R/si.data.R

In yimindai0521/MBalance: Mahalanobis balancing: a multivariate perspective on approximate covariate balancing

#' @title Generate data 
#' 
#' @details 
#' We first generate covariates X from standard normal distribution function.
#' Then we will use the following propensity score model with the binomial 
#' distribution to generate treatment indicator T:
#' 
#' \eqn{Pr(T = 1 | X) = 1/ (1 + exp(X_1 + ... + X_{10}))}
#' 
#' Finally, the outcome model will be generated by 
#' \eqn{Y = (X_1 + ... + X_{10}) + noise}, 
#' where noise follows standard normal distribution.
#' 
#' @description Generate data for testing our main function \code{MB} and \code{hdMB}..
#' @param sample.size sample size for data, Default: 500,
#' @param dimension dimension for covariate, it should be larger than 10, Default: 10.
#' 
#' @return data object with the following attributes:
#' \itemize{
#' \item{X:}{ covariates.}
#' \item{Tr:}{ treatment indicator.}
#' \item{Y:}{ observed outcome.}
#' }
#' 
#' @examples 
#' data <- si.data()
#' @rdname si.data
#' @export 
si.data <- function(sample.size = 500, dimension = 10){
  X   <- matrix(rnorm(sample.size * dimension), nrow = sample.size, ncol = dimension)
  ps1 <- rep(1,sample.size)
  ps2 <- exp(1 * apply(X[,1:10],1,sum))
  ps  <- cbind(ps1 / (ps1 + ps2),ps2 / (ps1 + ps2))
  uniform <- runif(sample.size)
  index   <- cbind(ps[,1] > uniform)
  noise   <- rnorm(sample.size)
  Tr      <- apply(index,1,sum)
  Y       <- 1 * apply(X[,1:10],1,sum) + noise
  return(list(X = X, Tr = Tr, Y = Y))
}