如何在R的分组ggplot图上添加配对t检验的p值

示例数据 #

问题背景 #

我拥有大量上述格式的fish_ID数据，需要针对每种颜色组，对inst_0和inst_3的time均值做配对t检验。已用以下代码绘制分组箱线图：

library(ggplot2)
library(ggprism)

ggplot(mydata , aes(x = factor(inst), y = time, fill = color)) +
 geom_boxplot(position = position_dodge(width = 0.75)) +
 scale_fill_manual(values = c("blue" = "blue", "green" = "green", "red" = "red", "yellow" = 
 "yellow")) +
 theme_prism()

现在需要将各颜色组的配对t检验p值添加到图中，不清楚如何针对每个颜色组完成inst_0与inst_3的两两比较。

实现方案 #

步骤1：计算各颜色组的配对t检验p值 #

先将数据转换为宽格式，确保同一fish_ID+color组合下同时存在inst_0和inst_3的观测，再按颜色分组执行配对t检验：

library(dplyr)
library(tidyr)

# 转换为宽格式，保留配对数据
wide_data <- mydata %>%
  pivot_wider(names_from = inst, values_from = time) %>%
  na.omit()  # 移除缺失配对的样本

# 按颜色分组计算配对t检验p值，并格式化标签
p_values <- wide_data %>%
  group_by(color) %>%
  summarize(
    p_val = t.test(inst_0, inst_3, paired = TRUE)$p.value,
    # 生成可读性强的p值标签，支持星号或数值显示
    p_label = case_when(
      p_val < 0.001 ~ "***",
      p_val < 0.01 ~ "**",
      p_val < 0.05 ~ "*",
      TRUE ~ "ns"
      # 若需要显示具体数值，替换为：sprintf("p = %.3f", p_val)
    )
  )

步骤2：将p值添加到箱线图中 #

推荐使用ggsignif包手动指定比较位置，或用ggpubr包自动适配分组布局：

方法1：用ggsignif手动控制标记位置

install.packages("ggsignif")
library(ggsignif)

# 定义每个颜色组的配对比较位置（适配position_dodge(width=0.75)的布局）
color_x_pos <- c(blue = 0.8, green = 0.95, red = 1.1, yellow = 1.25)
y_offset <- max(mydata$time) + c(10, 15, 20, 25)  # 避免标记与箱线图重叠

ggplot(mydata, aes(x = factor(inst), y = time, fill = color)) +
  geom_boxplot(position = position_dodge(width = 0.75)) +
  scale_fill_manual(values = c("blue" = "blue", "green" = "green", "red" = "red", "yellow" = "yellow")) +
  theme_prism() +
  # 添加配对t检验p值标记
  geom_signif(
    data = p_values,
    aes(
      xmin = color_x_pos, 
      xmax = color_x_pos + 1, 
      annotations = p_label, 
      y_position = y_offset
    ),
    manual = TRUE,
    tip_length = 0.01,
    textsize = 4
  )

方法2：用ggpubr自动适配分组

install.packages("ggpubr")
library(ggpubr)

ggplot(mydata, aes(x = factor(inst), y = time, fill = color)) +
  geom_boxplot(position = position_dodge(width = 0.75)) +
  scale_fill_manual(values = c("blue" = "blue", "green" = "green", "red" = "red", "yellow" = "yellow")) +
  theme_prism() +
  stat_compare_means(
    aes(group = color),
    method = "t.test",
    paired = TRUE,
    label = "p.signif",  # 显示星号，若要数值用"p.format"
    position = position_dodge(width = 0.75),
    hide.ns = TRUE  # 可选：隐藏无显著性的标记
  )

注意事项 #

• 确保数据为配对设计：每个fish_ID在同一color下同时有inst_0和inst_3的观测，否则配对t检验不适用。
• 可根据图的高度调整y_offset参数，避免p值标记与箱线图或异常值重叠。
• 若需要校正多重比较，可在计算p值时添加p.adjust()函数，比如p_val = p.adjust(t.test(inst_0, inst_3, paired = TRUE)$p.value, method = "bonferroni")。

内容的提问来源于stack exchange，提问作者Balina