Causal Mediation Analysis (Natural Direct and Indirect Effects)
Also known as: natural direct effect, natural indirect effect, NDE / NIE decomposition, counterfactual mediation, Nedensel Arabuluculuk Analizi (NDE / NIE)
Causal mediation analysis is a counterfactual framework that splits a treatment's total effect into a Natural Direct Effect (NDE) and a Natural Indirect Effect (NIE) that runs through a mediator. The modern general approach was formalised by Pearl (2001) and Imai, Keele and Tingley (2010), giving the decomposition a precise causal interpretation.
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.
+1 more
When to use it
Use causal mediation when you want to know not just whether a treatment affects an outcome but how much of that effect passes through a specific mediator, with at least about 100 observations. It rests on strong no-unmeasured-confounding assumptions: no treatment-outcome confounding, no mediator-outcome confounding, no treatment-mediator confounding, and the cross-world assumption that lets you compare mediator values across counterfactual treatment levels. It suits cross-sectional or longitudinal data with continuous, binary, or categorical treatments. It is unreliable when the sample is small or when the confounding assumptions are implausible.
Strengths & limitations
- Gives a precise counterfactual definition of direct and indirect effects rather than a purely descriptive product-of-coefficients story.
- Decomposes the total effect into NDE and NIE, so you can quantify how much of the treatment works through the mediator.
- Handles continuous, binary, or categorical treatments and works for cross-sectional or longitudinal designs.
- Requires a fairly large sample (at least about 100); below this the indirect-effect estimate becomes unstable and underpowered.
- Relies on four strong identification assumptions, including the untestable cross-world assumption, so the causal interpretation can be invalid if confounding is present.
- The causal interpretation only holds if all the no-confounding assumptions are credible; otherwise the decomposition is misleading.
Frequently asked
What is the difference between NDE and NIE?
The Natural Direct Effect (NDE) is the part of the treatment's effect that does not pass through the mediator, while the Natural Indirect Effect (NIE) is the part that flows through it. Together they sum to the total effect, telling you how much of the treatment operates directly versus through the mediator.
What is the cross-world assumption?
It is the assumption that lets you compare an outcome under one treatment level while the mediator takes the value it would have had under a different treatment level. It cannot be tested from observed data, which is why the causal interpretation of mediation depends on untestable identification conditions.
How large a sample do I need?
At least about 100 observations. With fewer, there is not enough statistical power to separate the NDE and NIE components, and the indirect-effect estimate becomes unstable; a DAG-based identification approach is a safer fallback.
How is this different from classical Baron-Kenny mediation?
Classical mediation multiplies regression coefficients and is descriptive. The causal approach defines direct and indirect effects through counterfactuals with explicit no-confounding assumptions, so the decomposition carries a genuine causal meaning rather than just a correlational path interpretation.
Sources
- Pearl, J. (2001). Direct and Indirect Effects. In Proceedings of the Seventeenth Conference on Uncertainty in Artificial Intelligence (UAI), 411-420. link ↗
- Imai, K., Keele, L., & Tingley, D. (2010). A General Approach to Causal Mediation Analysis. Psychological Methods, 15(4), 309-334. DOI: 10.1037/a0020761 ↗
How to cite this page
ScholarGate. (2026, June 1). Causal Mediation Analysis (Natural Direct and Indirect Effects). ScholarGate. https://scholargate.app/en/causal-inference/causal-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.
- Conditional Process AnalysisCausal inference↔ compare
- DAG Causal IdentificationCausal inference↔ compare
- Logistic RegressionResearch Statistics↔ compare
- Moderation AnalysisCausal inference↔ compare
- OLS RegressionEconometrics↔ compare