Activity Number:
|
250
- The Future of Designed Experiments in the Era of Big Data
|
Type:
|
Invited
|
Date/Time:
|
Tuesday, August 4, 2020 : 1:00 PM to 2:50 PM
|
Sponsor:
|
Quality and Productivity Section
|
Abstract #309387
|
|
Title:
|
The Role of Additivity in Causal Inference Under Interference
|
Author(s):
|
Daniel L Sussman* and Kelly Kung
|
Companies:
|
Boston University and Boston University
|
Keywords:
|
interference;
causal inference;
network
|
Abstract:
|
Minimizing assumptions in causal inference procedures is crucial to ensure valid conclusions but in the presence of interference, such minimal assumptions can lead to very small effective sample sizes. Incorporating additivity assumptions can enable the use of far more units. In this talk, we propose a framework for additivity across multiple components of a unit's exposure to treatments and demonstrate the potential improvements when these assumptions are valid. We explore the bias-variance trade-off inherent in these assumptions and argue that additivity assumptions can lead to smoother estimates of effect sizes. We conclude by considering how these assumptions impact the designs of experiments under network and other forms of interference.
|
Authors who are presenting talks have a * after their name.