README.md

In kota7/rddsigma: Sharp Regression Discontinuity Design with Running Variable Measured with Errors

rddsigma

Sharp Regression Discontinuity Design with Running Variable Measured with Error

This package provides estimators for the standard deviation of measurement errors in running variables for sharp regression discontinuity designs. The estimation requires the data for observed (mismeasured) running variables and assigment variables.

Installation

Install from GitHub using devtools::install_github function.

devtools::install_github("kota7/rddsigma")

Quick Start

Use estimate_sigma function.

library(rddsigma)
set.seed(875)

# generate simulation data
dat <- gen_data(500, 0.3, 0)

estimate_sigma(dat$d, dat$w, cutoff=0)
#> * RDD sigma Estimate *
#> 
#>        Estimate Std. Error z value Pr(>|t|)    
#> sigma  0.286483   0.025592  11.194   <2e-16 ***
#> mu_x  -0.010614   0.046339  -0.229   0.4094    
#> sd_x   0.995779   0.034677  28.716   <2e-16 ***
#> sd_w   1.036170   0.032767  31.623   <2e-16 ***
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#> 
#>  n obs  :  500 
#>  Method :  Two Step Gaussian 
#>  x dist :  Gaussian 
#>  u dist :  Gaussian 
#>  value  :  -0.195047 
#>  convergence: yes

Estimation Methods

The package support two estimators for estimating sigma. The first estimator assumes that both the true runnning variable and the measurement error follow the Gaussian distribution. This is the default estimator of estimate_sigma function, or specified explicitly by setting method="tsgauss"

estimate_sigma(dat$d, dat$w, cutoff=0, method="tsgauss")
#> * RDD sigma Estimate *
#> 
#>        Estimate Std. Error z value Pr(>|t|)    
#> sigma  0.286483   0.025592  11.194   <2e-16 ***
#> mu_x  -0.010614   0.046339  -0.229   0.4094    
#> sd_x   0.995779   0.034677  28.716   <2e-16 ***
#> sd_w   1.036170   0.032767  31.623   <2e-16 ***
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#> 
#>  n obs  :  500 
#>  Method :  Two Step Gaussian 
#>  x dist :  Gaussian 
#>  u dist :  Gaussian 
#>  value  :  -0.195047 
#>  convergence: yes

The second estimator relaxes the Gaussian assumption and allows the true runnning variable and the measurement error to follow some parametric distribution. Currently, the package supports the Gaussian distribution for the running variable and the Gaussian and Laplace distributions for the measurement error. This estimator is called with method="emparam"

estimate_sigma(dat$d, dat$w, cutoff=0, method="emparam", x_dist="gauss", u_dist="gauss")
#> * RDD sigma Estimate *
#> 
#>        Estimate Std. Error z value Pr(>|t|)    
#> sigma  0.281109   0.024359  11.540   <2e-16 ***
#> mu_x  -0.010614   0.045546  -0.233   0.4079    
#> sd_x   0.998358   0.031221  31.978   <2e-16 ***
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#> 
#>  n obs  :  500 
#>  Method :  EM Parametric 
#>  x dist :  Gaussian 
#>  u dist :  Gaussian 
#>  value  :  -1.650564 
#>  convergence: yes
estimate_sigma(dat$d, dat$w, cutoff=0, method="emparam", x_dist="gauss", u_dist="lap")
#> * RDD sigma Estimate *
#> 
#>        Estimate Std. Error z value  Pr(>|t|)    
#> sigma  0.299957   0.038387  7.8141 2.769e-15 ***
#> mu_x  -0.010614   0.045546 -0.2330    0.4079    
#> sd_x   0.991398   0.034457 28.7724 < 2.2e-16 ***
#> ---
#> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#> 
#>  n obs  :  500 
#>  Method :  EM Parametric 
#>  x dist :  Gaussian 
#>  u dist :  Laplace 
#>  value  :  -1.653077 
#>  convergence: yes

kota7/rddsigma documentation built on May 20, 2019, 1:11 p.m.