Overview

Precognition means knowing about the future before it can be known by any ordinary means. The information cannot come from memory, reasoning, or the five senses. That is what makes the claim so radical. If true, an effect somewhere in the present depends on a cause that has not happened yet.

Researchers split the topic into several types. Spontaneous cases are premonitions or dreams people report on their own, outside any experiment. Forced-choice studies ask a person to predict which of a few targets a machine will later pick. Free-response studies ask a person to describe a future scene in their own words. Presentiment studies skip conscious guessing entirely and measure whether the body reacts before a stimulus appears.²

The modern laboratory work traces back to J. B. Rhine’s Duke laboratory in the 1930s, which ran card-guessing tests. Helmut Schmidt later moved the targets onto electronic machines. Daryl Bem’s 2011 paper reignited the whole debate by running it inside mainstream psychology journals.¹⁴

Why This Is Hard to Study

The first problem is the size of the claimed effect. If precognition is real, it is tiny. A forced-choice meta-analysis pegged the average effect at roughly one part in fifty above chance. That is far smaller than most everyday psychology effects.¹² An effect this small can be drowned out by noise. It can also be faked by very minor flaws in how a study is run.

Effect size is the standard way to describe how big a result is. A large effect is obvious at a glance. A near-zero effect like this one only becomes detectable when you add many studies together. So researchers lean heavily on meta-analysis, a method that pools many separate experiments into one combined estimate. Pooling lets small signals show up. It also pools every weakness of every included study.

The second problem is telling precognition apart from ordinary explanations. Maybe the person picked up a faint cue from the equipment. Maybe the random target selection was not truly random and a clever participant learned the pattern. Maybe the experimenter’s expectations leaked in. These are the central confounds. A confound is any hidden factor that could produce the result without any psychic ability.²

The third problem is replication. A finding only counts in science if independent labs can reproduce it. Precognition results are inconsistent across labs. Defenders say the effect is real but weak and elusive. Skeptics say weak and elusive is exactly what a non-existent effect looks like. This disagreement runs through every section below.

The Experiments and How They Work

The oldest design is forced-choice. The person tries to predict which target a machine will randomly select next. Because the target is chosen after the guess, there is no normal way to know it. Across many decades and many labs, the combined result sits slightly above chance.¹²

Random target selection matters enormously here. If the sequence of targets is even slightly predictable, a participant could learn it and beat chance with no psychic ability at all. Modern studies use electronic random number generators and pick the target only after the response is locked in.¹²

Bem’s 2011 study took a different route. He took well-known psychology effects and ran them backwards in time, so the cause came after the response. For example, people seemed to respond faster to a word they would only be shown a fraction of a second later.¹⁸ The designs were familiar to mainstream psychologists. That is exactly why the paper caused such a storm.

Dream studies form a fourth strand. Participants try to dream about an image that will be randomly chosen later, then rate how closely a set of pictures matches the dream. One recent home-dream study reported that people rated the future target image as more similar to their dream than the decoys.⁹

Individual studies in the reference library

The studies listed below are individual precognition results currently held in the ESP-Nexus reference library beyond those already discussed above. What share of the published literature on precognition they represent has not been measured, so the table summarizes what the library holds rather than counting what has been published. The studies also report different kinds of number — hit rates, standardized effect sizes, z values, Bayes factors and p values alone — and those cannot be added together into one bottom-line figure.

Both directions are well represented in the table. The positive results include the two meta-analytic pools — 309 forced-choice studies cumulating to z = 11.41,¹⁹ and 22 precognition–clairvoyance study pairs in which both task types cumulated to a significant effect²⁰ — along with Schmidt’s selected subjects,²¹ Vernon’s preregistered precall result,²⁵ and the DAX stock-index prediction series.²⁶ The null results include the largest preregistered tests in the set: Walleczek and colleagues’ three confirmatory replications, whose first study returned 49.48% against a 50% chance level across 37,836 trials,³³ Muhmenthaler, Dubravac, and Meier’s three replications with effect sizes of d = 0.001, 0.034, and −0.03,²⁹ the joint proponent–skeptic priming replications whose preplanned confirmatory hypotheses failed to reach significance,²⁸ the multi-lab retroactive-avoidance replication in which the data favored the null,²⁷ and an associative remote viewing project predicting currency moves that ended at a 48% hit rate.²³

