Shortcourses

SCI Shortcourses will take place at the 2025 American Causal Inference Conference (ACIC), held in Detroit, MI, on Tuesday, May 13th.

RATES

Student Member – $80
Non-Student Member – $110
Student Non-Member – $180
Non-Student Non-Member – $210

For information on shortcourses please visit our Submissions page!

Proximal Causal Inference with Proxies: Valid and Invalid

DATE: May 13, 2025
TIME: 8:30 A.M. – 12:30 P.M.

This course introduces proximal causal inference, a recently proposed framework for valid causal inference in presence of hidden confounding factors for which proxies are available. The lecture consists of two parts, focusing on both valid and invalid proxies in causal inference. Read more

INSTRUCTORS:

Prabisha Rakshit

Chan Park

Causal Graphical Methods For Handling Nonignorable Missing Data

DATE: May 13, 2025
TIME: 8:30 A.M. – 12:30 P.M.

It is often said that the fundamental problem of causal inference is a missing data problem. The focus of this course is on the implications of the converse view: that missing data problems are a form of causal inference. Read more

We start off by providing a quick overview of classical approaches to missing data and move on to redefining missing data models using the terminology of causal models, where missingness indicators are viewed as treatment variables that can be intervened on, and the underlying variables are viewed as counterfactuals, i.e., variables had we (possibly contrary to fact) been able to observe them. We focus on missing data models where restrictions may be represented by a factorization of the full data law with respect to a directed acyclic graph (DAG) with potentially latent variables. In addition to providing concise representations of statistical models, missing data DAGs (m-DAGs) illustrate the causal mechanisms responsible for missingness and provide a natural interpretation of such mechanisms in applied settings. Using the m-DAG representations, we discuss various approaches for identifying a given target parameter as a function of the observed data distribution. We discuss how to carry out non/semi-parametric inference for the identified targets, primarily under missing-not-at-random mechanisms. Finally, we introduce methods for conducting sensitivity analysis and illustrate, via a case study, how the methods can be employed in practice.

Course Objectives:

With the growing popularity of graphical modeling approaches in statistics and the ever-lasting issue of missing data in practice, we believe this short course will provide a unique opportunity to educate scholars with different scientific backgrounds in (1) formalizing the missingness mechanisms more precisely, (2) evaluating identification and estimation assumptions more appropriately, and (3) selecting the best inference strategies. Upon conclusion of the course, attendees should be able to:

Represent MNAR mechanisms graphically, transferring expert knowledge into a concise graphical representation.
Interpret a missing data DAG model as a class of distributions with a set of independence restrictions.
Follow the steps in identifying a given query, write it down as a function of observed data, or prove the given query of interest is not identified without making further assumptions.
Follow the steps in deriving the non-parametric influence functions for an identified query.
Follow the steps in deriving the tangent space of the underlying full data and observed data distributions for a given missing data DAG.
Follow the general idea of assessing assumptions and interpreting findings using the discussed sensitivity analysis framework.

INSTRUCTORS:

Razieh Nabi

Daniel Scharfstein

Optimal Transport and Causal Inference: A Contemporary Perspective

DATE: May 13, 2025
TIME: 1:00PM – 5:00P.M.

Optimal Transport (OT) provides a unifying framework for solving problems of matching, distributional alignment, and robust optimization, making it a powerful tool for advancing causal inference. This course introduces the connections between OT and causal inference, covering both discrete and continuous OT frameworks. Read more

The first part focuses on Discrete OT, which generalizes classical matching methods, such as propensity-score matching and nearest-neighbor matching, by optimizing pairwise assignments under a cost structure. We examine state-of-the-art methods for large-scale matching, connections to balancing weights, and recent advancements in machine-learning-guided matching techniques. These approaches reduce bias and variance in causal estimates while offering scalable and flexible solutions for high-dimensional settings.

The second part explores Continuous OT, extending to transformations of probability distributions. We introduce Wasserstein distances and their relevance to causal inference under distributional bias, interpret TMLE as iterative updates along paths in Wasserstein space, and discuss how gradient flows in this space can optimize estimators in semiparametric models, with an emphasis on scalable computational techniques like entropic regularization. By integrating with the geometry of distributions, this course equips participants with cutting-edge tools for designing and analyzing robust causal studies, even in high-dimensional and dynamic settings.

Course Objectives:
1. Deep Integration: Identify cutting-edge approaches that unify discrete OT (matching) with classical and modern causal inference strategies (propensity-score matching, balancing weights, machine learning-based matching).
2. Geometry of Probability Measures: Explain the fundamental concepts of continuous OT (Wasserstein metrics, geodesics, gradient flows) with an eye toward causal inference challenges (robustness, distributional shifts).
3. Advanced Methodological Insights: Analyze how targeted maximum likelihood estimation (TMLE) can be interpreted as iteratively moving in probability space, drawing on recent theoretical connections to Wasserstein geometry.
4. Research Frontier: Assess recent researchâ€”both theoretical and empiricalâ€”that leverages OT-based optimization (e.g., entropic regularization, large-scale computational frameworks) for improved causal inference designs.
5. Open Problems and Future Directions: Propose next steps for harnessing OT in large-scale, high-dimensional, or dynamic treatment settings where classical matching or parametric submodels are inadequate.

INSTRUCTOR:

Kaiwen Hou

Beyond the Average Treatment Effect: Estimating the causal effects of binary, categorical, continuous, and multivariate exposures in R

DATE: May 13, 2025
TIME: 1:00PM – 5:00P.M.

We tend to be most familiar with estimating the effects of binary treatments or exposures. The classic average treatment effect (ATE), risk difference, risk ratio, and odds ratio are all examples of this. However, the exposure may be more complicated than a simple binary variable. Read more

SCI Shortcourses will take place at the 2025 American Causal Inference Conference (ACIC), held in Detroit, MI, on Tuesday, May 13th.

RATES

Proximal Causal Inference with Proxies: Valid and Invalid

INSTRUCTORS:

Prabisha Rakshit

Chan Park

Causal Graphical Methods For Handling Nonignorable Missing Data

INSTRUCTORS:

Razieh Nabi

Daniel Scharfstein

Optimal Transport and Causal Inference: A Contemporary Perspective

INSTRUCTOR:

Kaiwen Hou

Beyond the Average Treatment Effect: Estimating the causal effects of binary, categorical, continuous, and multivariate exposures in R

INSTRUCTORS:

Nicholas Williams

Kara Rudolph

Ivan Diaz