R语言如何最优可视化二分类变量的回归关联结果

问题描述 #

我所用数据集的所有变量（包含因变量、自变量）均为分类变量，仅存在1、2两种分类取值，示例数据集结构如下：

lp1=structure(list(a = c(2L, 2L, 1L, 2L, 2L, 1L, 2L, 2L, 1L, 2L, 
2L, 1L, 2L, 1L, 2L, 2L, 2L, 1L), b = c(1L, 1L, 1L, 2L, 2L, 2L, 
1L, 1L, 2L, 1L, 2L, 1L, 1L, 1L, 2L, 2L, 1L, 1L), c = c(1L, 1L, 
1L, 2L, 1L, 1L, 1L, 1L, 2L, 1L, 1L, 1L, 1L, 1L, 2L, 2L, 1L, 1L
), d = c(2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, NA, 2L, 2L, 
2L, 2L, 2L, 2L, 2L), e = c(1L, 2L, 1L, 2L, 1L, 2L, 2L, 2L, 2L, 
1L, NA, 2L, 2L, 2L, 2L, 1L, 2L, 1L), f = c(2L, 2L, 2L, 2L, 1L, 
1L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 1L, 2L, 1L, 2L, 1L), g = c(NA, 
1L, 2L, NA, 2L, 2L, 1L, 2L, 1L, NA, NA, NA, NA, NA, 1L, NA, NA, 
NA), h = c(2L, 2L, 2L, 2L, 2L, 2L, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 
2L, 2L, 2L, 2L, 2L), i = c(1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 1L, 
1L, 2L, 1L, 1L, 1L, 1L, 2L, 2L, 2L), j = c(2L, 2L, 2L, 2L, 2L, 
2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L, 2L), k = c(2L, 
1L, 1L, 2L, 2L, 1L, 1L, 1L, 2L, 1L, NA, 1L, 2L, 1L, 1L, 1L, 1L, 
1L), l = c(1L, 1L, 1L, 1L, 1L, 1L, 1L, 2L, 1L, 2L, NA, 2L, NA, 
1L, 1L, 1L, 1L, 2L), m = c(1L, 2L, 1L, 2L, 1L, 2L, 2L, 2L, 2L, 
1L, 1L, 1L, 2L, NA, 2L, 2L, 2L, 1L), n = c(1L, 2L, 2L, 2L, 1L, 
2L, 2L, 2L, 2L, NA, 1L, 1L, 2L, 2L, 2L, 2L, 2L, 1L), xxx = c(2L, 
1L, 1L, 1L, 2L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 1L, 1L, 1L, 1L, 1L, 
2L)), class = "data.frame", row.names = c(NA, -18L))

我尝试绘制两个变量的关联图，其中xxx为因变量，a为自变量，使用的ggplot2绘图代码如下：

library(ggplot2)

ggplot(lp1, aes(xxx, a)) +
  geom_point() +
  theme_minimal()

由于两个变量仅存在1、2两个取值，生成的散点图所有样本点完全重叠，信息量极低、展示效果差。请问是否存在更优的方案来可视化两个分类变量间的关联关系？是否需要对数据做归一化类的预处理？

解答 #

不需要做归一化这类预处理。二分类变量关联可视化的核心是展示每个交叉分类下的样本量/占比，普通散点图把所有重叠的点压在同一个坐标位置，自然没有参考价值，几个实用的实现方案如下：

• 扰动散点图
  给点增加微小的随机位置偏移，把重叠的点错开，就能直观看到每个分类组合的样本密度，改一行代码就能实现：

  # 提前把分类变量转成因子，避免被ggplot识别为连续数值
  lp1 <- lp1 |> 
    dplyr::mutate(across(c(a, xxx), as.factor))
  
  ggplot(lp1, aes(x = xxx, y = a)) +
    geom_jitter(width = 0.2, height = 0.2, alpha = 0.7, size = 2) +
    theme_minimal()
  

  参数width、height控制扰动幅度，不要设太大避免点偏离原本的分类位置，搭配半透明参数alpha可以更清晰体现点的重叠程度。

• 堆叠比例条形图
  直接展示自变量不同取值下，因变量的分布占比，是看分类变量关联最直观的形式：

  ggplot(lp1, aes(x = as.factor(a), fill = as.factor(xxx))) +
    geom_bar(position = "fill") +
    labs(x = "a", y = "占比", fill = "xxx取值") +
    theme_minimal()
  

• 马赛克图
  用矩形面积对应每个交叉分组的样本量，是分类变量关联分析的经典可视化方式，可以用ggmosaic包适配ggplot语法：

  # install.packages("ggmosaic")
  library(ggmosaic)
  ggplot(lp1) +
    geom_mosaic(aes(x = product(xxx, a), fill = as.factor(xxx))) +
    labs(x = "a", y = "xxx", fill = "xxx取值") +
    theme_minimal()
  

• 计数热图
  先统计每个分类组合的样本数，用颜色映射数量，搭配数值标注，信息展示最直接：

  library(dplyr)
  count_data <- lp1 |> 
    filter(!is.na(a), !is.na(xxx)) |> 
    count(a, xxx)
  
  ggplot(count_data, aes(x = as.factor(a), y = as.factor(xxx), fill = n)) +
    geom_tile() +
    geom_text(aes(label = n), color = "white", size = 5) +
    labs(x = "a", y = "xxx", fill = "样本数") +
    theme_minimal()
  

注意：所有分类变量传入ggplot前最好先转成因子类型，能避免很多坐标轴刻度、统计变换的异常问题，归一化、标准化这类针对连续变量的预处理操作，对二分类变量的可视化完全没有必要。

内容的提问来源于stack exchange，提问作者quest