Winter 2026 (in progress)

See our homepage for future talks.

• Tuesday, Feb 17, 2026: Young Researchers' Seminar
  Time: This event starts at 8:30 am PT/ 11:30 am ET/ 4:30 pm London time/ 00:30 am Beijing time(+1day)
  Zoom details: Zoom link (webinar ID: 968 8371 7451, password: 414559)
  - Speaker 1: Anna Guo (Emory University)
  - Title: Average Causal Effect Estimation in DAGs with Hidden Variables: Beyond Back-Door and Front-Door Criteria
  - Abstract: This talk focuses on flexible estimation of causal effects in hidden-variable graphical models. While the identification theory for causal effects in directed acyclic graphs (DAGs) with hidden variables is well developed, methods for estimation and inference of causal functionals beyond the g-formula remain limited.
  In the first part, we introduce novel one-step–corrected plug-in estimators and targeted minimum loss-based estimators (TMLE) for causal effects in a class of hidden-variable DAGs that extends beyond the classical back-door and front-door criteria, known in the literature as treatment primal fixability. These estimators leverage data-adaptive machine learning methods to reduce modeling assumptions while retaining key statistical guarantees, including double robustness, boundedness within the parameter space, and asymptotic linearity under $L^2(P)$-rate conditions on nuisance function estimation, yielding root-$n$ consistent causal effect estimators.
  In the second part, we discuss a model known as the Napkin graph, which falls outside this convenient class and requires a nonstandard identification strategy, necessitating special consideration in estimation. This model also encodes a generalized independence restriction, known as a Verma constraint. We show how such constraints can be exploited both to improve estimation efficiency and to test model assumptions. 
  [Paper 1, Paper 2, Paper 3][Slides][Video]
  - Speaker 2: Catharina Stoltenberg (University of Oslo)
  - Title: Single-world exchangeability conditions for a large class of regimes
  - Abstract: We present new sufficient exchangeability conditions for identifying causal effects under a large class of regimes, including dynamic regimes that depend on natural treatment values (NTVs). One of the proposed conditions is weaker and more parsimonious than the established condition in Richardson and Robins (2013). The proposed conditions can be verified via d-separation in one single-world intervention graph (SWIG). This simplifies the established graphical identification arguments for a large class of dynamic regimes. To motivate the theory, we introduce a type of NTV regime called the add-on regime. We show that this regime is practically relevant in many applied settings, including a real-world application on opioid use.
  This presentation will be based on joint work with Mats Stensrud.
  [Slides][Video]

• Tuesday, Feb 10, 2026:  OCIS+INI joint webinar
  - Speaker: Lin Liu (Shanghai Jiao Tong University (SJTU))
  - Time: This event starts at 8:30 am PT/ 11:30 am ET/ 4:30 pm London time/ 00:30 am Beijing time(+1day)
  - Zoom details: Zoom link,  meeting ID: 819 2387 7168​, passcode: Newton1
  - Title: Method-of-moments, U-statistics, and high-dimensional GLMs when $p = O (n)$
  - Abstract: In this talk, I will discuss several recent results in the direction of parameter estimation in high-dimensional GLMs when $p = O (n)$. In the first part, I will explain how a class of U-statistic-based estimators can be used for parameter estimation in high-dimensional linear regression when $p = O (n)$. In particular, I will also specialize the discussion to the problem of adjusting for high-dimensional baseline covariates in randomized experiments and demonstrate the advantageous statistical properties of U-statistic-based ATE estimators in various settings, compared to popular benchmarks. This class of estimators can cover popular study designs such as covariate adaptive randomization. In the second part, I will move to parameter estimation in high-dimensional GLMs. Many interesting root-n consistent estimators exist for linear or quadratic functionals of the regression coefficients of GLMs. However, I will argue that the same class of U-statistic-based estimators can also achieve similar theoretical properties, which, in contrast to other root-n consistent estimators, can be proved by almost elementary arguments.
  [Paper1, Paper2][Slides][Video]

