R/compute_IM_quantile.R

In cgaillac/MarginalFElogit: Estimation of Average Marginal Effects in Fixed Effect Logit Models

#' Computes the quantile suggested by Imbens & Manski 2004 in order to have
#' CI with asymptotic level of 1 - alpha. It solves Eq. (2) in Stoye (2009)
#'
#' @param alpha the level for the confidence interval
#' @param bounds1 a vector of length 2, the estimated bounds
#' @param std_bounds1 a vector of length 2, the estimated standard errors for
#' each bound
#'
#' @return the value of the quantile suggested by Imbens & Manski 2004
#' @export
#'
# @examples
compute_IM_quantile <- function(alpha, bounds1, std_bounds1) {

  term <- (bounds1[2]-bounds1[1]) / max(std_bounds1)
  binf <- qnorm(1-alpha)
  bsup <- qnorm(1-alpha/2)

  ff <- function(q) {return( (pnorm(q+term)-pnorm(-q)-1+alpha)^2)}
  res <- optimize(ff, c(binf,bsup))$minimum

  return(res)
}