set/reorder factor levels not working in ggplot (bar chart)

Question

I was trying to create a bar plot by the order of stay. I used reorder and factor but the order of bars in the plot did not change. Here is a reproducible example.

library(ggplot2)
speed <- c(300, 75, 100, 200, 150, 225, 75, 300, 150, 75, 200, 150, 225, 75, 350, 75, 300, 150, 150, 200, 225, 150, 75, 300, 100, 350, 100, 225, 350, 225, 150, 225, 200, 150)
stay <- c(0, runif(n = length(speed) - 1, min = 0, max = 25))
Vol <- c(0, 25, rnorm(n = length(speed) - 2, mean = -40, sd = 20))
Death <- sample(0:1, size = length(speed), replace = T)
fig.data <- data.frame(driver = 1:length(speed), speed = speed, stay = stay, Vol = Vol, Death = Death)
fig.data$speed <- as.factor(fig.data$speed)

# solution 1 (not work): use factor ------------------------------------------------------
# fig.data$driver <- factor(fig.data$driver, levels = order(-fig.data$stay))

# solution 2 (not work): use reorder -----------------------------------------------------
# fig.data$driver <- reorder(fig.data$driver, -fig.data$stay)

# solution 3 (work!): order the data frame -----------------------------------
# fig.data <- fig.data[order(-stay),]

I used the following code to produce ggplot

ggplot(data = fig.data, aes(x = driver, y = Vol)) +
  geom_col(aes(fill = speed), width = 0.5) +
  geom_point(data = fig.data %>% filter(Vol == 0),
             shape = 4,
             position = position_nudge(y = 0),
             size = 2,
             show.legend = FALSE) +
  geom_hline(yintercept = round(mean(Vol), 0)) +
  scale_x_discrete(limits = fig.data$driver,
                   breaks = fig.data$driver) +
  scale_y_continuous(limits = c(-100, 100), 
                     breaks = c(seq(-100,100,20), round(mean(fig.data$Vol), 0)),
                     minor_breaks = NULL) +
  coord_flip() + 
  guides(fill = FALSE)

The first driver has lowest stay, and thus this bar should be at the top of the figure. For convenience, I marked this driver using "X". However, both solution 1 and 2 gave me this

What I expect is this, (use solution 3)

Did I misunderstand the first two solutions?

Update

After reviewed the posted question, I found the reason was I used limits = fig.data$driver and breaks = fig.data$driver in scale_x_discrete. However, for solution 1, I should use limits = levels(fig.data$driver) and breaks = levels(fig.data$driver).

Some related questions:

1. Order Bars in ggplot2 bar graph

2. Reorder bars in geom_bar ggplot2

Answer 1

As suggested, I post my answer here.

After reviewed the posted question, I found the reason was I used limits = fig.data$driver and breaks = fig.data$driver in scale_x_discrete. However, for solution 1, they should be limits = levels(fig.data$driver) and breaks = levels(fig.data$driver).

Answer 2

Do not use scale_x_discrete() in the solution 1 and 2. By default, ggplot will order the axis labels according to the levels of the factor (if the axis is a factor). It will override the right order while using scale_x_discrete()

library(ggplot2)
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
speed <- c(300, 75, 100, 200, 150, 225, 75, 300, 150, 75, 200, 150, 225, 75, 350, 75, 300, 150, 150, 200, 225, 150, 75, 300, 100, 350, 100, 225, 350, 225, 150, 225, 200, 150)
stay <- c(0, runif(n = length(speed) - 1, min = 0, max = 25))
Vol <- c(0, 25, rnorm(n = length(speed) - 2, mean = -40, sd = 20))
Death <- sample(0:1, size = length(speed), replace = T)
fig.data <- data.frame(driver = 1:length(speed), speed = speed, stay = stay, Vol = Vol, Death = Death)
fig.data$speed <- as.factor(fig.data$speed)

fig.data$driver <- factor(fig.data$driver, levels = order(-fig.data$stay))

ggplot(data = fig.data, aes(x = driver, y = Vol)) +
  geom_col(aes(fill = speed), width = 0.5) +
  geom_point(data = fig.data %>% filter(Vol == 0),
             shape = 4,
             position = position_nudge(y = 0),
             size = 2,
             show.legend = FALSE) +
  geom_hline(yintercept = round(mean(Vol), 0)) +
  # scale_x_discrete(limits = fig.data$driver,
  #                 breaks = fig.data$driver) +
  scale_y_continuous(limits = c(-100, 100), 
                     breaks = c(seq(-100,100,20), round(mean(fig.data$Vol), 0)),
                     minor_breaks = NULL) +
  coord_flip() +
  guides(fill = FALSE)

^{Created on 2020-04-26 by the reprex package (v0.3.0)}