New effect size: Most Extreme (ECDF) Differences

NAMESPACE

@@ -170,6 +170,7 @@ export(logoddsratio_to_riskratio)
 export(mad_pooled)
 export(mahalanobis_d)
 export(means_ratio)
+export(med)
 export(nnt)
 export(nnt_to_arr)
 export(nnt_to_logoddsratio)

R/med.R

@@ -0,0 +1,124 @@
+#' Most Extreme Differences
+#'
+#' Compute the Most Extreme Differences effect sizes between two ECDFs. Pair
+#' with any reported Kolmogorov-Smirnov test ([`stats::ks.test()`]).
+#'
+#' @inheritParams cohens_d
+#' @param pdf a character string naming a cumulative distribution function or an
+#'   actual cumulative distribution function such as `pnorm`. Only continuous
+#'   CDFs are valid
+#' @param alternative a character string specifying the alternative hypothesis.
+#'   This _does not_ control the type of CI returned, but the effect calculated;
+#'   see [`stats::ks.test()`] for more details.
+#' @param ... for the default method, parameters of the distribution specified
+#'   (as a character string) by `pdf`.
+#'
+#' @details Something...
+#'
+#' @inheritSection effectsize_CIs CIs and Significance Tests
+#' @inheritSection print.effectsize_table Plotting with `see`
+#'
+#' @return A data frame with the effect size `D` and its CI (`CI_low` and
+#'   `CI_high`).
+#'
+#' @family standardized differences
+#'
+#' @examples
+#' x <- rnorm(50)
+#' y <- runif(30)
+#'
+#' # Do x and y come from the same distribution?
+#' ks.test(x, y)
+#' med(x, y)
+#'
+#' # Is the CDS of x larger than that of y?
+#' ks.test(x, y, alternative = "l")
+#' med(x, y, alternative = "l")
+#'
+#' plot(ecdf(x), xlim = range(c(x, y)), col = "blue")
+#' lines(ecdf(y), col = "red")
+#'
+#' # Does x come from a shifted gamma distribution with shape 3 and rate 2?
+#' ks.test(x+2, "pgamma", 3, 2)
+#' med(x+2, pdf = "pgamma", shape = 3, rate = 2)
+#'
+#' plot(ecdf(x + 2), xlim = range(x + 2), col = "blue")
+#' curve(pgamma(x, 3, 2), col = "red", add = TRUE)
+#'
+#'
+#' @export
+med <- function(
+    x,
+    y = NULL,
+    data = NULL,
+    pdf,
+    mu = 0,
+    reference = NULL,
+    ci = 0.95,
+    alternative = "two.sided",
+    verbose = TRUE,
+    ...
+) {
+  # if (.is_htest_of_type(x, "t-test")) {
+  #   return(effectsize(x, type = type, verbose = verbose, data = data, ...))
+  # } else if (
+  #   .is_BF_of_type(x, c("BFoneSample", "BFindepSample"), "t-squared")
+  # ) {
+  #   return(effectsize(x, ci = ci, verbose = verbose, ...))
+  # }
+
+  alternative <- .match.alt(alternative)
+  out <- .get_data_2_samples(
+    x,
+    y,
+    data,
+    reference = reference,
+    verbose = verbose,
+    ...
+  )
+  x <- out[["x"]]
+  y <- out[["y"]]
+
+  test <- stats::ks.test(x - mu, y %||% pdf, alternative = alternative, ...)
+
+  out <- data.frame(D = unname(test$statistic))
+
+  if (.test_ci(ci)) {
+    # Add cis
+    out$CI <- ci
+    ci_level <- .adjust_ci(ci, alternative)
+    alpha <- 1 - ci_level
+
+    if (is.null(y)) {
+      n <- length(x)
+      s <- sqrt(n)
+    } else {
+      n1 <- length(x)
+      n2 <- length(y)
+      s <- sqrt(n1 * n2 / (n1 + n2))
+    }
+
+    conf_int <- out[["D"]] + c(-1, 1) * stats::qnorm(1 - alpha / 2) / s
+    out$CI_low <- pmax(conf_int[1], 0)
+    out$CI_high <- pmin(conf_int[2], 1)
+    ci_method <- list(method = "normal")
+  } else {
+    ci_method <- alternative <- NULL
+  }
+
+  class(out) <- c(
+    "effectsize_difference",
+    "effectsize_table",
+    "see_effectsize_table",
+    class(out)
+  )
+  .someattributes(out) <- .nlist(
+    ci,
+    mu,
+    ci_method,
+    alternative,
+    approximate = FALSE
+  )
+  out
+}
+

play with med.R

@@ -0,0 +1,22 @@
+library(effectsize)
+
+x <- rnorm(50)
+y <- runif(30)
+
+# Do x and y come from the same distribution?
+ks.test(x, y)
+med(x, y)
+
+# Is the CDS of x larger than that of y?
+ks.test(x, y, alternative = "l")
+med(x, y, alternative = "l")
+
+plot(ecdf(x), xlim = range(c(x, y)), col = "blue")
+lines(ecdf(y), col = "red")
+
+# Does x come from a shifted gamma distribution with shape 3 and rate 2?
+ks.test(x+2, "pgamma", 3, 2)
+med(x+2, pdf = "pgamma", shape = 3, rate = 2)
+
+plot(ecdf(x + 2), xlim = range(x + 2), col = "blue")
+curve(pgamma(x, 3, 2), col = "red", add = TRUE)