Guido Imbens

This article analyzes the results of the 2024 ADIA Lab Causal Discovery Challenge, a large-scale competition designed to evaluate whether machine learning methods can infer causal roles from observational data when ground-truth causal structures are known. Participants classified variables into eight causal categories—including confounders, mediators, colliders, and causal antecedents or consequences of treatment and outcome variables—using 47,000 synthetic datasets generated from known causal directed acyclic graphs. The Challenge attracted 1,904 registered participants and 3,343 submitted solutions worldwide. Our results show that supervised learning approaches substantially outperformed traditional constraint-based causal discovery methods in this controlled setting. Top-performing solutions achieved multiclass balanced accuracies as high as 77%, compared with approximately 40% for baseline methods. The most successful approaches combined rich feature engineering, edge-centric representations, ensemble learning, and graph-based architectures inspired by graph neural networks. Performance differences across graph structures revealed that some causal environments are systematically more difficult for discovery algorithms than others. The findings suggest that, when large libraries of labeled dataset–graph pairs are available, causal-role recognition can be effectively approached as a supervised learning problem, with potential applications in factor research, macro-financial forecasting, stress testing, and portfolio construction.

Completely randomized experiments, originally developed by Fisher and Neyman in the 1930s, are still widely used in practice, even in online experimentation. However, such designs are of limited value for answering standard questions in marketplaces, where multiple populations of agents interact strategically, leading to complex patterns of spillover effects. In this article, we derive the finite-sample properties of tractable estimators for ‘Simple Multiple Randomization Designs’, a new class of experimental designs which account for complex spillover effects in randomized experiments. Our derivations are obtained under a natural and general form of cross-unit interference, which we call ‘local interference’. We discuss the estimation of main effects, direct effects, and spillovers, and present associated central limit theorems.

We propose a method for estimating long-term treatment effects with many short-term proxy outcomes: a central challenge when experimenting on digital platforms. We formalize this challenge as a latent variable problem where observed proxies are noisy measures of a low-dimensional set of unobserved surrogates that mediate treatment effects. Through theoretical analysis and simulations, we demonstrate that regularized regression methods substantially outperform naive proxy selection. We show in particular that the bias of Ridge regression decreases as more proxies are added, with closed-form expressions for the bias-variance tradeoff. We illustrate our method with an empirical application to the California GAIN experiment.

Causality and causal inference have emerged as core research areas at the interface of modern statistics and domains including biomedical sciences, social sciences, computer science, and beyond. The field's inherently interdisciplinary nature -- particularly the central role of incorporating domain knowledge -- creates a rich and varied set of statistical challenges. Much progress has been made, especially in the last three decades, but there remain many open questions. Our goal in this discussion is to outline research directions and open problems we view as particularly promising for future work. Throughout we emphasize that advancing causal research requires a wide range of contributions, from novel theory and methodological innovations to improved software tools and closer engagement with domain scientists and practitioners.

We study causal inference in settings characterized by interference with a bipartite structure. There are two distinct sets of units: intervention units to which an intervention can be applied and outcome units on which the outcome of interest can be measured. Outcome units may be affected by interventions on some, but not all, intervention units, as captured by a bipartite graph. Examples of this setting can be found in analyses of the impact of pollution abatement in plants on health outcomes for individuals, or the effect of transportation network expansions on regional economic activity. We introduce and discuss a variety of old and new causal estimands for these bipartite settings. We do not impose restrictions on the functional form of the exposure mapping and the potential outcomes, thus allowing for heterogeneity, non-linearity, non-additivity, and potential interactions in treatment effects. We propose unbiased weighting estimators for these estimands from a design-based perspective, based on the knowledge of the bipartite network under general experimental designs. We derive their variance and prove consistency for increasing number of outcome units. Using the Chinese high-speed rail construction study, analyzed in Borusyak and Hull [2023], we discuss non-trivial positivity violations that depend on the estimands, the adopted experimental design, and the structure of the bipartite graph.

Empirical research in economics often examines the behavior of agents located in a geographic space. In such cases, statistical inference is complicated by the interdependence of economic outcomes across locations. A common approach to account for this dependence is to cluster standard errors based on a predefined geographic partition. A second strategy is to model dependence in terms of the distance between units. Dependence, however, does not necessarily stop at borders and is typically not determined by distance alone. This paper introduces a method that leverages observations of multiple outcomes to adjust standard errors for cross-sectional dependence. Specifically, a researcher, while interested in a particular outcome variable, often observes dozens of other variables for the same units. We show that these outcomes can be used to estimate dependence under the assumption that the cross-sectional correlation structure is shared across outcomes. We develop a procedure, which we call Thresholding Multiple Outcomes (TMO), that uses this estimate to adjust standard errors in a given regression setting. We show that adjustments of this form can lead to sizable reductions in the bias of standard errors in calibrated U.S. county-level regressions. Re-analyzing nine recent papers, we find that the proposed correction can make a substantial difference in practice.

We study the identification and estimation of long-term treatment effects by combining short-term experimental data and long-term observational data subject to unobserved confounding. This problem arises often when concerned with long-term treatment effects since experiments are often short-term due to operational necessity while observational data can be more easily collected over longer time frames but may be subject to confounding. In this paper, we tackle the challenge of persistent confounding: unobserved confounders that can simultaneously affect the treatment, short-term outcomes, and long-term outcome. In particular, persistent confounding invalidates identification strategies in previous approaches to this problem. To address this challenge, we exploit the sequential structure of multiple short-term outcomes and develop several novel identification strategies for the average long-term treatment effect. Based on these, we develop estimation and inference methods with asymptotic guarantees. To demonstrate the importance of handling persistent confounders, we apply our methods to estimate the effect of a job training program on long-term employment using semi-synthetic data.

In this paper, we study the design and analysis of experiments conducted on a set of units over multiple time periods in which the starting time of the treatment may vary by unit. The design problem involves selecting an initial treatment time for each unit in order to most precisely estimate both the instantaneous and cumulative effects of the treatment. We first consider nonadaptive experiments, in which all treatment assignment decisions are made prior to the start of the experiment. For this case, we show that the optimization problem is generally NP-hard, and we propose a near-optimal solution. Under this solution, the fraction entering treatment each period is initially low, then high, and finally low again. Next, we study an adaptive experimental design problem, in which both the decision to continue the experiment and treatment assignment decisions are updated after each period’s data are collected. For the adaptive case, we propose a new algorithm, the precision-guided adaptive experiment algorithm, which addresses the challenges at both the design stage and the stage of estimating treatment effects, ensuring valid post-experiment inference, accounting for the adaptive nature of the design. Using realistic settings, we demonstrate that our proposed solutions can reduce the opportunity cost of the experiments by more than 50%, compared with static design benchmarks. This paper was accepted by George Shanthikumar, data science. Funding: S. Athey and G. Imbens were supported by the Office of Naval Research [Grant N00014-19-1-2468]. M. Bayati was supported by the National Science Foundation [Grant CMMI 1554140]. Supplemental Material: The e-companion and data are available at https://doi.org/10.1287/mnsc.2023.4928 .

Summary: In this survey we discuss the recent causal panel data literature. This recent literature has focused on credibly estimating causal effects of binary interventions in settings with longitudinal data, emphasising practical advice for empirical researchers. It pays particular attention to heterogeneity in the causal effects, often in situations where few units are treated and with particular structures on the assignment pattern. The literature has extended earlier work on difference-in-differences or two-way fixed effect estimators. It has more generally incorporated factor models or interactive fixed effects. It has also developed novel methods using synthetic control approaches.