Print  Close

Asymptotic Distribution of the IPW Estimator Trimmed with Estimated Propensity Scores

Abstract Number:

3641 

Submission Type:

Contributed Abstract 

Contributed Abstract Type:

Paper 

Participants:

Tianyu Guo (1), Sai Praneeth Karimireddy (2), Michael Jordan (3)

Institutions:

(1) N/A, N/A, (2) UC Berkeley, N/A, (3) Univ of California-Berkeley, N/A

Co-Author(s):

Sai Praneeth Karimireddy  
UC Berkeley
Michael Jordan  
Univ of California-Berkeley

First Author:

Tianyu Guo  
N/A

Presenting Author:

Tianyu Guo  
N/A

Abstract Text:

Inverse Propensity Score Weighting (\textsc{IPW}) is a fundamental methodological tool for causal inference in observational data. However, it can suffer from unbounded variance when encountering extreme propensity scores. Trimming these extremal scores is a widely used practice to avoid such a blowup of the variance, but its effect is poorly understood and remains an open theoretical problem in causal inference.
In this work, we derive the asymptotic distribution for the \textsc{IPW} estimate, using trimmed estimated propensity scores. We then show how our results can guide the choice of a trimming threshold, and systematically compare widely utilized methods for extreme propensity scores, suggesting different applicable scenarios for each method. Along the way, we develop a new technique to handle the non-smoothness in the delta method.

Keywords:

Inverse propensity score weighting|Limited overlap|Non-smoothness|Asymptotic distribution| |

Sponsors:

IMS

Tracks:

Statistical Theory

Can this be considered for alternate subtype?

Yes

Are you interested in volunteering to serve as a session chair?

No

I have read and understand that JSM participants must abide by the Participant Guidelines.

Yes

I understand that JSM participants must register and pay the appropriate registration fee by June 1, 2024. The registration fee is non-refundable.

I understand