Multilevel Mediation Analysis
Also known as: multilevel mediation, hierarchical mediation, cross-level mediation, 1-1-1 mediation, 2-1-1 mediation, 2-2-1 mediation, Çok Düzeyli Medyasyon Analizi
Multilevel mediation analysis is a parametric structural method that estimates indirect (mediated) effects within hierarchically nested data, such as students within schools or employees within organisations. Formalised for lower-level mediation in multilevel models by Kenny, Korchmaros and Bolger (2003), it simultaneously handles individual-level (1-1-1) and group-level (2-2-1 or 2-1-1) mediation pathways in a single coherent framework.
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 multilevel mediation when your data have a nested or hierarchical structure — for example individuals within groups, repeated measures within individuals, or classrooms within schools — and you want to test whether a mediator variable explains the path from a predictor to an outcome. Three conditions should hold: the intraclass correlation (ICC) must exceed 0.05, confirming meaningful between-group variance that makes multilevel modelling necessary; the indirect effect must be estimated via Monte Carlo simulation or bootstrapping rather than the Sobel test, because the product of two regression coefficients is not normally distributed; and the number of level-2 units (groups) should be at least 30 for stable estimation of random effects. The design type — whether the predictor, mediator and outcome are measured at level 1 or level 2 — must be specified before analysis, as it determines how the model is parameterised.
Strengths & limitations
- Correctly partitions mediation pathways into within-group and between-group components, preventing the ecological fallacy.
- Handles all three common hierarchical mediation designs (1-1-1, 2-1-1, 2-2-1) within a unified framework.
- Monte Carlo and bootstrap confidence intervals give accurate coverage for the non-normally distributed indirect effect.
- Accounts for non-independence among observations within groups, producing unbiased standard errors.
- Requires a minimum of approximately 100 observations and at least 30 level-2 units for reliable random-effect estimation.
- Model specification — particularly the choice of design type and which effects to treat as random — demands substantive theoretical justification.
- Computational demands are higher than single-level mediation, especially with bootstrapping or Monte Carlo methods.
- Does not establish causality; causal claims require random assignment or strong identification assumptions.
Frequently asked
Why can I not just run a standard mediation analysis on nested data?
Standard single-level mediation treats all observations as independent, which they are not when individuals share a group. This underestimates standard errors, inflates the Type I error rate, and conflates within-group and between-group mediation pathways. When ICC > 0.05 multilevel mediation is necessary to obtain unbiased estimates.
What is the difference between 1-1-1, 2-1-1, and 2-2-1 designs?
The three-number code indicates the level at which the predictor, mediator, and outcome are measured. In a 1-1-1 design all three variables vary at the individual level within groups. In a 2-1-1 design the predictor is a group-level variable while the mediator and outcome are individual-level. In a 2-2-1 design both the predictor and the mediator are group-level but the outcome is individual-level. Each design implies different model equations and should be chosen on theoretical grounds before data analysis.
Why should I use Monte Carlo or bootstrap intervals rather than the Sobel test?
The indirect effect ab is the product of two estimated quantities, and its sampling distribution is asymmetric, not normal. The Sobel test assumes normality and therefore produces confidence intervals that are too narrow and symmetric. Monte Carlo simulation and bootstrapping directly capture the shape of the distribution, yielding more accurate coverage probabilities, particularly in small samples or when the indirect effect is small.
How many level-2 units do I need?
The literature recommends at least 30 groups (level-2 units) for stable estimation of variance components and random effects. With fewer groups the random-effect variances and their standard errors become unreliable, which in turn affects the precision of the indirect-effect estimates. The total individual-level sample size should also be sufficient — a minimum of 100 observations is a common practical threshold.
Sources
- Kenny, D. A., Korchmaros, J. D., & Bolger, N. (2003). Lower level mediation in multilevel models. Psychological Methods, 8(2), 115–128. DOI: 10.1037/1082-989X.8.2.115 ↗
How to cite this page
ScholarGate. (2026, June 1). Multilevel Mediation Analysis. ScholarGate. https://scholargate.app/en/statistics/multilevel-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.
- Causal Mediation AnalysisCausal inference↔ compare
- Conditional Process AnalysisCausal inference↔ compare
- Hierarchical Linear ModelingStatistics↔ compare
- Mediation AnalysisStatistics↔ compare
- Mixed ANOVAStatistics↔ compare
- Moderation AnalysisCausal inference↔ compare
- Panel Fixed EffectsEconometrics↔ compare
- Structural Equation ModelingResearch Statistics↔ compare