指定分组独立刻度后factor()无法维持数据顺序的问题

问题根源 #

ggnewscale的new_scale_fill()会创建独立的填充刻度上下文，后续图层如果没有明确指定y轴变量的因子水平顺序，ggplot会默认把字符型变量转为因子并按字母排序，直接覆盖你之前用factor()设置的原始分组顺序。

修复方案 #

1. 全局预处理设定因子顺序 #

在绘图前统一把所有数据集里的y轴分组变量转为指定顺序的因子，这样所有图层都会继承这个顺序，不用在每个图层重复设置：

# 假设你的y轴分组变量是sample_group，从原始数据中提取要保留的顺序
target_levels <- unique(your_main_data$sample_group)

# 给所有用到的数据集统一设置因子水平
df1$sample_group <- factor(df1$sample_group, levels = target_levels)
df2$sample_group <- factor(df2$sample_group, levels = target_levels)
df3$sample_group <- factor(df3$sample_group, levels = target_levels)

2. 每个图层明确指定因子顺序 #

如果不想修改原始数据，就在每个使用y轴变量的图层里，显式指定因子的levels参数，确保新图层不会重置排序：

# 先提取要保留的原始顺序
target_levels <- unique(your_main_data$sample_group)

ggplot() +
  # 第一个图层
  geom_col(data = df1, aes(x = value, y = factor(sample_group, levels = target_levels), fill = group)) +
  new_scale_fill() +
  # 第二个图层，必须重复指定y轴的因子顺序
  geom_col(data = df2, aes(x = value + 25, y = factor(sample_group, levels = target_levels), fill = group)) +
  new_scale_fill() +
  # 第三个图层同理
  geom_col(data = df3, aes(x = value + 50, y = factor(sample_group, levels = target_levels), fill = group))

代码示例对比 #

出错的原始代码（排序混乱） #

library(ggplot2)
library(ggnewscale)

# 模拟数据
df1 <- data.frame(sample = c("Cluster2", "Cluster1", "Cluster3"), val = c(12, 18, 15), type = "TypeA")
df2 <- data.frame(sample = c("Cluster2", "Cluster1", "Cluster3"), val = c(8, 10, 7), type = "TypeB")
df3 <- data.frame(sample = c("Cluster2", "Cluster1", "Cluster3"), val = c(5, 6, 4), type = "TypeC")

# 单刻度时正常，y轴保留Cluster2/Cluster1/Cluster3顺序
ggplot(df1, aes(x = val, y = factor(sample, levels = unique(df1$sample)))) +
  geom_col(aes(fill = type))

# 加ggnewscale后y轴变成字母排序（Cluster1/Cluster2/Cluster3）
ggplot() +
  geom_col(data = df1, aes(x = val, y = factor(sample, levels = unique(df1$sample)), fill = type)) +
  new_scale_fill() +
  geom_col(data = df2, aes(x = val + 20, y = sample, fill = type)) +
  new_scale_fill() +
  geom_col(data = df3, aes(x = val + 40, y = sample, fill = type))

修正后的代码（保留原始顺序） #

library(ggplot2)
library(ggnewscale)

df1 <- data.frame(sample = c("Cluster2", "Cluster1", "Cluster3"), val = c(12, 18, 15), type = "TypeA")
df2 <- data.frame(sample = c("Cluster2", "Cluster1", "Cluster3"), val = c(8, 10, 7), type = "TypeB")
df3 <- data.frame(sample = c("Cluster2", "Cluster1", "Cluster3"), val = c(5, 6, 4), type = "TypeC")

# 全局设置因子顺序
target_levels <- unique(df1$sample)
df1$sample <- factor(df1$sample, levels = target_levels)
df2$sample <- factor(df2$sample, levels = target_levels)
df3$sample <- factor(df3$sample, levels = target_levels)

# 绘图，所有图层自动继承因子顺序
ggplot() +
  geom_col(data = df1, aes(x = val, y = sample, fill = type)) +
  new_scale_fill() +
  geom_col(data = df2, aes(x = val + 20, y = sample, fill = type)) +
  new_scale_fill() +
  geom_col(data = df3, aes(x = val + 40, y = sample, fill = type))

核心注意点 #

• 只要后续图层的y轴变量是未转换的字符型，ggplot就会自动按字母排序，必须确保所有图层的y轴变量都是带有固定levels的因子。
• 如果多个数据集的分组顺序不一致，需要手动指定统一的levels列表，比如target_levels <- c("Cluster2", "Cluster1", "Cluster3")。

内容的提问来源于stack exchange，提问作者Jen1984