Identification of Causal Mechanisms from Randomized Experiments

Posted: 16 Dec 2019 Last revised: 11 Feb 2020

See all articles by Jing Peng

Jing Peng

University of Connecticut - Department of Operations & Information Management

Date Written: February 2, 2020

Abstract

Experimental research in the business disciplines often focuses on the overall treatment effect and the heterogeneity therein. While this type of research allows us to understand the strength and direction of the treatment effect under different conditions, it does not directly speak to the generative mechanisms, namely, why and how the effect arises. A standard procedure to identify the mechanisms behind a treatment effect is mediation analysis, but extant mediation analysis frameworks either have no causal interpretation or require the involved mediators to be unconfounded. Since mediators are post-treatment variables that typically cannot be pre-assigned beforehand, the endogeneity of mediators remains to be a serious concern even in randomized experiments. In response to this issue, we present an endogenous mediation analysis framework that can provide causal interpretations for endogenous mediators. We then discuss the identification conditions for different types of endogenous mediators, including latent ones, under this framework. Contrary to conventional wisdom on experimental design, we show that, by explicitly modeling the data generation process, the identification of a mediation process does not necessarily require a separate manipulation that affects the mediator but not the outcome. Finally, leveraging field experiments on two e-commerce platforms, we illustrate how endogenous mediation analysis can help identify generalizable causal mechanisms from different contexts. Our work has important implications for the design and analysis of experiments.

Keywords: experiment, mechanism, mediation, endogeneity, identification, methodology

Suggested Citation

Peng, Jing, Identification of Causal Mechanisms from Randomized Experiments (February 2, 2020). Available at SSRN: https://ssrn.com/abstract=3494856 or http://dx.doi.org/10.2139/ssrn.3494856

Jing Peng (Contact Author)

University of Connecticut - Department of Operations & Information Management ( email )

368 Fairfield Road
Storrs, CT 06269-2041
United States

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Abstract Views
217
PlumX Metrics