Roger D. Nelson, PhD Sources: off on

PEAR Lab: Operator Intention and REG/RNG Anomalies

For more than a decade, Roger Nelson worked alongside Robert G. Jahn and Brenda J. Dunne at Princeton’s Engineering Anomalies Research (PEAR) laboratory, coordinating a systematic program to test whether human intention could produce statistically detectable deviations in the output of electronic random event generators (REGs). The resulting dataset, spanning millions of binary trials across hundreds of operators, constitutes one of the largest and most methodologically scrutinized bodies of evidence in experimental parapsychology. Nelson’s contributions ranged from experimental design and data architecture to specialized statistical analyses that probed the structure of anomalous effects.

Deeper dives — Nelson:

Back to hub Healing Intention and Energy Medicine Consciousness-Correlated Anomalies Meta-Analysis and Statistical Methods FieldREG and Collective Consciousness Mass Consciousness Events: September 11 Global Consciousness Project: Network 📖 Connected

Key findings

The 12-year PEAR REG program (91 operators, over 1,000 experimental series) produced a cumulative effect in the high-intention direction with z ≈ 3.8, p ≈ 0.0001, and a mean effect size of approximately 1 part in 10,000 bits per trial.¹
Anomalous deviations were consistently asymmetric: high-intention and low-intention series diverged from baseline in the predicted directions, while calibration (baseline) runs remained statistically flat.¹
Individual operator contributions were highly variable; a small subset of operators accounted for a disproportionate share of the cumulative effect, a pattern documented in large-database analyses.²
The Random Mechanical Cascade (RMC), a physical, non-electronic device, showed operator-related anomalies consistent with the electronic REG results, providing a cross-device replication within the same laboratory.³
The MegaREG experiment, designed with a vastly larger per-session trial count to improve statistical power, produced null results, raising unresolved questions about whether high-volume protocols suppress the effect.⁴
ANOVA decomposition of the REG database identified operator intention as the primary source of anomalous variance, with secondary parameters (run length, feedback type, local sidereal time) contributing smaller but detectable modulations.⁵

Overview

The PEAR laboratory was established at Princeton University‘s School of Engineering and Applied Science in 1979 under the direction of Robert G. Jahn, with Nelson joining in 1980 as research coordinator. The program’s founding statement articulated a rigorous engineering-science approach to anomalous human-machine interactions, framing the research question in terms of measurable statistical deviations rather than metaphysical claims.⁶ The core experimental paradigm placed human operators, unselected volunteers with no claimed psychic ability, in front of electronic REGs and asked them to hold a mental intention for the device to produce more high bits (HI condition), more low bits (LO condition), or to run without intention (BL baseline). Operators worked in their own time, at their own pace, generating data over months or years. This design choice, unselected operators, self-paced sessions, pre-stated intention direction, was central to the program’s claim that any observed effects were not artifacts of experimenter selection of gifted subjects.

A key Type-II vulnerability for this research area is the small effect size involved: approximately 1 part in 10,000 bits per trial. At this magnitude, even large datasets require careful power analysis, and individual studies are chronically underpowered to detect the effect reliably. This means that null replications, including the MegaREG experiment, cannot straightforwardly falsify the hypothesis; they may simply reflect insufficient sensitivity under altered protocol conditions.

Program Origins and Experimental Philosophy

The 1980 program statement⁶ outlined PEAR’s commitment to engineering-grade instrumentation and statistical rigor. The REG devices used hardware noise sources (typically electronic shot noise or radioactive decay) to generate binary sequences at rates of hundreds to thousands of bits per second. Operators were not pre-screened for psychic ability; the program explicitly sought to characterize the distribution of effects across an unselected population. The first large-database REG report⁷ established the data-collection infrastructure that would support the subsequent decade of research, including automated logging, operator-blind calibration runs, and pre-stated intention protocols designed to address optional stopping, operators declared their intended direction before each series began, eliminating post-hoc assignment of outcomes to conditions.

