| 研究課題/領域番号 |
21K01419
|
|---|
| 研究機関 | 筑波大学 |
|---|
研究代表者 |
YU ZHENGFEI 筑波大学, 人文社会系, 准教授 (40774758)
|
|---|
| 研究期間 (年度) |
2021-04-01 – 2025-03-31
|
| キーワード | Regression discontinuity / Empirical likelihood / Covariate adjustment / Moment restrictions / Efficiency gain / Coverage error / Uniform in bandwidth |
|---|
| 研究実績の概要 |
The regression discontinuity (RD) design has become one of the most popular methods for causal inference in social sciences. This project makes four contributions to the inference of regression discontinuity (RD). First, it resolves the indeterminacy in the literature regarding the asymptotic efficiency gain from incorporating covariates to the RD estimator. This project shows that covariates adjustment to the RD estimator achieves efficiency gain as long as the projection coefficients of some covariates are nonzero. Second, this project develops a new framework to incorporate covariates into RD by representing the covariate balance condition as over-identifying moment restrictions. This framework naturally calls for GMM or empirical likelihood (EL) estimation. Third, this project proposes a corrected EL confidence interval that achieves the parametric coverage error decay rate even though the point estimator converges at a nonparametric rate. Fourth, this project proves a uniform-in-bandwidth result for the EL ratio statistic, which is useful in sensitivity analysis with respect to the bandwidth choice.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
The main theoretical results have been obtained and proven as initially planned. The proposed empirical likelihood inference method is shown to have several theoretical advantages over the Wald-type counterpart.
In terms of implementation, an initial weakness has been overcome by introducing a correction term to the empirical likelihood ratio statistic.
|
| 今後の研究の推進方策 |
The research will proceed in the following directions: First, to combine the robust corrected empirical likelihood confidence interval with the uniform-in-bandwidth theory to obtain a confidence band for sensitivity analysis. Second, to study the power property of the proposed inference method. Third, to conduct Monte Carlo simulations in order to examine the finite sample performance of the proposed method. Fourth, to apply the proposed method to real datasets. Fifth, to present the results at conferences.
|
| 次年度使用額が生じた理由 |
Conferences and visits were cancelled due to the prolonged Covid pandemic. As a result, travel expenses and personnel expenditure were not used in the last year. This year I have several online or face-to-face conferences scheduled. There is a high probability that they will be conducted as planned.
|
