The Causal Inference across Fields: Methods, Insights, and Applications Workshop
The Causal Inference Across Fields: Methods, Insights, and Applications Workshop will explore how causal inference bridges theory and practice across disciplines, and how data science methods apply to real-world problems and foster greater connections between applied researchers and data science audiences.
The Workshop is part of the DSI Causal Inference Emerging Data Science Emergent Data Science Program that aims to facilitate cross-disciplinary exchange, where applied researchers from different disciplines can present their research questions and methodological issues. In turn, data science and causality researchers explore new and existing methods while promoting their research agendas.
Join us in-person to foster collaborative exploration, amplifying the impact of causal inference and data science research on real-world policy challenges.
| April 9, 2025 | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
9:15-9:45 |
Registration |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
9:45-10:00 |
Opening Remarks Rahul G. Krishnan Assistant Professor, Department of Computer Science, Faculty of Arts & Science, and Department of Laboratory Medicine and Pathobiology, Temerty Faculty of Medicine, University of Toronto |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
10:00-11:00 |
Elias Bareinboim
Associate Professor, Department of Computer Science and Director, Causal Artificial Intelligence Lab, Columbia University |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
11:00-12:00
|
Putting School Surveys to the Test
Parag Pathak Professor, Department of Economics, MIT School districts increasingly gauge school quality with surveys that ask about school climate and student engagement. We use data from New York City's middle and high schools to compare the long-run predictive validity of surveys with conventional test score value-added models (VAMs). Our analysis leverages the New York school match, which includes an element of random assignment, to validate a wide range of school quality estimates. We contrast the predictiveness of survey- and test-based measures of school quality for school effects on consequential outcomes related to high school graduation and college enrollment. Survey data generate better predictions of school impacts on high school graduation than test scores. School effects on advanced high school diplomas and college attainment are better predicted by test score VAMs than surveys. The practical value of test-based and survey-based school quality measures is quantified by simulating the effects of access to one or both types of information for parents. Parents interested in boosting their children's college attainment benefit more from test score value-added than from survey data. |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
12:00-1:00 |
Networking and Lunch (poster session) |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
1:00-2:00 |
Where the Rubber Meets the Road: Examining Efficiency and Equity in Designing Summer Youth Employment Programs
Alicia Modestino Associate Professor, School of Public Policy and Urban Affairs and Department of Economics, Northeastern University Summer Youth Employment Programs in the U.S. have been shown to have significant impacts on youth outcomes such as reducing violent crime, increasing high school graduation, and boosting subsequent employment and wages. Much of this research is based on lotteries from oversubscribed programs from a handful of cities which randomly assign youth to jobs. But most cities cannot use simple random assignment to allocate youth to jobs due to heterogeneous preferences of employers and youth participants. What happens to efficiency (e.g., number of jobs filled) and equity (e.g., which youth are selected for employment) under alternative assignment mechanisms? Using administrative data from the City of Boston’s hiring platform, we study youth application and employer selection behavior to explore these design challenges. Our findings reveal that despite having clearly stated goals of reducing inequality, workforce development programs are likely to result in job placements that perpetuate inequities found in the broader labor market in the absence of simple random assignment. Yet relying solely on automated placements to improve equity in the absence of additional in-person supports may ultimately reduce program efficiency. |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
2:00-3:00 |
Unmasking Time-Varying Unmeasured Confounding using Latent Variables Kuan Liu Institute of Health Policy, Management and Evaluation Dalla Lana School of Public Health, University of Toronto In observational research, time-varying unmeasured confounding poses a central challenge to accurately estimating causal effects. In this talk, we present two works that are designed or have the potential to address time-varying unmeasured confounding. In the first work, we present two causal sensitivity analysis approaches, namely the latent variable approach and the sensitivity confounding function approach, for causal estimation with unmeasured time-varying confounding. The latent variable approach incorporates unobserved confounding as hidden variables in the causal models. This approach allows researchers to integrate external data or