Does AI help humans make better decisions? A methodological framework for experimental evaluation

The use of Artificial Intelligence (AI) based on 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 as compared to a human alone or AI an alone. We introduce a new methodological framework that can be used to answer experimentally this question with no additional 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 experimental design, in which the provision of AI-generated recommendations is randomized across cases with a human making final decisions. Under this experimental design, we show how to compare the performance of three alternative decision-making systems--human-alone, human-with-AI, and AI-alone. We apply the proposed methodology to the data from our own randomized controlled trial of a pretrial risk assessment instrument. We find that AI recommendations do not improve the classification accuracy of a judge's decision to impose cash bail. Our analysis also shows that AI-alone decisions generally perform worse than human decisions with or without AI assistance. Finally, AI recommendations tend to impose cash bail on non-white arrestees more often than necessary when compared to white arrestees.

The Cram Method for Efficient Simultaneous Learning and Evaluation

We introduce the "cram" method, a general and efficient approach to simultaneous learning and evaluation using a generic machine learning (ML) algorithm. In a single pass of batched data, the proposed method repeatedly trains an ML algorithm and tests its empirical performance. Because it utilizes the entire sample for both learning and evaluation, cramming is significantly more data-efficient than sample-splitting. The cram method also naturally accommodates online learning algorithms, making its implementation computationally efficient. To demonstrate the power of the cram method, we consider the standard policy learning setting where cramming is applied to the same data to both develop an individualized treatment rule (ITR) and estimate the average outcome that would result if the learned ITR were to be deployed. We show that under a minimal set of assumptions, the resulting crammed evaluation estimator is consistent and asymptotically normal. While our asymptotic results require a relatively weak stabilization condition of ML algorithm, we develop a simple, generic method that can be used with any policy learning algorithm to satisfy this condition. Our extensive simulation studies show that, when compared to sample-splitting, cramming reduces the evaluation standard error by more than 40% while improving the performance of learned policy. We also apply the cram method to a randomized clinical trial to demonstrate its applicability to real-world problems. Finally, we briefly discuss future extensions of the cram method to other learning and evaluation settings.

Individualized Policy Evaluation and Learning under Clustered Network Interference

While there now exists a large literature on policy evaluation and learning, much of prior work assumes that the treatment assignment of one unit does not affect the outcome of another unit. Unfortunately, ignoring interference may lead to biased policy evaluation and yield ineffective learned policies. For example, treating influential individuals who have many friends can generate positive spillover effects, thereby improving the overall performance of an individualized treatment rule (ITR). We consider the problem of evaluating and learning an optimal ITR under clustered network (or partial) interference where clusters of units are sampled from a population and units may influence one another within each cluster. Under this model, we propose an estimator that can be used to evaluate the empirical performance of an ITR. We show that this estimator is substantially more efficient than the standard inverse probability weighting estimator, which does not impose any assumption about spillover effects. We derive the finite-sample regret bound for a learned ITR, showing that the use of our efficient evaluation estimator leads to the improved performance of learned policies. Finally, we conduct simulation and empirical studies to illustrate the advantages of the proposed methodology.

Statistical Performance Guarantee for Selecting Those Predicted to Benefit Most from Treatment

Across a wide array of disciplines, many researchers use machine learning (ML) algorithms to identify a subgroup of individuals, called exceptional responders, who are likely to be helped by a treatment the most. A common approach consists of two steps. One first estimates the conditional average treatment effect or its proxy using an ML algorithm. They then determine the cutoff of the resulting treatment prioritization score to select those predicted to benefit most from the treatment. Unfortunately, these estimated treatment prioritization scores are often biased and noisy. Furthermore, utilizing the same data to both choose a cutoff value and estimate the average treatment effect among the selected individuals suffer from a multiple testing problem. To address these challenges, we develop a uniform confidence band for experimentally evaluating the sorted average treatment effect (GATES) among the individuals whose treatment prioritization score is at least as high as any given quantile value, regardless of how the quantile is chosen. This provides a statistical guarantee that the GATES for the selected subgroup exceeds a certain threshold. The validity of the proposed methodology depends solely on randomization of treatment and random sampling of units without requiring modeling assumptions or resampling methods. This widens its applicability including a wide range of other causal quantities. A simulation study shows that the empirical coverage of the proposed uniform confidence bands is close to the nominal coverage when the sample is as small as 100. We analyze a clinical trial of late-stage prostate cancer and find a relatively large proportion of exceptional responders with a statistical performance guarantee.

Bayesian Safe Policy Learning with Chance Constrained Optimization: Application to Military Security Assessment during the Vietnam War

Algorithmic and data-driven decisions and recommendations are commonly used in high-stakes decision-making settings such as criminal justice, medicine, and public policy. We investigate whether it would have been possible to improve a security assessment algorithm employed during the Vietnam War, using outcomes measured immediately after its introduction in late 1969. This empirical application raises several methodological challenges that frequently arise in high-stakes algorithmic decision-making. First, before implementing a new algorithm, it is essential to characterize and control the risk of yielding worse outcomes than the existing algorithm. Second, the existing algorithm is deterministic, and learning a new algorithm requires transparent extrapolation. Third, the existing algorithm involves discrete decision tables that are common but difficult to optimize over. To address these challenges, we introduce the Average Conditional Risk (ACRisk), which first quantifies the risk that a new algorithmic policy leads to worse outcomes for subgroups of individual units and then averages this over the distribution of subgroups. We also propose a Bayesian policy learning framework that maximizes the posterior expected value while controlling the posterior expected ACRisk. This framework separates the estimation of heterogeneous treatment effects from policy optimization, enabling flexible estimation of effects and optimization over complex policy classes. We characterize the resulting chance-constrained optimization problem as a constrained linear programming problem. Our analysis shows that compared to the actual algorithm used during the Vietnam War, the learned algorithm assesses most regions as more secure and emphasizes economic and political factors over military factors.

