We propose strategies to estimate and make inference on key features of heterogeneous effects in randomized experiments. These key features include best linear predictors of the effects using machine learning proxies, average effects sorted by impact groups , and average characteristics of most and least impacted units . The approach is valid in high‐dimensional settings, where the effects are proxied (but not necessarily consistently estimated) by predictive and causal machine learning methods. We post‐process these proxies into estimates of the key features. Our approach is generic; it can be used in conjunction with penalized methods, neural networks, random forests, boosted trees, and ensemble methods, both predictive and causal. Estimation and inference are based on repeated data splitting to avoid overfitting and achieve validity. We use quantile aggregation of the results across many potential splits, in particular taking medians of p ‐values and medians and other quantiles of confidence intervals. We show that quantile aggregation lowers estimation risks over a single split procedure, and establish its principal inferential properties. Finally, our analysis reveals ways to build provably better machine learning proxies through causal learning: we can use the objective functions that we develop to construct the best linear predictors of the effects, to obtain better machine learning proxies in the initial step. We illustrate the use of both inferential tools and causal learners with a randomized field experiment that evaluates a combination of nudges to stimulate demand for immunization in India

Many events and policies (treatments) occur at specific spatial locations, with researchers interested in their effects on nearby units. I approach the spatial treatment setting from an experimental perspective: What ideal experiment would we design to estimate the causal effects of spatial treatments? This perspective motivates a comparison between units near realized treatment locations and units near counterfactual (unrealized) candidate locations, which differs from current empirical practice. I derive design‐based standard errors that are straightforward to compute. For observational data, I propose machine learning methods to find counterfactual candidate locations when observable characteristics, rather than potential outcomes, determine treatment probabilities. To accommodate methods for high‐dimensional data in the theory, I extend a double machine learning result to the design‐based framework with spatial correlations. I apply the proposed methods to study the causal effects of grocery stores on foot traffic to nearby businesses during COVID‐19 shelter‐in‐place policies, finding a large positive effect at very short distances, with no effect at larger distances

Many economic and causal parameters depend on nonparametric or high dimensional first steps. We give a general construction of locally robust/orthogonal moment functions for GMM, where first steps have no effect, locally, on average moment functions. Using these orthogonal moments reduces model selection and regularization bias, as is important in many applications, especially for machine learning first steps. Also, associated standard errors are robust to misspecification when there is the same number of moment functions as parameters of interest. We use these orthogonal moments and cross‐fitting to construct debiased machine learning estimators of functions of high dimensional conditional quantiles and of dynamic discrete choice parameters with high dimensional state variables. We show that additional first steps needed for the orthogonal moment functions have no effect, globally, on average orthogonal moment functions. We give a general approach to estimating those additional first steps. We characterize double robustness and give a variety of new doubly robust moment functions. We give general and simple regularity conditions for asymptotic theory

Many causal and structural effects depend on regressions. Examples include policy effects, average derivatives, regression decompositions, average treatment effects, causal mediation, and parameters of economic structural models. The regressions may be high‐dimensional, making machine learning useful. Plugging machine learners into identifying equations can lead to poor inference due to bias from regularization and/or model selection. This paper gives automatic debiasing for linear and nonlinear functions of regressions. The debiasing is automatic in using Lasso and the function of interest without the full form of the bias correction. The debiasing can be applied to any regression learner, including neural nets, random forests, Lasso, boosting, and other high‐dimensional methods. In addition to providing the bias correction, we give standard errors that are robust to misspecification, convergence rates for the bias correction, and primitive conditions for asymptotic inference for estimators of a variety of estimators of structural and causal effects. The automatic debiased machine learning is used to estimate the average treatment effect on the treated for the NSW job training data and to estimate demand elasticities from Nielsen scanner data while allowing preferences to be correlated with prices and income

This paper proposes a valid bootstrap‐based distributional approximation for M ‐estimators exhibiting a Chernoff (1964)‐type limiting distribution. For estimators of this kind, the standard nonparametric bootstrap is inconsistent. The method proposed herein is based on the nonparametric bootstrap, but restores consistency by altering the shape of the criterion function defining the estimator whose distribution we seek to approximate. This modification leads to a generic and easy‐to‐implement resampling method for inference that is conceptually distinct from other available distributional approximations. We illustrate the applicability of our results with four examples in econometrics and machine learning

We examine the split‐sample robust inference (SSRI) methodology introduced by Chernozhukov, Demirer, Duflo, and Fernandez‐Val for quantifying uncertainty in heterogeneous treatment effect estimates produced by machine learning (ML) models. Although SSRI properly accounts for the additional variability due to sample splitting, its computational cost becomes prohibitive with complex ML models. We propose an alternative approach based on randomization inference (RI) that preserves the broad applicability of SSRI while eliminating the need for repeated sample splitting. Leveraging cross‐fitting and design‐based inference, the RI procedure yields valid confidence intervals with substantially reduced computational burden. Simulation studies demonstrate that the RI method preserves the statistical efficiency of SSRI while scaling to much larger applications and more complex settings

We study the problem of measuring group differences in choices when the dimensionality of the choice set is large. We show that standard approaches suffer from a severe finite‐sample bias, and we propose an estimator that applies recent advances in machine learning to address this bias. We apply this method to measure trends in the partisanship of congressional speech from 1873 to 2016, defining partisanship to be the ease with which an observer could infer a congressperson's party from a single utterance. Our estimates imply that partisanship is far greater in recent years than in the past, and that it increased sharply in the early 1990s after remaining low and relatively constant over the preceding century

This paper investigates the effect of terrorism financing and recruitment on attacks. I exploit a Sharia‐compliant institution in Pakistan, which induces unintended and quasi‐experimental variation in the funding of terrorist groups through their religious affiliation. The results indicate that higher terrorism financing, in a given location and period, generate more attacks in the same location and period. Financing exhibits a complementarity in producing attacks with terrorist recruitment, measured through data from Jihadist‐friendly online fora and machine learning. A higher supply of terror is responsible for the increase in attacks and is identified by studying groups with different affiliations operating in multiple cities. These findings are consistent with terrorist organizations facing financial frictions to their internal capital market

Nearest neighbor (NN) matching is widely used in observational studies for causal effects. Abadie and Imbens (2006) provided the first large‐sample analysis of NN matching. Their theory focuses on the case with the number of NNs, M fixed. We reveal something new out of their study and show that once allowing M to diverge with the sample size an intrinsic statistic in their analysis constitutes a consistent estimator of the density ratio with regard to covariates across the treated and control groups. Consequently, with a diverging M , the NN matching with Abadie and Imbens' (2011) bias correction yields a doubly robust estimator of the average treatment effect and is semiparametrically efficient if the density functions are sufficiently smooth and the outcome model is consistently estimated. It can thus be viewed as a precursor of the double machine learning estimators