Bayesian Moderated Mediation
Bayesian Moderated Mediation Analysis · Also known as: Bayesian conditional process analysis, Bayesian mediated moderation, Bayesian PROCESS model, Bayesian conditional indirect effect
Bayesian moderated mediation estimates how a mediator transmits the effect of a predictor onto an outcome, and whether that indirect effect varies in size depending on a moderator variable — all within a Bayesian framework that quantifies uncertainty via posterior distributions rather than p-values and confidence intervals.
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 moderated mediation when you have a theoretically motivated conditional process model and want richer uncertainty quantification than bootstrap confidence intervals provide. It is especially valuable when the sample is moderate in size (say n = 100–300), when you have substantive prior knowledge to incorporate, or when you need to compare effect sizes across multiple moderator values simultaneously using the posterior distribution. It is also appropriate when the indirect effect distribution is likely non-normal — a case where frequentist bootstrap intervals can have poor coverage. Do NOT use it when the causal ordering of X, M, and W is ambiguous (the design must justify the directionality), when data are very sparse for the number of parameters (prior choice then dominates), or when a simpler frequentist bootstrap approach (PROCESS macro) adequately addresses the research question and reviewers are unfamiliar with Bayesian reporting conventions.
Strengths & limitations
- Provides a full posterior distribution for the conditional indirect effect, enabling direct probability statements about its size at any value of W.
- Naturally propagates uncertainty through the product-of-coefficients without requiring a separate bootstrap or delta-method step.
- Can incorporate prior knowledge from previous studies, improving estimation precision in small to moderate samples.
- Handles non-normal indirect effect distributions more gracefully than asymptotic methods.
- Readily extended to multilevel or multivariate outcome structures within the same Bayesian framework.
- Requires explicit prior specification; poorly chosen priors can distort results, particularly in small samples.
- Computationally intensive — Hamiltonian Monte Carlo can take minutes to hours for complex models.
- Bayesian reporting conventions (HDI, Bayes factors, posterior probabilities) are less familiar to many applied researchers and reviewers.
- Causal interpretation of indirect effects depends entirely on the research design; the Bayesian framework does not resolve confounding.
Frequently asked
How does Bayesian moderated mediation differ from the PROCESS macro approach?
Both estimate conditional indirect effects, but PROCESS uses ordinary least squares and bootstrapped confidence intervals, while Bayesian moderated mediation samples from the full posterior. The Bayesian approach yields a direct probability statement about the indirect effect and handles non-normality of the a×b product more naturally, at the cost of requiring prior specification and greater computational effort.
Do I need a large sample for Bayesian moderated mediation?
Not necessarily. One advantage of the Bayesian approach is that with informative priors based on prior literature, estimation can be stable with samples as small as n = 80–100. However, with flat priors the practical requirements are similar to those of frequentist bootstrap mediation, typically n ≥ 100–200 for reliable indirect effect estimation.
Which software implements Bayesian moderated mediation?
The brms R package (an interface to Stan) allows moderated mediation models to be specified as multivariate regression systems with full Bayesian inference. Alternatively, the model can be coded directly in Stan or JAGS. For simpler cases, BayesMed and the mediation R package with Bayesian options are also available.
How do I report Bayesian moderated mediation results?
Report the posterior median (or mean) and 95% highest-density interval for each path coefficient and for the conditional indirect effect at substantively meaningful values of the moderator. Include the R-hat convergence statistic and effective sample size. A plot of the conditional indirect effect against W with its HDI ribbon is highly informative.
What is the difference between moderated mediation and mediated moderation?
Both describe a model where mediation and moderation co-occur, but they emphasise different quantities. Moderated mediation focuses on how the indirect effect changes across levels of W (the conditional indirect effect is the primary estimand). Mediated moderation focuses on whether a moderation effect is itself explained by a mediator. In practice both can arise from the same statistical model; the distinction is conceptual and guides which parameters to highlight.
Sources
- Yuan, Y. & MacKinnon, D. P. (2009). Bayesian mediation analysis. Psychological Methods, 14(4), 301–322. DOI: 10.1037/a0016972 ↗
- Hayes, A. F. (2013). Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. Guilford Press. ISBN: 978-1609182304
How to cite this page
ScholarGate. (2026, June 3). Bayesian Moderated Mediation Analysis. ScholarGate. https://scholargate.app/en/statistics/bayesian-moderated-mediation
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 Moderation AnalysisStatistics↔ compare
- Moderated MediationStatistics↔ compare
- Path AnalysisStatistics↔ compare
- Structural Equation ModelingResearch Statistics↔ compare