Demand-side interventions, like subsidies or information treatments, are often considered to improve welfare in centralized markets. For these interventions, estimating global treatment effects is challenging because of spillover effects that occur through the market mechanism. Global treatment effects are non-parametrically identified when the mechanism is strategy-proof and data from an RCT in a single market is observed. For a class of cutoff mechanisms, we propose an estimator that runs a weighted and perturbed version of the mechanism. Under a continuum market approximation, the estimator is asymptotically normal and semi-parametrically efficient. We extend this approach to learn spillover-aware targeting rules with vanishing asymptotic regret. Empirically, adjusting for spillover effects notably diminishes the estimated effect of information on inequality in the Chilean school system.

We consider the problem of learning how to optimally allocate treatments whose cost is uncertain and can vary with pre-treatment covariates. This setting may arise in medicine if we need to prioritize access to a scarce resource that different patients would use for different amounts of time, or in marketing if we want to target discounts whose cost to the company depends on how much the discounts are used. Here, we show that the optimal treatment allocation rule under budget constraints is a thresholding rule based on priority scores, and we propose a number of practical methods for learning these priority scores using data from a randomized trial. Our formal results leverage a statistical connection between our problem and that of learning heterogeneous treatment effects under endogeneity using an instrumental variable. We find our method to perform well in a number of empirical evaluations.

Policy-relevant treatment effect estimation in a marketplace setting requires assessing both the direct treatment benefit and spillovers induced by changes to the market equilibrium. We show how to identify and estimate policy-relevant treatment effects using a unit-randomized trial run within a single large market. A Bernoulli-randomized trial allows consistent estimation of direct effects and of treatment-heterogeneity measures that enable welfare-improving targeting. Estimating spillovers—and providing confidence intervals for the direct effect—requires estimates of price elasticities, which we provide using an augmented experimental design. We illustrate our results using a simulation calibrated to a conditional cash-transfer experiment in the Philippines.

There is increasing interest in allocating treatments based on observed individual characteristics: examples include targeted marketing, individualized credit offers, and heterogenous pricing. Treatment personalization introduces incentives for individuals to modify their behavior to obtain a better treatment. This shifts the distribution of covariates, which means the Conditional Average Treatment Effect (CATE) now depends on how treatments are targeted. The optimal rule without strategic behavior allocates treatments only to those with a positive CATE. With strategic behavior, we show that the optimal rule can involve randomization, allocating treatments with less than 100% probability even to those with a positive CATE induced by that rule. We propose a sequential experiment based on Bayesian Optimization that converges to the optimal treatment rule without parametric assumptions on individual strategic behavior.

We discuss using Wasserstein Generative Adversarial Networks (WGANs) as a method for systematically generating artificial data that mimic closely any given real data set without the researcher having many degrees of freedom. We apply the methods to compare in three different settings twelve different estimators for average treatment effects under unconfoundedness.

We propose using a Bayesian hierarchical latent class model to summarize and interpret observed heterogeneity in categorical expectations data. We show that the statistical model corresponds to an economic structural model of information acquisition, which guides interpretation and estimation of the model parameters.

In online platforms, the impact of a treatment on an observed outcome may change over time as 1) users learn about the intervention, and 2) the system personalization, such as individualized recommendations, change over time. We introduce a non-parametric causal model of user actions in a personalized system. We show that the Cookie-Cookie-Day (CCD) experiment, designed for the measurement of the user learning effect, is biased when there is personalization. We derive new experimental designs that intervene in the personalization system to generate the variation necessary to separately identify the causal effect mediated through user learning and personalization. Making parametric assumptions allows for the estimation of long-term causal effects based on medium-term experiments. In simulations, we show that our new designs successfully recover the dynamic causal effects of interest.

We study the problem of a planner who wants to reduce inequality by awarding prizes to the worst contestants in a tournament without incentivizing shirking. We design an approximately optimal, incentive-compatible mechanism that targets low-ranked contestants based on the tournament's history up to an endogenous stopping time. We describe applications to eligibility for remedial education, retraining benefits for the unemployed, and draft lotteries in sports.