Allow choosing quantile definition in boxplots (#6820)

* update: geom_boxplot to add quantile type argument

* update: change quantile_type to quantile.type

* update: NEWS.md

* update: document()

* tiny doc polish

Author: munoztd0

NEWS.md

@@ -2,6 +2,10 @@
 
 * Adapt to changes in `rlang::warn_dots_used()` (@lionel-, #6830).
 * Fixed bug where `guide_axis_theta()` didn't anticipate old-style text elements (#6803).
+* `geom_boxplot()`/`stat_boxplot()` gain a `quantile.type` parameter (default `7`) 
+   to control the percentile definition used for hinges and median; set `quantile.type = 2` 
+   to match SAS's default `PCTLDEF = 5`, enabling parity with SAS boxplots out-of-the-box. 
+   (@munoztd0, #6819)
 
 # ggplot2 4.0.2
 

R/stat-boxplot.R

@@ -53,15 +53,16 @@ StatBoxplot <- ggproto("StatBoxplot", Stat,
 
   extra_params = c("na.rm", "orientation"),
 
-  compute_group = function(data, scales, width = NULL, na.rm = FALSE, coef = 1.5, flipped_aes = FALSE) {
+  compute_group = function(data, scales, width = NULL, na.rm = FALSE, coef = 1.5, flipped_aes = FALSE, quantile.type = 7) {
     data <- flip_data(data, flipped_aes)
     qs <- c(0, 0.25, 0.5, 0.75, 1)
 
     if (!is.null(data$weight)) {
       mod <- quantreg::rq(y ~ 1, weights = weight, data = data, tau = qs)
       stats <- as.numeric(stats::coef(mod))
     } else {
-      stats <- as.numeric(stats::quantile(data$y, qs))
+      # Follow base R default (type = 7) unless overridden by user
+      stats <- as.numeric(stats::quantile(data$y, qs, type = quantile.type))
     }
     names(stats) <- c("ymin", "lower", "middle", "upper", "ymax")
     iqr <- diff(stats[c(2, 4)])
@@ -99,6 +100,8 @@ StatBoxplot <- ggproto("StatBoxplot", Stat,
 
 #' @rdname geom_boxplot
 #' @param coef Length of the whiskers as multiple of IQR. Defaults to 1.5.
+#' @param quantile.type An integer between 1 and 9 setting the quantile algorithm
+#'   per [`stats::quantile(type)`][stats::quantile]. Defaults to `7`
 #' @inheritParams stat_identity
 #' @export
 #' @eval rd_computed_vars(

man/geom_boxplot.Rd

@@ -109,3 +109,32 @@ test_that("boxplot draws correctly", {
       )
   )
 })
+
+test_that("quantile.type changes hinges for small samples (unweighted)", {
+  df <- data_frame(x = 1, y = c(1, 2, 3, 4))
+
+  p_default <- ggplot(df, aes(x, y)) + stat_boxplot()
+  d_default <- get_layer_data(p_default)
+
+  p_t2 <- ggplot(df, aes(x, y)) + stat_boxplot(quantile.type = 2)
+  d_t2 <- get_layer_data(p_t2)
+
+  # Lower/upper hinges should differ under different quantile definitions
+  expect_false(isTRUE(all.equal(d_default$lower, d_t2$lower)))
+  expect_false(isTRUE(all.equal(d_default$upper, d_t2$upper)))
+})
+
+test_that("quantile.type = 7 matches default behavior (backward compatible)", {
+  set.seed(123)
+  df <- data_frame(x = 1, y = rnorm(25))
+
+  p_default <- ggplot(df, aes(x, y)) + stat_boxplot()
+  d_default <- get_layer_data(p_default)
+
+  p_t7 <- ggplot(df, aes(x, y)) + stat_boxplot(quantile.type = 7)
+  d_t7 <- get_layer_data(p_t7)
+
+  expect_equal(d_default$lower, d_t7$lower)
+  expect_equal(d_default$middle, d_t7$middle)
+  expect_equal(d_default$upper, d_t7$upper)
+})