Discovery Lab Answers

RNG psychokinesis research over the past ~50 years — what the results show

Coverage first. This reflects the RNG/micro-PK studies currently in the ESP-Nexus library, not the complete published literature on this question. How much of that literature the library holds has not been measured — so treat what follows as a summary of what we hold, not a settled account of the field.

The metrics don’t reduce to one number. The studies here report different, non-comparable quantities — standardized effect sizes (Cohen’s d, Pearson r), raw hit rates, bare p-values, and z-scores. There is no single “overall RNG effect size” that can honestly be pooled across them, so the picture is best read as a distribution of outcomes rather than one bottom line.

The distribution of results is genuinely split

Across the roughly 40 studies / 44 result rows in the evidence set, directions disagree — some positive, some null, and at least one large meta-analysis below chance. Fourteen of the result rows report a null or a below-chance (negative) effect. That heterogeneity is the central finding, not a footnote.

Positive results appear across several individual experiments and pooled estimates:

Null results are equally represented, including several preregistered primary tests:

Notably, some research groups that reported positive results in one study (e.g. Maier and colleagues) reported nulls in their own subsequent or combined analyses — Maier (2018) shows a positive Study 1 but a null combined Studies 1+2. That is a same-group inconsistency, not just cross-lab variability.

Below chance. The largest single aggregation here, Bösch, Steinkamp and Boller (2006), pooled 380 intentional studies and reported an overall fixed-effects index below chance expectation (z = −3.67, p < .001), alongside two features the authors emphasize: a strong inverse relation between effect size and sample size (a small-study effect) and extreme heterogeneity. Bösch et al. read publication bias as the most parsimonious account of that combination.

Skeptical critiques and discussion

What critics argue. Bösch, Steinkamp and Boller (2006) argued that the RNG database shows a small-study effect and extreme heterogeneity consistent with selective publication rather than a genuine mental-influence effect — concluding the effect was “not proven”.

What other analyses report. Kugel (2011) challenged the Bösch et al. meta-analysis on inclusion grounds, arguing that merging ESP studies into a PK meta-analysis pulls the pooled result toward chance, and that removing certain data changes the recomputed figure — Kugel’s own recomputation excluding the 2004 Dobyns et al. data returned z = 3.59. The 2021 Journal of Scientific Exploration review and other primary reports in the set catalogue the positive individual and pooled results above.

Analysis. The disagreement is concrete and traceable to method choices: Bösch, Steinkamp and Boller (2006) and Kugel (2011) analyze overlapping databases and reach opposing pooled figures because they include and exclude different studies and treat the small-study relationship differently. Independent large preregistered confirmatory tests within the set land on both sides — Dechamps (2021) positive, Dechamps (2019) Study 3 null, Maier (2022) preregistered primary null. What has not been established here is a single pooled figure that survives across authors’ differing inclusion rules.

For the forensic side of this — how datasets that look “too strong” as well as “too weak” get audited — see Forensic critique of psi datasets.