My research focuses on solving failure modes of causal inference. By immersing myself in the actual work of policy evaluation in a variety of contexts, I encounter issues that statisticians tend to abstract away from. My job market paper offers a concrete example of this approach. Donuts in regression discontinuity designs represent a collision between the manipulation-free world of statistics and the reality that many empirical designs encounter limited degrees of manipulation. My work in that setting is to identify issues that can arise, and work to solve them, with an eye towards utility for current practice.
- JMP: Partial Identification for Regression Discontinuity Donuts
- Inference in Sythetic Controls with Spillovers – with Jianfei Cao
- Synthetic Control designs rely on SUTVA for inference. In settings (like spread of covid), where one state’s actions affect outcomes in neighboring states, this may not be reasonable.
- Our paper allows valid inference under weaker assumptions that the spillovers are contained to a known group of states.
- Slides, Paper (and on arXiv)
- Code: R, Matlab
- A new ECDF Two-Sample Test Statistic
Work in Progress
- High Dimensional F-tests – with Panos Toulis and Wenxuan Guo
- F-tests, based on the R-squared, lose validity and power in high-dimensional settings where R-squared = 1. We propose an alternative which is valid under standard conditions, is easily calculated, and has good power properties.
- Waiting for the Hot Hand – with Sam Hirshman and Prof. Nick Polson
- Using sequence wait-time metrics to find the anti-hot-hand in NBA shooting data.
- Credit Card Repayment Strategies
- Looking at strategies that minimize downside risk in the Covid era using transunion data.
- Second Earners and Kindergarten entry – with Tim Dowd
- Looking at household formation consequences of kindergarten entry using RD Donuts and IRS data.