Search code examples
rggplot2tidyverse

make calculation inside ggplot with only subset from data according to aes

Hey everyone I want to label outliers with the unique id in my geom_boxplot. In general I use this formula (whatever is the data column) is the id I want:

plot +
  geom_text(aes(label = ifelse((whatever > 
                                  (quantile(whatever, 0.75) + 1.5 * IQR(whatever, 0.25))) |
                                 (whatever < 
                                    (quantile(whatever, 0.25) - 1.5 * IQR(whatever, 0.25))),
                               id, '')),
            hjust = -1)

This works fine, if I only have one boxplot (plot_all). However, if I separate the boxplots by a value (e.g. sex), the outliers are calculated based on the whole date and not on the subset of data. Is there anyway I can forward the selected data to calculate for those only?

rm(list=ls())

library(tidyverse)
library(ggplot2)
library(ggbeeswarm)

set.seed(1)

male <- tibble(sex = sample('male', 1000, replace=TRUE),
               id = as.character(1:1000),
               whatever = c(runif(998, min = .75, max = 1.25), 0, 2))

female <- tibble(sex = sample('female', 1000, replace=TRUE),
                 id = as.character(1001:2000),
                 whatever = c(runif(998, min = 1.25, max = 1.75), .5, 2.5))

data <- bind_rows(male, female)

plot_all <- ggplot(data, aes("all", whatever)) +
  geom_boxplot() +
  geom_quasirandom(size = .3, alpha = .3) +
  geom_text(aes(label = ifelse((whatever > 
                                  (quantile(whatever, 0.75) + 1.5 * IQR(whatever, 0.25))) |
                                 (whatever < 
                                    (quantile(whatever, 0.25) - 1.5 * IQR(whatever, 0.25))),
                               id, '')),
            hjust = -1)

plot_sex <- ggplot(data, aes(sex, whatever)) +
  geom_boxplot() +
  geom_quasirandom(size = .3, alpha = .3) +
  geom_text(aes(label = ifelse((whatever > 
                                  (quantile(whatever, 0.75) + 1.5 * IQR(whatever, 0.25))) |
                                 (whatever < 
                                    (quantile(whatever, 0.25) - 1.5 * IQR(whatever, 0.25))),
                               id, '')),
            hjust = -1)
Solution

  • It is possible to do this by manipulating the layer data, but why would you? Just modify the data on the way into ggplot:

    data %>%
  group_by(sex) %>%
  mutate(label = ifelse(whatever > max(boxplot.stats(whatever)$stats) | 
                        whatever < min(boxplot.stats(whatever)$stats), 
                        id, '')) %>%
  ggplot(aes(sex, whatever)) +
  geom_boxplot() +
  geom_quasirandom(size = .3, alpha = .3) +
  geom_text(aes(label = label), hjust = -1)

    enter image description here

    This is tidier than altering the layer data (which gives the same result)

      ggplot(data, aes(sex, whatever)) +
  geom_boxplot() +
  geom_quasirandom(size = .3, alpha = .3) +
  geom_text(data = . %>% group_by(sex) %>% filter((whatever > 
                                (quantile(whatever, 0.75) + 1.5 * IQR(whatever, 0.25))) |
                               (whatever < 
                               (quantile(whatever, 0.25) - 1.5 * IQR(whatever, 0.25)))), 
            aes(label = id), hjust = -1)