Psi Effects as a Result of Implicit Expectations About Probabilities – Investigating Micro-PK with a Biased Baseline

Dechamps, M. C., Iovine, C. G. N., & Maier, M. A. (2025). Psi effects as a result of implicit expectations about probabilities – Investigating micro-PK with a biased baseline. Journal of Scientific Exploration, 39(3), 279–285. https://doi.org/10.31275/20253571

AI Assessment

A pilot micro-PK test with a split result: strong Bayesian evidence in the ‘lucky’ condition, a moderate null in the ‘unlucky’ condition. Using a masked online coin-toss game with secretly biased win odds and a quantum random number generator, the study tested whether participants’ implicit expectation of a fair 50% coin would pull outcomes toward 50% and away from each group’s hidden baseline. The ‘lucky’ (60% baseline) group’s hits drifted toward 50% as predicted with strong evidence, while the ‘unlucky’ (40% baseline) group showed no such effect, with moderate evidence for the null. Every figure on this page was verified verbatim against the open-access primary article.

Provenance

DOI. 10.31275/20253571. The article is Platinum Open Access under a Creative Commons CC-BY-NC 4.0 license (submitted 24 October 2024, accepted 21 December 2024, published 15 October 2025).

Study type. A pilot Bayesian micro-psychokinesis (micro-PK) study using a quantum random number generator (qRNG), run online as a masked coin-toss game with a biased baseline and a between-participants two-condition design. All three authors are at Ludwig-Maximilians-Universität München.

Funding and approval. The procedure was approved by the ethical board of the Department of Psychology at LMU München. Psychology students received course credit; all other participants were uncompensated.

Data availability. The open data, materials, and analysis scripts are posted at OSF (osf.io/2zgp5). Analyses were run in R version 4.4.1.

Source basis. Every figure on this page was confirmed against the primary article as published in the Journal of Scientific Exploration, Volume 39, Number 3 (Fall 2025), pp. 279–285.

What the paper reports

Dechamps, Iovine, and Maier set out to test whether participants’ implicit expectation of a fair, 50% coin would bias the outcomes of a secretly weighted coin-toss game toward 50%.¹ Participants played a 20-round game whose win odds were quietly pre-set: a ‘lucky’ condition (LC) with a 60% baseline win probability, and an ‘unlucky’ condition (UC) with a 40% baseline. The hypothesis predicted that both groups would drift toward 50%, meaning fewer than 12 hits in the LC and more than 8 hits in the UC. In the ‘lucky’ condition the result was strong Bayesian evidence for the prediction (BF₁₀ = 10.87, Cohen’s d = .09; M = 11.80, SD = 2.13): participants averaged fewer hits than the 60% baseline would yield, in the direction of 50%. In the ‘unlucky’ condition the data instead favored the null hypothesis with moderate evidence (BF₀₁ = 3.76; M = 8.00, SD = 2.18), aligning with the 40% baseline and showing no micro-PK effect.

The authors read the ‘lucky’ result as suggesting that aligning outcomes with participants’ implicit 50% expectation may help avoid the so-called decline effect, the tendency for psi effects to weaken on repeated testing.¹ For the ‘unlucky’ null they offer a post-hoc moderator: emotional detachment, the idea that frustration-driven disengagement in the losing group may have hindered any effect. A direct replication of the ‘lucky’ condition is described as already in planning. The interpretive frame draws on a long-running micro-PK literature, including meta-analytic claims of an overall effect²³ and prior subconscious-resistance and decline-effect accounts.⁴⁵

Bayesian analysis revealed strong micro-PK effects towards 50% in the ‘lucky’ group but no effects in the ‘unlucky’ group.

