arXiv subjects

Eli Ben-Michael

Publications and source records attributed to Eli Ben-Michael.

At least 19 recordsLinked to original sources

arXiv2026

Learning Who to Treat When Treatment is Missing

Policy learning methods are increasingly used to inform treatment allocation under budget constraints. Most proposed methods assume complete treatment data, yet applications frequently suffer from missingness that can bias estimates and lead to suboptimal policies. We address this gap by extending efficient estimators for average treatment effect (ATE) estimation to policy value and conditional average treatment effect (CATE) estimation under missing at random (MAR) and missing completely conditionally at random (MCCAR) treatment data. Through asymptotic efficiency analysis, we prove that the MAR estimator, which leverages partially-observed units, is both valid and more efficient than the MCCAR estimator when MCCAR assumptions hold. This result provides formal justification for preferring MAR-based estimation in policy learning under both missing data settings. Our comprehensive experiments using synthetic and semi-synthetic datasets confirm that correctly specifying the missingness mechanism is crucial: misspecified estimators remain biased regardless of sample size, while our estimators achieve near-oracle performance when assumptions are satisfied. Our work provides practitioners with theoretically grounded, empirically validated tools for robust policy learning in the presence of missing treatment data.

cs.LG↗

arXiv2026

Bridging Predictions and Interventions: An Integrated Framework for Automated Decision-Systems

Automated decision systems (ADS) leverage predictions about individual future outcomes to inform consequential decision-making in organizational settings. Across various settings - including criminal pretrial release, clinical triage, student support, and more - it is often assumed that improved predictive accuracy is the priority consideration in determining better downstream outcomes upon the deployment of ADS. In practice, real-world case studies reveal that this is far from the case: introducing individual predictions into decision-making modifies organizational workflows, assessment, and decision-making processes in ways that require a complete re-consideration of our approach to the design, evaluation, and deployment of ADS. As a result, this Perspective develops an integrated framework for studying ADS in social systems, shifting current priorities from a purely prediction-based paradigm towards an intervention-oriented view that accounts for real-world conditions. Our aim is to improve our understanding of ADS and more meaningfully anticipate its downstream societal and organizational consequences.

cs.CY↗

arXiv2026

AI-Assisted Variance Reduction in Randomized Experiments

Generative AI and large language models can produce realistic predictions of human behavior from rich, unstructured inputs with little to no task-specific training data. Recent work uses these ``digital twin'' predictions to supplement human responses in surveys and experiments. We study the special case of using AI-generated predictions to reduce variance in randomized experiments. We argue that doing so requires no new estimators and that researchers can simply include AI predictions as covariates in standard regression adjustment, analogous to adjusting for a prognostic score. A benefit of this approach is a ``do no harm'' property whereby the adjusted estimator reverts to the unadjusted difference in means when predictions are uninformative. Other methods, such as variants of prediction-powered inference, do not have this guarantee. We provide implementation guidance, including how to obtain continuous scores from discrete LLM outputs and how to use LLMs to featurize unstructured inputs as auxiliary covariates. We demonstrate these ideas in simulations and three empirical applications: a survey mega-study, an email marketing A/B test, and a large-scale technology platform experiment. Overall, efficiency gains are real if modest, with greater benefits in studies that contain substantial text and other unstructured data. We also confirm the do no harm property empirically. Given these gains and limited costs, we recommend adjusting for AI-generated predictions as a regular empirical practice.

econ.EM↗

arXiv2026

Discussion of "Matrix Completion When Missing Is Not at Random and Its Applications in Causal Panel Data Models"

Choi and Yuan (2025) propose a novel approach to applying matrix completion to the problem of estimating causal effects in panel data. The key insight is that even in the presence of structured patterns of missing data -- i.e. selection into treatment -- matrix completion can be effective if the number of treated observations is small relative to the number of control observations. We applaud the authors for their insightful and interesting paper. We discuss this proposal from two complementary perspectives. First, we situate their proposal as an example of a "split-apply-combine" strategy that underlies many modern panel data estimators, including difference-in-differences and synthetic control approaches. Second, we discuss the issue of the statistical "last mile problem" -- the gap between theory and practice -- and offer suggestions on how to partially address it. We conclude by considering the challenges of estimating the impacts of public policies using panel data and apply the approach to a study on the effect of right to carry laws on violent crime.