The PEAR Operator-REG Program

Nelson’s earliest published contributions to the PEAR REG program focused on building the data infrastructure and characterizing operator-level variability. The 1984 technical report on large-database REG anomalies documented operator-related deviations across a growing pool of participants, establishing that the effect, while small, was reproducible within the laboratory across different operators and different experimental runs.⁸ By 1988, analyses of individual operator contributions revealed that the aggregate effect was not uniformly distributed: some operators produced consistent, replicable deviations while others produced near-chance results, and a small number produced effects in the direction opposite to intention.²

Individual Operator Variability in the Large Database

The 1988 analysis of individual operator contributions² examined the PEAR database as it stood at that point, decomposing the aggregate effect by operator. The analysis found that the distribution of operator-level z-scores was broader than expected by chance, consistent with genuine individual differences rather than a uniform population effect. A subset of operators, described as producing “consistent” results, accounted for a substantial portion of the cumulative deviation. The authors noted that this heterogeneity was itself an anomalous finding: if the effect were artifactual (e.g., due to a systematic equipment bias), it would not be expected to vary systematically with operator identity across sessions conducted months apart. The competing non-psi explanation considered was equipment drift or calibration artifact; this was addressed by comparing operator runs against interleaved calibration (BL) runs on the same device, which remained statistically flat. The artifact was mitigated but not fully eliminated, as the calibration runs were not always conducted in identical physical conditions to the intention runs.

The 1991 ANOVA technical report extended this analysis to the full database structure, applying analysis-of-variance models to decompose the sources of variance across operators, intentions, run lengths, and secondary parameters.⁹ This work was methodologically significant because it moved beyond simple mean-shift testing to ask whether the anomalous variance was structured, whether it appeared in theoretically predicted places (the HI-LO contrast) rather than randomly distributed across conditions.

ANOVA Decomposition of the REG Database

The 1991 ANOVA report⁹ applied multi-factor ANOVA to the accumulated PEAR REG database, treating operator intention (HI, LO, BL), operator identity, run length, and feedback type as independent variables. The primary finding was that the HI-LO contrast, the difference between high-intention and low-intention means, was the dominant source of anomalous variance, with F-ratios exceeding chance expectation. Secondary parameters (run length, feedback modality) showed smaller modulations. The analysis addressed the multiple-comparisons artifact by distinguishing pre-specified primary contrasts (HI vs. LO) from exploratory secondary analyses; the primary contrast remained significant after this distinction. The 2000 follow-up ANOVA paper⁵ extended these models to the full 12-year database, confirming the structure of the effect and providing specialized subsidiary analyses of secondary parameters including local sidereal time, an exploratory variable whose inclusion was not pre-specified and whose significance should be treated as hypothesis-generating rather than confirmatory.

The Twelve-Year Review

The 1997 review paper, co-authored by Jahn, Dunne, Nelson, Dobyns, and Bradish, synthesized the entire 12-year PEAR REG program into a single comprehensive analysis covering 91 operators, over 1,000 experimental series, and more than 100 million binary trials.¹ The cumulative result showed a small but consistent mean shift in the high-intention direction, with the HI-LO separation achieving statistical significance at a level that would be extremely unlikely by chance. The researcher’s preferred interpretation was that human intention produced a genuine, if small, influence on quantum-mechanical noise processes; alternative interpretations, including publication bias, optional stopping, and equipment artifact, were discussed but not fully resolved.

Twelve-Year Cumulative Statistics and Effect Structure

The 1997 review¹ reported that across 91 operators and 1,262 experimental series, the mean bit-score in the HI condition exceeded the theoretical mean of 100 by approximately 0.0001 bits per trial (roughly 1 part in 10,000), while the LO condition fell below by a comparable margin. The cumulative z-score for the HI-LO contrast was approximately 3.8 (p ≈ 0.0001, two-tailed). Calibration runs, conducted by operators without intention, remained statistically indistinguishable from chance (z ≈ 0.2), addressing the equipment-drift artifact by demonstrating that the same devices, operated by the same people in the same room, produced flat distributions when intention was absent. The optional-stopping artifact was addressed by the pre-stated intention protocol: operators declared HI, LO, or BL before each series began, so outcomes could not be retrospectively assigned to favorable conditions. However, the stopping rule for the overall 12-year program was not pre-specified, leaving open the question of whether the program would have been reported differently had the cumulative effect been smaller at an earlier stopping point.

MegaREG and Replication Attempts

The MegaREG experiment was designed explicitly to test whether the PEAR REG effect would replicate under conditions of dramatically increased statistical power per session.⁴ By generating vastly more bits per session than the standard PEAR protocol, MegaREG was intended to either confirm the effect with high confidence or rule it out at the effect sizes previously reported. The result was null, no significant deviation from chance in either the HI or LO condition, a finding the authors themselves reported and analyzed in detail, considering both the possibility that the effect is real but suppressed by high-volume protocols and the possibility that the original results were statistical artifacts.

