The rdd package provides tools for regression discontinuity design estimation and analysis. It allows for both sharp and fuzzy designs using local linear regression. Key functions include IKbandwidth to calculate optimal bandwidth, RDestimate for estimation, and DCdensity for McCrary sorting tests. The package is designed to undertake the main steps of RDD analysis within R.
The rdd package provides tools for regression discontinuity design estimation and analysis. It allows for both sharp and fuzzy designs using local linear regression. Key functions include IKbandwidth to calculate optimal bandwidth, RDestimate for estimation, and DCdensity for McCrary sorting tests. The package is designed to undertake the main steps of RDD analysis within R.
The rdd package provides tools for regression discontinuity design estimation and analysis. It allows for both sharp and fuzzy designs using local linear regression. Key functions include IKbandwidth to calculate optimal bandwidth, RDestimate for estimation, and DCdensity for McCrary sorting tests. The package is designed to undertake the main steps of RDD analysis within R.
The rdd package provides tools for regression discontinuity design estimation and analysis. It allows for both sharp and fuzzy designs using local linear regression. Key functions include IKbandwidth to calculate optimal bandwidth, RDestimate for estimation, and DCdensity for McCrary sorting tests. The package is designed to undertake the main steps of RDD analysis within R.
Maintainer Drew Dimmery <[email protected]> Author Drew Dimmery Version 0.57 License Apache License (== 2.0) Title Regression Discontinuity Estimation Description Provides the tools to undertake estimation in Regression Discontinuity Designs. Both sharp and fuzzy designs are supported. Estimation is accomplished using local linear regression. A provided function will utilize Imbens-Kalyanaraman optimal bandwidth calculation. A function is also included to test the assumption of no-sorting effects. Type Package Date 2016-03-14 Depends R (>= 2.15.0), sandwich, lmtest, AER, Formula Collate 'kernelwts.R' 'DCdensity.R' 'IKbandwidth.R' 'RDestimate.R' 'plot.RD.R' 'print.RD.R' 'rdd-package.R' 'summary.RD.R' RoxygenNote 5.0.1 NeedsCompilation no Repository CRAN Date/Publication 2016-03-14 23:46:03
Details rdd supports both sharp and fuzzy RDD utilizing the AER package for 2SLS regression under the fuzzy design. Local linear regressions are performed to either side of the cutpoint using the Imbens-Kalyanamaran optimal bandwidth calculation, IKbandwidth.
Arguments runvar numerical vector of the running variable cutpoint the cutpoint (defaults to 0) bin the binwidth (defaults to 2*sd(runvar)*length(runvar)^(-.5)) bw the bandwidth to use (by default uses bandwidth selection calculation from Mc- Crary (2008)) verbose logical flag specifying whether to print diagnostic information to the terminal. (defaults to FALSE) plot logical flag indicating whether to plot the histogram and density estimations (de- faults to TRUE). The user may wrap this function in additional graphical options to modify the plot. DCdensity 3
ext.out logical flag indicating whether to return extended output. When FALSE (the de- fault) DCdensity will return only the p-value of the test. When TRUE, DCdensity will return the additional information documented below. htest logical flag indicating whether to return an "htest" object compatible with base R’s hypothesis test output.
Value
If ext.out is FALSE, only the p value will be returned. Additional output is enabled when ext.out is TRUE. In this case, a list will be returned with the following elements:
theta the estimated log difference in heights at the cutpoint
se the standard error of theta z the z statistic of the test p the p-value of the test. A p-value below the significance threshhold indicates that the user can reject the null hypothesis of no sorting. binsize the calculated size of bins for the test bw the calculated bandwidth for the test cutpoint the cutpoint used data a dataframe for the binning of the histogram. Columns are cellmp (the mid- points of each cell) and cellval (the normalized height of each cell)
McCrary, Justin. (2008) "Manipulation of the running variable in the regression discontinuity de- sign: A density test," Journal of Econometrics. 142(2): 698-714. http://dx.doi.org/10.1016/ j.jeconom.2007.05.005
X a numerical vector which is the running variable
Y a numerical vector which is the outcome variable cutpoint the cutpoint verbose logical flag indicating whether to print more information to the terminal. Default is FALSE. kernel string indicating which kernel to use. Options are "triangular" (default and recommended), "rectangular", "epanechnikov", "quartic", "triweight", "tricube", "gaussian", and "cosine".
Imbens, Guido and Karthik Kalyanaraman. (2009) "Optimal Bandwidth Choice for the regression discontinuity estimator," NBER Working Paper Series. 14726. http://www.nber.org/papers/ w14726 kernelwts 5
kernelwts Kernel Weighting function
Description
This function will calculate the appropriate kernel weights for a vector. This is useful when, for instance, one wishes to perform local regression.
Usage
kernelwts(X, center, bw, kernel = "triangular")
Arguments
X input x values. This variable represents the axis along which kernel weighting should be performed. center the point from which distances should be calculated. bw the bandwidth. kernel a string indicating the kernel to use. Options are "triangular" (the default), "epanechnikov", "quartic", "triweight", "tricube", "gaussian", and "cosine".
Value
A vector of weights with length equal to that of the X input (one weight per element of X).
Plot the relationship between the running variable and the outcome
Usage
## S3 method for class 'RD'
plot(x, gran = 400, bins = 100, which = 1, range, ...)
