clips (30 seconds) were far shorter than the 10-to-40-minute exposures typical of ganzfeld work; the clips were all positive in valence and similar in content, where more varied or negative material might evoke stronger responses; the sender’s arousal was assumed from static image ratings rather than objectively measured; the reversed sender/receiver experiential states differ from a classic ganzfeld; and the narrow range of relationship closeness (nine of eleven pairs were friends or colleagues) may have masked any pairing effect. They argue a properly powered design (around 80 pairs) with longer, more varied clips and stronger receiver isolation would be needed to test the effect fairly.