Several of the papers qualify their own results directly. Watt and Valášek report that the hit rate of the original Watt (2014) dream study is inflated by the omission of the dropouts’ data, and that the combined 30.6% hit rate is significant by the planned one-tailed test.²² Kugel reports that the betting program scored above chance while the human subjects scored below chance.²⁴ Bancel, Boban, Bensahra, and Varvoglis report that their two preregistered cohort studies found no direct evidence for a psi effect, with the elevated session variance appearing in tryout data that were collected with the same software and procedures but preregistered as exploratory.³² Walleczek and colleagues note that their Study 2 result ran significantly below chance and that Study 3, with 217,800 trials, did not replicate it.³³

The Body Reacting Early

Presentiment is the most physiological strand. Instead of asking people to guess, researchers wire them up and measure the body. The idea is simple to state. If a computer will randomly show either a calm or an emotional picture, does the body react differently in the seconds before the calm or emotional picture appears?¹

The measures include skin conductance, heart rate, pupil size, and brain electrical activity. Skin conductance rises when you are aroused, even without you noticing. The presentiment claim is that this rise begins before an emotional image, not just after it.¹ Because the picture is randomly chosen after the recording starts, there should be no way for the body to know what is coming.

For the two most recent classes of studies, 85 of 101 studies from 25 different laboratories produced results in the direction predicted by a retrocausal effect.³

Researchers have extended presentiment to other measures and even other species. Eye-tracking work reported anticipatory pupil and blink changes before emotional images.⁴ One animal study reported that Bengalese finches showed alarm behavior seconds before a snake video appeared.⁷ These extensions are intriguing but rest on small samples.

What Researchers Think Is Happening

It is important to separate the data from the explanations. The data are the measured effects, with their sizes and p-values and flaws. The explanations are guesses about why the effects might exist. These are very much contested, and no consensus interpretation has emerged.

One proposed mechanism is retrocausation: information genuinely flowing backward in time. Defenders argue that time symmetry already exists in the equations of physics, so a future event influencing the present is not as absurd as it sounds.³ One review proposed that short-lead presentiment and long-lead conscious precognition may need different mechanisms entirely.²

A second family of explanations invokes quantum mechanics. Some authors argue presentiment fits interpretations where conscious observation collapses a quantum state, producing only retrospective correlations rather than real-time prediction.¹⁰ These models are mathematical proposals, not settled physics. The full quantum formalism is beyond what can be translated faithfully here, but the central move is the same: the future event is fixed only when an observer’s mind perceives it.¹¹

A third proposal reframes all psi as nonlocal correlations rather than signals. On this view, classical experiments mistake a correlation for a transmitted signal, which would explain why effects fade on replication.¹⁵ Each of these is a candidate explanation. None is established.

Skeptical Critiques and Discussion

Critique 1: The presentiment meta-analyses include low-quality studies and statistical artifacts that manufacture the effect.

Schwarzkopf argues the pooled studies contain serious flaws, including an fMRI study with multiple-comparison errors and conference papers that were never peer-reviewed.⁸

The meta-analysts say they excluded post-hoc experiments to avoid hand-picked data, computed a large file-drawer threshold, and named expectation bias and multiple analyses as the key challenges they tried to control for.⁶

Analysis. Schwarzkopf points to specific weaknesses: a 2:1 imbalance between control and target trials that could let participants learn the odds, and baseline-correction procedures that may create apparent pre-stimulus effects in slow physiological signals.⁸ Where a pooled study carries a documented flaw such as an uncorrected multiple-comparison error, that study yields no conclusion either way. The Mossbridge group reports controls against expectation bias; whether those controls fully address the baseline-artifact concern is something the two papers describe differently.

Critique 2: Bem’s 2011 result vanishes under a proper Bayesian reanalysis.

Critics, including the Wagenmakers reanalysis referenced in the field disputes, argued that a Bayesian analysis (which weighs how strongly the data shift belief, given a prior) gave little or no support for precognition once realistic priors were used.¹⁷

A later meta-analysis of 90 Bem-style experiments reported a Bayes factor far above the threshold for “decisive evidence,” both with and without Bem’s own studies.¹³ A defender’s commentary noted that 69 independent replications combined to a significant outcome, disputing the claim that replications uniformly failed.¹⁶

Analysis. The dispute turns on choices that are not purely empirical. A Bayesian conclusion depends on the prior probability assigned before seeing the data, and skeptics assign precognition a very low prior because it appears to violate known physics. Bem, Tressoldi, Rabeyron, and Duggan report decisive Bayes factors under their chosen priors; the skeptical reanalyses report weak support under theirs. An independent, adversarially preregistered replication run by both camps together has not been published.

