Bayesian Two-Way ANOVA
Bayesian Two-Way Analysis of Variance · Also known as: Bayesian factorial ANOVA, Bayes factor two-way ANOVA, Bayesian 2×k ANOVA, Bayesian two-factor ANOVA
Bayesian two-way ANOVA extends the classical two-way analysis of variance by replacing p-values with Bayes factors and posterior distributions. It quantifies evidence for or against main effects and their interaction using prior-weighted model comparison, yielding conclusions that are directly interpretable in probabilistic terms rather than relying on a fixed significance threshold.
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 Bayesian two-way ANOVA when you have a continuous outcome, exactly two categorical independent variables, and independent observations in each cell. It is especially valuable when you want to quantify evidence for the null hypothesis (something classical ANOVA cannot do), when sample sizes are modest and you want honest uncertainty propagation, or when repeated use of classical testing across many studies has caused concern about p-hacking. Do not use it when you lack a defensible prior or when reviewers or regulators require a classical p-value framework. Avoid it if cell counts are very unequal and you have no expertise in handling unbalanced Bayesian designs.
Strengths & limitations
- Can provide evidence in favour of the null hypothesis, distinguishing 'no effect found' from 'evidence of no effect'.
- Bayes factors are naturally calibrated for model comparison without multiple-testing corrections.
- Posterior distributions provide full uncertainty quantification, not just a binary reject/retain decision.
- Default JZS priors remove the need for subjective prior elicitation in most applied settings.
- Accumulation of evidence is coherent across studies — Bayes factors from independent replications multiply.
- Computationally more demanding than classical ANOVA; marginal likelihoods require numerical integration.
- Results depend on the choice of prior; sensitivity analysis is recommended to verify robustness.
- Less familiar to reviewers in many applied fields; may require additional justification in manuscripts.
- Software support, while growing (JASP, BayesFactor R package), is narrower than for classical ANOVA.
Frequently asked
How is this different from classical two-way ANOVA?
Classical two-way ANOVA produces F-statistics and p-values that quantify how surprising the data would be if an effect were absent, but cannot confirm the null. Bayesian two-way ANOVA produces Bayes factors that directly compare the evidence for competing models, and posterior distributions that describe likely effect sizes — including the possibility of concluding that an effect is probably absent.
Do I need to specify priors manually?
For most applications, no. The default JZS (Jeffreys-Zellner-Siow) prior implemented in packages like JASP and the R BayesFactor package provides a principled, data-scaled default. You should nonetheless verify that conclusions do not change substantially when the prior scale is varied.
What Bayes factor threshold should I use for significance?
Jeffreys's conventional scale is often used: BF10 > 3 = moderate evidence, > 10 = strong, > 30 = very strong, > 100 = decisive. These are guidelines, not hard cutoffs. Always report the actual BF and let readers judge the strength of evidence.
Can I use this with unbalanced designs?
Yes, but unbalanced cell counts complicate prior specification and interpretation. Main-effects Bayes factors can be sensitive to imbalance in ways that differ from classical Type III sums of squares. Consult Rouder et al. (2012) or JASP documentation for guidance on unbalanced Bayesian factorial designs.
Which software can run Bayesian two-way ANOVA?
JASP provides a point-and-click interface with the default JZS prior. The BayesFactor package in R (functions anovaBF and lmBF) offers the same analysis programmatically. Stan and brms support full custom Bayesian models for more complex factorial structures.
Sources
- Rouder, J. N., Morey, R. D., Speckman, P. L., & Province, J. M. (2012). Default Bayes factors for ANOVA designs. Journal of Mathematical Psychology, 56(5), 356–374. DOI: 10.1016/j.jmp.2012.08.001 ↗
- Jeffreys, H. (1961). Theory of Probability (3rd ed.). Oxford University Press. ISBN: 978-0198503682
How to cite this page
ScholarGate. (2026, June 3). Bayesian Two-Way Analysis of Variance. ScholarGate. https://scholargate.app/en/statistics/bayesian-two-way-anova
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 one-way ANOVAStatistics↔ compare
- Bayesian t-TestBayesian↔ compare
- Mixed ANOVAStatistics↔ compare
- Two-Way ANOVAStatistics↔ compare