R语言用ifelse基于多列生成新变量结果异常如何解决？

背景 #

我尝试基于tu_aftermatthew数据集的三个字段tu_aftermatthew$FLD_ZONE、tu_aftermatthew$SFHA_TF、tu_aftermatthew$ZONE_SUBTY生成新的风险分级变量，各字段交叉统计结果如下：

> with(tu_aftermatthew,table(FLD_ZONE, SFHA_TF, useNA = "always"))
                   SFHA_TF
FLD_ZONE                   F        T     <NA>
  A                        0     2494        0
  AE                       0   286245        0
  AH                       0      621        0
  AO                       0      669        0
  AREA NOT INCLUDED     8257        0        0
  VE                       0     5710        0
  X                 12626557        0        0
  <NA>                     0        0  7241489
> with(tu_aftermatthew,table(ZONE_SUBTY, SFHA_TF, useNA = "always"))
                                                       SFHA_TF
ZONE_SUBTY                                                     F        T     <NA>
  0.2 PCT ANNUAL CHANCE FLOOD HAZARD                      159698        0        0
  1 PCT ANNUAL CHANCE FLOOD HAZARD CONTAINED IN CHANNEL        0       13        0
  1 PCT CONTAINED IN STRUCTURE, COMMUNITY ENCROACHMENT         8        0        0
  1 PCT CONTAINED IN STRUCTURE, FLOODWAY                       6        0        0
  1 PCT FUTURE CONDITIONS                                   4192        0        0
  1 PCT FUTURE CONDITIONS CONTAINED IN STRUCTURE               1        0        0
  AREA OF MINIMAL FLOOD HAZARD                          12459618        0        0
  AREA WITH REDUCED FLOOD RISK DUE TO LEVEE                 3034        0        0
  COMMUNITY ENCROACHMENT AREA                                  0     1007        0
  FLOODWAY                                                     0    17059        0
  STATE ENCROACHMENT AREA                                      0      294        0
  <NA>                                                      8257   277366  7241489

新变量FLOOD_RISK生成规则 #

• 当SFHA_TF取值为"T"且FLD_ZONE不为"VE"时，赋值为"SFHA_INLAND"，总计数应为290029
• 当FLD_ZONE取值为"VE"时，赋值为"SFHA_COASTAL"，总计数应为5710
• 当ZONE_SUBTY取值为"0.2 PCT ANNUAL CHANCE FLOOD HAZARD"时，赋值为"500_FLOOD"，该类记录与SFHA_TF=="T"的记录无重叠，不会和前两类冲突

预期输出结果 #

> table(tu_aftermatthew$FLOOD_RISK) 

  500_FLOOD SFHA_INLAND   SFHA_COASTAL  
     159698      90029           5710         

问题现象 #

我使用ifelse语句实现上述逻辑时结果不符合预期，代码如下：

> tu_aftermatthew$FLOOD_RISK <- ifelse(tu_aftermatthew$SFHA_TF == "T", "SFHA_INLAND", NA)
> tu_aftermatthew$FLOOD_RISK <- ifelse(tu_aftermatthew$FLD_ZONE  == "VE", "SFHA_COASTAL", tu_aftermatthew$FLOOD_RISK)
> tu_aftermatthew$FLOOD_RISK <- ifelse(tu_aftermatthew$ZONE_SUBTY  == "0.2 PCT ANNUAL CHANCE FLOOD HAZARD", "500_FLOOD", tu_aftermatthew$FLOOD_RISK)
> 
> table(tu_aftermatthew$FLOOD_RISK, useNA = "always" ) 

  500_FLOOD SFHA_INLAND        <NA> 
     159698       18373    19993971 

拆分步骤验证发现，第三个ifelse语句似乎覆盖了前一步生成的SFHA_INLAND、SFHA_COASTAL赋值，但按照数据特征，0.2%概率洪泛区的记录和SFHA_TF为T的记录没有重叠，不该出现该问题。我构造了如下测试数据复现逻辑，运行却完全正常，目前无法定位问题原因：

FLD<-c("A","A","V","X","X")
SFHA<-c("T","T","T","F","F")
DESC<-c("1","1","1","0", ".2")
df<-data.frame(FLD, SFHA, DESC)
df

df$FLOOD_RISK <- ifelse(df$SFHA == "T", "SFHA_INLAND", NA)
df$FLOOD_RISK <- ifelse(df$FLD  == "V", "SFHA_COASTAL", df$FLOOD_RISK)
df$FLOOD_RISK <- ifelse(df$DESC  == ".2", "500_FLOOD", df$FLOOD_RISK)
df

  FLD SFHA DESC   FLOOD_RISK
1   A    T    1  SFHA_INLAND
2   A    T    1  SFHA_INLAND
3   V    T    1 SFHA_COASTAL
4   X    F    0         <NA>
5   X    F   .2    500_FLOOD

问题根因 #

问题核心是R中ifelse函数对NA判断的处理逻辑：当判断条件的计算结果为NA时，ifelse会直接返回NA，而非你期望的保留原有值。
你的原数据中存在大量ZONE_SUBTY为NA的记录：对于已经被赋值为SFHA_INLAND、SFHA_COASTAL的记录，在第三个ifelse判断时，ZONE_SUBTY == "0.2 PCT ANNUAL CHANCE FLOOD HAZARD"的结果为NA，因此这些记录会被错误覆盖为NA。你构造的测试数据中没有NA值，因此不会复现该问题。

修复方案 #

将第三个ifelse的判断条件补充NA校验即可，修改后代码如下：

tu_aftermatthew$FLOOD_RISK <- ifelse(tu_aftermatthew$SFHA_TF == "T", "SFHA_INLAND", NA)
tu_aftermatthew$FLOOD_RISK <- ifelse(tu_aftermatthew$FLD_ZONE  == "VE", "SFHA_COASTAL", tu_aftermatthew$FLOOD_RISK)
# 补充NA判断，只有明确满足条件时才赋值，否则保留原有值
tu_aftermatthew$FLOOD_RISK <- ifelse(!is.na(tu_aftermatthew$ZONE_SUBTY) & tu_aftermatthew$ZONE_SUBTY  == "0.2 PCT ANNUAL CHANCE FLOOD HAZARD", "500_FLOOD", tu_aftermatthew$FLOOD_RISK)

也可以使用更易读的dplyr::case_when实现，避免多层ifelse嵌套的逻辑问题：

library(dplyr)
tu_aftermatthew <- tu_aftermatthew %>%
  mutate(FLOOD_RISK = case_when(
    FLD_ZONE == "VE" ~ "SFHA_COASTAL",
    SFHA_TF == "T" ~ "SFHA_INLAND",
    ZONE_SUBTY == "0.2 PCT ANNUAL CHANCE FLOOD HAZARD" ~ "500_FLOOD",
    TRUE ~ NA_character_
  ))

内容的提问来源于stack exchange，提问作者tchoup