ggplot2 stat_summary 首先转换数据，在对数刻度中标记错误条

Question

这个问题是从 Cross Validated 迁移而来的，因为它可以在 Stack Overflow 上回答。 6 小时前迁移。

我有一个匹配的“前”和“后”数据集，并希望 plot 几何平均值和 SD 在对数刻度的 plot 行中（见下图）。 由于stat_summary() function 转换数据然后进行计算，所以左图中绘制的几何平均值和 SD 不正确。 几何平均 SD 在对数尺度上应该是对称的，而在 plot（左图中的“pre”组）中则不是。

我知道coord_trans()不进行计算，应该完成这项工作。 然而，对数刻度中的连接线不是直的，这对于可视化来说看起来有点奇怪。

是否有针对 plot 几何平均值和 SD 的解决方案，该方法是根据原始数据以及对数刻度的直线连接线计算得出的？

data_raw = data.frame(ID=c(1,2,3,4,5,6,7,8,9,10,11,12), 
                      Group=c(rep("before",12),rep("post",12)),
                      Values=c(15,60,70,300,40,35,100,1520,102,172,141,103,1200,130,
                               118,158,199,5804,1258,4582,4052,3332,2202,5129))

data_sorted <- data_raw %>% arrange(ID, Group)

left=ggplot(data_sorted, aes(Group,Values))+
  geom_line(aes(group = ID),colour = "gray",linetype= 2,position = position_jitter(width = 0.25, seed = 1))+
  geom_point(size = 1.2, position = position_jitter(width = 0.25, seed = 1))+
  stat_summary(fun = function(x) {exp(mean(log(x)))}, geom="crossbar")+
  stat_summary(fun = function(x) {exp(mean(log(x)))*exp(sd(log(x)))}, geom="crossbar", width=0.4, size=0.1)+
  stat_summary(fun = function(x) {exp(mean(log(x)))/exp(sd(log(x)))}, geom="crossbar", width=0.4, size=0.1)+
  scale_y_log10(breaks = trans_breaks("log10", function(x) 10^x), labels = trans_format("log10", math_format(10^.x)))+
  theme(text = element_text(size = 20))

right=ggplot(data_sorted, aes(Group,Values))+
  geom_line(aes(group = ID),colour = "gray",linetype= 2,position = position_jitter(width = 0.25, seed = 1))+
  geom_point(size = 1.2, position = position_jitter(width = 0.25, seed = 1))+
  stat_summary(fun = function(x) {exp(mean(log(x)))}, geom="crossbar")+
  stat_summary(fun = function(x) {exp(mean(log(x)))*exp(sd(log(x)))}, geom="crossbar", width=0.4, size=0.1)+
  stat_summary(fun = function(x) {exp(mean(log(x)))/exp(sd(log(x)))}, geom="crossbar", width=0.4, size=0.1)+
  coord_trans(y="log10")+
  scale_y_continuous(breaks = trans_breaks("log10", function(x) 10^x), labels = trans_format("log10", math_format(10^.x)))+
  theme(text = element_text(size = 20))

ggarrange(left,right)

只是为您指出错误，“post”组的实际几何平均值（粗横线）> 1000（右图）。 但是，它在左图中显示 <1000。

Answer 1

欧几里得空间中的几何平均值与对数空间中的算术平均值相同。 因为 y 尺度对数转换您的数据，所以您只需将exp(mean(log(x)))*exp(sd(log(x)))等函数转换为mean(x) + sd(x) (回想一下exp(log(A) + log(B)) == A * B )。

library(ggplot2)
library(ggpubr)
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
library(scales)

data_raw = data.frame(ID=c(1,2,3,4,5,6,7,8,9,10,11,12), 
                      Group=c(rep("before",12),rep("post",12)),
                      Values=c(15,60,70,300,40,35,100,1520,102,172,141,103,1200,130,
                               118,158,199,5804,1258,4582,4052,3332,2202,5129))

data_sorted <- data_raw %>% arrange(ID, Group)

left=ggplot(data_sorted, aes(Group,Values))+
  geom_line(aes(group = ID),colour = "gray",linetype= 2,position = position_jitter(width = 0.25, seed = 1))+
  geom_point(size = 1.2, position = position_jitter(width = 0.25, seed = 1))+
  stat_summary(fun = function(x) {mean(x)}, geom="crossbar")+
  stat_summary(fun = function(x) {mean(x) + sd(x)}, geom="crossbar", width=0.4, size=0.1)+
  stat_summary(fun = function(x) {mean(x) - sd(x)}, geom="crossbar", width=0.4, size=0.1)+
  scale_y_log10(breaks = trans_breaks("log10", function(x) 10^x), labels = trans_format("log10", math_format(10^.x)))+
  theme(text = element_text(size = 20))

right=ggplot(data_sorted, aes(Group,Values))+
  geom_line(aes(group = ID),colour = "gray",linetype= 2,position = position_jitter(width = 0.25, seed = 1))+
  geom_point(size = 1.2, position = position_jitter(width = 0.25, seed = 1))+
  stat_summary(fun = function(x) {exp(mean(log(x)))}, geom="crossbar")+
  stat_summary(fun = function(x) {exp(mean(log(x)))*exp(sd(log(x)))}, geom="crossbar", width=0.4, size=0.1)+
  stat_summary(fun = function(x) {exp(mean(log(x)))/exp(sd(log(x)))}, geom="crossbar", width=0.4, size=0.1)+
  coord_trans(y="log10")+
  scale_y_continuous(breaks = trans_breaks("log10", function(x) 10^x), labels = trans_format("log10", math_format(10^.x)))+
  theme(text = element_text(size = 20))

ggarrange(left,right)

^{由代表 package (v2.0.1) 于 2022 年 9 月 27 日创建}