rddensity: Manipulation Testing Using Local Polynomial Density...

In rddensity: Manipulation Testing Based on Density Discontinuity

Manipulation Testing Using Local Polynomial Density Estimation

Description

rddensity implements manipulation testing procedures using the local polynomial density estimators proposed in Cattaneo, Jansson and Ma (2020), and implements graphical procedures with valid confidence bands using the results in Cattaneo, Jansson and Ma (2022, 2023). In addition, the command provides complementary manipulation testing based on finite sample exact binomial testing following the esults in Cattaneo, Frandsen and Titiunik (2015) and Cattaneo, Frandsen and Vazquez-Bare (2017). For an introduction to manipulation testing see McCrary (2008).

A companion Stata package is described in Cattaneo, Jansson and Ma (2018).

Companion commands: rdbwdensity for data-driven bandwidth selection, and rdplotdensity for density plots.

Related Stata and R packages useful for inference in regression discontinuity (RD) designs are described in the website: https://rdpackages.github.io/.

Usage

rddensity(
  X,
  c = 0,
  p = 2,
  q = 0,
  fitselect = "",
  kernel = "",
  vce = "",
  massPoints = TRUE,
  h = c(),
  bwselect = "",
  all = FALSE,
  regularize = TRUE,
  nLocalMin = NULL,
  nUniqueMin = NULL,
  bino = TRUE,
  binoW = NULL,
  binoN = NULL,
  binoWStep = NULL,
  binoNStep = NULL,
  binoNW = 10,
  binoP = 0.5
)

Arguments

X	Numeric vector or one dimensional matrix/data frame, the running variable.
c	Numeric, specifies the threshold or cutoff value in the support of X, which determines the two samples (e.g., control and treatment units in RD settings). Default is 0.
p	Nonnegative integer, specifies the local polynomial order used to construct the density estimators. Default is 2 (local quadratic approximation).
q	Nonnegative integer, specifies the local polynomial order used to construct the bias-corrected density estimators. Default is p+1 (local cubic approximation for default p=2).
fitselect	String, specifies the density estimation method. "unrestricted" for density estimation without any restrictions (two-sample, unrestricted inference). This is the default option. "restricted" for density estimation assuming equal distribution function and higher-order derivatives.
kernel	String, specifies the kernel function used to construct the local polynomial estimators. "triangular": K(u)=(1-|u|)*(|u|<=1). This is the default option. "epanechnikov": K(u) = 0.75*(1-u^2)*(|u|<=1). "uniform": K(u) = 0.5 * (|u|<=1).
vce	String, specifies the procedure used to compute the variance-covariance matrix estimator. "plugin" for asymptotic plug-in standard errors. "jackknife" for jackknife standard errors. This is the default option.
massPoints	TRUE (default) or FALSE, specifies whether to adjust for mass points in the data.
h	Numeric, specifies the bandwidth used to construct the density estimators on the two sides of the cutoff. If not specified, the bandwidth h is computed by the companion command rdbwdensity If two bandwidths are specified, the first bandwidth is used for the data below the cutoff and the second bandwidth is used for the data above the cutoff.
bwselect	String, specifies the bandwidth selection procedure to be used. "each" based on MSE of each density estimator separately (two distinct bandwidths, hl and hr). "diff" based on MSE of difference of two density estimators (one common bandwidth, hl=hr). "sum" based on MSE of sum of two density estimators (one common bandwidth, hl=hr). "comb" bandwidth is selected as a combination of the alternatives above. This is the default option. For fitselect="unrestricted", it selects median(each,diff,sum). For fitselect = "restricted", it selects min(diff,sum).
all	TRUE or FALSE (default), if specified, will report two testing procedures: conventional test statistic (not valid when using MSE-optimal bandwidth choice) and robust bias-corrected statistic.
regularize	TRUE (default) or FALSE, specifies whether to conduct local sample size checking. When set to TRUE, the bandwidth is chosen such that the local region includes at least nLocalMin observations and at least nUniqueMin unique observations.
nLocalMin	Nonnegative integer, specifies the minimum number of observations in each local neighborhood. This option will be ignored if set to 0, or if regularize=FALSE is used. Default is 20+p+1.
nUniqueMin	Nonnegative integer, specifies the minimum number of unique observations in each local neighborhood. This option will be ignored if set to 0, or if regularize=FALSE is used. Default is 20+p+1.
bino	TRUE (default) or FALSE, specifies whether to conduct binomial tests. By default, the initial (smallest) window contains at least 20 observations on each side, and its length is also used as the increment for subsequent windows. This feature is based on the binom.test function.
binoW	Numeric, specifies the half length(s) of the initial window. If two values are provided, they will be used for the data below and above the cutoff separately.
binoN	Nonnegative integer, specifies the minimum number of observations on each side of the cutoff used for the binomial test. This option will be ignored if binoW is provided.
binoWStep	Numeric, specifies the increment in half length(s).
binoNStep	Nonnegative integer, specifies the minimum increment in sample size (on each side of the cutoff). This option will be ignored if binoWStep is provided.
binoNW	Nonnegative integer, specifies the total number of windows. Default is 10.
binoP	Numeric, specifies the null hypothesis of the binomial test. Default is 0.5.

