This event will take place in the Physics/Astronomy Auditorium 102 (PAA A102) on the University of Washington campus.

“Identifying and estimating causal effects with incomplete causal information”

Abstract: Questions of cause and effect are ideally answered by intervening in a system through a randomized controlled experiment. However, these experiments can often be costly, time-consuming, unethical, or impossible to conduct. On the other hand, observational data and specific domain or background knowledge may still be available. In this talk, we consider how partial knowledge of causal relationships can be combined with observational data to assess a causal effect while providing efficient estimators in certain settings.

Suppose the causal system can be represented by a directed acyclic graph (DAG) encoding causal relationships. This causal DAG is a priori unknown to us. Instead, we have access to a class of potential causal DAGs representing the same set of observed conditional independencies and background knowledge. We present a necessary and sufficient graphical criterion to uniquely identify a causal effect given such a class. When the causal effect cannot be uniquely identified given the class of possible graphical models, we consider the identification of a set of possible total causal effects and devise a minimal and complete approach to solving this problem. This result resolves an issue with existing methods, which often report possible total effects with duplicates, namely those numerically distinct due to sampling variability but causally identical. Next, for a causal effect that is identified from the partial knowledge of the causal relationships, we devise an estimator based on recursive least squares. Under the linearity of the causal system, this estimator consistently estimates the causal effect while achieving a minimal asymptotic variance among a broad class of established estimators. We conclude the presentation by discussing further research directions.

Bio: Emilija Perkovic joined the Department of Statistics at the University of Washington in Autumn 2018 as an Acting Assistant Professor and was promoted to a tenure-track Assistant Professor role in Autumn 2020.  Before coming to UW, she completed a Ph.D. in Statistics at ETH Zürich in 2018 under the supervision of Professor Marloes Maathuis, an M.Sc. in Statistics from ETH Zürich in 2014, and a B.Sc. in Mathematics from the University of Belgrade in 2012.  Her research interests are focused on causal inference from the perspective of graphical models. A large part of her Ph.D. thesis was on causal inference through covariate adjustment. She hopes to learn some new perspectives on causal inference while she is here.

The UW Data Science Seminar is an annual lecture series at the University of Washington that hosts scholars working across applied areas of data science, such as the sciences, engineering, humanities and arts along with methodological areas in data science, such as computer science, applied math and statistics. Our presenters come from all domain fields and include occasional external speakers from regional partners, governmental agencies and industry.

The 2022-2023 seminars will be held in person, and are free and open to the public.