Discovery Lab Answers

Dean I. Radin, PhD — Data Summary

Coverage note: The studies in the evidence table reflect what the ESP-Nexus library currently holds — the library’s share of the published literature on Radin’s work has not been measured. Treat what follows as a summary of the library’s holdings, not a settled account of his complete research record or of the broader field.

The evidence table below your answer covers 66 studies and 81 result rows spanning multiple research programs. The metrics across those rows are not comparable — standardized effect sizes, raw hit rates, z-scores, p-values, and other correlational measures appear together — so no single pooled figure can be drawn across them.

Experiments

Radin’s work in the sources retrieved spans several distinct lines of research:

• Global Consciousness Project (GCP) analyses — examining whether network-wide random number generator (RNG) behavior shows anomalous deviations during large-scale collective human events, specifically New Year’s Eve transitions. A 2025 paper analyzed GCP data from 1998–2025 across all time zones at the stroke of midnight; a 2026 corrigendum corrected a variance inflation error in the PCA composite used in that analysis.

• Water pH “excess correlation” studies — testing whether isolated beakers of water show correlated pH shifts when one is subjected to magnetic field stimulation, with a human intention component. The 2025 paper was framed explicitly as a preliminary independent replication of a previously published claim.

• Double-slit optical mind-matter interaction — a cumulative analysis across 30 experiments at five laboratories, cited in the evidence table via Kauffman (2023).

• Electroencephalographic (EEG) and physiological correlate studies — multiple rows in the evidence table concern EEG slope differences (labeled EH3/EH4 trend analyses) comparing observed versus randomized conditions, and a 2022–2023 set of studies examining EEG deconvolution and signal analysis across sessions.

• Language and population-scale sentiment analyses — examining whether RNG deviation slopes correlate with “sad day” markers across ten languages.

• Collaborations — several rows involve co-investigators. Helané Wahbeh is a co-author on multiple entries in the evidence table. Alibalaei (2025) and Yu (2024) appear as lead-author studies in the table, with Radin as a collaborator. King (2022) similarly reflects collaborative work. Kauffman (2023) references a multi-lab cumulative analysis.

Methodology

GCP analyses: Data from the GCP network’s distributed RNGs are aggregated by UTC time zone, weighted by active RNG count per second, and analyzed using Stouffer Z-score combination across time zones. A key feature is the nonparametric permutation baseline: each second’s deviation is compared against the same second across all other days of the year (up to 366 comparisons), with a sequential randomness test used to exclude days with temporal dependence before analysis. Smoothing (30-day moving window for long-term trends; shorter windows for the midnight analysis) and a principal components analysis composite across multiple complexity measures are also used. The 2026 corrigendum identified that the PCA composite had been constructed from correlated components, inflating variance by approximately 1.67-fold; the corrected analysis used a parametric bootstrap to adjust the p-value.

Water pH studies: Experimental and control runs were conducted using beakers separated by varying distances (1 m to 10 m), with magnetic field stimulation applied via Arduino-controlled halos. A third non-stimulated control beaker was placed between local and remote beakers in some configurations. Analysis used linear mixed-effects modeling, with a bias-corrected and accelerated (BCa) nonparametric bootstrap applied to address autocorrelation concerns flagged by Durbin-Watson statistics. Autocorrelation was a recurring issue: the BCa procedure substantially changed results in some run sets, converting a significant interaction to a non-significant one.

Double-slit and EEG studies: The evidence rows indicate large trial counts (e.g., 85,792,278 trials in one experiment) and session-based designs (up to 470 sessions across 47 participants), suggesting automated data collection at scale.

Data

The 81 result rows span incompatible metrics, so the table below organizes by study rather than pooling. Nine rows report null or below-chance results; the remainder report positive directions. This is the distribution the library holds — not a claim about the full literature.

Unsettled signals the evidence requires surfacing:

• Nine rows report null or below-chance outcomes. The Alibalaei (2025) yoga-course precognition study returned a near-zero effect (Cohen’s d = −0.027) at p = .783. The Wahbeh (2021) quick remote viewing study found overall null performance across 534 participants. Wahbeh (2022) found below-chance extended perception with a negative Hedges’ g.
• The Radin (2023) single-participant concentration study went in the below-chance direction (z = −3.0), a psi-missing result the evidence table flags directly.
• Effect sizes and metrics are heterogeneous across programs and not reducible to a single summary figure.
• The 2026 corrigendum is itself a direct example of within-program correction: the originally reported PCA midnight result was based on inflated variance; after correction, the p-value moved from the originally reported figure to p = 0.0013, which Radin describes as “not as significant as originally reported” while noting the midnight segment remained the most statistically significant of all 100 time points tested.

  Skeptical critiques

  The sources retrieved for this question do not include named published criticisms of Radin’s work with the specific citation form required to attribute them here. The water pH study records one methodological concern internally: the Durbin-Watson statistic repeatedly indicated excess autocorrelation in the ΔpH measures across run sets, and when the BCa bootstrap correction was applied to address this, at least one Phase × Condition interaction that initially showed significance (p < 0.001) was reduced to a non-significant result (p = 0.111). Radin flags this directly in the paper as a limitation bearing on interpretation. Similarly, the 2026 corrigendum documents a self-identified error in PCA construction — correlated components inflating variance — and presents the corrected, weaker figures in place of the original.

  For published external criticism of Radin’s research programs, the ESP-Nexus page for Dean I. Radin, PhD at https://esp-nexus.org/scientists/dean-i-radin/ carries the fuller treatment of that material.