Testing the Effects of Personality-Related Beliefs on Micro-PK

Jakob, M.-J., Dechamps, M. C., & Maier, M. A. (2024). Testing the effects of personality-related beliefs on micro-PK. Journal of Anomalous Experience and Cognition, 4(1), 34–59. https://doi.org/10.31156/jaex.23809

AI Assessment

A large preregistered micro-PK study that confirmed one of its three target hypotheses, with the headline result resting on a disclosed preregistration typo. Using a quantum random number generator and a Bayesian analysis, the study tested whether high scorers on three Cluster C personality traits would bias which of two on-screen sentences a quantum device selected. One of the three target groups reached the preregistered strong-evidence threshold, the other two did not, and one control group unexpectedly showed moderate evidence for an effect. The authors disclose that the prior reported in the preregistration was a typo, and that under the prior as written the headline effect reverses. Every figure on this page was verified verbatim against the open-access primary article.

Provenance

DOI. 10.31156/jaex.23809. The article is open access under a Creative Commons CC-BY license, published in the Journal of Anomalous Experience and Cognition, 2024, Volume 4, Number 1, pp. 34–59.

Study type. Preregistered Bayesian micro-psychokinesis (micro-PK) study, run online as a quasi-experiment with a quantum random number generator (QRNG). It tests an observer effect mediated by personality traits rather than by deliberate intention. The first two authors share first authorship, and all three authors are at Ludwig-Maximilians-Universität München.

Funding. The article names no external funding source. Participants recruited through the online platforms MTurk and Prolific were paid a small fee ($1.5 or 1.80 GBP); university students received course credit. The procedure was approved by the ethical board of the Department of Psychology at LMU Munich.

Data availability. The study was preregistered at OSF (osf.io/gw98t), and the data and analyses are openly available at the OSF project repository (osf.io/qxu3s).

Source basis. Every figure on this page was confirmed against the primary article as published in the Journal of Anomalous Experience and Cognition, Volume 4, Number 1 (2024), pp. 34–59.

What the paper reports

Jakob, Dechamps, and Maier set out to test whether intentional observation can bias quantum outcomes in line with an observer’s implicit, fear-based concerns.¹ Working within the micro-PK tradition,² they measured three Cluster C personality traits (PTs), dependent, avoidant, and obsessive-compulsive, and predicted that high scorers would observe more trait-related sentences, selected by a quantum random number generator, than expected by chance. The confirmatory prediction was deliberately weak: that at least one of the three target (high-scorer) groups would reach a Bayes factor BF₁₀ greater than 10 during data collection.

That prediction was met for one of the three traits. For the dependent-PT high scorers (n = 1,400), the final Bayes factor was BF₁₀ = 10.41 (M = 15.18, SD = 2.68, against a chance expectation of 15 of 30), reported as strong evidence for the hypothesis (frequentist t(1399) = 2.51, p < .001). The other two target groups returned anecdotal evidence for the null, and one control group, the avoidant-PT low scorers, unexpectedly showed moderate evidence for an effect. The authors frame the result as “confirmatory evidence for a preregistered micro-PK effect for one out of three PTs” while noting that “hypothesizing at least one of three tests performed with the target groups to show a result is a relatively weak postulate.”

The results revealed strong evidence (Bayes Factor > 10) for a micro-PK effect in the dependent PT group, with high scorers observing more sentences addressing their concerns than expected by chance. We did not find strong evidence for the other PT groups or low scorers.

The headline figure carries an important load-bearing caveat that the authors themselves disclose. The analyses used an informed prior of delta ~ Cauchy(.05, .05), but the preregistration text mistakenly specified Cauchy(0.5, 0.5). The authors address this in their limitations, argue with supporting detail that it was a typo, and concede that it “diminishes the empirical strength of the data.”¹ Under the prior as written in the preregistration, the dependent-PT high-scorer result reverses to BF₀₁ = 1.31, that is, no support for the effect.