• Tuesday, Feb 3rd, 2026: Young Researchers' Seminar
  - Speaker 1: Herb Susmann (New York University)
  - Title: Non-overlap Average Treatment Effect Bounds
  - Abstract: The average treatment effect (ATE), the mean difference in potential outcomes under treatment and control, is a canonical causal effect. Overlap, which says that all subjects have non-zero probability of either treatment status, is necessary to identify and estimate the ATE. When overlap fails, the standard solution is to change the estimand, and target a trimmed effect in a subpopulation satisfying overlap; however, this no longer addresses the original goal of estimating the ATE. When the outcome is bounded, we demonstrate that this compromise is unnecessary. We derive non-overlap bounds: partial identification bounds on the ATE that do not require overlap. They are the sum of a trimmed effect within the overlap subpopulation and worst-case bounds on the ATE in the non-overlap subpopulation. Non-overlap bounds have width proportional to the size of the non-overlap subpopulation, making them informative when overlap violations are limited -- a common scenario in practice. Since the bounds are non-smooth functionals, we derive smooth approximations of them that contain the ATE but can be estimated using debiased estimators leveraging semiparametric efficiency theory. Specifically, we propose a Targeted Minimum Loss-Based estimator that is root-n consistent and asymptotically normal under nonparametric assumptions on the propensity score and outcome regression. We then show how to obtain a uniformly valid confidence set across all trimming and smoothing parameters with the multiplier bootstrap. This allows researchers to consider many parameters, choose the tightest confidence interval, and still attain valid coverage. We demonstrate via simulations that non-overlap bound estimators can detect non-zero ATEs with higher power than traditional doubly-robust point estimators. We illustrate our method by estimating the ATE of right heart catheterization on mortality.
  [Paper][Slides][Video]
  - Speaker 2: Juraj Bodík (University of Lausanne, Switzerland)
  - Title: Causality and Extreme Events: Why Additive Models Can Be Dangerous and What to Do Instead
  - Abstract: Extreme events, outliers, and heavy-tailed behavior are the rule rather than the exception in many causal systems. Although additive-noise models of the form Y=f(X)+e massively dominate the literature, they can be a poor choice in such systems, when causal effects act primarily through the tail or the variance. This can potentially lead to seriously misleading conclusions. In this talk, I discuss two complementary approaches for causal reasoning in such regimes. For i.i.d. data, I introduce Conditionally Parametric Causal Models (CPCM), which move beyond mean-oriented additive formulations to accommodate heteroscedasticity and heavy tails. For time series, where temporal structure can be an ally rather than a nuisance, I present a framework for Granger-type causality in extremes, designed to detect and characterize causal links solely from extreme events.
  [Paper1, Paper2, Paper3][Slides][Video]

• Tuesday, Jan 27, 2025: OCIS+INI joint webinar
  - Speaker: Victor Chernozhukov (MIT)
  - Title: Adventures in Demand Analysis Using AI
  - Abstract: This paper advances empirical demand analysis by integrating multimodal product representations derived from artificial intelligence (AI). Using a detailed dataset of toy cars on Amazon.com, we combine text descriptions, images, and tabular covariates to represent each product using transformer-based embedding models. These embeddings capture nuanced attributes, such as quality, branding, and visual characteristics, that traditional methods often struggle to summarize. Moreover, we fine-tune these embeddings for causal inference tasks. We show that the resulting embeddings substantially improve the predictive accuracy of sales ranks and prices and that they lead to more credible causal estimates of price elasticity. Notably, we uncover strong heterogeneity in price elasticity driven by these product-specific features. Our findings illustrate that AI-driven representations can enrich and modernize empirical demand analysis. The insights generated may also prove valuable for applied causal inference more broadly.

• Tuesday, Jan 20, 2025: OCIS+INI joint webinar
  - Speaker: Paul Rosenbaum (University of Pennsylvania)
  - Title: Being Realistic About Unmeasured Biases in Observational Studies
  - Abstract: Observational studies of the effects caused by treatments are always subject to the concern that an ostensible treatment effect may reflect a bias in treatment assignment, rather than an effect actually caused by the treatment. The degree of legitimate concern is strongly affected by simple decisions that an investigator makes during the design and analysis of an observational study. Poor choices lead to heightened concern; that is, poor choices make a study sensitive to small unmeasured biases where better choices would correctly report insensitivity to larger biases. Indeed, perhaps surprisingly, unambiguous evidence of the presence of unmeasured bias may increase insensitivity to unmeasured bias. These issues are discussed with the aid of some theory and a simple example of an observational study.

• Tuesday, Jan 13, 2026: Zoom link (webinar ID: 968 8371 7451, password: 414559)  
  - Speaker: Yilin Song (Columbia University) & Richard Guo (University of Michigan)
  - Title: The Categorical Instrumental Variable Model: Characterization, Partial Identification, and Statistical Inference
  - Abstract: We study categorical instrumental variable (IV) models with instrument, treatment, and outcome taking finitely many values. We derive a simple closed-form characterization of the set of joint distributions of potential outcomes that are compatible with a given observed data distribution in terms of a set of inequalities. These inequalities unify several different IV models defined by versions of the independence and exclusion restriction assumptions and are shown to be non-redundant. Finally, given a set of linear functionals of the joint counterfactual distribution, such as pairwise average treatment effects, we construct confidence intervals with simultaneous finite-sample coverage, using a tail bound on the Kullback--Leibler divergence. We illustrate our method using data from the Minneapolis Domestic Violence Experiment.
  - Discussant: Desire Kedagni (University of North Carolina - Chapel Hill)
  [Slides][Paper][Video][Discussion slides]