Simons Institute Receives Alfred P. Sloan Foundation Grant for Causality Program
We’re delighted to announce that the Simons Institute has received a grant from the Alfred P. Sloan Foundation in support of our Spring 2022 research program on Causality.
“We’re grateful to the Alfred P. Sloan Foundation for their visionary support of our research programs exploring the intersection of algorithmic science with economics and society,” says Simons Institute Director Shafi Goldwasser. “This semester’s interdisciplinary program on Causality promises to be particularly interesting, and we’re thrilled to be once again partnering with the Sloan Foundation.”
This research program aims to integrate advances and techniques from theoretical computer science into methods for causal inference and discovery. Discovering causal mechanisms from data is at the heart of the scientific method. In many scientific endeavors, causal knowledge is seen as the ultimate goal of the scientific investigation. Causal knowledge not only provides the basic understanding of how a system works, but is crucial to predicting how a system will behave when subject to intervention or perturbation. Consequently, it is central to just about everything from informing policy decisions on the basis of social and economic data to finding cures for diseases using optimal experimental regimes. Ultimately, causal notions underlie most of our deliberate decisions that we make about our daily lives.
Attempts to characterize causal relations can be found in some of the oldest written records, yet over the past 100 years, the history of the usage of causal concepts within scientific discussions has been rocky, varying from the outright denial of any role of causality in mature scientific theories to a disingenuous usage of alternative terms that play the role of cause and effect without mentioning the “c-word” (consider, for example, link, impact, strong connection, dependent vs. independent variables in regression, etc.).
It then comes as no surprise that the absence of an integration of such notions as causation, intervention, and counterfactuals underlies much of the criticism of purely predictive approaches, such as deep neural nets. Prediction from observed data about a system and prediction given an intervention on the system are distinct tasks, but they are not unrelated. But while the field of causal inference provides us with a rich formal language for expressing the effect of variables on one another via interventions, a major limitation of the field is that it does not yet yield efficient algorithms for discovering causal structure or the computation of causal effects.
Against this background, the recent award of the Nobel Memorial Prize in economics to David Card, Joshua Angrist, and Guido Imbens, who have "shown what conclusions about cause and effect can be drawn from natural experiments," is a striking recognition of the need for estimation techniques for causal quantities in economics, and the social and natural sciences more generally. All three researchers made major contributions to the “instrumental variable” techniques that allow researchers to identify causal effects in observational data. For example, in 1991, Angrist and Krueger, using U.S. census data, studied the causal effect of education on income, which is challenging to measure due to unobserved confounders. Is it education that causes the higher income, or are there background factors, such as socioeconomic status, that explain the correlation between education and income? Angrist and Krueger proposed the use of the quarter of the year in which a child was born as an “instrument” — that is, a variable that is independent of other potential confounders, such as socioeconomic status — of the education and income relation, but that still is dependent on education. The upshot is that this sort of “instrument” provides variation in education that forms a sort of natural experiment, which in turn allows for the development of statistical estimators of the causal effect of education on income. This reasoning relies on several assumptions that have been subject to scrutiny and debate, but this pioneering work has provided the basis for a principled and applicable approach to causal inference, which has been adopted in numerous studies in economics and the social sciences.
“Every traditional statistics text includes a warning not to mistake correlation for causality. But until very recently, there has been almost nothing else to say about causality,” observes Daniel Goroff, director of the Sloan Foundation’s economics program. “Research on the theory and practice of causal inference, as facilitated by the Simons Institute, will be critical to progress on everything from machine learning to evidence-based policymaking.”
The Alfred P. Sloan Foundation is a not-for-profit, mission-driven grant-making institution dedicated to improving the welfare of all through the advancement of scientific knowledge. Established in 1934 by Alfred Pritchard Sloan Jr., then the president and CEO of the General Motors Corp., the foundation makes grants in four broad areas: direct support of research in science, technology, engineering, mathematics, and economics; initiatives to increase the quality, equity, diversity, and inclusiveness of scientific institutions and the science workforce; projects to develop or leverage technology to empower research; and efforts to enhance and deepen public engagement with science and scientists.
The Simons Institute’s grant from the Sloan Foundation supports long-term participants in the Causality program, with an eye toward advancing the intersections of foundational research on causality with the foundation’s particular interest in empirical economics. Previous grants to the Simons Institute from the Sloan Foundation have supported our research programs on Data Privacy (2019) and Lattices (2020).