prior knowledge about the unmeasured confounding and is easier to interpret and conceptualize in applications. The sensitivity function approach directly characterizes the net bias caused by unmeasured time-varying confounding without explicitly introducing latent variables. This approach avoids making distributional assumptions about the unmeasured confounding but can be difficult to interpret in practice. We investigate the performance of these methods in a series of simulation studies and provide a brief discussion on their strengths and limitations. In the second work, we introduce a latent confounding class framework that has the potential to handle time-varying unmeasured confounding. This approach presents a causal setting that utilizes a time-varying latent confounding class to represent a patient’s time-varying confounding profile. The introduction of these latent confounder classes permits a full Bayesian estimation of the causal effects via g-computation. We examine and illustrate this approach using simulated and clinical data. |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
3:00-3:30 |
Networking and coffee break
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
3:30-4:30 |
Student Round Table |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
4:30-5:00 |
Closing remarks Rahul G. Krishnan |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Speakers
Elias Bareinboim
Associate Professor, Department of Computer Science and Director, Causal Artificial Intelligence Lab, Columbia University
Alicia Modestino
Associate Professor, School of Public Policy and Urban Affairs and Department of Economics, Northeastern University
Dr. Modestino is an Associate Professor with appointments in the School of Public Policy and Urban Affairs and the Department of Economics at Northeastern University, where she also serves as the Research Director of the Dukakis Center for Urban and Regional Policy. Her research primarily focuses on labor market dynamics including skills mismatch, youth labor market attachment, and career pathways. She currently leads a multi-year Institutional Challenge Grant funded by the William T. Grant Foundation to evaluate the Boston Summer Youth Employment Program.
Last year, Modestino launched a new $4.5M initiative across Northeastern’s global campus network: Community to Community (C2C): Policy Equity for All. Working in partnership with city departments, state agencies, and community based organizations C2C provides rigorous data and analysis to find solutions to the most urgent public problems at each campus location. In 2024, she was the researcher recipient of JPAL’s Evidence Champion award for her “extraordinary contributions to the field of evidence-based policymaking.”
Dr. Modestino received her Ph.D. in Economics from Harvard University where she was also a fellow in the Inequality and Social Policy Program. She is currently a Nonresident Fellow at the Brookings Institution and an Affiliated Researcher of the Abdul Latif Jameel Poverty Action Lab (J-PAL). Previously, she was a Senior Economist at the Federal Reserve Bank of Boston where she conducted research on regional economic and policy issues for over a decade.
Parag Pathak
Professor, Department of Economics, MIT
Parag Pathak is the Class of 1922 Professor of Economics at MIT and founding co-director of MIT’s Blueprint Labs. He is also the founding co-director of the NBER Working Group on Market Design. Pathak’s research has garnered several recognitions including a Presidential Early Career Award for Scientists and Engineers and the 2016 Social Choice and Welfare Prize (with Fuhito Kojima). In 2018, he was awarded the John Bates Clark Medal by the American Economic Association as the best American economist under the age of 40.
Kuan Liu
Assistant Professor, Institute of Health Policy, Management and Evaluation
Dalla Lana School of Public Health, University of Toronto
Kuan Liu is an Assistant Professor at the Institute of Health Policy, Management & Evaluation (IHPME) at the University of Toronto. She holds a Master of Mathematics in Statistics-Biostatistics from the University of Waterloo and a PhD in Biostatistics from the University of Toronto. At IHPME, Kuan leads a research and training program on causal inference and Bayesian analysis in clinical and public health research and her program has attracted funding support from CIHR, NSERC, the Data Sciences Institute, and the Institute for Pandemics at the University of Toronto. Kuan is the current Chair of the Student and Recent Graduate Committee of the Statistical Society of Canada and the Executive Secretary of the Health Policy Statistics Section of the American Statistical Association.
Rahul G. Krishnan
Assistant Professor, Department of Computer Science, Faculty of Arts & Science, and Department of Laboratory Medicine and Pathobiology, Temerty Faculty of Medicine, University of Toronto
Rahul Krishnan received his PhD from MIT advised by Prof. David Sontag where he was part of the Clinical ML Group. He received a BASc in Computer Engineering at the University of Toronto and an MS from New York University. Rahul was previously a senior researcher at Microsoft Research New England. He is expanding the Artificial Intelligence education in LMP (alongside Dr. Bo Wang) and has recently launched a new graduate course on Machine Learning for Healthcare.