R/kde.algo.R

In smicd: Statistical Methods for Interval-Censored Data

#' @title Estimation of Statistical Indicators from Interval-Censored Data
#'
#' @description The function applies an iterative kernel density
#' algorithm for the estimation of a variety of statistical indicators
#' (e.g. mean, median, quantiles, gini) from interval-censored data. The estimation
#' of the standard errors is facilitated by a non-parametric bootstrap.
#' @param xclass interval-censored values; factor with ordered factor values,
#' as in \code{\link[Kernelheaping]{dclass}}
#' @param classes numeric vector of classes; Inf as last value is allowed,
#' as in \code{\link[Kernelheaping]{dclass}}
#' @param burnin burn-in sample size, as in \code{\link[Kernelheaping]{dclass}}
#' @param samples sampling iteration size, as in \code{\link[Kernelheaping]{dclass}}
#' @param bootstrap.se if \code{TRUE} standard errors for the statistical
#'  indicators are estimated
#' @param b number of bootstrap iterations for the estimation of the standard errors
#' @param bw bandwidth selector method, defaults to "nrd0", as in
#' \code{\link[stats]{density}}
#' @param evalpoints number of evaluation grid points, as in
#' \code{\link[Kernelheaping]{dclass}}
#' @param adjust the user can multiply the bandwidth by a certain factor such
#' that bw=adjust*bw as in \code{\link[stats]{density}}
#' @param threshold used for the Head-Count Ratio and Poverty Gap, default is 60\%
#' of the median e.g. \code{threshold=0.6}
#' @param custom_indicator a list of functions containing the indicators to be
#' additionally calculated.
#' Such functions must only depend on the target variable \code{y} and the
#' \code{threshold}. For the estimation of weighted custom indicators the function
#' must also depend on \code{weights}. Defaults to \code{NULL}.
#' @param upper if the upper bound of the upper interval is \code{Inf} e.g.
#' \code{(15000,Inf)}, then \code{Inf} is replaced by \code{15000*upper}
#' @param weights any kind of survey or design weights that will be used for the
#' weighted estimation of the statistical indicators
#' @param oecd weights for equivalized household size
#' @return An object of class "kdeAlgo" that provides estimates for statistical indicators
#' and optionally, corresponding standard error estimates. Generic
#' functions such as, \code{\link{print}},
#' and \code{\link{plot}} have methods that can be used
#' to obtain further information. See \code{\link{kdeAlgoObject}} for a description
#' of components of objects of class "kdeAlgo".
#' @details The statistical indicators are estimated using pseudo samples as
#' proxy for the interval-censored variable. The object \code{resultX} returns the
#' pseudo samples for each iteration step of the KDE-algorithm.
#' @references
#' Walter, P. (2019). A Selection of Statistical Methods for Interval-Censored
#' Data with Applications to the German Microcensus, PhD thesis,
#' Freie Universitaet Berlin\cr \cr
#' Groß, M., U. Rendtel, T. Schmid, S. Schmon, and N. Tzavidis (2017).
#' Estimating the density of ethnic minorities and aged people in Berlin:
#' Multivariate
#' Kernel Density Estimation applied to sensitive georeferenced administrative
#' data
#' protected via measurement error. Journal of the Royal Statistical Society:
#' Series A
#' (Statistics in Society), 180.
#' @seealso \code{\link[Kernelheaping]{dclass}}, \code{\link{print.kdeAlgo}},
#' \code{\link{plot.kdeAlgo}}
#' @export
#' @importFrom ineq ineq
#' @importFrom Hmisc wtd.quantile
#' @importFrom laeken gini
#' @importFrom laeken arpr
#' @importFrom laeken qsr
#' @importFrom stats weighted.mean
#' @importFrom stats weights
#' @import formula.tools
#' @return NULL
#' @examples
#' \dontrun{
#' # Generate data
#' x <- rlnorm(500, meanlog = 8, sdlog = 1)
#' classes <- c(0, 500, 1000, 1500, 2000, 2500, 3000, 4000, 5000, 6000, 8000, 10000, 15000, Inf)
#' xclass <- cut(x, breaks = classes)
#' weights <- abs(rnorm(500, 0, 1))
#' oecd <- rep(seq(1, 6.9, 0.3), 25)
#'
#' # Estimate statistical indicators with default settings
#' Indicator <- kdeAlgo(xclass = xclass, classes = classes)
#'
#' # Include custom indicators
#' Indicator_custom <- kdeAlgo(
#'   xclass = xclass, classes = classes,
#'   custom_indicator = list(quant5 = function(y, threshold) {
#'     quantile(y, probs = 0.05)
#'   })
#' )
#'
#' # Indclude survey and oecd weights
#' Indicator_weights <- kdeAlgo(
#'   xclass = xclass, classes = classes,
#'   weights = weights, oecd = oecd
#' )
#' }
#' \dontshow{
#' # Generate data
#' x <- rlnorm(500, meanlog = 8, sdlog = 1)
#' classes <- c(0, 500, 1000, 1500, 2000, 2500, 3000, 4000, 5000, 6000, 8000, 10000, 15000, Inf)
#' xclass <- cut(x, breaks = classes)
#'
#' # Estimate statistical indicators
#' Indicator <- kdeAlgo(xclass = xclass, classes = classes, burnin = 10, samples = 40)
#' }
#'
kdeAlgo <- function(xclass, classes, threshold = 0.6, burnin = 80, samples = 400,
                    bootstrap.se = FALSE, b = 100,
                    bw = "nrd0", evalpoints = 4000, adjust = 1,
                    custom_indicator = NULL, upper = 3, weights = NULL,
                    oecd = NULL) {
  Standard.Error <- NULL
  density.est <- dclassICD(
    xclass = xclass, classes = classes,
    burnin = burnin, samples = samples,
    bw = bw, evalpoints = evalpoints, adjust = adjust,
    upper = upper, weights = weights, oecd = oecd
  )

  Indicators.run <- NULL
  for (i in 1:dim(density.est$resultX)[2]) {
    Indicators.run <- cbind(Indicators.run, indicators.est(
      x = density.est$resultX[, i],
      threshold = threshold,
      custom_indicator = custom_indicator,
      weights = density.est$resultW[, i]
    ))
  }
  Indicators <- rowMeans(Indicators.run[, (burnin + 1):dim(Indicators.run)[2]])

  if (bootstrap.se == TRUE) {
    Standard.Error <- standardError.est(
      b = b, xclass = xclass,
      classes = classes, burnin = burnin,
      samples = samples, bw = bw,
      evalpoints = evalpoints, adjust = adjust,
      threshold = threshold,
      custom_indicator = custom_indicator, upper = upper,
      weights = weights, oecd = oecd
    )
  }

  results <- list(
    Point_estimate = Indicators, Standard_Error = Standard.Error,
    Mestimates = density.est$Mestimates,
    resultDensity = density.est$resultDensity,
    resultX = density.est$resultX,
    xclass = xclass, gridx = density.est$gridx,
    classes = classes, burnin = burnin,
    samples = samples, Point_estimate.run = Indicators.run,
    oecd = oecd, weights = weights, upper = upper
  )

  class(results) <- "kdeAlgo"
  return(results)
}