How it was run

• Design. A preregistered two-group quasi-experiment, comparing high versus low scorers on three Cluster C personality traits (dependent, avoidant, obsessive-compulsive), each tested in a separate within-participant block presented in randomized order.
• Quantum randomness. A “Quantis” QRNG by ID Quantique generated the per-trial outcome, sending a photon through a semi-conductive mirror-like prism with a 50% chance of deflection in either direction, following the structure of a double-slit experiment (it transforms 8 bits into 1 byte). The device passed the DIEHARD and NIST test batteries and used no post-correction, providing a true quantum source per trial.
• Blocks and trials. Each of the three PT blocks comprised 30 on-screen trials. On each trial, immediately before presentation, the QRNG selected between a trait-related and a neutral stimulus (short sentences) at a baseline 50/50 probability. Each trial ran fixation (700 ms), stimulus (3500 ms), then a black-screen inter-stimulus interval (400 ms). Observation was passive, with no manual response.
• Outcome measure. The dependent variable was the number of trait-related stimuli displayed per block, with a chance expectation of 15 of 30. Personality was measured with the VDS-30 (Sulz, 2000): 10 items per trait on a 4-point scale, with a preregistered cut-off of mean greater than or equal to 1.00 defining the high (target) group.
• Sample. N = 2,403 participants were tested (56% female, 44% male, 0.2% unspecified), recruited through a department board, psychology classes, university Facebook groups, MTurk (n = 545), and Prolific (n = 120). An attention check was added later, at n = 1,105, as an exploratory, non-preregistered measure.
• Analysis. Bayesian inference following Wagenmakers et al. (2011):³ a one-sided Bayesian one-sample t-test for the high (target) groups (testing for more than 50%, i.e., more than 15 of 30) and a two-sided test for the low (control) groups. The informed prior was delta ~ Cauchy(.05, .05). The a priori stopping rule was BF = 10 (strong evidence in either direction), with a preregistered maximum N = 1,000 as a backstop.

Results, as reported

The Bayes factors are reported without confidence intervals, consistent with the Bayesian framework. The final row reproduces the paper’s own footnote: under the prior the preregistration actually specified (later disclosed as a typo), the dependent-PT high-scorer effect that headlines the study reverses to no support (BF₀₁ = 1.31).

Eleven-dimension audit

Pre-registration

The study was preregistered at OSF (osf.io/gw98t), with the cut-off value, the confirmatory hypothesis, the Bayesian stopping rule, and the maximum-N backstop all specified in advance; data and analyses are posted at osf.io/qxu3s. The preregistration is honored in the analysis, but it contains a consequential error: the informed prior was written as delta ~ Cauchy(0.5, 0.5) when delta ~ Cauchy(.05, .05) was used. The authors disclose this openly as a typo and argue the point in their limitations rather than quietly substituting the analyzed prior.

Randomization

The per-trial outcome was set by a hardware quantum RNG (Quantis by ID Quantique), which generates randomness from photon deflection through a semi-conductive prism in a double-slit-like configuration, with no post-correction step, and which the authors state passed the DIEHARD and NIST batteries. The block order was also randomized using the QRNG. The randomization source for the dependent measure is therefore a true quantum source and is well specified.

Sensory leakage

Micro-PK has no sender, so the classical sensory-leakage concern does not arise; the stimulus that the QRNG selects is shown openly to the participant, and the dependent variable is the count of trait-related sentences, not a guess. The relevant analogue is whether ordinary semantic preference, rather than any anomalous influence, could drive the count; the chance baseline of 15 of 30 and the passive-observation design address this at the level of the device output rather than participant behavior.

Blinding

The task was a passive online watching task with no experimenter in contact with participants, and the outcome was determined by the QRNG rather than by any rater, so participant-side blinding is not the operative issue. The authors instead flag the analyst as an unmasked observer: the data were inspected during collection, the analysis was not masked, and the group held a documented “moderate belief,” which they treat as a possible source of an experimenter effect rather than concealing it.

Optional stopping

Stopping followed a preregistered Bayesian rule of BF = 10 with a preregistered maximum N = 1,000 backstop. The authors report transparently that the maximum-N criterion was ignored for the dependent-PT data, which continued to n = 1,400, because a clear trend was visible at n = 1,000; they argue this conditional extension was mentioned in the preregistration and is consistent with the Bayesian approach. Because BF was monitored during collection, the disclosed prior typo is especially load-bearing: under the prior as written, the dependent-PT effect is BF₀₁ = 1.31 rather than BF₁₀ = 10.41.

Outcome measure

The confirmatory outcome was defined in advance as the number of trait-related stimuli selected per 30-trial block, tested against a chance value of 15 with a one-sided Bayesian t-test for the high groups and a two-sided test for the low groups. The personality split used the preregistered VDS-30 cut-off (mean greater than or equal to 1.00). The primary endpoint is unambiguous and matches the preregistration.

