R/huber2.R

In tobiasschoch/robsurvey: Robust Survey Statistics Estimation

# Weighted Huber proposal 2 estimator of location and scale
huber2 <- function(x, w, k = 1.5, na.rm = FALSE, maxit = 50, tol = 1e-4,
                   info = FALSE, k_Inf = 1e6, df_cor = TRUE)
{
    stopifnot(k_Inf > 0, is.numeric(k))
    dat <- .check_data_weights(x, w, na.rm)
    if (is.null(dat))
        return(NA)

    kk <- if (is.finite(k))
        k
    else
        k_Inf

    tmp <- .C(C_whuber2, x = as.double(dat$x), w = as.double(dat$w),
        robwgt = as.double(numeric(dat$n)), k = as.double(kk),
        loc = as.double(numeric(1)), scale = as.double(numeric(1)),
        n = as.integer(dat$n), maxit = as.integer(maxit), tol = as.double(tol),
        df_cor = as.integer(df_cor), success = as.integer(0))

    if (tmp$success == 0) {
        warning("Initial estimate of scale (IQR) is zero\n", call. = FALSE)
        return(NA)
    }

    if (tmp$maxit == maxit) {
        warning(paste0("Failure of convergence\n"), call. = FALSE)
        tmp$loc <- NA; tmp$scale <- NA; tmp$robwgt <- rep(NA, dat$n)
    }

    # return
    if (info) {
        res <- list(characteristic = "mean",
	        estimator = paste0("Weighted Huber proposal 2 estimator (k=", k,
                ")"),
            estimate = tmp$loc, variance = NA, scale = tmp$scale,
            robust = list(k = k, robweights = tmp$robwgt),
            optim = list(converged = (tmp$maxit < maxit), niter = tmp$maxit,
                tol = tmp$tol),
            residuals = dat$x - tmp$loc,
            model = list(y = dat$x, w = dat$w), design = NA,
            call = match.call())
        res
    } else {
        tmp$loc
    }
}