stat.ME↗

arXiv2026

Omitted Variable Bias in Language Models Under Distribution Shift

Despite their impressive performance on a wide variety of tasks, modern language models remain susceptible to distribution shifts, exhibiting brittle behavior when evaluated on data that differs in distribution from their training data. In this paper, we describe how distribution shifts in language models can be separated into observable and unobservable components, and we discuss how established approaches for dealing with distribution shift address only the former. Importantly, we identify that the resulting omitted variable bias from unobserved variables can compromise both evaluation and optimization in language models. To address this challenge, we introduce a framework that maps the strength of the omitted variables to bounds on the worst-case generalization performance of language models under distribution shift. In empirical experiments, we show that using these bounds directly in language model evaluation and optimization provides more principled measures of out-of-distribution performance, improves true out-of-distribution performance relative to standard distribution shift adjustment methods, and further enables inference about the strength of the omitted variables when target distribution labels are available.

cs.LG↗

arXiv2026

A Weighting Framework for Clusters as Confounders in Observational Studies

When units in observational studies are clustered in groups, such as students in schools or patients in hospitals, researchers often address confounding by adjusting for cluster-level covariates or cluster membership. In this paper, we develop a unified weighting framework that clarifies how different estimation methods control two distinct sources of imbalance: global balance (differences between treated and control units across clusters) and local balance (differences within clusters). We show that inverse propensity score weighting (IPW) with a random effects propensity score model -- the current standard in the literature -- targets only global balance and constant level shifts across clusters, but imposes no constraints on local balance. We then present two approaches that target both forms of balance. First, hierarchical balancing weights directly control global and local balance through a constrained optimization problem. Second, building on the recently proposed Generalized Mundlak approach, we develop a novel Mundlak balancing weights estimator that adjusts for cluster-level sufficient statistics rather than cluster indicators; this approach can accommodate small clusters where all units are treated or untreated. Critically, these approaches rest on different assumptions: hierarchical balancing weights require only that treatment is ignorable given covariates and cluster membership, while Mundlak methods additionally require an exponential family structure. We then compare these methods in a simulation study and in two applications in education and health services research that exhibit very different cluster structures.

stat.ME↗

arXiv2025

Forest Kernel Balancing Weights: Outcome-Guided Features for Causal Inference

While balancing covariates between groups is central for observational causal inference, selecting which features to balance remains a challenging problem. Kernel balancing is a promising approach that first estimates a kernel that captures similarity across units and then balances a (possibly low-dimensional) summary of that kernel, indirectly learning important features to balance. In this paper, we propose forest kernel balancing, which leverages the underappreciated fact that tree-based machine learning models, namely random forests and Bayesian additive regression trees (BART), implicitly estimate a kernel based on the co-occurrence of observations in the same terminal leaf node. Thus, even though the resulting kernel is solely a function of baseline features, the selected nonlinearities and other interactions are important for predicting the outcome -- and therefore are important for addressing confounding. Through simulations and applied illustrations, we show that forest kernel balancing leads to meaningful computational and statistical improvement relative to standard kernel methods, which do not incorporate outcome information when learning features.

stat.ME↗

arXiv2025

Partial identification via conditional linear programs: estimation and policy learning

