star twitter facebook envelope linkedin instagram youtube alert-red alert home left-quote chevron hamburger minus plus search triangle x

New Approach to Research Estimates Binary Outcomes That Assume Only Randomization of Treatment

In experiments that study social phenomena (e.g., peer influence, herd immunity), the treatment of one unit may influence the outcomes of others. Such interference between units violates traditional approaches for causal inference, so additional assumptions are often imposed to model or limit the underlying social mechanism.

In a new article, a Carnegie Mellon University researcher proposes a novel way to estimate binary outcomes without such assumptions, allowing for interval estimates that assume only the randomization of treatment. However, the causal implications of these are more limited than are those attainable under stronger assumptions.

“In experiments where the assumption of ‘no interference between units’ does not apply, the outcome of each unit may depend not only on their own treatment assignment, but also on the treatment of others,” explains David Choi, assistant professor of statistics and information systems at Carnegie Mellon’s Heinz College, who wrote the article. “Examples include social networks in which units may be influenced by the actions of others and vaccination studies in which units receiving the placebo may be protected by the vaccine through herd immunity.”

Choi proposes a new approach involving new estimands that are more limited in their causal interpretation but can be interval estimated with no assumptions on interference. This new method shows whether the treatment effects under the observed assignment vary systematically as a function of each unit’s direct and indirect exposure to treatment, while also lower bounding the number of units affected.

In the article, he uses to examples to illustrate his proposal—one related to social networks and the decision to insure and the other related to a simulated vaccine trial. The new approach goes beyond hypothesis testing by measuring the extent to which an observed contrast in outcomes may be attributed to the relative effects of treatment.


Summarized from an article in Journal of the American Statistical Association, New Estimands for Experiments with Strong Interference by Choi, D (Carnegie Mellon University). Copyright 2023 American Statistical Association. All rights reserved.. 

About Heinz College of Information Systems and Public Policy
The Heinz College of Information Systems and Public Policy is home to two internationally recognized graduate-level institutions at Carnegie Mellon University: the School of Information Systems and Management and the School of Public Policy and Management. This unique colocation combined with its expertise in analytics set Heinz College apart in the areas of cybersecurity, health care, the future of work, smart cities, and arts & entertainment. In 2016, INFORMS named Heinz College the #1 academic program for Analytics Education. For more information, please visit