Technology firms conduct randomized controlled experiments ("A/B tests") to learn which actions to take to improve business outcomes.In firms with mature experimentation platforms, experimentation programs can consist of many thousands of tests.To effectively scale experimentation, firms rely on decision rules: standard operating procedures for mapping the results of an experiment to a choice of treatment arm to launch to the general user population.Despite the critical role of decision rules in translating experimentation into business decisions, rigorous guidance on how to evaluate and choose decision rules is scarce.This paper proposes to evaluate decision rules based on their cumulative returns to business north star metrics.Although intuitive and easy to explain to decisionmakers, this quantity can be difficult to estimate, especially when experiments have weak signal-to-noise ratios.We develop a crossvalidation estimator that is much less biased than the naive plug-in estimator under conditions realistic to digital experimentation.We demonstrate the efficacy of our approach via a case study of 123 historical A/B tests at Netflix, where we used it to show that a new decision rule would have increased cumulative returns to the north star metric by an estimated 33%, directly leading to the adoption of the new rule.

Multi-arm bandit experimental designs are increasingly being adopted over standard randomized trials due to their potential to improve outcomes for study participants, enable faster identification of the best-performing options, and/or enhance the precision of estimating key parameters. Current approaches for inference after adaptive sampling either rely on asymptotic normality under restricted experiment designs or underpowered martingale concentration inequalities that lead to weak power in practice. To bypass these limitations, we propose a simulation-based approach for conducting hypothesis tests and constructing confidence intervals for arm specific means and their differences. Our simulation-based approach uses positively biased nuisances to generate additional trajectories of the experiment, which we call \textit{simulation with optimism}. Using these simulations, we characterize the distribution potentially non-normal sample mean test statistic to conduct inference. We provide guarantees for (i) asymptotic type I error control, (ii) convergence of our confidence intervals, and (iii) asymptotic strong consistency of our estimator over a wide variety of common bandit designs. Our empirical results show that our approach achieves the desired coverage while reducing confidence interval widths by up to 50\%, with drastic improvements for arms not targeted by the design.

We study inference on the long-term causal effect of a continual exposure to a novel intervention, which we term a long-term treatment, based on an experiment involving only short-term observations. Key examples include the long-term health effects of regularly-taken medicine or of environmental hazards and the long-term effects on users of changes to an online platform. This stands in contrast to short-term treatments or "shocks," whose long-term effect can reasonably be mediated by short-term observations, enabling the use of surrogate methods. Long-term treatments by definition have direct effects on long-term outcomes via continual exposure, so surrogacy conditions cannot reasonably hold. We connect the problem with offline reinforcement learning, leveraging doubly-robust estimators to estimate long-term causal effects for long-term treatments and construct confidence intervals.

When primary objectives are insensitive or delayed, experimenters may instead focus on proxy metrics derived from secondary outcomes. For example, technology companies often infer the long-term impacts of product interventions from their effects on short-term user engagement signals. We consider the meta-analysis of many historical experiments to learn the covariance of treatment effects on these outcomes, which can support the construction of such proxies. Even when experiments are plentiful, if treatment effects are weak, the covariance of estimated treatment effects across experiments can be highly biased. We overcome this with techniques inspired by weak instrumental variable analysis. We show that Limited Information Maximum Likelihood (LIML) learns a parameter equivalent to fitting total least squares to a transformation of the scatterplot of treatment effects, and that Jackknife Instrumental Variables Estimation (JIVE) learns another parameter computable from the average of Jackknifed covariance matrices across experiments. We also present a total covariance estimator for the latter estimand under homoskedasticity, which is equivalent to a k-class estimator. We show how these parameters can be used to construct unbiased proxy metrics under various structural models. Lastly, we discuss the real-world application of our methods at Netflix.

Contextual bandit algorithms are increasingly replacing non-adaptive A/B tests in e-commerce, healthcare, and policymaking because they can both improve outcomes for study participants and increase the chance of identifying good or even best policies. To support credible inference on novel interventions at the end of the study, nonetheless, we still want to construct valid confidence intervals on average treatment effects, subgroup effects, or value of new policies. The adaptive nature of the data collected by contextual bandit algorithms, however, makes this difficult: standard estimators are no longer asymptotically normally distributed and classic confidence intervals fail to provide correct coverage. While this has been addressed in non-contextual settings by using stabilized estimators, variance stabilized estimators in the contextual setting pose unique challenges that we tackle for the first time in this paper. We propose the Contextual Adaptive Doubly Robust (CADR) estimator, a novel estimator for policy value that is asymptotically normal under contextual adaptive data collection. The main technical challenge in constructing CADR is designing adaptive and consistent conditional standard deviation estimators for stabilization. Extensive numerical experiments using 57 OpenML datasets demonstrate that confidence intervals based on CADR uniquely provide correct coverage.

Empirical risk minimization (ERM) is the workhorse of machine learning, whether for classification and regression or for off-policy policy learning, but its model-agnostic guarantees can fail when we use adaptively collected data, such as the result of running a contextual bandit algorithm. We study a generic importance sampling weighted ERM algorithm for using adaptively collected data to minimize the average of a loss function over a hypothesis class and provide first-of-their-kind generalization guarantees and fast convergence rates. Our results are based on a new maximal inequality that carefully leverages the importance sampling structure to obtain rates with the good dependence on the exploration rate in the data. For regression, we provide fast rates that leverage the strong convexity of squared-error loss. For policy learning, we provide regret guarantees that close an open gap in the existing literature whenever exploration decays to zero, as is the case for bandit-collected data. An empirical investigation validates our theory.

We propose the Generalized Policy Elimination (GPE) algorithm, an oracle-efficient contextual bandit (CB) algorithm inspired by the Policy Elimination algorithm of Dudik et al. [2011]. We prove the first regret-optimality guarantee theorem for an oracle-efficient CB algorithm competing against a nonparametric class with infinite VC-dimension. Specifically, we show that GPE is regret-optimal (up to logarithmic factors) for policy classes with integrable entropy. For classes with larger entropy, we show that the core techniques used to analyze GPE can be used to design an $\varepsilon$-greedy algorithm with regret bound matching that of the best algorithms to date. We illustrate the applicability of our algorithms and theorems with examples of large nonparametric policy classes, for which the relevant optimization oracles can be efficiently implemented.

We study the problem of off-policy evaluation (OPE) in Reinforcement Learning (RL), where the aim is to estimate the performance of a new policy given historical data that may have been generated by a different policy, or policies. In particular, we introduce a novel doubly-robust estimator for the OPE problem in RL, based on the Targeted Maximum Likelihood Estimation principle from the statistical causal inference literature. We also introduce several variance reduction techniques that lead to impressive performance gains in off-policy evaluation. We show empirically that our estimator uniformly wins over existing off-policy evaluation methods across multiple RL environments and various levels of model misspecification. Finally, we further the existing theoretical analysis of estimators for the RL off-policy estimation problem by showing their $O_P(1/\sqrt{n})$ rate of convergence and characterizing their asymptotic distribution.

Off-policy evaluation is the problem of estimating the value of a target policy using data collected under a different policy. Given a base estimator for bandit off-policy evaluation and a parametrized class of control variates, we address the problem of computing a control variate in that class that reduces the risk of the base estimator. We derive the population risk as a function of the class parameters and we establish conditions that guarantee risk improvement. We present our main results in the context of multi-armed bandits, and we propose a simple design for contextual bandits that gives rise to an estimator that is shown to perform well in multi-class cost-sensitive classification datasets.