Value

hat	left/right: density estimate to the left/right of cutoff; diff: difference in estimated densities on the two sides of cutoff.
sd_asy	left/right: standard error for the estimated density to the left/right of the cutoff; diff: standard error for the difference in estimated densities. (Based on asymptotic formula.)
sd_jk	left/right: standard error for the estimated density to the left/right of the cutoff; diff: standard error for the difference in estimated densities. (Based on the jackknife method.)
test	t_asy/t_jk: t-statistic for the density discontinuity test, with standard error based on asymptotic formula or the jackknife; p_asy/p_jk: p-value for the density discontinuity test, with standard error based on asymptotic formula or the jackknife.
hat_p	Same as hat, without bias correction (only available when all=TRUE).
sd_asy_p	Same as sd_asy, without bias correction (only available when all=TRUE).
sd_jk_p	Same as sd_jk, without bias correction (only available when all=TRUE).
test_p	Same as test, without bias correction (only available when all=TRUE).
N	full: full sample size; left/right: sample size to the left/right of the cutoff; eff_left/eff_right: effective sample size to the left/right of the cutoff (this depends on the bandwidth).
h	left/right: bandwidth used to the left/right of the cutoff.
opt	Options passed to the function.
bino	Binomial test results. leftWindow/rightWindow: window lengths. leftN/rightN: number of observations. pval: p-values.
X_min	left/right: the samllest observation to the left/right of the cutoff.
X_max	left/right: the largest observation to the left/right of the cutoff.

Author(s)

Matias D. Cattaneo, Princeton University [e-mail].

Michael Jansson, University of California Berkeley. [e-mail].

Xinwei Ma (maintainer), University of California San Diego. [e-mail].

References

Cattaneo, M. D., B. Frandsen, and R. Titiunik. 2015. Randomization Inference in the Regression Discontinuity Design: An Application to the Study of Party Advantages in the U.S. Senate. Journal of Causal Inference 3(1): 1-24. doi: 10.1515/jci-2013-0010

Cattaneo, M. D., M. Jansson, and X. Ma. 2018. Manipulation Testing based on Density Discontinuity. Stata Journal 18(1): 234-261. doi: 10.1177/1536867X1801800115

Cattaneo, M. D., M. Jansson, and X. Ma. 2020. Simple Local Polynomial Density Estimators. Journal of the American Statistical Association, 115(531): 1449-1455. doi: 10.1080/01621459.2019.1635480

Cattaneo, M. D., M. Jansson, and X. Ma. 2022. lpdensity: Local Polynomial Density Estimation and Inference. Journal of Statistical Software, 101(2), 1–25. doi: 10.18637/jss.v101.i02

Cattaneo, M. D., M. Jansson, and X. Ma. 2023. Local Regression Distribution Estimators. Journal of Econometrics, forthcoming. doi: 10.1016/j.jeconom.2021.01.006

Cattaneo, M. D., R. Titiunik and G. Vazquez-Bare. 2017. Comparing Inference Approaches for RD Designs: A Reexamination of the Effect of Head Start on Child Mortality. Journal of Policy Analysis and Management 36(3): 643-681. doi: 10.1002/pam.21985

McCrary, J. 2008. Manipulation of the Running Variable in the Regression Discontinuity Design: A Density Test. Journal of Econometrics 142(2): 698-714. doi: 10.1016/j.jeconom.2007.05.005

See Also

rdbwdensity, rdplotdensity

Examples

### Continuous Density
set.seed(42)
x <- rnorm(2000, mean = -0.5)
rdd <- rddensity(X = x, vce = "jackknife")
summary(rdd)

### Bandwidth selection using rdbwdensity()
rddbw <- rdbwdensity(X = x, vce = "jackknife")
summary(rddbw)

### Plotting using rdplotdensity()
# 1. From -2 to 2 with 25 evaluation points at each side
plot1 <- rdplotdensity(rdd, x, plotRange = c(-2, 2), plotN = 25)

# 2. Plotting a uniform confidence band
set.seed(42) # fix the seed for simulating critical values
plot2 <- rdplotdensity(rdd, x, plotRange = c(-2, 2), plotN = 25, CIuniform = TRUE)

### Density discontinuity at 0
x[x > 0] <- x[x > 0] * 2
rdd2 <- rddensity(X = x, vce = "jackknife")
summary(rdd2)
plot3 <- rdplotdensity(rdd2, x, plotRange = c(-2, 2), plotN = 25)

rddensity documentation built on Jan. 22, 2023, 1:26 a.m.