Causal Mediation Analysis with ACME and ADE Estimation

Performs causal mediation analysis to decompose the total treatment effect into the Average Causal Mediation Effect (ACME / indirect effect) and the Average Direct Effect (ADE). Wraps mediation package functionality with enhanced output and visualization.

Usage

mediation_analysis(
  data,
  outcome,
  treatment,
  mediator,
  covariates = NULL,
  treat_mediator_interaction = FALSE,
  family_mediator = "gaussian",
  family_outcome = "gaussian",
  sims = 1000,
  conf_level = 0.95,
  seed = 42,
  plot = TRUE
)

Arguments

data

A data frame.

outcome

Character. Name of the outcome variable.

treatment

Character. Name of the binary treatment variable.

mediator

Character. Name of the mediator variable.

covariates

Character vector. Additional covariates in both models. Default NULL.

treat_mediator_interaction

Logical. Whether to include a treatment × mediator interaction in the outcome model. Default FALSE.

family_mediator

Character. GLM family for the mediator model. Default "gaussian". Use "binomial" for binary mediators.

family_outcome

Character. GLM family for the outcome model. Default "gaussian".

sims

Integer. Number of quasi-Bayesian Monte Carlo simulations. Default 1000.

conf_level

Numeric. Confidence level for CIs. Default 0.95.

seed

Integer. Random seed. Default 42.

plot

Logical. If TRUE (default), produce a decomposition plot.

Value

A list with:

acme

Numeric. Average Causal Mediation Effect (indirect effect).

ade

Numeric. Average Direct Effect.

total

Numeric. Total treatment effect.

prop_mediated

Numeric. Proportion of effect mediated.

summary_df

Data frame with ACME, ADE, Total, and their CIs.

mediation_obj

The mediate object (if mediation is available).

plot

ggplot2 decomposition bar chart.

Details

This function implements the approach from Imai et al. (2010) via the mediation package. When mediation is not available, a simple product-of-coefficients (Baron-Kenny) approach is used.

The sequential ignorability assumption (no unmeasured confounding of mediator-outcome relationship) is required for causal interpretation. Use sensitivity_plot() or the sens.analysis output from the mediation package to assess robustness.

References

Imai, K., Keele, L., & Tingley, D. (2010). A general approach to causal mediation analysis. Psychological Methods, 15(4), 309–334.

Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological research. Journal of Personality and Social Psychology, 51(6), 1173–1182.

Examples

set.seed(123)
n <- 200
df <- data.frame(
  treat = rbinom(n, 1, 0.5),
  X     = rnorm(n)
)
df$mediator <- 0.5 * df$treat + 0.3 * df$X + rnorm(n, 0, 0.5)
df$outcome  <- 0.3 * df$mediator + 0.2 * df$treat + rnorm(n)

result <- mediation_analysis(
  data      = df,
  outcome   = "outcome",
  treatment = "treat",
  mediator  = "mediator"
)
#> Registered S3 method overwritten by 'lme4':
#>   method           from
#>   na.action.merMod car 
result$summary_df
#>            effect   estimate     ci_lower  ci_upper
#> 1 ACME (Indirect) 0.09632680  0.006451494 0.2015267
#> 2    ADE (Direct) 0.09566813 -0.193571221 0.4071730
#> 3    Total Effect 0.19199493 -0.078644711 0.4779045
result$plot