Activity Number:
|
563
- Mechanisms of Interference: New Strategies for Identification and Estimation
|
Type:
|
Invited
|
Date/Time:
|
Wednesday, August 1, 2018 : 2:00 PM to 3:50 PM
|
Sponsor:
|
Social Statistics Section
|
Abstract #326709
|
Presentation
|
Title:
|
Causal Inference Under Unmodeled and All-Encompassing Interference
|
Author(s):
|
Fredrik Sävje*
|
Companies:
|
Yale University
|
Keywords:
|
treatment effects;
causal inference;
interference;
SUTVA
|
Abstract:
|
It has been established that causal effects can be estimated under interference either when the structure of the interference is known or when interference is restricted so that a limited number of units are interfering with each other. This paper investigates whether causal inference is possible when neither of those conditions hold. I ask whether we can accurately estimate causal effects in randomized experiments when all pairs of units are interfering and the structure of the interference is left unmodeled. I show that common experimental estimators are consistent with respect to an average treatment effect if we partially restrict the strength of the interference but otherwise allow it to have an arbitrary and unknown form. In particular, consistency follows if the number of pairs of units with "strong" interference grows at a sufficiently slow rate. The results do not require us to identify these pairs.
|
Authors who are presenting talks have a * after their name.