yujian-wu / ctsdoseresponse

ctsDoseResponse

Overview

ctsDoseResponse is a tool for nonparametric estimation and inference of a continuous dose-response curve, as well as for null hypothesis testing of whether a dose-response is flat or not. It has two main functions ctsCausal and ctsCausalTest.

The function for nonparametric estimation and inference, ctsCausal, is implemented with three different methods. The method dr.isoreg is for causal isotonic regression, the method dr.loclin is for doubly robust estimation of continuous treatment effects and the method dr.debiased is for debiased doubly robust estimation of continuous treatment effects.

ctsCausalTest has two different methods for null hypothesis testing. dr.nd is for nonparametric tests of causal null with nondiscrete exposures and dr.cts is for nonparametric doubly robust test for a continuous treatment effect.

Installation

Please use the R command devtools::install_github("https://github.com/Yujian-Wu/ctsDoseResponse/tree/main", dependencies = T) to install the package and dependent packages. You might be prompted to updated some existing packages. If you wish to update the package RcppArmadillo and the version of your R is not most updated, please be advised to install it from sources that the package does NOT need compilation.

Example

library(ctsDoseResponse)

set.seed(12345)

n <- 200
cols <- 3

W <- matrix(runif(n*cols), ncol = cols) # a 200 * 3 matrix of covariates
A <- rnorm(n, mean=W%*%rnorm(cols)) # a 200 * 1 vector of treatment variable
Y <- rnorm(n, mean = a^2 + rnorm(n)) # a 200 * 1 vector of response variable

### causal isotonic regression
iso.reg <- ctsCausal(Y, A, W, method='dr.isoreg', cross.fit = F)

### continuous dose-response curve with smoothing treatment
loc.lin <- ctsCausal(Y, A, W, method='dr.loclin')

### debiased continuous dose-response curve with smoothing treatment
bc.loc.lin <- ctsCausal(Y, A, W, method='dr.debiased')

### null hypothesis test for non-diescrete treatment
test.nd <- ctsCausalTest(Y, A, W, method='dr.nd')

### null hypothesis test for continuous treatment
test.cts <- ctsCausalTest(Y, A, W, method='dr.cts')