Contextual bandit algorithms have transformed modern experimentation by enabling real-time adaptation for personalized treatment. Yet these advantages create challenges for statistical inference due to adaptivity. We study inference with contextual-bandit data without assuming a well-specified outcome model. In this setting, we show a previously overlooked issue: standard algorithms such as LinUCB may fail to stabilize under misspecified working models, leading to non-Gaussian estimator behavior and invalid inference. This issue is practically important, as misspecified working models -- such as approximations of complex dynamical systems -- are often employed by online agents in real-world adaptive experiments to balance reward, computational tractability, and robustness. We develop an inverse-probability-weighted Z-estimation framework for a broad class of marginal moment targets, including projection parameters, structural parameters with noisy contexts, and off-policy values. We identify a stability condition tailored to this framework, scaled inverse-propensity convergence, under which the IPW-Z estimator is consistent and asymptotically normal with a consistent sandwich variance estimator. We further establish sufficient conditions for scaled inverse-propensity convergence for several policy classes, including multi-armed bandit algorithms and smooth contextual allocation policies. Simulations and a HeartSteps V1 real-data-calibrated application show reliable coverage and competitive performance across multiple targets. Overall, our results highlight the importance of stability-aware adaptive design for valid post-experiment inference.

Yet, modern machine learning pipelines offer a promising solution--provided their predictions yield correct conclusions. We focus on Prediction-Powered Causal Inferences (PPCI), i.e., estimating the treatment effect in an unlabeled target experiment, relying on training data with the same outcome annotated but potentially different treatment or effect modifiers.

Single-arm trials are an important study design for evaluating drug efficacy and safety without enrolling patients into a control arm. Although they do not provide the gold-standard evidence of randomized controlled trials, they are increasingly used in clinical development as they offer an efficient, ethical, and practical alternative. A wide variety of approaches can be used to construct control comparators and estimate treatment effects, from fixed comparators informed by clinical knowledge to data-based and model-based patient-level comparators, also known as synthetic controls. Powerful and flexible machine learning models can allow outcome-model-based synthetic controls to overcome key limitations of direct data-based approaches, yield more robust estimates of treatment effects, and provide a principled way to incorporate corrections or encode additional assumptions when external data are not directly comparable. In this work, we argue that outcome-model-based synthetic control arms are an important tool for single-arm trials. We focus on digital twins, personalized predictions of disease progression generated from machine learning models trained on historical datasets, which naturally leverage these flexible approaches. We review doubly robust estimators, present power and sample size formulas, and discuss trade-offs in selecting historical data for training and analysis. We also outline practical considerations for deploying digital twins within the framework of recent FDA draft guidance on the use of artificial intelligence in drug development. Finally, we reanalyze data from trials in amyotrophic lateral sclerosis and Huntington's disease to demonstrate the proposed methods.

To design an efficient intervention, decision-makers need not only to estimate the total effect of the intervention, but also understand the underlying causal mechanisms driving this effect.

Applied work with interference typically models outcomes as functions of own treatment and a low-dimensional exposure mapping of others' treatments, even when that mapping may be misspecified. This raises a basic question: what policy object are exposure-based estimands implicitly targeting, and how should we interpret their direct and spillover components relative to the underlying policy question? We take as primitive the marginal policy effect, defined as the effect of a small change in the treatment probability under the actual experimental design, and show that any researcher-chosen exposure mapping induces a unique pseudo-true outcome model. This model is the best approximation to the underlying potential outcomes that depends only on the user-chosen exposure. Utilizing that representation, the marginal policy effect admits a canonical decomposition into exposure-based direct and spillover effects, and each component provides its optimal approximation to the corresponding oracle objects that would be available if interference were fully known. We then focus on a setting that nests important empirical and theoretical applications in which both local network spillovers and global spillovers, such as market equilibrium, operate. There, the marginal policy effect further decomposes asymptotically into direct, local, and global channels. An important implication is that many existing methods are more robust than previously understood once we reinterpret their targets as channel-specific components of this pseudo-true policy estimand. Simulations and a semi-synthetic experiment calibrated to a large cash-transfer experiment show that these components can be recovered in realistic experimental designs.

Evaluating novel contextual bandit policies using logged data is crucial in applications where exploration is costly, such as medicine. But it usually relies on the assumption of no unobserved confounders, which is bound to fail in practice. We study the question of policy evaluation when we instead have proxies for the latent confounders and develop an importance weighting method that avoids fitting a latent outcome regression model. Surprisingly, we show that there exist no single set of weights that give unbiased evaluation regardless of outcome model, unlike the case with no unobserved confounders where density ratios are sufficient. Instead, we propose an adversarial objective and weights that minimize it, ensuring sufficient balance in the latent confounders regardless of outcome model. We develop theory characterizing the consistency of our method and tractable algorithms for it. Empirical results validate the power of our method when confounders are latent.