使用marginaleffects绘图函数处理多重插补数据时遇错误

使用marginaleffects包的plot_predictions()和plot_comparisons()函数处理基于多重插补数据的线性模型时，出现两类不同的错误，但同包的avg_predictions()和avg_comparisons()生成数值结果完全正常。

可复现代码 #

完整数据测试（正常运行） #

# 加载包与设置种子
library(marginaleffects)
library(ggplot2)
library(mice)
set.seed(1024)

# 完整数据模型
dat <- iris
mod <- lm(Petal.Width ~ Sepal.Length * Sepal.Width + Species, data = dat)

# 数值预测（正常）
avg_predictions(
  mod,
  variables = "Species",
  newdata = "balanced"
)

# 绘图（正常）
plot_predictions(
  mod,
  condition = "Species",
  newdata = "balanced"
)

# 数值对比（正常）
avg_comparisons(
  mod,
  variables = "Sepal.Length",
  by = "Species",
  newdata = "balanced"
)

# 绘图（正常）
plot_comparisons(
  mod,
  variables = "Sepal.Length",
  by = "Species",
  newdata = "balanced"
)

多重插补数据测试（绘图函数出错） #

# 创建含缺失值的数据
dat_imp <- iris
dat_imp$Sepal.Length[sample(seq_len(nrow(iris)), 40)] <- NA
dat_imp$Sepal.Width[sample(seq_len(nrow(iris)), 40)] <- NA
dat_imp$Species[sample(seq_len(nrow(iris)), 40)] <- NA

# 多重插补与模型拟合
dat_mice <- mice(dat_imp, m = 20, printFlag = FALSE, .Random.seed = 1024)
mod_mice <- with(dat_mice, lm(Petal.Width ~ Sepal.Length * Sepal.Width + Species))

# 数值预测（正常）
avg_predictions(
  mod_mice,
  variables = "Species",
  newdata = "balanced"
)

# 绘图函数出错
plot_predictions(
  mod_mice,
  condition = "Species",
  newdata = "balanced"
)

错误信息：

Error: Entries in the `condition` argument must be element of: 

# 数值对比（正常）
avg_comparisons(
  mod_mice,
  variables = "Sepal.Length",
  by = "Species",
  newdata = "balanced"
)

# 绘图函数出错
plot_comparisons(
  mod_mice,
  variables = "Sepal.Length",
  by = "Species",
  newdata = "balanced"
)

错误信息：

Error in subset.default(modeldata, select = cl) : 
  argument "subset" is missing, with no default

尝试添加subset参数后的新错误：

plot_comparisons(
  mod_mice,
  variables = "Sepal.Length",
  by = "Species",
  newdata = subset(dat_imp, Species == "setosa"),
)

错误信息：

Error in 1 - conf_level : non-numeric argument to binary operator

会话信息 #

> sessionInfo()
R version 4.4.3 (2025-02-28 ucrt)
Platform: x86_64-w64-mingw32/x64
Running under: Windows 11 x64 (build 26100)

other attached packages:
[1] mice_3.18.0            ggplot2_3.5.2          marginaleffects_0.28.0

解决思路 #

1. 手动生成绘图数据（推荐）
   既然avg_predictions()和avg_comparisons()能正常输出数值结果，可直接用这些结果结合ggplot2手动绘图，绕开plot_*函数的兼容性问题：

   • 预测图示例：

     pred_df <- avg_predictions(
       mod_mice,
       variables = "Species",
       newdata = "balanced"
     )
     ggplot(pred_df, aes(x = Species, y = Estimate)) +
       geom_col(fill = "steelblue") +
       geom_errorbar(aes(ymin = `2.5 %`, ymax = `97.5 %`), width = 0.2) +
       labs(title = "Predicted Petal.Width by Species", y = "Estimate") +
       theme_minimal()
     

   • 对比图示例：

     comp_df <- avg_comparisons(
       mod_mice,
       variables = "Sepal.Length",
       by = "Species",
       newdata = "balanced"
     )
     ggplot(comp_df, aes(x = Species, y = Estimate)) +
       geom_point(size = 3) +
       geom_errorbar(aes(ymin = `2.5 %`, ymax = `97.5 %`), width = 0.2) +
       geom_hline(yintercept = 0, linetype = "dashed", color = "red") +
       labs(title = "Comparison of Sepal.Length Effect by Species", y = "Estimate") +
       theme_minimal()
     

2. 更新marginaleffects包
   当前使用的marginaleffects_0.28.0可能存在针对mice插补模型的绘图函数兼容性bug，尝试更新到最新版本：

   install.packages("marginaleffects")
   

3. 手动构建newdata
   避免使用newdata = "balanced"字符串参数，手动创建平衡数据框传给newdata，确保函数能正确解析变量：

   # 手动创建平衡的Species数据
   newdat <- expand.grid(
     Species = unique(dat_imp$Species[!is.na(dat_imp$Species)]),
     Sepal.Length = mean(dat_imp$Sepal.Length, na.rm = TRUE),
     Sepal.Width = mean(dat_imp$Sepal.Width, na.rm = TRUE)
   )
   # 尝试绘图
   plot_predictions(mod_mice, condition = "Species", newdata = newdat)
   

内容的提问来源于stack exchange，提问作者EdP