MegaREG Protocol, Results, and Interpretive Challenges

The MegaREG experiment⁴ used a REG device generating bits at a substantially higher rate than the standard PEAR protocol, with each session accumulating orders of magnitude more trials than a typical PEAR series. The rationale was straightforward: if the effect size is approximately 1 part in 10,000, then a sufficiently large N should detect it reliably. The null result (effect sizes near zero, z-scores near chance) was interpreted by the authors in two ways. First, they considered the “protocol sensitivity” hypothesis: that the standard PEAR protocol’s pacing, short bursts of trials with natural pauses, may be a necessary condition for the effect, and that continuous high-volume generation disrupts whatever process underlies it. Second, they acknowledged that the null result is consistent with the original effects being statistical artifacts of optional stopping or multiple comparisons in the smaller-N standard protocol. The authors did not resolve this ambiguity, and the MegaREG null result remains an unresolved challenge to the PEAR REG literature. The competing non-psi explanation, that the original PEAR results reflected optional stopping across the 12-year accumulation, was partially addressed by the pre-stated intention protocol but not fully eliminated, as the program-level stopping rule was never pre-specified.

Mechanical and Physical Extensions

Nelson and colleagues extended the operator-intention paradigm beyond electronic REGs to physical devices, most notably the Random Mechanical Cascade (RMC) and a linear pendulum apparatus. The RMC, a device in which polystyrene balls fall through a matrix of pegs and accumulate in bins, producing a physical approximation of a binomial distribution, showed operator-related anomalies consistent with the electronic REG results.³ The pendulum experiment tested whether operators could influence the damping rate of a physical oscillator through intention.¹⁰ These extensions were methodologically important because they addressed the artifact hypothesis that the electronic REG effects were due to subtle electromagnetic interference from operators, a mechanism that would not apply to purely mechanical systems.

Random Mechanical Cascade: Cross-Device Replication

The 1988 RMC study³ reported data from operators attempting to shift the distribution of balls in the cascade toward the right (HI), left (LO), or without intention (BL). The device used approximately 9,000 polystyrene balls per run, falling through a 330-peg matrix into 19 collecting bins. The distribution of balls across bins was compared to the theoretical binomial distribution and to the BL calibration runs. The reported effect was a small but statistically significant shift in the mean bin position in the intended direction, with the HI-LO contrast achieving p-values comparable to those in the electronic REG program. The electromagnetic-interference artifact was addressed by the purely mechanical nature of the device, no electronic components were involved in the randomization process itself, only in the counting of balls in bins. This artifact was substantially mitigated, though not fully eliminated, as the counting electronics could in principle be influenced. The authors noted that the RMC effect size was consistent with the electronic REG effect size, suggesting a common underlying phenomenon rather than device-specific artifacts.

Linear Pendulum Experiment: Damping Rate Effects

The 1994 pendulum study¹⁰ tested whether operators could influence the damping rate of a linear pendulum, the rate at which oscillation amplitude decays over time, through pre-stated intention. Operators attempted to increase damping (faster decay), decrease damping (slower decay), or held no intention. The study reported small but statistically detectable differences in damping rate between the HI and LO conditions. The mechanical nature of the system addressed electromagnetic-interference artifacts. The study was smaller in scale than the main REG program, and the effect size and sample context are not fully specified in the available pool documentation; the result should be treated as exploratory and hypothesis-generating rather than confirmatory.

Statistical Architecture and Variance Analysis

A distinctive feature of Nelson’s contributions to the PEAR program was sustained attention to the statistical architecture of the database, not merely asking whether an effect existed, but asking where in the data structure the anomalous variance was located and what secondary parameters modulated it.⁵ The 2000 ANOVA paper represented the most comprehensive application of this approach, applying specialized subsidiary analyses to the full 12-year dataset to characterize the effect’s dependence on run length, feedback type, operator experience, and other secondary variables.

