Bayesian Wilcoxon Signed-Rank Test
Also known as: Bayesian signed-rank test, Bayesian nonparametric paired comparison, Benavoli signed-rank Bayesian test, signed-rank Bayesian hypothesis test
The Bayesian Wilcoxon signed-rank test is a Bayesian nonparametric method for comparing two paired or related samples. Rather than returning a single p-value, it produces posterior probabilities that one condition is better, practically equivalent, or worse than the other, enabling richer and more interpretable inference for paired continuous or ordinal data without assuming normality.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use this test when you have paired or repeated-measures data and either the normality assumption for a paired t-test is questionable or sample sizes are small, and you want a Bayesian interpretation rather than a classical p-value. It is ideal when practical equivalence matters — when knowing whether the groups are 'essentially the same' is as important as detecting a difference. Do not use it when observations are independent (use Bayesian Mann-Whitney U instead), when you need a strict frequentist p-value for regulatory or pre-registered hypothesis testing contexts, or when the sample is so small that the prior dominates and sensitivity analysis becomes unwieldy.
Strengths & limitations
- Produces intuitive posterior probabilities for 'better', 'equivalent', and 'worse', avoiding binary accept/reject dichotomy.
- Nonparametric: does not require normality of differences, suitable for ordinal or skewed continuous data.
- Explicitly models the region of practical equivalence (ROPE), integrating practical and statistical significance.
- Provides a coherent Bayesian update: results can be sequentially refined as more data arrive.
- Closed-form Dirichlet posterior makes computation fast and transparent.
- The ROPE boundaries must be chosen by the researcher and results can be sensitive to that choice.
- The Dirichlet prior concentration parameter requires justification; default non-informative choices may be inappropriate for small samples.
- Less widely implemented than classical tests; requires specialist software or custom coding in R or Python.
- The interpretation of posterior probabilities as 'evidence' can still be misunderstood by readers unfamiliar with Bayesian inference.
Frequently asked
What is the ROPE and how do I set it?
The Region of Practical Equivalence (ROPE) defines a range of signed-rank weight differences so small that you consider the two conditions practically the same. The boundaries are set by the researcher based on domain knowledge — for example, a performance difference below 1% in classification accuracy. Sensitivity analysis across several ROPE widths is strongly recommended.
How does this differ from the classical Wilcoxon signed-rank test?
The classical test produces a p-value testing whether the median difference is zero, using a fixed significance threshold. The Bayesian version produces three posterior probabilities (left, rope, right), incorporates prior information, and allows a richer conclusion including explicit probability of practical equivalence.
What prior should I use?
The symmetric Dirichlet prior with a small concentration parameter (e.g., 0.5 or 1) is commonly used to reflect near-ignorance. When previous studies exist, an informative prior can be elicited. Always report what prior was used and run a sensitivity analysis.
Can I use it with ordinal data?
Yes. Because the method operates on ranks rather than raw values, it is appropriate for ordinal paired data, provided the signed-rank structure is meaningful — i.e., the direction of each paired difference is interpretable.
Is software available?
The 'bayesplot' and 'BayesianFirstAid' R packages and the Python 'baycomp' library (by Benavoli and colleagues) implement this test. JASP also supports Bayesian nonparametric paired comparison.
Sources
- Benavoli, A., Corani, G., & Mangili, F. (2014). Should we really use post-hoc tests based on mean-ranks? Journal of Machine Learning Research, 17(5), 1–10. link ↗
- Benavoli, A., Corani, G., Demsar, J., & Zaffalon, M. (2017). Time for a change: a tutorial for comparing multiple classifiers through Bayesian analysis. Journal of Machine Learning Research, 18(77), 1–36. link ↗
How to cite this page
ScholarGate. (2026, June 3). Bayesian Wilcoxon Signed-Rank Test. ScholarGate. https://scholargate.app/en/statistics/bayesian-wilcoxon-signed-rank-test
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Bayesian Mann-Whitney U testStatistics↔ compare
- Paired samples t-testStatistics↔ compare