Distributionally Robust Causal Inference with Observational Data

We consider the estimation of average treatment effects in observational studies without the standard assumption of unconfoundedness. We propose a new framework of robust causal inference under the general observational study setting with the possible existence of unobserved confounders. Our approach is based on the method of distributionally robust optimization and proceeds in two steps. We first specify the maximal degree to which the distribution of unobserved potential outcomes may deviate from that of obsered outcomes. We then derive sharp bounds on the average treatment effects under this assumption. Our framework encompasses the popular marginal sensitivity model as a special case and can be extended to the difference-in-difference and regression discontinuity designs as well as instrumental variables. Through simulation and empirical studies, we demonstrate the applicability of the proposed methodology to real-world settings.

Race and ethnicity data for first, middle, and last names

We provide the largest compiled publicly available dictionaries of first, middle, and last names for the purpose of imputing race and ethnicity using, for example, Bayesian Improved Surname Geocoding (BISG). The dictionaries are based on the voter files of six Southern states that collect self-reported racial data upon voter registration. Our data cover a much larger scope of names than any comparable dataset, containing roughly one million first names, 1.1 million middle names, and 1.4 million surnames. Individuals are categorized into five mutually exclusive racial and ethnic groups -- White, Black, Hispanic, Asian, and Other -- and racial/ethnic counts by name are provided for every name in each dictionary. Counts can then be normalized row-wise or column-wise to obtain conditional probabilities of race given name or name given race. These conditional probabilities can then be deployed for imputation in a data analytic task for which ground truth racial and ethnic data is not available.

Policy learning with asymmetric utilities

Data-driven decision making plays an important role even in high stakes settings like medicine and public policy. Learning optimal policies from observed data requires a careful formulation of the utility function whose expected value is maximized across a population. Although researchers typically use utilities that depend on observed outcomes alone, in many settings the decision maker's utility function is more properly characterized by the joint set of potential outcomes under all actions. For example, the Hippocratic principle to ``do no harm'' implies that the cost of causing death to a patient who would otherwise survive without treatment is greater than the cost of forgoing life-saving treatment. We consider optimal policy learning with asymmetric utility functions of this form. We show that asymmetric utilities lead to an unidentifiable social welfare function, and so we first partially identify it. Drawing on statistical decision theory, we then derive minimax decision rules by minimizing the maximum regret relative to alternative policies. We show that one can learn minimax decision rules from observed data by solving intermediate classification problems. We also establish that the finite sample regret of this procedure is bounded by the mis-classification rate of these intermediate classifiers. We apply this conceptual framework and methodology to the decision about whether or not to use right heart catheterization for patients with possible pulmonary hypertension.

Addressing Census data problems in race imputation via fully Bayesian Improved Surname Geocoding and name supplements

Prediction of an individual's race and ethnicity plays an important role in social science and public health research. Examples include studies of racial disparity in health and voting. Recently, Bayesian Improved Surname Geocoding (BISG), which uses Bayes' rule to combine information from Census surname files with the geocoding of an individual's residence, has emerged as a leading methodology for this prediction task. Unfortunately, BISG suffers from two Census data problems that contribute to unsatisfactory predictive performance for minorities. First, the decennial Census often contains zero counts for minority racial groups in the Census blocks where some members of those groups reside. Second, because the Census surname files only include frequent names, many surnames -- especially those of minorities -- are missing from the list. To address the zero counts problem, we introduce a fully Bayesian Improved Surname Geocoding (fBISG) methodology that accounts for potential measurement error in Census counts by extending the na\"ive Bayesian inference of the BISG methodology to full posterior inference. To address the missing surname problem, we supplement the Census surname data with additional data on last, first, and middle names taken from the voter files of six Southern states where self-reported race is available. Our empirical validation shows that the fBISG methodology and name supplements significantly improve the accuracy of race imputation across all racial groups, and especially for Asians. The proposed methodology, together with additional name data, is available via the open-source software package wru.

Statistical Inference for Heterogeneous Treatment Effects Discovered by Generic Machine Learning in Randomized Experiments

Researchers are increasingly turning to machine learning (ML) algorithms to investigate causal heterogeneity in randomized experiments. Despite their promise, ML algorithms may fail to accurately ascertain heterogeneous treatment effects under practical settings with many covariates and small sample size. In addition, the quantification of estimation uncertainty remains a challenge. We develop a general approach to statistical inference for heterogeneous treatment effects discovered by a generic ML algorithm. We apply the Neyman's repeated sampling framework to a common setting, in which researchers use an ML algorithm to estimate the conditional average treatment effect and then divide the sample into several groups based on the magnitude of the estimated effects. We show how to estimate the average treatment effect within each of these groups, and construct a valid confidence interval. In addition, we develop nonparametric tests of treatment effect homogeneity across groups, and rank-consistency of within-group average treatment effects. The validity of our methodology does not rely on the properties of ML algorithms because it is solely based on the randomization of treatment assignment and random sampling of units. Finally, we generalize our methodology to the cross-fitting procedure by accounting for the additional uncertainty induced by the random splitting of data.