方法对比

	Multilevel Regression and Poststratification	因果中介分析（自然直接效应和自然间接效应）
领域	Political Science	因果推断
方法族	Regression model	Regression model
起源年份	2004	2010
提出者	Gelman and Little (method); Park, Gelman & Bafumi (state-level application)	Pearl (2001); general framework by Imai, Keele & Tingley (2010)
类型	Survey small-area estimation model combining multilevel regression with census poststratification	Counterfactual causal decomposition
开创性文献	Park, D. K., Gelman, A., & Bafumi, J. (2004). Bayesian Multilevel Estimation with Poststratification: State-Level Estimates from National Polls. Political Analysis, 12(4), 375–385.	Pearl, J. (2001). Direct and Indirect Effects. In Proceedings of the Seventeenth Conference on Uncertainty in Artificial Intelligence (UAI), 411-420.
别名	MRP, Mister P, Multilevel regression with poststratification, Small-area opinion estimation	natural direct effect, natural indirect effect, NDE / NIE decomposition, counterfactual mediation
相关	5	5
摘要	Multilevel regression and poststratification (MRP) estimates opinion or behavior in small subpopulations — states, districts, demographic groups — from a single national survey that is far too small to support direct estimates in each unit. It first fits a multilevel model that predicts the outcome from individual demographic and geographic characteristics, borrowing strength across units through partial pooling, and then poststratifies the predicted values to known population counts of demographic-by-geographic cells. Introduced for state-level opinion by Park, Gelman, and Bafumi (2004) and shown by Lax and Phillips (2009) to outperform disaggregation, MRP has become the standard tool for subnational opinion estimation.	Causal mediation analysis is a counterfactual framework that splits a treatment's total effect into a Natural Direct Effect (NDE) and a Natural Indirect Effect (NIE) that runs through a mediator. The modern general approach was formalised by Pearl (2001) and Imai, Keele and Tingley (2010), giving the decomposition a precise causal interpretation.