Critique 3: Parapsychology’s claims cannot be true because they violate physics, so the data are irrelevant.

Reber and Alcock argued psi is impossible a priori and stated plainly that they did not examine the data because “the data are irrelevant.”¹⁶

Williams responds that this dismisses evidence without examining it, and notes that remote viewing reportedly works inside Faraday cages and underwater, which argues against a simple electromagnetic-signal model and against inverse-square-law objections.¹⁶

Analysis. This exchange is about method, not a single result. Reber and Alcock argue from physical principle; Williams argues that refusing to look at data is not a scientific stance. Williams also notes that parapsychology effect sizes (roughly 0.02 to 0.28) fall in the same range as many mainstream psychology effects.¹⁶ The reader can weigh whether an a-priori impossibility argument or an empirical-examination argument carries more force here.

Critique 4: Scientific method itself cannot settle the question, so persistent small effects prove nothing.

Rabeyron argues that if psi is real, the observer cannot be cleanly separated from the observed, which breaks the assumptions the scientific method needs.¹⁷

The same paper reports that meta-analyses still show small significant effects for precognitive dreams, telepathy, and presentiment, and suggests the decline effect may reflect psi’s elusive nature rather than fraud.¹⁷

Analysis. This is an unusual argument: it is partly sympathetic to psi yet pessimistic about ever proving it. Rabeyron describes a paradox in which replicated hypotheses suppress the very effect under study. Skeptics read the decline effect as evidence the phenomenon was never there. Both readings are present in the published exchange, and the decline-effect interpretation is not resolved by the data alone.

Critique 5: The specific protocols behind the positive meta-analyses produce null results when rerun as large preregistered replications.

Walleczek and colleagues ran three preregistered confirmatory replications of Bem’s erotic-image experiment, totaling 26,483 participants and 420,472 critical trials, and report that Study 1 failed to replicate the effect: 49.48% correct against a 50% chance level (p = 0.979), where Bem had reported 53.1%.³³ Muhmenthaler, Dubravac, and Meier replicated three of Bem’s experiments with samples of 727, 1,414, and 1,395 participants and report no evidence for precognition in any of them, while the forward control versions produced the classic psychology effects.²⁹

The Bem, Tressoldi, Rabeyron, and Duggan meta-analysis reports that the pooled 90-experiment database is positive on both its frequentist and Bayesian measures.¹³ Schlitz and coauthors — a joint proponent–skeptic team that included Bem — report an exploratory result consistent with the original effect among Experiment 1’s English-language participants (t(177) = 2.08, p = .02, d = 0.16), even though their preplanned confirmatory hypotheses failed to reach significance.²⁸

Analysis. The published record splits along protocol lines. Databases pooled across decades report small positive aggregates,¹⁹¹³ while the preplanned confirmatory analyses of the large preregistered replications — the multi-lab retroactive-avoidance test (BF₀₁ = 4.38),²⁷ the two joint priming replications,²⁸ the three Bem replications by Muhmenthaler’s group,²⁹ and Walleczek’s three erotic-trial studies³³ — report null outcomes. Walleczek’s own sequence adds a caution against reading any single significant result too quickly: Study 2 returned a significantly below-chance rate (49.65%, p = 0.013) that Study 3, with 217,800 trials, did not reproduce (50.07%, p = 0.496).³³

Open Questions: What Would Settle This

The cleanest test would be a large, preregistered, multi-lab presentiment study designed jointly by proponents and skeptics. Preregistration means the hypotheses, sample size, and analysis are locked in publicly before any data are collected. That removes the worry that an interesting result was found by trying many analyses after the fact.

Such a study should fix the trial imbalance Schwarzkopf flagged, so that target and control trials are equally likely and participants cannot learn the odds.⁸ It should also settle on a single baseline-correction method in advance, since that procedure is one of the contested sources of apparent pre-stimulus effects. An adversarial design, where both camps agree the result will count, is the missing piece.

A second open question is whether the body’s pre-response tracks the future external stimulus or the person’s own future internal reaction. The EEG finding that pre-responses predicted a future button press but not the stimulus type points toward the second possibility.² A preregistered study that separates these two could clarify what presentiment even is, before anyone argues about retrocausation.

A third question is the experimenter effect. The planarian study found avoidance only when a person actively ran the apparatus.⁵ A design that blinds and rotates experimenters, while logging who ran each session, could test whether human presentiment effects depend on the participant or the experimenter. Until experiments like these are run and published, the small consistent effects and the methodology disputes will continue to sit side by side.