wunt: Data-driven Uniform transformer.

In lakerwsl/WUNT: Covariate balancing by uniform transformer

Description

This function constructs weights for each subject in the control group with a data-driven Uniform transformer, and estimates the average treatment effect on the treated group with the constructed weights.

Usage

wunt(x,z,y,type='constant',marginal=FALSE,extra=FALSE,xextra=NULL,
method='Kernel',h=NULL,kernel='gaussian',degree=2,interact=TRUE)

Arguments

x	A data matrix/vector
z	A vector of treamtment assignment, with 1 for treated and 0 for control.
y	A vector of observed outcomes, with length(y)=length(z).
type	Type of the smooth kernel function used to approximate the density of the covariates in the control group. type should be one of 'constant', 'linear' and 'quadratic'.
marginal	An indicator for whether the Uniform transformer is constructed based on the marginal distribution or joint distribution. By defaul, marginal is FALSE, i.e., the Uniform transformer is constructed based on the joint distribution of x.
extra	An indicator for whether there is extra unlabeled data to construct the Uniform transformer.
xextra	Extra unlabeled data to construct the Uniform transformer. xextra must be specified if extra=T
method	Method to estimate the density of the transformed data in the treated group. method should be 'Kernel' or 'Projection'
h	The bandwidth to be used for the Kernel method.
kernel	Type of kernel. kernel can be chosen from 'gaussian','gaussian4','gaussian6','gaussian8','gaussian10','epanechnikov','uniform','triangular','triweight','tricube','biweight',cosine','silverman'
degree	Degree of polynomials for the basis functions of the Projection method.
interact	An indicator for whether interactions would be included for the basis functions.

Value

est	Estimated average treatment effect on the treated group.
weight	Constructed weights for the each subject in the control group.
control	Transformed data for the control group.
treated	Tranformed data for the treated group.

Examples

x1=rnorm(1000)
x2=rnorm(1000, mean=0.5)
x=cbind(x1,x2)
ix=sample(1:1000,300)
z=numeric(1000)
z[ix]=1
y=x1*x2
res1=wunt(x,z,y,type='constant',marginal=FALSE,method='Kernel')
res1$est
res2=wunt(x,z,y,type='constant',marginal=FALSE,method='Projection')
res2$est

lakerwsl/WUNT documentation built on Sept. 13, 2020, 12:15 a.m.