How it was run

• Design. A pilot single-group design with two conditions assigned by between-participants random assignment: a ‘lucky’ condition (LC) with a pre-set 60% baseline win probability and an ‘unlucky’ condition (UC) with a pre-set 40% baseline. Participants were unaware of this biasing.
• The game. A masked micro-PK coin-toss game of 20 rounds. Each round the participant guessed ‘heads’ or ‘tails’ on a made-up coin (an AI-generated image) and pressed a button; a quantum random number generator (qRNG) connected to the experiment server produced a number from 1 to 100 to settle the round. In the LC a qRNG result of 60 or below counted as a win; in the UC a result of 40 or below counted as a win. A 4000 ms spinning-coin animation, “Well done!” or “Unfortunately, wrong guess.” feedback, and a running cumulative win/loss tally accompanied play.
• Outcome measure. The dependent variable was the number of successful hits across the 20-round game.
• Hypothesis. Participants’ implicit expectation of a fair (50%) coin was predicted to bias outcomes toward 50%: fewer than 12 hits in the LC (below the 60% baseline of 12 of 20) and more than 8 hits in the UC (above the 40% baseline of 8 of 20), both toward 10 of 20.
• Sample. Of 1,602 who took part, 21 requested their data be excluded, 18 were underage, and 38 did not answer the data-integrity question, leaving a final N = 1,526 (1,012 female, 491 male, 23 diverse; mean age 32.47 years, SD = 13.94). Condition sizes were LC N = 801 and UC N = 725. The study ran in German (89%) and English (11%).
• Analysis. Bayesian one-sample t-tests with an uninformed prior delta ~ Cauchy(0, 0.1) defined a priori, one-tailed (LC: fewer than 12 hits; UC: more than 8 hits), with a Bayes-factor threshold of 10 for strong evidence. Analyses used R version 4.4.1.

Results, as reported

The ‘lucky’ result rests on a small effect size (Cohen’s d = .09) and a Bayes factor (10.87) only just past the authors’ threshold of 10; a prior-robustness check found strong support at medium prior widths but only anecdotal support at larger priors. The exploratory comparisons are reported by the authors as secondary and were not the confirmatory test.

Eleven-dimension audit

Pre-registration

This is a pilot study and is not reported as a formal preregistration of the full protocol. The authors state that the prior (delta ~ Cauchy(0, 0.1)) was defined a priori before data collection, and that the data-exclusion criteria were consensually stated before data collection and applied before the analyses, with the data, materials, and scripts openly posted at OSF (osf.io/2zgp5). What is absent is a registered record of the complete analysis plan in advance, so the a-priori prior and exclusion rules carry the advance-commitment weight here rather than a preregistration document.

Randomization

Round outcomes were determined by a quantum random number generator connected to the experiment server, which produced a value from 1 to 100 per round; a result at or below the condition’s hidden threshold (60 in the LC, 40 in the UC) counted as a win. Assignment to the ‘lucky’ or ‘unlucky’ condition was by random between-participants allocation. The randomization source for the dependent measure is therefore specified, though the paper does not report a hardware-level randomness test battery for the qRNG.

Sensory leakage

Micro-PK has no sender, so the classical sensory-leakage concern does not apply in the telepathy sense. The relevant analogue is whether participants could detect the biasing: the win thresholds were masked and participants were unaware that the odds were weighted, so the manipulation was concealed by design. The authors note, however, that many participants questioned the game’s fairness, which bears on how well the cover held.

Blinding

The game was a self-administered online task with no experimenter in contact with participants during play, and outcomes were set by the qRNG rather than by any rater, so rater-blinding in the conventional sense is not the operative issue. Participants were blind to the baseline biasing of their condition, which is the design feature that makes the implicit-expectation test meaningful.

Optional stopping

The paper reports a fixed Bayes-factor threshold of 10 for strong evidence but describes the work as a pilot with a single large sample rather than a sequential design with a pre-specified stopping rule tied to that threshold. The analysis is presented on the final assembled sample (N = 1,526) after the stated exclusions, so there is no indication of data-peeking-driven stopping, but neither is a formal optional-stopping protocol documented.

