如何阻止R导出SPSS时字符串变量转为因子并保留标签？

I get it—frustrating when your text responses turn into factors and you lose labels with other methods. Let's solve this using tools you're already working with, plus a more reliable alternative:

Option 1: Tweak sjlabelled::write_spss() Parameters #

Since sjlabelled::write_spss() uses the haven package under the hood, we can pass a key parameter to stop string-to-factor conversion:

# First, double-check your string variable has the correct SPSS format attribute (yours already shows "A255", so this is optional)
attr(SPSS_new$StringVariable, "format.spss") <- "A255"

# Export with string preservation enabled
sjlabelled::write_spss(
  SPSS_new,
  path = "your_final_data.sav",
  string_as_factor = FALSE  # This is the magic line!
)

This will keep your StringVariable as raw text instead of converting it to a factor, while retaining all variable labels, value labels, and SPSS format attributes you set up.

Option 2: Use haven::write_sav() (Recommended) #

The haven package is the gold standard for SPSS file handling in the tidyverse, and it handles labels and variable types more consistently. Since you're already using sjlabelled (which plays nicely with haven), this is a smooth switch:

library(haven)

# Export directly with string preservation
write_sav(
  SPSS_new,
  path = "your_final_data.sav",
  string_as_factor = FALSE
)

Verify the Result #

To make sure everything worked, re-import the file and check the structure:

check_data <- read_sav("your_final_data.sav")
str(check_data)

You'll see StringVariable stays as a character type (shown as chr (S3: haven_char) in the structure output), and all your labels (variable labels for every column, value labels for your factor/ordinal variables) are fully preserved.

Why Your Previous Methods Failed #

• sjlabelled::write_spss() defaults to string_as_factor = TRUE (inherited from haven), which converts text to factors automatically.
• foreign::write.foreign() doesn't support the label attributes added by sjlabelled, so it drops all that metadata entirely.

内容的提问来源于stack exchange，提问作者Rasul89