Keynote Speakers

Title: Towards Complete Causal Explanation with Expert Knowledge

Abstract: We study the problem of restricting Markov equivalence classes of maximal ancestral graphs(MAGs) containing certain edge marks, which we refer to as expert knowledge. MAGs forming a Markov equivalence class can be uniquely represented by an essential ancestral graph. We seek to learn the restriction of the essential ancestral graph containing the proposed expert knowledge. Our contributions are several-fold. First, we prove certain properties for the entire Markov equivalence class, including a conjecture from Ali et al. (2009). Second, we present three sound graphical orientation rules, two of which generalize previously known rules, for adding expert knowledge to an essential graph. We also show that some orientation rules of Zhang (2008) are unnecessary for restricting the Markov equivalence class with expert knowledge. We provide an algorithm for including this expert knowledge and show that our algorithm is complete in certain settings. Meaning that, in these settings, the output of our algorithm is a restricted essential ancestral graph. We conjecture this algorithm is complete generally. Outside of our specified settings, we provide an algorithm for checking whether a graph is a restricted essential graph and discuss its runtime. This work can be seen as a generalization of Meek (1995).

Title: Causal AI to Generate New Managerial Insights

Abstract: In today’s data-driven economy, businesses gather huge amounts of data, largely influenced by datafication and the Internet of Things (IoT). Business managers are increasingly creating comprehensive data acquisition strategies to collect big data across a wide range of features or variables. This trend is further emphasized by the emergence of data-sharing platforms with advanced data clean rooms, which allow organizations to make use of shared data while safeguarding privacy. However, as businesses gain access to large datasets, the complexity of analyzing potential connections between variables becomes overwhelming. For example, N variables can have N*(N-1) directional connections, resulting in exponential growth of possible connections as more variables are included. This complexity can lead to situations where businesses have a lot of data but not enough insights. To address this complexity, causal AI becomes a critical tool. While traditional machine learning techniques identify patterns in data, these patterns are merely correlational and not sufficient for strategic decision-making. Managers need to identify causal relationships to make informed interventions. Traditional methods like A/B testing are expensive and time-consuming, especially with high-dimensional data. Causal AI tackles this challenge by identifying causal patterns in observational data, enabling businesses to prioritize experiments and validate causality more efficiently and reliably. This paper explores the transformative potential of causal AI in turning abundant data into actionable insights. 

Title: Higher order influence functions and the minimaxity and admissibility of double machine learning (DML) estimators under minimal assumptions

Abstract: For many functionals that arise in causal inference, DML estimators are the state-of-the-art, incorporating the good predictive performance of black-box machine learning algorithms; the decreased bias of doubly robust estimators; and the analytic tractability and bias reduction of sample splitting with cross fitting. Recently Balakrishnan, Wasserman and Kennedy (BWK) introduced a novel assumption-lean model that formalizes the problem of functional estimation when no complexity reducing assumptions (such as smoothness or sparsity) are imposed on the nuisance functions occurring in the functional’s first order influence function (IF 1 ). Then, for the integrated squared density and the expected conditional variance functionals, they showed that first-order estimators, which include DML estimators, based on IF 1 are rate minimax under squared error loss. However, earlier Liu, Mukherjee, and Robins (2020) had shown that, for these functionals, higher-order influence function (HOIF) based estimators (ie estimators that add a debiasing mth-order U-statistic to a first -order estimator) could have smaller risk (mean squared error) than the first order estimator. In this talk, I resolve this apparent paradox. I show that, although minimax, DML estimators of these functionals are (asymptotically) inadmissible under the BWK model because the risk of any first-order estimator is never less than that of the corresponding HOIF estimator and, under many laws, may be much greater. As a consequence, under many data generating laws, HOIF estimators can be used to show that actual coverage of nominal 1-alpha Wald confidence intervals centered at a DML estimator is less than nominal.

Title:  Targeted Machine Learning with Missing & Dependent Data

Abstract: Despite best intentions, data for causal inference are often subject to complex missingness and dependence. We illustrate with 3 real-world examples from the SEARCH Consortium to (1) assess disease burden and control; (2) evaluate effects among those known to be "at risk", and (3) estimate total effects when measurement is the key mediator. For each, we use hierarchical causal models to define Counterfactual Strata Effects, occurring when the subgroup of interest is impacted by the exposure/intervention. For estimation and inference of these effects, we use TMLE with Super Learner, harnessing recent advances in machine learning and appropriately accounting for uncertainty. We conclude with practical recommendations and areas of ongoing work.