Effect size

Effect sizes are small and reported in their natural terms: target-group means hover just above the chance value of 15 (15.18, 15.06, 15.05) with standard deviations near 2.7, and the authors note that micro-PK effects in their prior work run around d = .1 or lower. Even the confirmed dependent-PT result (M = 15.18) is a fraction of a sentence above chance across 30 trials, with the Bayes factor (BF₁₀ = 10.41) sitting just over the strong-evidence threshold.

Multiple comparisons

Six Bayesian t-tests were run, three target and three control groups, and the confirmatory prediction required only that at least one of the three target tests reach BF₁₀ greater than 10, which the authors acknowledge is “a relatively weak postulate.” One of three target tests cleared the threshold; one control test (AV-PT-low) unexpectedly returned moderate evidence for an effect, which is awkward for the directional hypothesis. The prior-typo footnote shows how sensitive the verdict across all six tests is to the prior chosen.

Internal replication

The three traits and their high and low subsamples provide internal cross-checks within the single study, and they did not move together: only one of three target groups confirmed, two target groups favored the null, and one control group unexpectedly showed an effect. There is no fresh within-study replication of the dependent-PT result beyond the single preregistered test reported.

External replication

The authors situate the dependent-PT finding as corroborating their own earlier studies, Jakob et al. (2020) and Maier and Dechamps (2018),⁴⁵ and the broader micro-PK literature includes two meta-analyses reporting an overall effect (Radin & Nelson, 1989; Bösch et al., 2006).⁶⁷ The same group also reports that micro-PK results commonly decline or fail to replicate directly,⁸ a pattern their theoretical model (von Lucadou et al., 2007) actively predicts.⁹ This single confirmed trait awaits independent replication.

Transparency

Transparency is a strength of the paper, which is open access with an open preregistration and open data and analyses. Most notably, the authors enumerate their own limitations rather than burying them. They are: (1) the prior was incorrectly reported in the preregistration as Cauchy(0.5, 0.5) instead of the Cauchy(.05, .05) used, disclosed as a typo that they concede “diminishes the empirical strength of the data,” and under which the dependent-PT effect flips to BF₀₁ = 1.31; (2) the data were checked regularly during collection by an unmasked analyst holding a “moderate belief,” so an experimenter-psi effect “might have contributed to the results”; (3) the preregistered maximum N = 1,000 was ignored, with collection continuing to n = 1,400 because a clear trend was seen at n = 1,000; (4) the attention check was not implemented from the start (only from n = 1,105), reducing experimental control online, with 8% of participants showing reaction times larger than 10 seconds; and (5) the three Cluster C traits are highly correlated, raising the question of whether a combined index would be a better independent factor.

The adversarial record

• Precursor work. The design extends the group’s own micro-PK program, including Jakob et al. (2020) and Maier and Dechamps (2018), and sits within the longer QRNG micro-PK tradition reviewed by Varvoglis and Bancel (2015). The theoretical scaffolding draws on the Unus Mundus model and the Model of Pragmatic Information (von Lucadou et al., 2007), the latter of which predicts that such effects must decline over repeated confirmation attempts.
• Contested-literature context. Micro-PK is among the most disputed areas of parapsychology. The two meta-analyses the paper cites report an overall effect, but the Bösch, Steinkamp, and Boller (2006) meta-analysis is itself well known for concluding that the effect, while statistically present, is not clearly distinguishable from small-study or publication bias. Skeptics of the field argue that initial effects are artifacts of error, random fluctuation, or questionable research practices (Wagenmakers et al., 2011).
• Reading against the study. Three features weaken the headline. First, under the prior the preregistration actually specified, the dependent-PT effect reverses from BF₁₀ = 10.41 to BF₀₁ = 1.31, so the result rests on the authors’ after-the-fact reading of an error as a typo, which they concede diminishes the empirical strength. Second, only 1 of 3 target tests confirmed, against a self-described “relatively weak” disjunctive prediction. Third, a control group (AV-PT-low) returned moderate evidence for an effect it was not supposed to show, which the directional hypothesis does not predict. The confirmed dependent-PT mean is also a fraction of a sentence above chance.
• Source-fidelity note. The page reproduces the paper’s reported statistics verbatim, including the frequentist t and p values as the authors stated them and the full footnote of Bayes factors under the erroneously preregistered prior. The two BF subscripts (BF₁₀ for evidence favoring H1, BF₀₁ for evidence favoring H0) are carried exactly as the paper uses them.