Simultaneous selection of optimal bandwidths for the sharp regression discontinuity estimator

Yoichi Arai, Hidehiko Ichimura

Research output: Contribution to journalArticle

6 Citations (Scopus)

Abstract

A new bandwidth selection method that uses different bandwidths for the local linear regression estimators on the left and the right of the cut-off point is proposed for the sharp regression discontinuity design estimator of the average treatment effect at the cut-off point. The asymptotic mean squared error of the estimator using the proposed bandwidth selection method is shown to be smaller than other bandwidth selection methods proposed in the literature. The approach that the bandwidth selection method is based on is also applied to an estimator that exploits the sharp regression kink design. Reliable confidence intervals compatible with both of the proposed bandwidth selection methods are also proposed as in the work of Calonico, Cattaneo, and Titiunik (2014a). An extensive simulation study shows that the proposed method's performances for the samples sizes 500 and 2000 closely match the theoretical predictions. Our simulation study also shows that the common practice of halving and doubling an optimal bandwidth for sensitivity check can be unreliable.

Original languageEnglish
Pages (from-to)441-482
Number of pages42
JournalQuantitative Economics
Volume9
Issue number1
DOIs
Publication statusPublished - 2018 Mar 1

Fingerprint

Estimator
Regression discontinuity
Bandwidth
Simulation study
Mean squared error
Regression discontinuity design
Average treatment effect
Confidence interval
Linear regression
Sample size
Prediction

Keywords

  • Bandwidth selection
  • confidence interval
  • local linear regression
  • regression discontinuity design
  • regression kink design

ASJC Scopus subject areas

  • Economics and Econometrics

Cite this

Simultaneous selection of optimal bandwidths for the sharp regression discontinuity estimator. / Arai, Yoichi; Ichimura, Hidehiko.

In: Quantitative Economics, Vol. 9, No. 1, 01.03.2018, p. 441-482.

Research output: Contribution to journalArticle

@article{6ed9e25c80374863898c362f87abf5cc,
title = "Simultaneous selection of optimal bandwidths for the sharp regression discontinuity estimator",
abstract = "A new bandwidth selection method that uses different bandwidths for the local linear regression estimators on the left and the right of the cut-off point is proposed for the sharp regression discontinuity design estimator of the average treatment effect at the cut-off point. The asymptotic mean squared error of the estimator using the proposed bandwidth selection method is shown to be smaller than other bandwidth selection methods proposed in the literature. The approach that the bandwidth selection method is based on is also applied to an estimator that exploits the sharp regression kink design. Reliable confidence intervals compatible with both of the proposed bandwidth selection methods are also proposed as in the work of Calonico, Cattaneo, and Titiunik (2014a). An extensive simulation study shows that the proposed method's performances for the samples sizes 500 and 2000 closely match the theoretical predictions. Our simulation study also shows that the common practice of halving and doubling an optimal bandwidth for sensitivity check can be unreliable.",
keywords = "Bandwidth selection, confidence interval, local linear regression, regression discontinuity design, regression kink design",
author = "Yoichi Arai and Hidehiko Ichimura",
year = "2018",
month = "3",
day = "1",
doi = "10.3982/QE590",
language = "English",
volume = "9",
pages = "441--482",
journal = "Quantitative Economics",
issn = "1759-7323",
publisher = "The Economic Society",
number = "1",

}

TY - JOUR

T1 - Simultaneous selection of optimal bandwidths for the sharp regression discontinuity estimator

AU - Arai, Yoichi

AU - Ichimura, Hidehiko

PY - 2018/3/1

Y1 - 2018/3/1

N2 - A new bandwidth selection method that uses different bandwidths for the local linear regression estimators on the left and the right of the cut-off point is proposed for the sharp regression discontinuity design estimator of the average treatment effect at the cut-off point. The asymptotic mean squared error of the estimator using the proposed bandwidth selection method is shown to be smaller than other bandwidth selection methods proposed in the literature. The approach that the bandwidth selection method is based on is also applied to an estimator that exploits the sharp regression kink design. Reliable confidence intervals compatible with both of the proposed bandwidth selection methods are also proposed as in the work of Calonico, Cattaneo, and Titiunik (2014a). An extensive simulation study shows that the proposed method's performances for the samples sizes 500 and 2000 closely match the theoretical predictions. Our simulation study also shows that the common practice of halving and doubling an optimal bandwidth for sensitivity check can be unreliable.

AB - A new bandwidth selection method that uses different bandwidths for the local linear regression estimators on the left and the right of the cut-off point is proposed for the sharp regression discontinuity design estimator of the average treatment effect at the cut-off point. The asymptotic mean squared error of the estimator using the proposed bandwidth selection method is shown to be smaller than other bandwidth selection methods proposed in the literature. The approach that the bandwidth selection method is based on is also applied to an estimator that exploits the sharp regression kink design. Reliable confidence intervals compatible with both of the proposed bandwidth selection methods are also proposed as in the work of Calonico, Cattaneo, and Titiunik (2014a). An extensive simulation study shows that the proposed method's performances for the samples sizes 500 and 2000 closely match the theoretical predictions. Our simulation study also shows that the common practice of halving and doubling an optimal bandwidth for sensitivity check can be unreliable.

KW - Bandwidth selection

KW - confidence interval

KW - local linear regression

KW - regression discontinuity design

KW - regression kink design

UR - http://www.scopus.com/inward/record.url?scp=85045465070&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85045465070&partnerID=8YFLogxK

U2 - 10.3982/QE590

DO - 10.3982/QE590

M3 - Article

VL - 9

SP - 441

EP - 482

JO - Quantitative Economics

JF - Quantitative Economics

SN - 1759-7323

IS - 1

ER -