在R中构建虚拟变量后lm回归失效的问题求助

Let's walk through the likely issues causing your regression to fail and fix them step by step:

1. Incorrect Factor Creation #

Your current code for creating CNAME uses Combined.data[6], which returns a single-column data frame instead of a numeric vector. The factor() function requires an atomic vector as input, so this is a critical mistake that's probably breaking your model.

Fix: #

Use double brackets ([[ ]]) or comma indexing to extract the column as a vector, then add the factor to your finalresults dataset (since that's what your regression uses):

# Extract the 6th column as a vector and convert to factor
CNAME <- factor(Combined.data[[6]], 
                levels = 1:20, 
                labels = c("AUSTRIA", "BELGIUM", "DENMARK", "FINLAND", "FRANCE", 
                           "GERMANY", "IRELAND", "ISLE OF MAN", "ITALY", "LUXEMBOURG", 
                           "NETHERLANDS", "NORWAY", "POLAND", "PORTUGAL", "SPAIN", 
                           "SWEDEN", "SWITZERLAND", "TURKEY", "UNITED KINGDOM", "UNITED STATES"))

# Add the factor to your regression dataset
finalresults$CNAME <- CNAME

2. Duplicate Variable in Regression Formula #

Your formula includes ROA twice:

Tax_Avoidance ~ ... + ROA + MTB + ROA + ...

This introduces perfect multicollinearity, which will either trigger an error or produce unreliable coefficient estimates.

Fix: #

Remove one instance of ROA from the formula:

# Corrected regression formula
results <- lm(Tax_Avoidance ~ ENVSCORE + CGVSCORE + SOCSCORE + ECNSCORE + Size + Leverage + ROA + MTB + RND + AUD + PPE + Intang + CDP + CHS + NET + CNAME, 
              data = finalresults)
summary(results)

Additional Troubleshooting Steps #

If the regression still fails after fixing the above, check these common issues:

• Missing Values: Use colSums(is.na(finalresults)) to identify variables with missing data. lm() drops rows with NA values by default, but excessive missingness can leave too few observations to estimate the model.
• Mismatched Factor Levels: Verify that the values in Combined.data[[6]] are all within 1-20 with table(Combined.data[[6]]). Any values outside this range will become NA in CNAME, which can break the regression.
• No Variation in Factor: If all observations belong to a single country, CNAME has no variability, and lm() will throw an error. Use table(finalresults$CNAME) to confirm there are multiple countries represented.

内容的提问来源于stack exchange，提问作者VishJ