Many important quantities of interest are only partially identified from observable data: the data can limit them to a set of plausible values, but not uniquely determine them. This paper develops a unified framework for covariate-assisted estimation, inference, and decision making in partial identification problems where the parameter of interest satisfies a series of linear constraints, conditional on covariates. In such settings, bounds on the parameter can be written as expectations of solutions to conditional linear programs that optimize a linear function subject to linear constraints, where both the objective function and the constraints may depend on covariates and need to be estimated from data. Examples include estimands involving the joint distributions of potential outcomes, policy learning with inequality-aware value functions, and instrumental variable settings. We propose two de-biased estimators for bounds defined by conditional linear programs. The first directly solves the conditional linear programs with plugin estimates and uses output from standard LP solvers to de-bias the plugin estimate, avoiding the need for computationally demanding vertex enumeration of all possible solutions for symbolic bounds. The second uses entropic regularization to create smooth approximations to the conditional linear programs, trading a small amount of approximation error for improved estimation and computational efficiency. We establish conditions for asymptotic normality of both estimators, show that both estimators are robust to first-order errors in estimating the conditional constraints and objectives, and construct Wald-type confidence intervals for the partially identified parameters. These results also extend to policy learning problems where the value of a decision policy is only partially identified. We apply our methods to a study on the effects of Medicaid enrollment.

stat.ME↗

arXiv2025

Bridging Prediction and Intervention Problems in Social Systems

Many automated decision systems (ADS) are designed to solve prediction problems -- where the goal is to learn patterns from a sample of the population and apply them to individuals from the same population. In reality, these prediction systems operationalize holistic policy interventions in deployment. Once deployed, ADS can shape impacted population outcomes through an effective policy change in how decision-makers operate, while also being defined by past and present interactions between stakeholders and the limitations of existing organizational, as well as societal, infrastructure and context. In this work, we consider the ways in which we must shift from a prediction-focused paradigm to an intervention-oriented paradigm when considering the impact of ADS within social systems. We argue this requires a new default problem setup for ADS beyond prediction, to instead consider predictions as decision support, final decisions, and outcomes. We highlight how this perspective unifies modern statistical frameworks and other tools to study the design, implementation, and evaluation of ADS systems, and point to the research directions necessary to operationalize this paradigm shift. Using these tools, we characterize the limitations of focusing on isolated prediction tasks, and lay the foundation for a more intervention-oriented approach to developing and deploying ADS.

cs.LG↗

arXiv2025

Isolated Causal Effects of Natural Language

As language technologies become widespread, it is important to understand how changes in language affect reader perceptions and behaviors. These relationships may be formalized as the isolated causal effect of some focal language-encoded intervention (e.g., factual inaccuracies) on an external outcome (e.g., readers' beliefs). In this paper, we introduce a formal estimation framework for isolated causal effects of language. We show that a core challenge of estimating isolated effects is the need to approximate all non-focal language outside of the intervention. Drawing on the principle of omitted variable bias, we provide measures for evaluating the quality of both non-focal language approximations and isolated effect estimates themselves. We find that poor approximation of non-focal language can lead to bias in the corresponding isolated effect estimates due to omission of relevant variables, and we show how to assess the sensitivity of effect estimates to such bias along the two key axes of fidelity and overlap. In experiments on semi-synthetic and real-world data, we validate the ability of our framework to correctly recover isolated effects and demonstrate the utility of our proposed measures.

cs.CL↗

arXiv2025

Sensitivity Analysis for Clustered Observational Studies with an Application to the Effectiveness of Magnet Nursing Hospitals

In a clustered observational study, treatment is assigned to groups and all units within the group are exposed to the treatment. Here, we use a clustered observational study (COS) design to estimate the effectiveness of Magnet Nursing certificates for emergency surgery patients. Recent research has introduced specialized weighting estimators for the COS design that balance baseline covariates at the unit and cluster level. These methods allow researchers to adjust for observed confounders, but are sensitive to unobserved confounding. In this paper, we develop new sensitivity analysis methods tailored to weighting estimators for COS designs. We provide several key contributions. First, we introduce a key bias decomposition, tailored to the specific confounding structure that arises in a COS. Second, we develop a sensitivity framework for weighted COS designs that constrain the error in the underlying weights. We introduce both a marginal sensitivity model and a variance-based sensitivity model, and extend both to accommodate multiple estimands. Finally, we propose amplification and benchmarking methods to better interpret the results. Throughout, we illustrate our proposed methods by analyzing the effectiveness of Magnet nursing hospitals.

