Bootstrapped indirect effects

Description

Testing a mediated (indirect) effect by resampling the data thousands of times to build a confidence interval for the product of paths, without assuming normality.

An indirect effect is the product of the X to M path (a) and the M to Y path controlling for X (b). The sampling distribution of a times b is usually skewed, so the older Sobel test, which assumes normality, is underpowered and inaccurate. The bootstrap resamples cases with replacement (typically 5,000-10,000 times), re-estimates a and b in each sample and uses the percentiles of the resulting a times b distribution as a confidence interval. If the 95% interval excludes zero, the indirect effect is supported. Rigour comes from adequate resamples, a fixed seed, reporting the point estimate with CI, completely standardised effects where helpful, and remembering that a significant indirect effect does not prove causal mediation.

Attributes

Loss type
Siloed
Stage
Evaluation
Effort
Hours

Steps

  1. Estimate path a (X predicting M) and path b (M predicting Y controlling for X) in the full sample.
  2. Draw at least 5,000 bootstrap resamples of cases with replacement, with a recorded random seed.
  3. Re-estimate a times b in each resample and store the distribution.
  4. Form the 95% percentile confidence interval (2.5th and 97.5th percentiles); bias-corrected intervals are an option but can inflate Type I error.
  5. Conclude that mediation is supported if the CI excludes zero, and report the effect with CI and the proportion of the total effect.
  6. Discuss assumptions: temporal order, no unmeasured M-Y confounding and measurement reliability of M.

Sources

  • SPSS REGRESSION
  • PLUM
  • LOGISTIC REGRESSION procedures; Hayes PROCESS macro v4
  • Hayes, A. F. (2022) Introduction to Mediation, Moderation, and Conditional Process Analysis, 3rd ed.
  • Efron, B. & Tibshirani, R. J. (1993) An Introduction to the Bootstrap
  • Shrout, P. E. & Bolger, N. (2002) Mediation in experimental and nonexperimental studies. Psychological Methods
  • The Inquiry Series, C8 — SPSS for Regression, Moderation & Mediation

Also known as

  • Bootstrap mediation test
  • Percentile bootstrap CI for indirect effects

Categories

  • Analytics & Behaviour
  • Statistics & Modelling