Arguments
x rd object, typically the result of RDestimate
gran the granularity of the plot. This specifies the number of points to either side of the cutpoint for which the estimate is calculated. bins if the dependent variable is binary, include the number of bins within which to average which identifies which of the available plots to display. For a sharp design, the only possibility is 1, the plot of the running variable against the outcome variable. For a fuzzy design, an additional plot, 2, may also be displayed, showing the relationship between the running variable and the treatment variable. Both plots may be displayed with which=c(1,2). range the range of values of the running variable for which to plot. This should be a vector of length two of the format c(min,max). To plot from the minimum to the maximum value, simply enter c("min","max"). The default is a window 20 times wider than the first listed bandwidth from the rd object, truncated by the min/max values of the running variable from the data. ... unused
Details
It is important to note that this function will only plot the discontinuity using the bandwidth which is first in the vector of bandwidths passed to RDestimate
Description RDestimate supports both sharp and fuzzy RDD utilizing the AER package for 2SLS regression under the fuzzy design. Local linear regressions are performed to either side of the cutpoint using the Imbens-Kalyanaraman optimal bandwidth calculation, IKbandwidth.
Arguments formula the formula of the RDD. This is supplied in the format of y ~ x for a simple sharp RDD, or y ~ x | c1 + c2 for a sharp RDD with two covariates. Fuzzy RDD may be specified as y ~ x + z where x is the running variable, and z is the endogenous treatment variable. Covariates are then included in the same manner as in a sharp RDD. data an optional data frame 8 RDestimate
subset an optional vector specifying a subset of observations to be used
cutpoint the cutpoint. If omitted, it is assumed to be 0. bw a numeric vector specifying the bandwidths at which to estimate the RD. If omit- ted, the bandwidth is calculated using the Imbens-Kalyanaraman method, and then estimated with that bandwidth, half that bandwidth, and twice that band- width. If only a single value is passed into the function, the RD will similarly be estimated at that bandwidth, half that bandwidth, and twice that bandwidth. kernel a string specifying the kernel to be used in the local linear fitting. "triangular" kernel is the default and is the "correct" theoretical kernel to be used for edge es- timation as in RDD (Lee and Lemieux 2010). Other options are "rectangular", "epanechnikov", "quartic", "triweight", "tricube", "gaussian" and "cosine". se.type this specifies the robust SE calculation method to use. Options are, as in vcovHC, "HC3", "const", "HC", "HC0", "HC1", "HC2", "HC4", "HC4m", "HC5". This op- tion is overriden by cluster. cluster an optional vector specifying clusters within which the errors are assumed to be correlated. This will result in reporting cluster robust SEs. This option overrides anything specified in se.type. It is suggested that data with a discrete running variable be clustered by each unique value of the running variable (Lee and Card 2008). verbose will provide some additional information printed to the terminal. model logical. If TRUE, the model object will be returned. frame logical. If TRUE, the data frame used in model fitting will be returned.
Details Covariates are problematic for inclusion in the regression discontinuity design. This package allows their inclusion, but cautions against them insomuch as is possible. When covariates are included in the specification, they are simply included as exogenous regressors. In the sharp design, this means they are simply added into the regression equation, uninteracted with treatment. Likewise for the fuzzy design, in which they are added as regressors in both stages of estimation.
Value RDestimate returns an object of class "RD". The functions summary and plot are used to obtain and print a summary and plot of the estimated regression discontinuity. The object of class RD is a list containing the following components:
type a string denoting either "sharp" or "fuzzy" RDD.
est numeric vector of the estimate of the discontinuity in the outcome under a sharp design, or the Wald estimator in the fuzzy design for each corresponding band- width se numeric vector of the standard error for each corresponding bandwidth z numeric vector of the z statistic for each corresponding bandwidth p numeric vector of the p value for each corresponding bandwidth ci the matrix of the 95 for each corresponding bandwidth RDestimate 9
bw numeric vector of each bandwidth used in estimation
obs vector of the number of observations within the corresponding bandwidth call the matched call na.action the observations removed from fitting due to missingness model (if requested) For a sharp design, a list of the lm objects is returned. For a fuzzy design, a list of lists is returned, each with two elements: firststage, the first stage lm object, and iv, the ivreg object. A model is returned for each corresponding bandwidth. frame (if requested) Returns the model frame used in fitting.
Lee, David and Thomas Lemieux. (2010) "Regression Discontinuity Designs in Economics," Jour- nal of Economic Literature. 48(2): 281-355. http://www.aeaweb.org/articles.php?doi=10. 1257/jel.48.2.281 Imbens, Guido and Thomas Lemieux. (2010) "Regression discontinuity designs: A guide to prac- tice," Journal of Econometrics. 142(2): 615-635. http://dx.doi.org/10.1016/j.jeconom. 2007.05.001 Lee, David and David Card. (2010) "Regression discontinuity inference with specification error," Journal of Econometrics. 142(2): 655-674. http://dx.doi.org/10.1016/j.jeconom.2007.05. 003 Angrist, Joshua and Jorn-Steffen Pischke. (2009) Mostly Harmless Econometrics. Princeton: Princeton University Press.
x<-runif(1000,-1,1) cov<-rnorm(1000) y<-3+2*x+3*cov+10*(x>=0)+rnorm(1000) RDestimate(y~x) # Efficiency gains can be made by including covariates RDestimate(y~x|cov) 10 summary.RD
