Research

We propose a welfare-based analog of p-values to capture the value of future data collection.

We provide an economic framework to choose which experiment to run and how to design it to estimate general equilibrium effects or externally valid counterfactuals.

We provide a framework and method to learn when and how effects generalize across different contexts, and when instead they do not.

We provide an econometric framework for causal inference when researchers do not have direct access to the outcome, but only observe remote sensed outcomes such as satellite images or digital traces.

We study the design of scientific communication in settings where researchers have private incentives.

We propose an economic framework to explain when, how, and why multiple-hypothesis-testing adjustments should be done differently.

(See also related comment on FDA advisory for handling multiple endpoints in clinical trials.)

Can satellite-based predictions substitute for traditional outcome measurements in program evaluation? We study this question through calibrated semi-synthetic simulations studying environmental programs in developing countries.

We propose and study treatment allocation rules that are fair and Pareto optimal for applications in welfare programs.

We study when and how researchers should design clusters in experiments with network spillovers (with an application to the universe of Facebook users).

We provide a framework to estimate complementarities and substitution effects between social norms in the presence of spillovers.

We propose an experimental design for inference on and estimation of policies under network interference.

(with follow-up data collection in an implementation reaching over 400 000 participants.)

I propose a statistical framework for two-stage experimental designs under interference, with a focus on precise inference on treatment and spillover effects.

I design and study how to choose whom to treat in the presence of spillovers using data from existing experiments or observational studies.

We formalize the properties of Synthetic Control estimators with unobserved confounders.

We study inference on time-varying treatments with panel data in the presence of high-dimensional covariates and (unknown) dynamic selection into treatment.

We introduce a panel data estimator with strong robustness properties. We illustrate its properties across several applications.

We study estimation and inference with machine-learning estimators in the context of synthetic controls.