stat.ME↗

arXiv2025

Safe Policy Learning through Extrapolation: Application to Pre-trial Risk Assessment

Algorithmic recommendations and decisions have become ubiquitous in today's society. Many of these data-driven policies, especially in the realm of public policy, are based on known, deterministic rules to ensure their transparency and interpretability. We examine a particular case of algorithmic pre-trial risk assessments in the US criminal justice system, which provide deterministic classification scores and recommendations to help judges make release decisions. Our goal is to analyze data from a unique field experiment on an algorithmic pre-trial risk assessment to investigate whether the scores and recommendations can be improved. Unfortunately, prior methods for policy learning are not applicable because they require existing policies to be stochastic. We develop a maximin robust optimization approach that partially identifies the expected utility of a policy, and then finds a policy that maximizes the worst-case expected utility. The resulting policy has a statistical safety property, limiting the probability of producing a worse policy than the existing one, under structural assumptions about the outcomes. Our analysis of data from the field experiment shows that we can safely improve certain components of the risk assessment instrument by classifying arrestees as lower risk under a wide range of utility specifications, though the analysis is not informative about several components of the instrument.

stat.ML↗

arXiv2025

Longitudinal Causal Inference with Selective Eligibility

Dropout poses a significant challenge to causal inference in longitudinal studies with time-varying treatments. However, existing research does not simultaneously address dropout and time-varying treatments. We examine selective eligibility, an important yet overlooked source of non-ignorable dropout in such settings. This problem arises when a unit's prior treatment history influences its eligibility for subsequent treatments, a common scenario in medical and other settings. We propose a general methodological framework for longitudinal causal inference with selective eligibility. By focusing on a subgroup of units who would become eligible for treatment given a specific past treatment sequence, we define the time-specific eligible treatment effect and expected number of outcome events under a treatment sequence of interest. Under a generalized version of sequential ignorability, we derive two nonparametric identification formulae, each leveraging different parts of the observed data distribution. We then derive the efficient influence function of each causal estimand, yielding the corresponding doubly robust estimator. Finally, we apply the proposed methodology to an impact evaluation of a pre-trial risk assessment instrument in the criminal justice system, in which selective eligibility arises due to recidivism.

stat.ME↗

arXiv2025

Using Multiple Outcomes to Improve the Synthetic Control Method

When there are multiple outcome series of interest, Synthetic Control analyses typically proceed by estimating separate weights for each outcome. In this paper, we instead propose estimating a common set of weights across outcomes, by balancing either a vector of all outcomes or an index or average of them. Under a low-rank factor model, we show that these approaches lead to lower bias bounds than separate weights, and that averaging leads to further gains when the number of outcomes grows. We illustrate this via a re-analysis of the impact of the Flint water crisis on educational outcomes.

econ.EM↗

arXiv2024

Does AI help humans make better decisions? A statistical evaluation framework for experimental and observational studies

The use of Artificial Intelligence (AI), or more generally data-driven algorithms, has become ubiquitous in today's society. Yet, in many cases and especially when stakes are high, humans still make final decisions. The critical question, therefore, is whether AI helps humans make better decisions compared to a human-alone or AI-alone system. We introduce a new methodological framework to empirically answer this question with a minimal set of assumptions. We measure a decision maker's ability to make correct decisions using standard classification metrics based on the baseline potential outcome. We consider a single-blinded and unconfounded treatment assignment, where the provision of AI-generated recommendations is assumed to be randomized across cases with humans making final decisions. Under this study design, we show how to compare the performance of three alternative decision-making systems--human-alone, human-with-AI, and AI-alone. Importantly, the AI-alone system includes any individualized treatment assignment, including those that are not used in the original study. We also show when AI recommendations should be provided to a human-decision maker, and when one should follow such recommendations. We apply the proposed methodology to our own randomized controlled trial evaluating a pretrial risk assessment instrument. We find that the risk assessment recommendations do not improve the classification accuracy of a judge's decision to impose cash bail. Furthermore, we find that replacing a human judge with algorithms--the risk assessment score and a large language model in particular--leads to a worse classification performance.