Outcome measure

The confirmatory outcome was defined in advance as the number of hits in the 20-round game, tested one-tailed against each condition’s directional prediction (fewer than 12 hits in the LC, more than 8 in the UC) with Bayesian one-sample t-tests and an a-priori Cauchy(0, 0.1) prior. The primary endpoint is unambiguous and tied directly to the toward-50% hypothesis.

Effect size

The ‘lucky’ effect is reported as Cohen’s d = .09, a very small effect, with mean hits M = 11.80 (SD = 2.13) against a 60% baseline of 12 of 20. The ‘unlucky’ condition sat at the 40% baseline (M = 8.00, SD = 2.18) with no effect. The magnitudes are small and, in the LC, the Bayes factor (BF₁₀ = 10.87) only marginally exceeds the authors’ threshold of 10.

Multiple comparisons

The two confirmatory tests are the one-tailed LC and UC t-tests. Beyond these the paper reports several exploratory analyses, including the winning-importance comparison (t(1512) = 3.86, p < .001), the perceived-influence comparison (t(1514) = 1.75, p = .08), and a UC-only motivation-DV correlation (r(718) = .12, p < .001), along with gender and belief-scale breakdowns. These are presented as exploratory rather than corrected confirmatory tests, so the reader should weight them accordingly.

Internal replication

The two conditions function as a within-study contrast rather than a true internal replication, and they diverged: the ‘lucky’ condition produced strong evidence for the prediction while the ‘unlucky’ condition returned a moderate null. The headline effect therefore rests on one of two pre-specified tests, not on a replicated within-study finding.

External replication

No external replication is reported in this paper; the authors describe a direct replication of the ‘lucky’ condition as already in planning. The study situates itself within a contested micro-PK literature that includes meta-analyses reporting an overall effect²³ alongside the group’s own prior work documenting marked replication difficulty, including a failed replication of a related paradigm.⁶⁷

Transparency

The paper is open access, with data, materials, and analysis scripts openly posted at OSF, and it reports the split result without overstating the ‘unlucky’ null. The authors enumerate their own limitations: that external validity may be limited because the sample was not fully representative in age and gender distribution, mitigated somewhat by drawing on an online community sample rather than students alone; that many participants questioned the game’s fairness, which may have affected engagement; that displaying performance may have framed the game too competitively, leaving less room for subjective experience to shape outcomes; and that emotional investment and its frustration, especially in the ‘unlucky’ group, may have hindered any micro-PK effect. They further flag that their emotional-detachment moderator is post-hoc and that motivation was assessed after the game, a caveat on causal direction, and they attribute the smaller UC sample to higher dropout.

The adversarial record

• Precursor and decline-effect literature. The study’s framing engages a long micro-PK tradition and the decline-effect problem it is built to address, drawing on subconscious-resistance accounts⁴ and generalized-quantum-theory treatments of how psi correlations may decline under repeated observation.⁵
• Contested-literature context. Micro-PK is among the most disputed areas of parapsychology. The meta-analyses the field leans on report an overall effect,²³ but the Bösch, Steinkamp, and Boller (2006) meta-analysis is well known for concluding that the effect, while statistically present across 380 studies, is entangled with marked heterogeneity and a possible small-study or publication bias, a reading later contested by Kugel (2011).
• Reading against the study. The work is a pilot, not a preregistration of the full protocol; the headline rests on one of two pre-specified tests, with the ‘unlucky’ condition returning a moderate null; the confirmatory ‘lucky’ effect is very small (d = .09) with a Bayes factor (10.87) only just past the threshold of 10 and merely anecdotal at larger priors; the emotional-detachment moderator is post-hoc; and the authors’ own report that many participants suspected the game was unfair raises a possible confound between the implicit-expectation account and ordinary suspicion-driven responding.
• Source-fidelity note. This page reproduces the paper’s reported figures exactly, including both the strong ‘lucky’ result (BF₁₀ = 10.87, d = .09) and the ‘unlucky’ null (BF₀₁ = 3.76), and presents the authors’ interpretive claims as their own framing rather than as adjudicated findings.