2022 Fiscal Year Research-status Report
New Developments in Regression Discontinuity Designs: Covariates Adjustment and Coverage Optimal Inference
| Project/Area Number |
21K01419
|
|---|
| Research Institution | University of Tsukuba |
|---|
Principal Investigator |
YU ZHENGFEI 筑波大学, 人文社会系, 准教授 (40774758)
|
|---|
| Project Period (FY) |
2021-04-01 – 2025-03-31
|
| Keywords | Regression discontinuity / Empirical likelihood / Coverage error / Empirical likelihood / Local misspecification |
|---|
| Outline of Annual Research Achievements |
This project proposes a novel approach to incorporate covariates in regression discontinuity (RD) designs. It represents the covariate balance condition as over-identifying moment restrictions. The empirical likelihood (EL) RD estimator efficiently incorporates the information from covariate balance and has an asymptotic variance no larger than that of the standard estimator without covariates. This project then proposes a robust corrected EL confidence set which achieves a faster coverage error decay rate. The coverage accuracy of the proposed confidence interval is automatically robust against slight perturbation to the covariate balance condition, which may happen in cases such as data contamination and misspecified “unaffected” outcomes used as covariates. This project also propose a way to conduct sensitivity analysis in the bandwidth choice when the researcher wants to use regression discontinuity with covariate adjustment. This project conducts Monte Carlo simulations to assess the finite-sample performance of the proposed inference method and then applies it to a real dataset.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
As planned, this research resolves the indeterminacy raised by Calonico, Cattaneo, Farrell, and Titiunik (2019, Page 448) regarding the asymptotic efficiency gain from incorporating covariates to RD estimator, by showing that RD estimator with covariate adjustment has an asymptotic variance no larger than that of the standard estimator without covariates. In addition, the proposed empirical likelihood inference method is shown to have theoretical advantages such as a faster coverage error decay rate and robustness to local specification. It also performs well in Monte Carlo simulations.
I presented the findings in conferences and received good feedbacks.
|
| Strategy for Future Research Activity |
The next steps for this project is to focus on the following aspects:
1. Give examples to illustrate that the moment restrictions framework for regression discontinuity applies to nonlinear models, such as categorical outcomes.
2. Connect the current work to the literature on balancing under unconfoundedness. The idea that balancing-type estimators can be an alternative to regression adjustments with many attractive properties has been explored in depth in the
literature on treatment effect estimation under confoundedness.
|
| Causes of Carryover |
I attended a hybrid conference using the online mode, so saved some travel expenses.
|
