The GCP hypothesis predicts that inter-RNG correlations and deviations from expected entropy will increase during periods of global collective attention. The challenge is that “collective attention” is not operationally defined in advance with sufficient precision to eliminate post-hoc event selection as a confound. The formal experiment (1998–2015) used a pre-specified list of events, partially addressing this concern. The broader exploratory analyses, including the New Year’s Eve studies, use data-driven methods (entropy, fractal dimension, PCA) applied after the fact, which increases multiple-comparisons risk. Radin acknowledges the exploratory nature of these extended analyses.¹²

Radin’s 2023 analysis in Physics Essays examined the full GCP database through 2022, not just the pre-selected event windows.¹ The formal experiment (1998–2015) had already found a highly significant result; the extended analysis asked whether anomalous entropic effects were confined to event windows or present more broadly. The finding that non-random structure appeared beyond the formal event windows is exploratory and raises the possibility that either (a) collective attention effects are more pervasive than the event-window model predicts, or (b) the GCP network itself has systematic non-random properties not fully captured by calibration. The researcher’s preferred interpretation is (a); alternative interpretation (b) remains unresolved. No independent replication of the full-database analysis by a separate research group is available in the pool.

The Nelson, Radin, Shoup, and Bancel analysis compared daily inter-RNG correlations for 250 days classified as major news events in 2001 versus 115 non-eventful days, finding a larger average RNG correlation on news days (p = 0.011).⁵ The September 11 result, largest daily variance change and correlation of the year, was identified post-hoc within this dataset. The non-psi alternative of data-dredging (multiple implicit comparisons across 365 days to identify the most extreme) was partially addressed by the pre-specified news-day vs. non-news-day comparison, but the identification of September 11 as the single most extreme day was not pre-specified. The researchers’ preferred interpretation is that collective emotional focus on that date produced a genuine physical anomaly; the alternative is that the extreme value is a statistical outlier expected by chance in a large dataset.

The corrigendum published in the Journal of Scientific Exploration in 2026 identified that the PCA composite score reported in the 2025 New Year’s Eve paper had been computed under the assumption that the five constituent metrics, correlation dimension, permutation entropy, Higuchi fractal dimension, BDS statistic, and autocorrelation, were independent.³ Because these metrics are mutually inter-correlated, the first principal component (a weighted linear combination of correlated measures) does not follow a standard normal distribution with independent measures. The originally reported z ≈ −4.9 was therefore incorrect. The corrected statistic is smaller. This correction was identified by two external JSE readers and acknowledged promptly by Radin. The non-PCA results from the same paper, mean shifts, entropy, fractal dimension analyzed individually, were not affected by this correction. The episode illustrates both the value of post-publication review and the specific artifact of assuming independence among correlated composite metrics.

The Radin and Atwater study used three RNGs, two based on electronic noise and one on radioactive decay latencies, continuously recording during 14 workshops (coherence condition) and 8 control periods (building unoccupied) at the Monroe Institute.⁴ The specific artifact of building occupancy (vibration, electromagnetic interference from people present) was partially addressed by comparing workshop periods to control periods when the building was unoccupied rather than occupied by non-meditating groups, meaning the comparison confounds mental coherence with simple physical presence. The researchers acknowledged this limitation. No effect size or p-value is reported in the pool extract for this study; the paper is characterized as providing exploratory evidence. The researcher’s preferred interpretation is that binaural-beat-entrained coherence produced the anomalous deviations; the alternative of physical confounds from group presence was partially addressed but not eliminated.

The Mason, Patterson, and Radin study collected 94 hours of RNG data across 33,927 trials (each trial = 1,000 random bits in 10-second periods) during Transcendental Meditation group sessions in a large meditation hall.⁸ The study was designed to explore whether group meditation, as a venue for collective coherence, would produce anomalous RNG deviations. The specific artifact of demand characteristics (experimenters aware of meditation periods) was partially mitigated by automated data collection. The artifact of multiple implicit comparisons across 33,927 trials was not fully addressed. No pre-specified primary outcome or stopping rule is reported in the pool. The study is characterized as exploratory.

In a 2023 review essay, Radin examined dual-aspect monism, as developed by Atmanspacher and Rickles, as a metaphysical framework that proposes reality consists of a single undifferentiated substance splitting into mind and matter as dual aspects, linked by meaning through acausal correlations rather than direct causal influence.⁹ Radin’s preferred interpretation is that this framework is compatible with GCP-type results: if mind and matter are acausally correlated at a deep level, then collective mental states might be associated with anomalous physical patterns without requiring a classical causal mechanism. The proposed mechanism remains contested; no experimental test capable of distinguishing dual-aspect monism from other interpretations has been conducted. Alternative interpretations, including that GCP results are statistical artifacts, do not require any revision to standard physical theory.

Modern Context

The GCP’s statistical methodology sits within a broader mainstream debate about how to detect small effects in large, heterogeneous datasets and how to guard against post-hoc analysis inflation. The question of whether inter-RNG correlations during world events reflect genuine anomalies or artifacts of flexible event-window definition is structurally similar to debates in mainstream psychology about researcher degrees of freedom and p-hacking. The 2026 corrigendum to the New Year’s Eve PCA analysis³ illustrates precisely the kind of composite-metric independence assumption that mainstream statisticians have flagged as a recurring source of inflated significance claims across many fields. The GCP’s use of pre-specified event lists for its formal experiment (1998–2015) represents a partial response to these concerns, but the extended exploratory analyses remain vulnerable to the same critiques that apply to exploratory data mining in any domain.¹