RStudio中如何将exerany2为No的exerof1变量NA值转为0

Hey there! Great question—let's break this down for you, starting with whether your approach makes sense, then getting the code right.

首先：你的分析思路完全合理（甚至是最优的！） #

Your logic here is spot-on. The NAs in exerof1 for respondents who answered "No" to exerany2 aren't random missing data—they directly mean those people didn't exercise at all in the past 30 days. Converting these NAs to 0 is the most contextually accurate way to handle them, way better than:

• Deleting the rows (which would lose 30% of your data and weaken statistical power)
• Filling with mean/median (which would invent fake exercise counts and introduce bias)

Just one key note: don't touch the NAs in exerof1 for respondents who chose "拒绝回答" in exerany2—those are true unknowns, and you should leave them as NA or handle them separately (like excluding those rows from analysis).

实现方法：两种简单的R代码方案 #

Let's fix that ifelse issue you ran into, plus offer a more readable alternative for future work.

基础R（直接用ifelse）

This is straightforward and works without extra packages:

# Replace NA with 0 ONLY where exerany2 is "No"
df$exerof1 <- ifelse(
  condition = df$exerany2 == "No" & is.na(df$exerof1),
  yes = 0,
  no = df$exerof1
)

Quick check first: Run table(df$exerany2) to make sure the "No" label matches exactly (no typos like "NO" or " no").

Tidyverse/dplyr方法（更直观，适合新手维护）

If you use the tidyverse, case_when is easier to read than nested ifelse statements, especially if you add more conditions later:

library(dplyr)

df <- df %>%
  mutate(
    exerof1 = case_when(
      # Target condition: No exercise + NA count → set to 0
      exerany2 == "No" & is.na(exerof1) ~ 0,
      # Keep all other values (including NA for "拒绝回答")
      TRUE ~ exerof1
    )
  )

验证处理是否正确 #

Always double-check to make sure you didn't miss anything:

# This should return 0 if all target NAs were converted
sum(df$exerany2 == "No" & is.na(df$exerof1))

# Cross-tab to confirm "No" responses now have 0 for exerof1
table(df$exerany2, df$exerof1, useNA = "always")

后续分析小建议 #

For your correlation analysis between exercise and health:

• Since genhlth is a 5-level factor, consider using an ordered logistic regression (via MASS::polr) or non-parametric test (like Kruskal-Wallis) instead of linear regression—these methods are better suited for ordinal health outcomes.
• Decide upfront how to handle the "拒绝回答" group in exerany2: you can exclude them, or add them as a separate category in your models to control for non-response bias.

内容的提问来源于stack exchange，提问作者Rafael Rodrigues Santana