New developments in causal inference using balance optimization subset selection

Kwon, Hee Youn

Permalink

https://hdl.handle.net/2142/100968 Copy

Description

Title

New developments in causal inference using balance optimization subset selection

Author(s)

Kwon, Hee Youn

Issue Date

2018-04-13

Director of Research (if dissertation) or Advisor (if thesis)

Jacobson, Sheldon H.

Doctoral Committee Chair(s)

He, Niao

Committee Member(s)

• Chandrasekaran, Karthekeyan
• Nagi, Rakesh

Department of Study

Industrial&Enterprise Sys Eng

Discipline

Systems & Entrepreneurial Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Date of Ingest

2018-09-04T20:27:03Z

Keyword(s)

• Causal Analysis
• Optimization
• Subset Selection

Abstract

Causal inference with observational data has drawn attention across various fields. These observational studies typically use matching methods which find matched pairs with similar covariate values. However, matching methods may not directly achieve covariate balance, a measure of matching effectiveness. As an alternative, the Balance Optimization Subset Selection (BOSS) framework, which seeks the optimal covariate balance directly, has been proposed. This dissertation extends the BOSS framework in various ways and is composed of the following five parts. The first part of the dissertation investigates all the possible cases that may lead to bias in the context of BOSS and tries to mitigate the bias. Second, this dissertation then extends the BOSS by estimating and decomposing a treatment effect as a combination of heterogeneous treatment effects from a partitioned set using the BOSS. Third, the dissertation generalizes the BOSS framework from a binary treatment setting to a multi-treatment setting. A treatment effect estimate with multiple treatments can be computed by combining estimates obtained from BOSS with binary treatments. The fourth part discusses on how to handle missing data with BOSS. It includes a sensitivity analysis of BOSS studying how the estimated values are affected by violation of the conditional independence assumption and methods to apply BOSS after multiple imputation on missing covariates. In these discussions, the performances of BOSS estimators are compared to those of matching estimators. In the last part, BOSS is formulated as an LP by relaxing integer constraints in the original mixed integer programming formulation and properties of its dual problem are investigated.

Graduation Semester

2018-05

Type of Resource

text

Permalink

http://hdl.handle.net/2142/100968

Copyright and License Information

Copyright 2018 Hee Youn Kwon

Owning Collections