Identification, Inference and Sensitivity Analysis for Causal Mediation Effects
By Kosuke Imai, Luke Keele, Teppei Yamamoto. 2010
1,746 citations recorded by the source corpus.
Causal mediation analysis is routinely conducted by applied researchers in a variety of disciplines. The goal of such an analysis is to investigate alternative causal mechanisms by examining the roles of intermediate variables that lie in the causal paths between the treat- ment and outcome variables. In this paper we first prove that under a particular version of sequential ignorability assumption, the aver- age causal mediation effect (ACME) is nonparametrically identified. We compare our identification assumption with those proposed in the literature. Some practical implications of our identification result are also discussed. In particular, the popular estimator based on the linear structural equation model (LSEM) can be interpreted as an ACME estimator once additional parametric assumptions are made. We show that these assumptions can easily be relaxed within and outside of the LSEM framework and propose simple nonparametric estimation strate- gies. Second, and perhaps most importantly, we propose a new sensi- tivity analysis that can be easily implemented by applied researchers within the LSEM framework. Like the existing identifying assumptions, the proposed sequential ignorability assumption may be too strong in many applied settings. Thus, sensitivity analysis is essential in order to examine the robustness of empirical findings to the possible existence of an unmeasured confounder. Finally, we apply the proposed methods to a randomized experiment from political psychology. We also make easy-to-use software available to implement the proposed methods.
Subfields: Artificial Intelligence, Sociology and Political Science, Statistics and Probability, Causal Inference & Discovery
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Judea Pearl
- Yonatan Belinkov
- Elias Bareinboim
- Sebastian Gehrmann
- Bernhard Schölkopf
- Tyler J. VanderWeele
- Ilya Shpitser
- Mohit Bansal
In Sociology and Political Science
- Judea Pearl
- Dustin Tingley
- Martin Huber
- Tyler J. VanderWeele
- Stijn Vansteelandt
- Kristopher J Preacher
- Kosuke Imai
- Elias Bareinboim
In Statistics and Probability
- Judea Pearl
- Tyler J. VanderWeele
- Stijn Vansteelandt
- Martin Huber
- Ilya Shpitser
- Kosuke Imai
- Edward H. Kennedy
- Kristopher J Preacher
In Causal Inference & Discovery
- Judea Pearl
- Tyler J. VanderWeele
- Stijn Vansteelandt
- Ilya Shpitser
- E. T. Tchetgen Tchetgen
- Dustin Tingley
- Elias Bareinboim
- Martin Huber
Citations in the corpus
- Causal Mediation Analysis for Interpreting Neural NLP: The Case of Gender Bias (2020, methodology): “This equation parallels a previously-described assumption in the causal mediation analysis literature that ascertains that the NIE is the same regardless of the fixed value at which the intervention (the analogue of set-gender/null) is held, known as a no-interaction assumption (Imai et al., 2010b).”
- Fairness in Decision-Making - The Causal Explanation Formula (2018, background): “Pearl introduced the causal mediation formula for arbitrary non-parametric models, which decomposes the total effect into what was called the natural direct (NDE) and indirect (NIE) effects (Pearl 2001) (see also (Imai, Keele, and Yamamoto 2010; Imai et al. 2011; VanderWeele 2015).”
- Causal inference in statistics: An overview (2009, background): “…) = ∑ z E(Y |x, z)[P (z|x′)− P (z|x)] (52) This expression deserves the label Mediation Formula, due to its pivotal role in mediation analysis (Imai et al., 2008), which has been a thorny issue in several sciences (Shrout and Bolger, 2002; MacKinnon et al., 2007; Mortensen et al., 2009).”
- On Partial Identification of the Pure Direct Effect (2015, neutral): “…et al., 2007; Albert, 2008; Goetgeluk et al., 2008; van der Laan and Petersen, 2008; VanderWeele, 2009; VanderWeele and Vansteelandt, 2009, 2010; Imai et al., 2010a,b; Albert and Nelson, 2011; Tchetgen Tchetgen, 2011; VanderWeele, 2011; Albert, 2012; Tchetgen Tchetgen and Shpitser, 2012; Wang…”
- Total, Direct, and Indirect Effects in Logit and Probit Models (2013, methodology): “Using Monte Carlo simulations, Karlson et al. (2012) compared the method we propose with other methods for comparing coefficients across samesample nested logit or probit models.”
- CAUSAL INTERPRETATIONS OF BLACK-BOX MODELS (2019, neutral): “…i = 1, . . . , n } with fixed The consideration of partial effect of some independent variables X S on a dependent variable Y is common in social science when model parameters are not immediately interpretable (e.g., King, Tomz, and Wittenberg 2000; Imai, Keele, and Yamamoto 2010; Wooldridge 2015).”
- A Minimax Surrogate Loss Approach to Conditional Difference Estimation (2018, background): “Recently there has been a lot of work on subgroup identification [16, 18, 25] and other types of nonlinear predictors of causal effects [3, 19, 2, 36, 34].”
- The International Journal of Biostatistics Principal Stratification — a Goal or a Tool ? (2011, neutral): “…and identifying causal mechanisms (Joffe, Small, and Hsu, 2007, Albert and Nelson, 2011, Mortensen, Diderichsen, Smith, and Andersen, 2009, Imai, Keele, and Yamamoto, 2010, Robins et al., 2007, 2009, Pearl, 2009a, Petersen, Sinisi, and van der Laan, 2006, Hafeman and Schwartz, 2009,…”
- Semiparametric doubly robust targeted double machine learning: a review (2022, methodology): “The first strategy is somewhat commonplace and has been used and detailed for example in Kandasamy et al. [2015] and Hines et al.”
- EIGHT MYTHS ABOUT CAUSALITY AND STRUCTURAL EQUATION MODELS (2013, positive): “This development allowed researchers to cross the linear-nonlinear barrier and has spawned a rich literature in nonparametric mediation analysis (Imai et al., 2010; Muthen, 2011; Pearl, forthcoming, Intervention Science, VanderWeele & Vansteelandt, 2009).”
- Ratio of Mediator Probability Weighting for Estimating Natural Direct and Indirect Effects (2025, background): “Using the potential outcomes framework, researchers (Holland 1988; Imai et al, 2010; Pearl, 2001; Robins and Greenland, 1992; Sobel, 2008; VanderWeele & Vansteelandt, 2009) have clarified the identification assumptions under which the above path coefficients have causal meanings.”
- Mediation analysis allowing for exposure-mediator interactions and causal interpretation: theoretical assumptions and implementation with SAS and SPSS macros (2013, methodology): “Preacher and Hayes (2004) developed several macros for mediation mainly implementable in SAS, SPSS and Mplus (indirect, mediate, modmed, medcurve); Imai et al. (2009) and Imai, Keele, Tingley, & Yamamoto (2010) also developed a macro in R (mediate).”