为什么在我将 NA 分配给缺失值后，我的因子中的空级别会被制表？

Question

我有一个数据框 df，其中有一列 foo 包含类型因子的数据：

df <- data.frame("bar" = c(1:4), "foo" = c("M", "F", "F", "M"))

当我使用 str(df$foo) 检查结构时，我得到了：

因子 w/ 3 级 "","F",..: 2 2 2 2 2 2 2 2 2 2 ..

为什么在我的数据中只有 2 个级别时它会报告 3 个级别？

---

编辑：

似乎有一个缺失值""，我通过分配它NA 来清理它。 当我调用table(df$foo) 时，它似乎仍在计算“缺失值”级别，但没有发现任何情况：

  F M
0 2 2

但是，当我打电话给df$foo 时，我发现它只报告了两个级别：

Levels:  F M

table 怎么可能仍然计算空级别，我该如何解决这种行为？

Answer 1

检查您的数据框是否确实没有缺失值，因为看起来确实如此。试试这个：

# works because factor-levels are integers, internally; "" seems to be level 1
which(as.integer(df$MF) == 1)

# works if your missing value is just ""
which(df$MF == "") 

然后，您应该清理数据框以正确反映缺失值。 factor 将处理 NA：

df <- data.frame("rest" = c(1:5), "sex" = c("M", "F", "F", "M", ""))
df$sex[which(as.integer(df$sex) == 1)] <- NA

一旦你清理了你的数据，你将不得不删除未使用的级别以避免像table这样的列表计算空级别的出现。

观察这一系列步骤及其输出：

# Build a dataframe to reproduce your behaviour
> df <- data.frame("Restaurant" = c(1:5), "MF" = c("M", "F", "F", "M", ""))
# notice the empty level "" for the missing value
> levels(df$MF)
[1] ""  "F" "M"

# notice how a tabulation counts the empty level;
# this is the first column with a 1 (it has no label because
# there is no label, it is "")
> table(df$MF)

  F M 
1 2 2

# find the culprit and change it to NA
> df$MF[which(as.integer(df$MF) == 1)] <- as.factor(NA)

# AHA! So despite us changing the value, the original factor
# was not updated! I wonder what happens if we tabulate the column...
> levels(df$MF)
[1] ""  "F" "M"

# Indeed, the empty level is present in the factor, but there are
# no occurences!
> table(df$MF)

  F M 
0 2 2 

# droplevels to the rescue:
# it is used to drop unused levels from a factor or, more commonly,
# from factors in a data frame.
> df$MF <- droplevels(df$MF)

# factors fixed
> levels(df$MF)
[1] "F" "M"

# tabulation fixed
> table(df$MF)

F M 
2 2