Bayesian Analysis of Random Event Generator Data
⚡ Contested ↗📌 Appears in:
Plain English Summary
The PEAR lab at Princeton collected a massive dataset -- over 104 million trials -- claiming people could mentally nudge random number generators. The results looked statistically impressive using standard methods. But astronomer William Jefferys applied a different statistical lens called Bayesian analysis (which asks "how much should this evidence actually change my mind?") and found something striking: the evidence actually favors there being NO psychic effect. The classic statistical test was overestimating the strength of the evidence by at least 20 times. This is a famous statistical trap called the Jeffreys-Lindley paradox -- when you run enormous numbers of trials, tiny meaningless blips can look significant with standard tests. Jefferys argued these results shouldn't convince anyone who wasn't already a believer, and that parapsychology needs better statistical tools.
Research Notes
Foundational Bayesian critique of the REG/PK paradigm. Directly challenges the PEAR lab's flagship dataset and introduces the Jeffreys-Lindley paradox to parapsychological methodology debates. Key paper in controversies #8 (GCP/RNG) and #10 (meta-debate).
Applying Bayesian hypothesis testing to Jahn, Dunne & Nelson's (1987) PEAR random event generator dataset of 104.49 million trials reveals that the Jeffreys-Lindley paradox undermines the strong classical p-values reported. Under a uniform prior on the alternative hypothesis, the Bayes factor B = 12, actually increasing confidence in the null. For nearly all reasonable prior distributions on effect size, B exceeds 1 (favoring no anomaly). Even the most favorable prior yields B approximately 30 times larger than the classical p-value, showing the frequentist test overestimates significance by at least a factor of 20. Jefferys concludes these data are insufficient to shift the opinions of observers with even moderate priors, and advocates Bayesian methods as more appropriate for parapsychology.
Related Papers
Cites
Companion
- Parapsychological Research: A Tutorial Review and Critical Appraisal — Hyman, Ray (1986)
- False-Positive Effect in the Radin Double-Slit Experiment on Observer Consciousness as Determined With the Advanced Meta-Experimental Protocol — Walleczek, Jan (2019)
- Replication and Meta-Analysis in Parapsychology — Utts, Jessica (1991)
More in Skeptical
Cognitive Styles and Psi: Psi Researchers Are More Similar to Skeptics Than to Lay Believers
Searching for the Impossible: Parapsychology's Elusive Quest
False-Positive Effect in the Radin Double-Slit Experiment on Observer Consciousness as Determined with the Advanced Meta-Experimental Protocol
Cross-Examining the Case for Precognition: Comment on Mossbridge and Radin (2018)
N,N-Dimethyltryptamine and the Pineal Gland: Separating Fact from Myth
📋 Cite this paper
Jefferys, William H (1990). Bayesian Analysis of Random Event Generator Data. Journal of Scientific Exploration.
@article{jefferys_1990_bayesian_rng,
title = {Bayesian Analysis of Random Event Generator Data},
author = {Jefferys, William H},
year = {1990},
journal = {Journal of Scientific Exploration},
}