cs.AI↗

arXiv2024

Safe Policy Learning under Regression Discontinuity Designs with Multiple Cutoffs

The regression discontinuity (RD) design is widely used for program evaluation with observational data. The primary focus of the existing literature has been the estimation of the local average treatment effect at the existing treatment cutoff. In contrast, we consider policy learning under the RD design. Because the treatment assignment mechanism is deterministic, learning better treatment cutoffs requires extrapolation. We develop a robust optimization approach to finding optimal treatment cutoffs that improve upon the existing ones. We first decompose the expected utility into point-identifiable and unidentifiable components. We then propose an efficient doubly-robust estimator for the identifiable parts. To account for the unidentifiable components, we leverage the existence of multiple cutoffs that are common under the RD design. Specifically, we assume that the heterogeneity in the conditional expectations of potential outcomes across different groups vary smoothly along the running variable. Under this assumption, we minimize the worst case utility loss relative to the status quo policy. The resulting new treatment cutoffs have a safety guarantee that they will not yield a worse overall outcome than the existing cutoffs. Finally, we establish the asymptotic regret bounds for the learned policy using semi-parametric efficiency theory. We apply the proposed methodology to empirical and simulated data sets.

stat.ME↗

arXiv2024

Intervention effects based on potential benefit

Optimal treatment rules are mappings from individual patient characteristics to tailored treatment assignments that maximize mean outcomes. In this work, we introduce a conditional potential benefit (CPB) metric that measures the expected improvement under an optimally chosen treatment compared to the status quo, within covariate strata. The potential benefit combines (i) the magnitude of the treatment effect, and (ii) the propensity for subjects to naturally select a suboptimal treatment. As a consequence, heterogeneity in the CPB can provide key insights into the mechanism by which a treatment acts and/or highlight potential barriers to treatment access or adverse effects. Moreover, we demonstrate that CPB is the natural prioritization score for individualized treatment policies when intervention capacity is constrained. That is, in the resource-limited setting where treatment options are freely accessible, but the ability to intervene on a portion of the target population is constrained (e.g., if the population is large, and follow-up and encouragement of treatment uptake is labor-intensive), targeting subjects with highest CPB maximizes the mean outcome. Focusing on this resource-limited setting, we derive formulas for optimal constrained treatment rules, and for any given budget, quantify the loss compared to the optimal unconstrained rule. We describe sufficient identification assumptions, and propose nonparametric, robust, and efficient estimators of the proposed quantities emerging from our framework. Finally, we illustrate our methodology using data from a prospective cohort study in which we assess the impact of intensive care unit transfer on mortality.

stat.ME↗

arXiv2024

Optimizing Language Models for Human Preferences is a Causal Inference Problem

As large language models (LLMs) see greater use in academic and commercial settings, there is increasing interest in methods that allow language models to generate texts aligned with human preferences. In this paper, we present an initial exploration of language model optimization for human preferences from direct outcome datasets, where each sample consists of a text and an associated numerical outcome measuring the reader's response. We first propose that language model optimization should be viewed as a causal problem to ensure that the model correctly learns the relationship between the text and the outcome. We formalize this causal language optimization problem, and we develop a method--causal preference optimization (CPO)--that solves an unbiased surrogate objective for the problem. We further extend CPO with doubly robust CPO (DR-CPO), which reduces the variance of the surrogate objective while retaining provably strong guarantees on bias. Finally, we empirically demonstrate the effectiveness of (DR-)CPO in optimizing state-of-the-art LLMs for human preferences on direct outcome data, and we validate the robustness of DR-CPO under difficult confounding conditions.

cs.LG↗