ANOVA Models and Secondary Parameter Analysis

The 2000 ANOVA paper⁵ applied multi-factor ANOVA models to the complete PEAR REG database, treating the HI-LO-BL intention factor as the primary variable and a range of secondary parameters as additional factors. Key findings included: (1) the primary HI-LO contrast remained the dominant source of anomalous variance across all model specifications; (2) run length showed a modest interaction with the intention effect, with shorter runs producing slightly larger per-trial effect sizes, a pattern consistent with the MegaREG null result but not definitively explaining it; (3) feedback modality (visual, auditory, none) did not significantly modulate the effect; (4) local sidereal time showed a suggestive interaction in exploratory analyses, but this variable was not pre-specified and the finding should be treated as hypothesis-generating only. The multiple-comparisons artifact was addressed by distinguishing the pre-specified primary contrast from the exploratory secondary analyses, though the large number of secondary variables examined increases the risk of spurious findings among the secondary results.

Modern Context

The PEAR operator-intention program intersects mainstream methodology in three load-bearing places. First, the underlying physical-randomness testing regime against which any anomalous-influence claim is evaluated is the responsibility of the NIST Statistical Test Suite (SP 800-22), which defines the canonical battery of tests for random and pseudorandom generators used in cryptographic and scientific contexts. Any operator-influence claim must demonstrate departure from a baseline that itself passes the NIST suite over comparable run-lengths.¹¹ Second, the PEAR program’s long N×trial structure (hundreds of millions of bits across decades) addresses mainstream concerns about statistical power for small effect sizes — Cohen’s framework predicts that sub-percent-of-a-percent shifts require precisely the kind of cumulative-N scale PEAR amassed to reach reliable detection thresholds.¹² Third, the researcher-degrees-of-freedom literature (Simmons, Nelson, & Simonsohn 2011 and the broader open-science movement) raises the bar for credible inference: even with adequate N, undisclosed analytic flexibility can convert null effects into nominally significant ones. PEAR’s evolving documentation of protocol, pre-specification, and operator-blinding addresses this concern unevenly across the corpus, with Bancel & Nelson’s 2008 design-paper representing the most explicit pre-registration-like artifact.¹³

Skeptical Critiques and Discussion

Critique 1: Claim: The PEAR REG effects are artifacts of optional stopping and lack of pre-specified stopping rules at the program level

Skeptic source: A persistent methodological concern about the 12-year PEAR program is that the cumulative effect could reflect optional stopping at the program level: the program ran until a significant result accumulated, and would have been reported differently, or not at all, had the effect been smaller at an earlier point. This critique applies to the program-level stopping rule, which was never pre-specified.¹

Response: The PEAR team addressed optional stopping at the series level through the pre-stated intention protocol: operators declared their intended direction (HI, LO, or BL) before each series began, preventing post-hoc assignment of outcomes to favorable conditions.¹ The calibration runs, conducted by the same operators on the same devices without intention, remained statistically flat throughout the program,¹ addressing equipment-drift as an alternative explanation. However, the program-level stopping rule was not pre-specified, leaving the optional-stopping critique at the program level only partially addressed. The MegaREG null result⁴ is consistent with both the optional-stopping hypothesis and the protocol-sensitivity hypothesis; the authors did not resolve this ambiguity.

Analysis. The optional-stopping critique centers on whether the cumulative 12-year PEAR effect could reflect program-level stopping rules that were never pre-specified. The PEAR protocol cites the pre-stated-intention requirement at the series level (operators declared HI, LO, or BL direction before each series began) and the flat calibration-run record across the program as addressing the series-level form of the artifact and equipment drift. The program-level stopping question — and the interpretive ambiguity of the MegaREG null result, which is consistent both with the optional-stopping account and with the team’s protocol-sensitivity alternative — remains an active methodological item in the literature.

Critique 2: Claim: The MegaREG null result falsifies the PEAR REG effect at the reported effect size

Skeptic source: The MegaREG experiment was designed with sufficient statistical power to detect the effect size reported in the 12-year review. Its null result has been interpreted as evidence that the original PEAR effects were statistical artifacts rather than genuine phenomena.⁴

Response: The MegaREG authors themselves, including Nelson, reported and analyzed the null result, considering both the artifact interpretation and the protocol-sensitivity hypothesis.⁴ The ANOVA analysis of the standard-protocol database showed that run length interacted with the intention effect, with shorter runs producing slightly larger per-trial effect sizes,⁵ providing a specific mechanistic account of why high-volume protocols might suppress the effect. The cross-device replication in the Random Mechanical Cascade³, which used a fundamentally different randomization mechanism, is not easily explained by the artifact interpretation, though it too was conducted under standard-protocol (not high-volume) conditions.

