Estimation of the Average Treatment Effect with Missing Outcome Data

DOI

10.2991/csece-18.2018.81How to use a DOI?

Keywords

average treatment effects (ATE); asymptotic variance; delta-method; missing data

Abstract

Missing outcome data occurs often in the causal inference of observational studies. For example, in observational study on the safety of a Traditional Chinese Medicine (TCM) injection in market, some patients are missing safety outcome variables. In this paper we proposed a consistent estimator for the average treatment effect (ATE) with partially missing outcome data and derived the asymptotic variance of the proposed ATE estimator under the condition that the missing-data mechanism is missing at random (MAR). We then proceeded to assess the performance of the asymptotic variance estimator via a simulation study. The simulation study showed that the asymptotic variance estimator had good performance in finite sample sizes. This asymptotic variance could then be used to construct a confidence interval for the average treatment effect. We also compared the bias and mean squared error (MSE) of the ATE estimators based on the proposed method-dealing with the missing outcomes with that of the complete-data method, which means that directly deleting the samples with the missing data through a simulation study. The simulation study showed that the MSE and Bias of our method were smaller than the complete-data method under MAR. In addition, we also found that coverage of the confidence interval constructed on the ATE and its asymptotic variance from our method are better than those based on the traditional method.

Copyright

© 2018, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Feng Han
PY  - 2018/02
DA  - 2018/02
TI  - Estimation of the Average Treatment Effect with Missing Outcome Data
BT  - Proceedings of the 2018 International Conference on Computer Science, Electronics and Communication Engineering (CSECE 2018)
PB  - Atlantis Press
SP  - 377
EP  - 382
SN  - 2352-538X
UR  - https://doi.org/10.2991/csece-18.2018.81
DO  - 10.2991/csece-18.2018.81
ID  - Han2018/02
ER  -