I'm having trouble understanding what the 'fct_reorder2()' function does - r

Question

I'm trying to understand what the fct_reorder2() does but the R documentation is not clear on this. Below is copied from help for this function:
df <- tibble::tribble(
~color, ~a, ~b,
"blue", 1, 2,
"green", 6, 2,
"purple", 3, 3,
"red", 2, 3,
"yellow", 5, 1
)
df$color <- factor(df$color)
#> Levels: blue green purple red yellow
fct_reorder(df$color, df$a, min)
#>Levels: blue red purple yellow green
fct_reorder2(df$color, df$a, df$b)
#> Levels: purple red blue green yellow
Can someone help me understand why fct_reorder2() sorts the data this way?

Answer 1

The example in the docs is poor, because there is only one row of data for each factor level.
What helped me was a chapter of R4DS: tidyverse and beyond, which is based on an example in the original R4DS, which I reproduce below.
fct_reorder2() reorders the factor by the y values associated with the largest x values. This makes the plot easier to read because the line colours line up with the legend.
library(tidyverse)
by_age <- gss_cat %>%
filter(!is.na(age)) %>%
count(age, marital) %>%
group_by(age) %>%
mutate(prop = n / sum(n))
ggplot(by_age, aes(age, prop, colour = marital)) +
geom_line(na.rm = TRUE)
ggplot(by_age, aes(age, prop, colour = fct_reorder2(marital, age, prop))) +
geom_line() +
labs(colour = "marital")
Created on 2021-03-03 by the reprex package (v1.0.0)

Answer 2

The fct_reorder2 is using the .fun as last2. If we check the source code
last2
function (.x, .y)
{
.y[order(.x, na.last = FALSE)][length(.y)]
}
and now using last2
last2(df$a, df$b)
#[1] 2
df$b[order(df$a, na.last = FALSE)][length(df$b)]
#[1] 2
Now, if we check the source code of fct_reorder, it is calling lvls_reorder, which calls refactor
s1 <- tapply(seq_along(df$a), df$color, function(i) last2(df$a[i], df$b[i]))
forcats:::refactor(df$color, levels(df$color)[order(s1, decreasing = TRUE)])
#[1] blue green purple red yellow
#Levels: purple red blue green yellow
Code lineage
fct_reorder2
function (.f, .x, .y, .fun = last2, ..., .desc = TRUE)
{
f <- check_factor(.f)
stopifnot(length(f) == length(.x), length(.x) == length(.y))
ellipsis::check_dots_used()
summary <- tapply(seq_along(.x), f, function(i) .fun(.x[i],
.y[i], ...))
if (is.list(summary)) {
stop("`fun` must return a single value per group", call. = FALSE)
}
lvls_reorder(.f, order(summary, decreasing = .desc))
}
lvls_reorder
function (f, idx, ordered = NA)
{
f <- check_factor(f)
if (!is.numeric(idx)) {
stop("`idx` must be numeric", call. = FALSE)
}
if (!setequal(idx, lvls_seq(f)) || length(idx) != nlevels(f)) {
stop("`idx` must contain one integer for each level of `f`",
call. = FALSE)
}
refactor(f, levels(f)[idx], ordered = ordered)
}

Answer 3

I was also confused by this function and it toke me a lot of time to understand. Here below is my understanding. I hope it helps.
Imagine that you are trying to draw a line plot with multiple lines. The lines are distinguished by their colors. Allow me to use code from #nacnudus, for clarity.
library(tidyverse)
by_age <- gss_cat %>%
filter(!is.na(age)) %>%
count(age, marital) %>%
group_by(age) %>%
mutate(prop = n / sum(n)) %>%
ungroup()
# here we use a customized sequential palette
my_palette = c("#7a0177", "#c51b8a", "#f768a1", "#fa9fb5", "#fcc5c0", "#feebe2")
scales::show_col(my_palette)
my_palette
ggplot(by_age, aes(age, prop, colour = marital)) +
scale_color_manual(values = my_pallete) +
geom_line(na.rm = TRUE)
line plot without reorder
The color without reorder the marital variable looks messy. To make colors match the data, we generally hope the color is sequentially correspond to the value that on the rightest end of each line, which means the larger of the value at the rightest point, the darker the color.
So our next mission is to reorder the variable marital according to their vlaue that in the rightest point (i.e, the value of prop at each marital's largest age).
new_order = by_age %>%
group_by(marital) %>% # for each marital
arrange(age, prop) %>% # group first by age, and then prop
dplyr::filter(age == max(age)) %>% # find the largest age in each marital
ungroup() %>%
arrange(desc(prop)) %>% # reorder the data by prop
pull(marital) %>% # get the new order
as.character() # convert the new order to general character
# Then we plot the reordered line plot
ggplot(by_age, aes(age, prop, colour = fct_relevel(marital, new_order))) + # now we reorder the marital by new orders
geom_line() +
scale_color_manual(values = my_palette) +
labs(colour = "marital")
reordered line plot 1
Now it's colors look pretty reasonable.
And it's EXACTLY what fct_reorder2() does for you.
ggplot(by_age, aes(age, prop, colour = fct_reorder2(marital, age, prop))) + # now we use fct_reorder2 to reorder marital
geom_line() +
scale_color_manual(values = my_palette) +
labs(colour = "marital")
reordered line plot 2
What it does is for each marital, first order the data by age, then by prop, and then pick the last prop value (In this case, there is only one value). The picked values are used to reorder the factor level (by default .fun = last2()). If you define .fun = first2()， the function will reorder marital by the prop vlaue at each marital's smallest age:
ggplot(by_age, aes(age, prop, colour = fct_reorder2(marital, age, prop, .fun = first2))) +
geom_line() +
scale_color_manual(values = my_palette) +
labs(colour = "marital")
reordered line plot 3
I hope it helps.