R:通过将多个值与一个值进行比较来创建瀑布图

Question

主要问题:您能否找到一种更简洁的方法来将多个观测值与长向表中的单个值进行比较？

目标是创建一个瀑布图，显示政党(“A”)的投票并将其与竞争对手进行比较。

输入是一个列表 (tibble)，包含三列:neighbourhood、party 和 pct_votes。示例:

prop.long
# A tibble: 304 x 3
   neighbourhood    party       pct_votes
   <fct>            <chr>        <dbl>
 1 Region-A         A           0.0938
 2 Region-A         B           0.0672
 3 Region-A         C           0.0906
 4 Region-A         D           0.228
 5 Region-A         E           0.0318 
 6 Region-B         A           0.0932
 7 Region-B         B           0.118
 8 Region-B         C           0.0837
 9 Region-B         D           0.199
10 Region-B         E           0.0544

要显示 A 方比竞争对手做得更好还是更差，需要使用 direction 属性。 y 变量将用于显示条形。如果一方表现比 A 差，则应将 y_min 设置为该方的值，否则，应将 A 方的 pct_votes 值视为 y_min 。如果表现较差，则应将 y_max 设置为 A 方的 pct_votes 值，否则采用其自己的 pct_votes 值。 x 变量用于在图中以特定顺序并排显示政党。

这是所需的输出:

prop.wf
# A tibble: 76 x 10
# Groups:   neighbourhood [19]
   neighbourhood    party   pct_votes   tmp     direction   y_min y_max x_min x_max
   <fct>            <fct>   <dbl>       <dbl>   <chr>       <int> <int> <int> <int>
 1 Region-A         A       0.0938      0.0938  target      0     9     0     1
 2 Region-A         B       0.0672      0.0938  lower       6     9     1     2
 3 Region-A         C       0.0906      0.0938  lower       9     9     3     4
 4 Region-A         D       0.228       0.0938  higher      9    22     4     5
 5 Region-B         A       0.0932      0.0932  target      0     9     0     1
 6 Region-B         B       0.118       0.0932  higher      9    11     1     2
 7 Region-B         C       0.0837      0.0932  lower       8     9     3     4
 8 Region-B         D       0.199       0.0932  higher      9    19     4     5
# … with 68 more rows

根据需要生成输出的代码:

prop.wf <- prop.long %>%
  filter(party %in% c('A', 'B', 'C', 'D')) %>%
  group_by(neighbourhood) %>%
  mutate(tmp = pct_votes[party == 'A']) %>%
  mutate(party = factor(party, levels = c('A', 'B', 'C', 'D')),
    direction = ifelse(party == 'A', 'target', ifelse(pct_votes > tmp, 'higher', ifelse(pct_votes < tmp, 'lower', 'equal'))),
    y_min = as.integer((ifelse(party == 'A', 0, ifelse(direction == 'lower', pct_votes, tmp)) * 100)),
    y_max = as.integer((ifelse(party == 'A', pct_votes, ifelse(direction == 'lower', tmp, pct_votes)) * 100)),
    x_min = as.integer(ifelse(party == 'A', 0, ifelse(party == 'B', 1, ifelse(party == 'C', 2, ifelse(party == 'D', 3, 4))))),
    x_max = as.integer(ifelse(party == 'A', 1, ifelse(party == 'B', 2, ifelse(party == 'C', 3, ifelse(party == 'D', 4, 5)))))) # `x_min + 1` did not yield int, even after casting with `as.integer()

我的主要问题是:你能帮我将其重构为更清晰/可扩展的代码吗？ (例如，如果需要添加两方怎么办？最好这些 ifelse() 语句不被链接。)我无法停止思考:“应该有一种更简单的方法来制定这个”，但我想不出来。

我的最终输出示例(使用ggplot的geom_rect):

利用资源:

• https://www.r-bloggers.com/basic-waterfall-graphs-in-r/
• Convert Dataframe to make Waterfall Chart in ggplot2
• Calculate difference between values in consecutive rows by group

Answer 1

通过执行 geom_col 而不是 geom_rect ，您可以节省大量代码。这涉及欺骗 y 轴，但这意味着包括绘图在内的整个代码如下所示:

library(dplyr)
library(ggplot2)

df %>% 
  group_by(neighbourhood) %>%
  mutate(pct = ifelse(party == "A", -pct_votes, 
                      pct_votes - pct_votes[party == "A"]),
         fill = ifelse(party == "A", "#fd9826", 
                       c("#3eca3f", "", "#ca1f15")[sign(pct) + 2])) %>%
  ggplot(aes(party, pct, fill = fill)) +
  geom_col(color = "gray50", width = 1) +
  facet_grid(~neighbourhood) +
  scale_y_continuous(limits = c(min(-df$pct_votes[df$party == "A"]), 0.15),
                     breaks = seq(min(-df$pct_votes[df$party == "A"]), 0.2,  0.05),
                     labels = scales::percent(0:5 / 20)) +
  scale_fill_identity()

^{由reprex package于2020年8月9日创建(v0.3.0)}