Working Papers
Robust Inference for Weighted Estimands. [arXiv] [Current Draft]
Researchers often conduct inference on weighted estimands, defined as weighted averages of group-level effects. Example settings include event studies with cohort-level effects and experiments with site-level effects. Under heterogeneous effects, different weighting schemes yield estimands with distinct empirical and policy interpretations, leading to ambiguity and disagreement over the choice of weights. I establish bounds on differences between weighted estimands and confidence bounds on effect heterogeneity, which I use to construct estimators that minimize worst-case bias and confidence intervals that are uniformly valid over classes of weighted estimands. I apply these methods to an event study in Lakdawala, Nakasone, and Kho (2023), which studies the effects of school-based internet access on test scores. I find that results are robust to broad classes of weights. I then apply the methods to Tennessee’s Project STAR experiment and find that results are sensitive to small departures from baseline weights.
Integrating Diagnostic Checks into Estimation (with Reca Sarfati). [arXiv] [Current Draft]
Empirical researchers often use diagnostic checks to assess the plausibility of their modeling assumptions, such as testing for covariate balance in RCTs, pre-trends in event studies, or instrument validity in IV designs. While these checks are traditionally treated as external hurdles to estimation, we argue they should be integrated into the estimation process itself. In particular, we propose residualizing one’s baseline estimator against the vector of diagnostic check statistics to remove the component of baseline sampling variation explained by the diagnostic checks. This residualized estimator offers researchers a “free lunch,” delivering three properties simultaneously: (i) eliminating inference distortions from check-based selective reporting; (ii) reducing variance without changing the estimand when the baseline model is correctly specified; and (iii) minimizing worst-case bias under bounded local misspecification within the class of linear adjustments. We apply our method to the RCT in Kaur et al. (2024) and find that, even in a setting where all balance checks pass comfortably, residualization increases the magnitude of the baseline point estimate and reduces its standard error, equivalent to approximately a 10% increase in sample size.
"Post" Pre-Analysis Plans: Valid Inference for Non-Preregistered Specifications (with Reca Sarfati). [arXiv] [Current Draft]
Pre-analysis plans are now standard in experimental economics, but researchers often supplement preregistered analyses with non-prespecified results after observing the data. We study inference in this setting when the decision to report additional results depends on realized findings. We show that conventional confidence intervals and point estimates can be invalid even absent manipulation, because non-preregistered results are observed only in selected data realizations. We derive conditional inference procedures that account for selective reporting, characterize when the implied corrections are large or negligible, and apply the approach to Bessone et al. (2021). We conclude with guidance for empirical researchers and journals.
Publications
Interpreting TSLS Estimators in Information Provision Experiments (with Whitney Zhang). 2025. American Economic Review: Insights. [Journal] [Final Draft]
In information provision experiments, researchers often estimate the causal effects of beliefs on actions using two-stage least squares (TSLS). This paper formalizes exclusion and monotonicity conditions that ensure that TSLS recovers a positive-weighted average of causal effects. We assess common TSLS estimators for both passive and active control designs from the literature; we find that two commonly used passive control estimators generally allow for negative weights. The choice of passive control estimator affects the magnitude and significance of estimates in simulations and in an empirical application. We give practical recommendations for addressing these issues.
Work in Progress
Headline Estimation with Multiple Research Designs.
Federated Learning and Privacy Protection (with Dirk Bergemann, Alessandro Bonatti, and Mert Demirer).