Discovery Lab Answers

How good is the evidence for micro-PK?

Coverage note first: the studies below come from the ESP-Nexus library, and the library’s share of the full published literature on micro-PK (RNG/REG-based psychokinesis) has not been measured. What follows is a summary of what the library holds, not a settled account of the field.

For the broader picture, the evidence table beneath this answer covers 39 studies across 43 result rows, using several non-comparable metrics — so no single pooled bottom line is appropriate.

What the evidence shows

Pattern across studies

The results are genuinely mixed, and that mix is the honest headline:

• Positive results are present. Several individual experiments report statistically significant deviations in the intended direction. Williams (2021) computed a pooled estimate across 42 retro-PK RNG studies (1975–2021) and reported a z of 6.82 (p = 4.57 × 10⁻¹²). Dechamps (2021) found a hit rate of 0.505 (p = .002) across over 81,000 trials in a positive-priming condition. Collesso (2021) reported a Cohen’s d of 0.49 (p = .01) in a meditation-plus-visualization experiment. Mossbridge (2021) reported a pre-registered confirmatory result (p < 0.001) in a Heart Quest micro-PK task. Jakob (2024) found a hit rate of 0.506 (p < .001) in high-scoring participants. Dechamps (2025) found a Cohen’s d of 0.09 in a lucky-condition task (hit rate 0.59 across ~16,000 trials).

• Null results are also present — and numerous. Of the 43 result rows, 14 report null or below-chance outcomes. Maier (2022)’s preregistered primary analysis found a Pearson r of 0.01 across 2,052 participants — effectively zero. Penberthy (2024)’s meditation cohort produced z = −0.0013 (p = 0.4995). Pisoni (2024)’s re-run ANOVA returned p = 0.21. Grote (2017)’s combined pre-planned analysis returned p = 0.315. Alexander (2019)’s combined-operators correlation was ES = 0.024 (p = 0.3). Dechamps (2019)’s preregistered smoker study (Study 3) returned a hit rate of 0.498, directionally null.

• **One meta-analytic result runs below chance.** Bösch (2006) pooled 380 intentional RNG studies and obtained a proportion index (π) result with z = −3.67 (p < .001), flagged as below-chance / psi-missing — meaning the aggregate deviation, while significant, ran opposite to intention. This is a substantively important finding: a large meta-analysis finding a significant effect in the wrong direction is not a null; it is a disconfirmation of the simple “intention moves the RNG” hypothesis.

• Effect sizes, where reported on comparable metrics, are small. The standardized effect sizes available in the rows range from essentially zero (Pearson r = 0.01, Maier 2022) to moderate (Cohen’s d = 0.49, Collesso 2021), but the larger values come from smaller, less constrained studies. The Dechamps (2025) result of d = 0.09 and the Vares (2013) regression r = 0.185 give a sense of the smaller-study range.

Key methodological tensions

Preregistered vs. unregistered designs. A recurring pattern in the rows is that unregistered or exploratory analyses more often return positive results, while preregistered primary outcomes (Maier 2022, Dechamps 2019 Study 3, Penberthy 2024) more often return null. This is not universal — Mossbridge (2021) reports a preregistered confirmatory positive — but the gap is visible enough to matter.

Trial size and effect direction. Pallikari (2023) notes a feature in the micro-PK database: studies up to roughly 100,000 trials show broader score scatter (consistent with a claimed effect), while studies from roughly 100,000 to 100 billion trials scatter around 50% as expected from random data. Pallikari proposes a Markov model of correlated binary scores as a non-paranormal explanation for the broadening seen in smaller studies, and concludes that both a paranormal and a non-paranormal interpretation remain scientifically supportable — the evidence does not cleanly exclude either.

Geomagnetic and physical confounds. Stevens (2005) examines whether geomagnetic fluctuations could account for systematic deviations in REG outputs, particularly when combined with external noise sources, and proposes that micro-PK studies should control for this. This is an unresolved methodological question, not a settled refutation.

The “eraser” studies (Maier 2022) introduce an unusual design in which experimental data are recalled from memory after being permanently deleted, producing positive results in those sub-conditions despite the null preregistered primary. These results are directionally positive but methodologically distinct from standard RNG designs, and the interpretation is disputed.

Skeptical critiques

What critics argue. Bösch, Steinkamp, and Boller (2006) conducted a meta-analysis of 380 intentional RNG studies — the largest systematic review in the rows — and raised two interconnected concerns: (a) a significant file-drawer problem, with predominantly small unpublished studies inflating the apparent effect, and (b) the significant overall result running below chance (z = −3.67) rather than in the intended direction, which contradicts the simplest form of the micro-PK hypothesis. Pallikari (2023) identifies conformity bias and experimenter expectancy effects as likely contributors to the scatter pattern, and argues that a non-paranormal model (frequency-correlated binary sequences) can account for the aggregate data without invoking psi.

What the experimental data show. The positive pooled estimate from Williams (2021) (z = 6.82 across 42 studies) exists alongside Bösch (2006)’s below-chance aggregate across 380 studies — two meta-analytic summaries using overlapping literatures reach directionally different conclusions, which itself is informative about heterogeneity. Preregistered individual experiments (Maier 2022, Dechamps 2019 Study 3) have not consistently replicated positive findings.

Analysis. The two large meta-analytic estimates — Williams (2021) and Bösch (2006) — use different inclusion criteria and different aggregation methods, and their directional disagreement has not been resolved by a third independent analysis in the rows. Preregistered direct replications of specific positive protocols have returned mixed outcomes. No single experimental group’s positive finding has been independently replicated by a different laboratory using the same protocol with a preregistered primary outcome and a null-finding rate comparable to chance — at least not within what the library currently holds.

The Phenomena and Methodology sections of ESP-Nexus carry additional context on how micro-PK and related RNG designs are situated within broader psi research.