Here's the uncomfortable position precognition research has been in for fifteen years: the most rigorous attempt to replicate its flagship findings mostly failed, and yet the statistical anomaly at the center of the original work has never been fully explained away.
That tension is worth sitting with, because it tells you something real about where edge science gets stuck.
What the Original Claim Actually Was
In 2011, Daryl Bem published nine experiments in the Journal of Personality and Social Psychology — a mainstream, high-prestige outlet — reporting evidence for what he called "retroactive influence": the idea that future stimuli could measurably affect present responses. The FORRT-archived summary of that work describes the core finding: across more than 1,000 participants, the mean effect size across all nine experiments was d = 0.22, with all but one experiment reaching statistical significance. Participants who scored high on stimulus-seeking showed a mean effect size of d = 0.43.
Those are not enormous effects. But they're not trivial either, and they appeared in a peer-reviewed journal that doesn't typically publish parapsychology. The controversy that followed wasn't just about whether precognition exists — it was about what the findings revealed about psychological research methods more broadly.
A PeerJ analysis of the Bem controversy makes this explicit: critics like Wagenmakers and colleagues argued that the methodological practices Bem used — multiple analyses, flexible stopping rules, selective reporting — were common across mainstream psychology. If Bem's work was flawed, so was a lot of other published psychology. The precognition debate became a proxy war for the replication crisis.
The Meta-Analysis That Didn't Settle It
A subsequent meta-analysis by Bem and colleagues, covering 90 precognition studies, attempted to show cumulative evidence for a real effect. The headline number was striking: an overall effect exceeding six sigma, with a Bayes Factor described as "decisive evidence."
Daniel Lakens's methodological critique — written as a pre-publication peer review — is worth reading carefully, because it models exactly the kind of epistemic discipline this field needs. His core point: of 90 studies, only 18 produced statistically significant precognition effects, from just 7 different labs. Seventy-two studies found nothing. The meta-analytic signal depends heavily on how publication bias is handled, and Lakens argues the correction methods used were insufficient to rule it out as the primary driver of the apparent effect.
This is not a dismissal of the data. It's a demand for better accounting. And it's the right demand.
The Publication Bias Question Gets More Complicated
Here's where things get genuinely interesting. A 2026 study in the Journal of Scientific Exploration — published just weeks ago — directly challenges the assumption that publication bias is the main story in psi research. The researchers analyzed 165 published ESP studies (243 experiments) alongside 40 preregistered confirmatory experiments from the Koestler Parapsychology Unit Registry, then compared null-result rates to mainstream psychology.
Their finding: the rate of published null outcomes in ESP research is higher than in mainstream psychology, for both preregistered and non-preregistered experiments. Publication bias in parapsychology appears to be substantially smaller than in the field that spent years criticizing it.
The authors are careful not to overclaim. They write that while the bias is smaller, they "cannot conclusively show that the statistical evidence for ESP is not an artifact of that relatively minimal bias" — and they note that earlier file-drawer analyses complicate the picture further. This is the epistemic honesty the field needs more of: a result that cuts against a common assumption, presented without triumphalism.
What it does do is complicate the easy dismissal. If the standard critique — "it's all publication bias" — turns out to be less applicable to parapsychology than to psychology generally, then the residual anomaly in the data needs a different explanation. Either there's a real effect, or there's a methodological artifact we haven't correctly identified yet. Those are both scientifically interesting possibilities.
Where This Leaves the Field
Precognition research is in a peculiar position: it has a replication problem and a publication-bias problem that may be smaller than its critics assumed, a flagship study that generated a methodological crisis in mainstream psychology, and a statistical signal that hasn't been explained away despite fifteen years of trying.
That's not evidence for precognition. It's evidence that the question is harder to close than either side wants to admit.
The most useful thing the field could do now is exactly what the Journal of Scientific Exploration study models: preregistered designs, transparent null-result reporting, and methodological critiques that engage the data rather than dismissing the question. The Bem controversy didn't prove precognition exists. But it did prove that "this is obviously impossible" is not a research methodology.
Watch for whether the Koestler Unit's preregistered experiment registry continues to grow — that's the infrastructure that would actually let us distinguish signal from noise, if either is there to find.
