ReBayes

A trial reports P = 0.04. How likely is it that the effect is real? ReBayes takes the result as the paper printed it and answers that question — the one a P value cannot.

Coming to the App Store Use it in a browser Support

A trial's result as the paper printed it: an odds ratio of 0.8 with its 95% confidence interval, and what the trial set out to show. — Enter the estimate and its confidence interval — or paste, scan, or read a PDF

A 37% prior gives a 72% probability of a real effect. The curve shows how that answer changes with the prior, and the prior the result would need to reach 95%. — Is it real? The answer for your prior, and for every other prior

Two bell curves: what the data would look like if the null hypothesis were true and if the postulated effect were true, with the observed result marked between them. — Evidence: where the observed result sits between the two hypotheses

Under a sceptical prior there is a 99% probability of some benefit and a 76% probability it is at least as large as the smallest worthwhile benefit. — How big? The same trial read by a sceptic, an enthusiast and someone with no view

The same entry form for a continuous outcome: a difference in means of 2 with a confidence interval from 1 to 3. — Differences in means work the same way as ratios

Fragility index: 96 of 230 patients with the outcome in one arm, 67 of 235 in the other. Changing the outcome of 10 patients would make the result non-significant. — Fragility: how many patients' outcomes would have to change

Every patient as a dot, five per dot, with the ten whose outcomes would have to change ringed. — The whole trial on one screen, and the ten patients the result rests on

How the P value climbs as outcomes are flipped, for this trial and for the same trial at larger sizes. — How the result comes apart, and what a bigger trial would look like

The question a P value cannot answer

Ask a room of clinicians what P = 0.04 means and many will say there is a 4% chance the result is a fluke. It does not mean that. A P value is the probability of seeing results like these if there were no real effect; it says nothing about how probable a real effect is. To get from one to the other you need to say how plausible the effect was before the trial — and then let the trial's evidence update it. That is what ReBayes does.

What you give it

The three numbers every paper prints: the estimate, its confidence interval, and the effect the investigators said they were looking for when they planned the study. You can type them, paste a sentence from the paper, share a PDF into the app and choose from the results it finds, or point the camera at the page.

What you get back

Is it real?

The probability that the effect is real given the data, for whatever level of prior belief you choose — and the level of belief the result would need in order to convince you.

How strong is the evidence?

The likelihood ratio: how much more likely these results are under a real effect than under none, in plain words as well as numbers.

How big is it?

The effect restated in patients per hundred, and what a sceptic, an enthusiast and someone with no view at all would each conclude from the same trial.

How fragile is it?

How many patients' outcomes would have to change for the result to lose significance, and whether a negative trial has actually shown "no effect".

Written for the reader you have

ReBayes writes out what a result means in two versions: one in plain English for a colleague, a journal club or a note in the chart, and one in technical terms for a paper or a referee, with the citation. They are the same numbers, said to different readers. Copy whichever suits.

Where the method comes from

The calculations follow published methods — Colquhoun's false-positive risk, Held's reverse-Bayes analysis, Matthews' critical prior interval, Spiegelhalter's sceptical and enthusiastic priors, and the fragility index — and every one is cited inside the app, with the working shown. The method is described in Jones PM, Martin J. BJA Education 2026;26:10–19.

ReBayes is a teaching and reading aid for interpreting published research. It is not a medical device and does not make or support decisions about the care of any individual patient.

Support · Privacy policy