To Exclude Inattentive Respondents or Not? -- Informing Complier Average Treatment Effects with Post-treatment Variables

This paper proposes a new methodological framework to estimate the Complier Average Treatment Effects (CATE), when a researcher only observes treatment assignment fully but only a noisy post-treatment proxy for treatment take-up. Such empirical setting is common in both online and field experiments. Existing method is either estimating a different quantity of interest – Intention-to-Treat (ITT) effect, or is mistakenly instrumenting the proxy with the hope that the instrumental variable could reduce the measurement bias, which indeed leads to an inconsistent estimator for CATE. This new method considers whether a unit is a complier as a latent variable, and then estimates the probability of a unit being a complier with a Gaussian mixture model. With the aggregated proportion of compliers estimated by the model, researchers can recover CATE from the ITT. In addition, leveraging a full Bayesian framework, this framework allows researchers to adjust for observed covariates in the estimation for both compliance status and the CATE, and report credible intervals for these estimation. Simulations and the validation example show the effectiveness of the proposed method. This paper further illustrates the method with an online media experiment conducted on urban Chinese residents.


Shiyao “Sean” Liu is a PhD Candidate at Department of Political Science, Massachusetts Institute of Technology. He works in political methodology (causal inference, experimental studies, and machine learning), as well as comparative politics (political behavior), with a specific focus on China. Before that, he obtained his Bachelor of Economics in Philosophy, Politics and Economics and Bachelor of Science in Statistics from Peking University. His work is appearing in the Journal of Politics. He was awarded the Best Poster Award (methods) for graduate students by the Society for Political Methodology for 2019.

Speaker(s) Mr Shiyao (Sean) LIU
PhD Candidate, Department of Political Science, Massachusetts Institute of Technology
Date 25 Jan 2021 (Monday)
Time 11:00 am
Venue Online via ZOOM (link will be sent via email)

Please use your HKUST login to join the zoom meeting.
This meeting is being recorded. By joining, you are giving consent for this meeting to be recorded.

Hong Kong University of Science and Technology
Home  |  HKUST  |  The School  |  Jobs  |  Site Map  |  Contact Us