Complex Experiments for a Complex World
Abstract
Dating back to the 1930s randomized experiments or A/B tests have become the gold stan-dard for establishing causal effects. The bas,ic experimental designs proposed at that time, where units are divided randomly into a treatment and control group, continue to be w,idely used in drug trials, social science settings, and in the private sector. However, in many settings of current interest the key, assumption underlying the standard analyses of such experiments that there is no interference between units is not tenable: in soci,al sciences individuals interact strategically in response to treatment assignments, and through those interactions affect others, a,nd in market places the exchanges directly lead to spillovers. We propose to develop new ex-perimental designs that do justice to th,ese complex interactions, and to develop new statistical methods for analyzing them. These experiments consider structures on the sp,illovers through networks. We consider dynamic settings with sequential assignment mechanisms. We also analyze experimental designs,in settings with multiple populations as arises in market places.Approved for Public Release
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 03, 2022
- Source ID
- N000142212668
Entities
People
- Guido Imbens
Organizations
- Office of Naval Research
- Stanford University
- United States Navy