Discovery Lab Answers

Roger D. Nelson, PhD — Effect Sizes Across Research Programs

Coverage note: The evidence table below reflects studies currently held in the ESP-Nexus library. The library’s share of the published literature on Nelson’s work has not been measured, so treat this as a summary of what the library holds rather than a settled account of the field.

The evidence spans two distinct research programs — the Princeton Engineering Anomalies Research (PEAR) lab’s random event generator (REG) work and the Global Consciousness Project (GCP) — and the metrics reported across them are genuinely different and non-comparable. The chart below plots the four results reported on the z_over_sqrt_n metric against publication year; the other metrics must be read from the table separately.

Experiments

Nelson spent more than two decades at PEAR, where he coordinated studies examining whether human intention could shift the output of REG devices. That program produced a large accumulated database: sources retrieved for this question cite 108 operators, 1,262 replications, and more than 5.6 million trials. Structural questions addressed in that corpus included the role of operator pairing, serial-position effects, and whether deterministic versus quantum-based random sources responded differently to intention.

Nelson founded the Global Consciousness Project in 1997, which shifted the question from individual intention to whether globally shared attention — during major world events — correlates with network-wide departures from randomness across a geographically distributed array of RNGs. The formal GCP experiment accumulated 500 pre-registered events from 1998 through approximately 2015. A parallel FieldREG line of work deployed portable REG units at sites Nelson characterized as ritually or collectively significant, including Egyptian sacred sites measured in 1997 and published later, and a 2024 study in the Khufu Subterranean Chamber.

Methodology

The PEAR REG protocol ran operators against a commercial or laboratory REG under three instructed intentions (HI, LO, and baseline), with pre-specified analysis of the HI-LO separation as the primary outcome. The GCP protocol differed structurally: events were pre-registered in a public registry before data analysis, the outcome measure was inter-RNG network variance (correlation among nodes rather than individual device output), and the hypothesis was that variance would depart from chance expectation during qualifying events. Nelson and colleagues have described the statistical test as operating on the squared network variance second by second across the event window, with permutation analysis used in some FieldREG applications to correct for empirical variance.

FieldREG studies added a further layer: a portable REG ran continuously during group activity at a site, and segments were defined by an independent index mark logged at the time of activity, then tested post hoc.

Data

The metrics across Nelson’s programs are not poolable into a single effect size, so they are reported below by program and metric type.

PEAR REG program

The PEAR benchmark HI-LO effect is on the order of 10⁻⁴ bits deviation per bit processed — very small per trial, achieving statistical significance through accumulation across an extremely large trial base.

Global Consciousness Project formal series

Null result (UNSETTLED SIGNAL — must be surfaced): Nelson (2011) reported z = 1.143 (null) when comparing Nelson-sourced versus other-sourced events within the recategorized GCP formal series — a within-corpus test that did not reach significance.

FieldREG / site studies

The 1998 exploratory applications aggregate (p =.059) is the one below-threshold result in the FieldREG record held here; Nelson (1998) labels this work exploratory, and the result is in a mixed direction.

The GCP correlation-type effect sizes (reported as ~0.33–0.4 in z/√k units across different time-slices of the formal series) are not directly comparable to the PEAR per-trial standardized effect sizes, nor to the FieldREG z/√N figures. The chart below shows the four z/√N results plotted against year with a fitted trend.

Skeptical critiques

What critics argue. Bösch, Steinkamp, and Boller (2006) conducted an independent meta-analysis of the PEAR-era mind-machine REG literature and reported an overall significant effect but also substantial file-drawer sensitivity — they estimated that a relatively modest number of unpublished null studies could reduce the composite to non-significance. They also flagged that effect sizes were heterogeneous across labs in ways inconsistent with a uniform underlying effect. On the GCP specifically, Jeffers (2006) argued that the event-selection and scoring rules were sufficiently flexible that the GCP hypothesis could not be falsified in any single test, and that the composite statistic is therefore difficult to interpret as a confirmatory result.

What the experimental data show. The GCP protocol does pre-register events in a public registry before analysis, and Nelson has described the 500-event formal series as operating under that pre-specified framework, with the composite z = 7.31. The 2008 and 2020 summaries report the cumulative effect growing monotonically across replications. The within-corpus comparison of Nelson-sourced versus other-sourced events (Nelson 2011) yielded z = 1.143 — null — which Nelson’s own reporting includes in the formal record.

Analysis. The Bösch et al. (2006) file-drawer estimate and the heterogeneity finding apply to the PEAR REG literature broadly, not specifically to GCP network data, which uses a different dependent measure (inter-node correlation rather than individual device output). Independent replication of the GCP design by a separate research team using a separate RNG network has not been reported in the sources retrieved for this question.

For Nelson’s full research record, see Roger D. Nelson.