How sure would you want to be that the effect is real before acting on it? Pick a target, and the app works out the prior belief that would get you there.
Two bell curves: what the data would look like if the null were true, and if the postulated effect were. The red line is what this study actually observed — the closer it sits to a curve's peak, the better that hypothesis explains the result.
How the evidence depends on the postulated effect
Download as PDFExplanation of graph
Every probability below is read against this; set the wrong way, a 99% chance of benefit reads as a 1% one.
These probabilities usually run higher than PEGD, and both can be right: PEGD asks whether an effect exists at all, while this analysis assumes some effect and asks how big it is. Compare PEGD with P(worthwhile), not with P(benefit).
All three priors, in numbers
Vague: as if you brought no opinion at all. This row is the study's own result, restated.
Skeptical: starts by doubting the effect. This is the row to quote to a doubter.
Enthusiastic: starts by expecting the effect the trial was powered for.
In absolute terms: patients per 100
The same number read as an odds ratio, relative risk, or hazard ratio gives different absolute risks, so say which the paper reported — there is no default. The control-arm risk is usually in the abstract.
Interpretation
Statistical details
Likelihood ratio
PEGD (Probability of a [real] effect given the data) versus the null
Required prior
Please cite this page if you find it useful:
Jones PM. Bayesian Reanalysis of Biomedical Research, version 1.8, https://shiny.seaturtle.site/bayesian_reanalysis/ Last accessed 2026-08-29
The following ratios are for the experimental arm relative to the control arm
Odds ratio Relative riskAbove 1 the outcome was more common in the experimental arm and below 1 less common; which of those you want depends on the outcome, which this calculator does not know.
It is a floor: a reader who thought a real effect less likely than not beforehand faces a higher risk than this.
How many patients' outcomes would have to change for this result to stop being statistically significant?