手法を比較

	Four-Way Decomposition	Marginal Structural Model (IPTW)
分野	Social Epidemiology	Social Epidemiology
系統	Process / pipeline	Process / pipeline
提唱年	2014	2000
提唱者	Tyler J. VanderWeele	James M. Robins, Miguel A. Hernán & Babette Brumback
種類	Counterfactual decomposition pipeline for total effects	Reweighting pipeline for time-varying confounding affected by prior treatment
原典	VanderWeele, T. J. (2014). A unification of mediation and interaction: a four-way decomposition. Epidemiology, 25(5), 749-761.	Robins, J. M., Hernán, M. A., & Brumback, B. (2000). Marginal structural models and causal inference in epidemiology. Epidemiology, 11(5), 550-560.
別名	4-Way Decomposition, VanderWeele Four-Way Decomposition, Mediation-Interaction Decomposition, Unification of Mediation and Interaction	MSM with IPTW, Inverse-Probability-of-Treatment-Weighted Marginal Structural Model, IPTW Marginal Structural Model, Robins Marginal Structural Model
関連	3	3
概要	The four-way decomposition, introduced by Tyler VanderWeele in 2014, unifies the two great themes of effect analysis — mediation and interaction — into a single, exhaustive partition of a total causal effect. Any total effect of an exposure on an outcome can be split into exactly four pieces: a controlled direct effect (neither mediation nor interaction), a reference interaction (interaction but no mediation), a mediated interaction (both mediation and interaction at once), and a pure indirect effect (mediation but no interaction). These four components are mutually exclusive and add up to the total effect, and they nest the familiar two-way and three-way decompositions as special cases. Formalized in counterfactual notation and developed at book length in VanderWeele's 2015 Explanation in Causal Inference, the method gives social epidemiologists a precise vocabulary for asking how much of an exposure's effect runs through a mediator, how much depends on the exposure and mediator acting together, and how much is direct.	Marginal structural models, introduced by Robins, Hernán, and Brumback in 2000, are causal models for the mean of a counterfactual outcome under a treatment regime, estimated by inverse-probability-of-treatment weighting. They solve the same problem as the g-formula — estimating the effect of a time-varying exposure when time-varying confounders are themselves affected by prior treatment — but through a different device: instead of modeling the outcome and confounder processes, they reweight each person by the inverse of their probability of receiving the treatment history they actually received. This creates a pseudo-population in which treatment is, by construction, unconfounded by the measured covariates, so a simple weighted regression recovers the causal effect. The companion 2000 paper applying the method to zidovudine and HIV survival showed its practical payoff. In social epidemiology, MSMs with IPTW are standard for the cumulative effects of time-varying social exposures.