error bars in x and y (feature request)

[courtiol]

Dear all,

Following a discussion at community.rstudio with @eipi10 and @tp2750, I hereby file a feature request to help creating a double error bar (i.e. in x and y).

The idea could be to provide a new geom_* or stat_* that directly provide error bars in x and y, as it seems otherwise surprisingly tedious to code (unless one uses ggstance). The situation can be useful when users call geom_point on a dataset that has groups inside it (e.g. species in iris).

Here is a slightly modified version of @eipi10's code (using ggstance) which illustrates the point when error bars represent 95%CI:

library(ggplot2)
library(rlang)
library(dplyr)

fnc <- function(data, group, x, y, adj = 1) {
  
  group <- enquo(group)
  x <- enquo(x)
  y <- enquo(y)
  
  # Set size of whisker end caps
  w_vertical   <- data %>% pull(!!y) %>% range %>% diff/100*adj
  w_horizontal <- data %>% pull(!!x) %>% range %>% diff/100*adj
  
  # If grouping variable is numeric, turn it into a factor
  if(data %>% pull(!!group) %>% is.numeric) {
    data <- data %>% mutate(!!quo_name(group) := factor(!!group))
  } 
  
  # Generate column names for the x and y means that we'll calculate below
  x_mean <- paste0(quo_text(x), "_mean")
  y_mean <- paste0(quo_text(y), "_mean")
  
  left_join(data, 
            data %>%
              group_by(!!group) %>%
              summarise_at(vars(!!x, !!y), funs(mean = mean))
  ) %>% 
    ggplot(aes(colour = !!group)) +
    geom_point(aes(x = !!x, y = !!y), shape = 1) +
    geom_point(aes(x = !!sym(x_mean), y = !!sym(y_mean)), shape = 18, size = 3) +
    stat_summary(fun.data = mean_se, fun.args = list(mult = 1.96),
                 aes(x = !!sym(x_mean), y = !!y),
                 geom = "errorbar", width = w_horizontal) +
    ggstance::stat_summaryh(fun.data = ggstance::mean_se_h, fun.args = list(mult = 1.96),
                  aes(x = !!x, y = !!sym(y_mean)),
                  geom="errorbarh", width = w_vertical) +
    theme_bw()
}

fnc(iris, Species, Petal.Length, Sepal.Length)
![image](https://user-images.githubusercontent.com/7499666/55148649-e69c2a00-5148-11e9-8b6a-c34dbe4e80af.png)

fnc(iris, Species, Petal.Length, Sepal.Length) + facet_wrap(~ Species)
![image](https://user-images.githubusercontent.com/7499666/55148693-fca9ea80-5148-11e9-86d9-8621fef70135.png)

(not sure the figure will display, but it is easy to run)

Perhaps it could also be a feature request for ggstance instead... but I post this here only for now.

[ptoche]

reprex below for reference:

library(reprex)
library(ggplot2)
library(rlang)
library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union

fnc <- function(data, group, x, y, adj = 1) {
    
    group <- enquo(group)
    x <- enquo(x)
    y <- enquo(y)
    
    # Set size of whisker end caps
    w_vertical   <- data %>% pull(!!y) %>% range %>% diff/100*adj
    w_horizontal <- data %>% pull(!!x) %>% range %>% diff/100*adj
    
    # If grouping variable is numeric, turn it into a factor
    if(data %>% pull(!!group) %>% is.numeric) {
        data <- data %>% mutate(!!quo_name(group) := factor(!!group))
    } 
    
    # Generate column names for the x and y means that we'll calculate below
    x_mean <- paste0(quo_text(x), "_mean")
    y_mean <- paste0(quo_text(y), "_mean")
    
    left_join(data, 
              data %>%
                  group_by(!!group) %>%
                  summarise_at(vars(!!x, !!y), funs(mean = mean))
    ) %>% 
        ggplot(aes(colour = !!group)) +
        geom_point(aes(x = !!x, y = !!y), shape = 1) +
        geom_point(aes(x = !!sym(x_mean), y = !!sym(y_mean)), shape = 18, size = 3) +
        stat_summary(fun.data = mean_se, fun.args = list(mult = 1.96),
                     aes(x = !!sym(x_mean), y = !!y),
                     geom = "errorbar", width = w_horizontal) +
        ggstance::stat_summaryh(fun.data = ggstance::mean_se_h, fun.args = list(mult = 1.96),
                                aes(x = !!x, y = !!sym(y_mean)),
                                geom="errorbarh", width = w_vertical) +
        theme_bw()
}

fnc(iris, Species, Petal.Length, Sepal.Length)
#> Warning: funs() is soft deprecated as of dplyr 0.8.0
#> please use list() instead
#> 
#> # Before:
#> funs(name = f(.)
#> 
#> # After: 
#> list(name = ~f(.))
#> This warning is displayed once per session.
#> Joining, by = "Species"

fnc(iris, Species, Petal.Length, Sepal.Length) + facet_wrap(~ Species)
#> Joining, by = "Species"

^{Created on 2019-03-28 by the reprex package (v0.2.1)}

[courtiol]

Thanks @ptoche... I should have done that.

[hadley]

I think this would be most natural in an extension package (maybe in ggstance?); otherwise I think we'd need to make bidirectional versions of geom_crossbar(), geom_errorbar(), geom_linerange() and geom_pointrange().

[courtiol]

For reference: @tp2750 has recently coded the discussed feature in a separate 📦 : ggerr

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/