Mediation Analysis

Mediation analysis estimates how much of the effect of an independent variable on an outcome is transmitted through an intermediate variable, separating the indirect effect from the direct effect.

매개분석은 독립변수가 결과변수에 미치는 효과 중 중간 변수를 거쳐 전달되는 부분을 추정하여 간접효과와 직접효과를 분리하는 방법이다.

How do I test whether a variable explains the mechanism between a predictor and an outcome?

어떤 변수가 예측변수와 결과변수 사이의 작동 기제를 설명하는지 어떻게 검정하나요?

Mediation analysis decomposes the total effect of a predictor on an outcome into an indirect effect through a mediator and a direct effect. It fits questions about mechanisms, and the indirect effect is usually tested with a bootstrap confidence interval rather than the causal-steps approach.

When to use Mediation Analysis

• Testing whether X affects Y through an intermediate variable M
• Comparing direct and indirect effects in a theoretical process model
• Models with one or several parallel or serial mediators

When to use something else

• You want to know whether the X–Y relationship changes across levels of a third variable: use moderation analysis
• Many latent constructs and mediators with measurement error: use structural equation modeling

Assumptions and requirements

• Temporal or theoretical ordering of X, M and Y
• No unmeasured confounding of the X–M, M–Y and X–Y relationships
• Correctly specified regression models
• Mediator measured without substantial error

What the analysis reports

• Path coefficients a, b, c and c′
• Indirect effect (a × b) with bootstrap confidence interval
• Direct and total effects
• Proportion mediated
• Path diagram

How to report it

The indirect effect of [X] on [Y] through [M] was significant, ab = [value], 95% bootstrap CI [[lower], [upper]].

Related methods

• Moderation Analysis
• Structural Equation Modeling (SEM)
• Conditional Process Analysis

Run Mediation Analysis on Quorant