Analysis. Critics question whether the null result from the high-volume MegaREG protocol falsifies the original PEAR effect size. The MegaREG paper (which includes Nelson among the authors) reports the null result and discusses both the artifact interpretation and the protocol-sensitivity hypothesis; an ANOVA of the standard-protocol database is cited as showing a run-length-by-intention interaction in which shorter runs yielded slightly larger per-trial effects. Whether the protocol-sensitivity account survives a preregistered adversarial test — and how to weight the Random Mechanical Cascade replication, which used a different randomization mechanism under standard-protocol conditions — are open questions in the published record.

References

Jahn, R. G., Dunne, B. J., Nelson, R. D., Dobyns, Y. H., & Bradish, G. J. (1997). Correlations of random binary sequences with pre-stated operator intention: A review of a 12-year program. Journal of Scientific Exploration, 11(3), 345–367. R001 [Jahn 1997] ↩︎
Brenda J. Dunne, Roger D. Nelson, & York H. Dobyns (1988). Individual Operator Contributions in Large Data Base Anomalies Experiments. Princeton Engineering Anomalies Research, Princeton University, School of Engineering/Applied Science. R002 [Dobyns 1988] ↩︎
Dunne, B., Nelson, R., & Jahn, R. G. (1988). Operator-Related Anomalies in a Random Mechanical Cascade. Journal of Scientific Exploration, 2(1), 155–179. R003 [Dunne 1988] ↩︎
Dobyns, Y. H., Dunne, B. J., Jahn, R. G., & Nelson, R. D. (2004). The MegaREG Experiment: Replication and interpretation. Journal of Scientific Exploration, 18, 369–397. R004 [Dobyns 2004] ↩︎
Nelson, R. D., Jahn, R. G., Dobyns, Y. H., & Dunne, B. J. (2000). Contributions to variance in REG experiments: ANOVA models and specialized subsidiary analyses. Journal of Scientific Exploration, 14(1), 73-89. https://www.scientificexploration.org/docs/14/jse_14_1_nelson.pdf R005 [Nelson 2000] ↩︎
Jahn, R. G., Dunne, B. J., & Nelson, R. D. (1980). Princeton Engineering Anomalies Research. Program Statement. Princeton University, School of Engineering/Applied Science. R006 [Jahn 1980] ↩︎
Dunne, B. J., Jahn, R. G., & Nelson, R. D. (1981). An REG Experiment with Large Data-Base Capability. Princeton Engineering Anomalies Research, Princeton University, School of Engineering/Applied Science. R007 [Dunne 1981] ↩︎
Nelson, R. D., B. J. Dunne, & R. G. Jahn (1984). An REG experiment with large data base capability, III: Operator related anomalies. Princeton Engineering Anomalies Research Laboratory, Princeton Univ. School of Engineering/Applied Science. R008 [Nelson 1984] ↩︎
Nelson, R.D., Dobyns, Y.H., Dunne, B.J., & Jahn, R.G (1991). Analysis of variance of REG experiments: operator intentions, secondary parameters data base structure. Princeton Engineering Anomalies Research. R009 [Nelson 1991] ↩︎
Nelson, R. D., Bradish, G. J., Jahn, R. G., & Dunne, B. J. (1994). A linear pendulum experiment: Effects of operator intention on damping rate. Journal of Scientific Exploration, 8(4), 471-489. https://global-mind.org/pend.html R010 [Nelson 1994] ↩︎
Rukhin, A., Soto, J., Nechvatal, J., Smid, M., Barker, E., Leigh, S., et al. (2010). A statistical test suite for random and pseudorandom number generators for cryptographic applications (NIST Special Publication 800-22 Rev 1a). National Institute of Standards and Technology. https://nvlpubs.nist.gov/nistpubs/Legacy/SP/nistspecialpublication800-22r1a.pdf R011 [NIST 2010] ↩︎
Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences (2nd ed.). Lawrence Erlbaum Associates. https://doi.org/10.4324/9780203771587 R012 [Cohen 1988] ↩︎
Simmons, J. P., Nelson, L. D., & Simonsohn, U. (2011). False-positive psychology: Undisclosed flexibility in data collection and analysis allows presenting anything as significant. Psychological Science, 22(11), 1359–1366. https://doi.org/10.1177/0956797611417632 R013 [Simmons 2011] ↩︎

Deeper dives — Nelson: