A Bayesian evaluation of the experimental evidence for parapsychology

Overview

It was not a field investigation, a witness narrative, or a new experimental demonstration of paranormal ability. Rather, it used Bayesian reasoning to reassess how experimental results offered in support of parapsychological effects should be interpreted, particularly in the dispute surrounding Daryl Bem’s reported precognition experiments. The central issue was whether small frequentist p-values or a collection of nominally significant studies should count as strong evidence when the proposed effect conflicts with well-established expectations and when flexible analytic decisions, selective reporting, and publication bias are plausible concerns.

Bayesian analyses express comparison between models through Bayes factors and require explicit prior assumptions; consequently, they make the role of prior plausibility visible instead of leaving it implicit. This can lower the apparent evidential force of positive psi results under skeptical priors, but it does not mechanically prove that psi is impossible. Critics have argued that strongly skeptical priors prejudge the question, while proponents have argued that repeated experimental anomalies deserve more neutral or empirically informed priors.

The controversy therefore turns on model specification, the relevant evidential threshold, the quality and independence of the underlying experiments, and what should be inferred from later replication attempts. Its importance is historical and methodological: the case became a widely discussed illustration of the replication-era challenge to conventional null-hypothesis significance testing. No paranormal effect should be treated as established on the basis of this reanalysis or the contested experiments it discussed.

Chronology.

The relevant immediate background was the publication and circulation in 2011 of experimental claims that some participants’ responses appeared to anticipate, or be influenced by, events selected or presented after the response. These claims were controversial because the proposed causal direction conflicted with ordinary temporal assumptions, while their reported statistical support was expressed largely through familiar significance-testing conventions.

During 2011, Wagenmakers and collaborators were associated with a Bayesian critique or reassessment of the experimental evidence invoked in this debate. The recalled subject title should be checked against the original text, because closely related titles, commentaries, supplements, and later discussions may be conflated in secondary retellings.

In the following years, the debate broadened beyond the particular psi experiments into arguments about replication, optional stopping, analytic flexibility, publication bias, prior distributions, and the limits of p-values. Later accounts often use the episode as a methodological case study, which can obscure distinctions between a critique of one evidential analysis, a criticism of a research program, and a claim about whether psi phenomena exist.

People, organisations, and setting.

The recalled institutional setting is the University of Amsterdam in the Netherlands, where Wagenmakers and some collaborators were associated with research on Bayesian statistical methods and psychological measurement. The setting matters principally as an academic and disciplinary one: the work concerned published datasets, statistical models, and interpretive standards rather than a haunted location, laboratory séance, or community tradition.

Daryl Bem is relevant as the psychologist whose precognition-related experiments formed an important public focus of the surrounding controversy. Wagenmakers and collaborators are relevant as methodological critics or reanalysts, while later commentators include psychologists, statisticians, replication researchers, parapsychologists, and philosophers of science. Journals, peer review, conference discussion, blogs, teaching materials, and press coverage were also part of the transmission environment in which the analysis acquired influence.

Reported experimental phenomena.

The contested phenomena were reported behavioral deviations in controlled psychological tasks, not uncontrolled sensory experiences reported by witnesses. Recalled descriptions of the broader program include participants making choices, reacting to visual or affective stimuli, recalling material, or completing priming-style tasks in arrangements where an allegedly relevant stimulus, selection, or training event occurred later in the experimental sequence. The claimed interpretation was that later events could influence earlier behavior or physiological response, although ordinary statistical fluctuation, expectancy effects, procedural artifacts, and selective analysis provide non-paranormal alternatives.

Sensory material in the disputed paradigms could include images, words, screen-based prompts, emotionally valenced content, and response cues, depending on the particular experiment. Observable behavior was typically recorded as a choice, reaction time, memory score, preference, accuracy rate, or aggregate difference between conditions. The Bayesian reanalysis did not independently observe a participant sensing the future; it evaluated how much the reported numerical patterns should shift belief between a null account and one or more effect accounts.

Descriptions of these experiments can become rhetorically compressed into claims that participants literally perceived future events. That wording exceeds what the measurements directly establish. At most, the studies reported group-level associations interpreted as anomalous, with the causal and ontological explanation remaining disputed.

Analytic and investigation history.

The investigation was a statistical reanalysis rather than a fresh replication. Its likely workflow involved taking reported test statistics or experimental outcomes, defining competing models, assigning prior distributions to effect-size parameters, and calculating Bayes factors or related posterior quantities. A Bayes factor compares the relative predictive performance of models for observed data; it is not itself a direct probability that psi exists, and its meaning depends on the models and priors selected.

The key methodological contribution was to force questions that conventional significance testing can leave underdescribed. These include how plausible a nonzero anomalous effect was before data collection, how large such an effect was expected to be, whether multiple studies were independent confirmations, and whether the reported literature was conditioned by unreported tests or analytic choices. The conclusions attributed to the analysis should be verified from the original article, including exactly which experiments, summary statistics, priors, and sensitivity checks it used.

Later replication activity and methodological discussion are relevant contextual evidence but should not be collapsed into the original Bayesian calculation. Failed or attenuated replications can arise from a nonexistent original effect, inflated initial estimates, design differences, inadequate power, questionable research practices, or genuine context sensitivity. Conversely, a positive replication would still require scrutiny of preregistration, blinding, data handling, and cumulative evidence.

Disagreements and contested inferences.

The main disagreement concerns prior probability. Skeptical analysts may reason that apparent information transfer from the future conflicts with extensive background knowledge and therefore demands exceptionally diagnostic evidence. Critics of that position reply that assigning an extremely low prior can make any feasible laboratory result look unconvincing and can turn methodological evaluation into an undeclared metaphysical decision.

Both positions require their assumptions to be stated rather than treated as self-evident.

A second disagreement concerns the proper alternative model. A simple point null of exactly zero may not represent the practical alternatives of measurement error, laboratory artifacts, researcher degrees of freedom, selective reporting, or small conventional effects. Likewise, a broad psi alternative may spread its predictive probability across many possible effect sizes and thereby compare unfavorably even if it is philosophically possible.

Sensitivity analysis over reasonable priors and models is therefore more informative than quoting one Bayes factor as a final verdict.

There is also a dispute over communication. Popular retellings sometimes portray the Bayesian critique as a mathematical disproof of paranormal phenomena, or portray the original experiments as an empirical overthrow of causality. Neither simplification accurately reflects a contested inference from limited psychological data.

The responsible conclusion is that the claimed effects and the evidential methods remain open to critical examination.

Transmission, genre, and commercial influences.

This case belongs to the genre of methodological controversy and replication-era scientific debate. It spread through scholarly articles and replies, preprints or supplementary analyses, statistics education, professional discussion, popular science reporting, blogs, and online social media. Each layer favors different compression: technical discussions focus on priors and likelihoods, while mass-media accounts often favor the striking phrase that people can feel or know the future.

Commercial and reputational incentives may have shaped transmission without establishing misconduct by any named party. Extraordinary claims draw public attention, citations, invitations, sales of accessible science writing, and media coverage, while forceful debunking can also attract attention and professional visibility. Journal novelty preferences and academic career rewards for positive findings can amplify initial reports, whereas later null replications or corrections may receive less attention outside specialist communities.

Subsequent narratives should distinguish the original experiments, the Bayesian reassessment, responses by advocates and critics, and later replication literature. A citation to this episode may function as a shorthand for broader concerns about psychology’s statistical culture rather than as a careful account of the exact analysis. Researchers should recover the primary texts before treating a retelling as evidence for any particular technical assertion.

Cross-case connections and comparative motifs.

The strongest comparative motifs are extraordinary-claim evidential thresholds, disputes over prior plausibility, significance testing versus model comparison, and the difference between statistical anomaly and demonstrated mechanism. These motifs connect the case to other contested literatures involving weak effects, high researcher flexibility, and strong prior theoretical constraints, but such analogies do not make those subjects duplicates.

A further motif is temporal reversal. Claims that a later event affects an earlier response recur in precognition, presentiment, retroactive priming, and time-reversed learning narratives. In comparative work, it is important to record whether the evidence is behavioral, physiological, anecdotal, or merely interpretive, and whether the proposed anomaly survives preregistered prospective testing.

Another motif is the contrast between controlled experimental settings and later folklore-like transmission. Although the source material is laboratory research, the headline implication can become a culturally portable story about people sensing the future. This transmission pathway can magnify certainty, erase statistical qualifications, and encourage links to paranormal traditions that were not directly tested by the analysis.

Limits, alternative explanations, and research cautions.

This dossier is a recalled synthesis and does not establish the exact contents, numerical results, or wording of the 2011 analysis. The canonical title may refer to a publication, a section, a commentary, or a commonly used description of an associated Bayesian critique. The original article, data provenance, target experiments, and subsequent responses must be checked before extracting quantitative conclusions.

Mundane explanations for apparently anomalous positive results include sampling variation, regression to the mean, multiple testing, optional stopping, selective exclusion rules, analytic flexibility, coding or randomization problems, expectancy effects, publication bias, and ordinary design weaknesses. These explanations are not assertions that any particular error occurred in a named experiment. They are rival mechanisms that a cumulative evidence review should actively test.

Bayesian methods improve transparency when models and priors are fully reported, but they do not remove judgment. A defensible assessment should preregister analytic choices where possible, show robustness across justified prior ranges, examine raw data and protocols, include registered replications, report null findings, and avoid treating an isolated evidential metric as a direct measurement of reality. No reliable basis is present here for asserting verified paranormal causation.

Chronology

People and roles

Connections to explore

Unretrieved reference leads

1. A Bayesian evaluation of the experimental evidence for parapsychology.

   E.-J. Wagenmakers and collaborators. · Suggested methodological analysis.

   This is the canonical recalled lead and should be checked for its exact title, authorship, datasets, models, priors, and conclusions.

   Suggested search: Wagenmakers Bayesian evaluation experimental evidence parapsychology 2011 article.

2. Why Psychologists Must Change the Way They Analyze Their Data: The Case of Psi.

   E.-J. Wagenmakers and collaborators. · Suggested related commentary.

   This likely related item may clarify the precise relationship between the Bayesian critique and Bem’s experiments.

   Suggested search: Wagenmakers Wetzels Borsboom van der Maas case of psi Bem 2011.

3. Feeling the Future: Experimental Evidence for Anomalous Retroactive Influences on Cognition and Affect.

   Daryl Bem. · Suggested target experimental report.

   This provides the experimental claims that formed an important context for the later evidential dispute.

   Suggested search: Daryl Bem Feeling the Future anomalous retroactive influences cognition affect.

4. Later replies and replication discussions concerning Bayesian analyses of psi.

   Various authors. · Suggested debate record.

   Responses are needed to assess disagreements about priors, model choices, data selection, and the interpretation of cumulative evidence.

   Suggested search: Bayesian psi Bem Wagenmakers reply replication prior sensitivity.