Hello
I am conducting a linear regression analysis for categorical variables using dummy coding in SPSS. I am then cross-checking the results in R. However, I have encountered a discrepancy where one of the dummy coding results is inconsistent with R, showing completely different outputs although the other dummy variables are consistent between the two software programs.
I am trying to identify the possible issues that could be causing this discrepancy, but I cannot able to find a solution thus far. Can someone help me with this problem?
I tried this at home (ClassNew has 4 categories):
glm(BRS_Total ~ as.factor(ClassNew), data=m1, family=gaussian)
versus
GENLIN BRS_Total BY ClassNew (ORDER=DESCENDING)
/MODEL ClassNew INTERCEPT=YES DISTRIBUTION=NORMAL LINK=IDENTITY
which resulted in (almost) identical estimates after selecting the correct reference category in R and SPSS.
Thank you. Yes, I also got all the same results (nearly) for other categorical variables but was troubled with only that one. I did the same dataset but I copied the computed categorical variable from SPSS into the R working file, then I got different results. Do you think that is the problem? Now I completely export my SPSS dataset into R, then I got the same results.
If the original variable had 4 categories, then you need 3 dummy variables to represent that. Is that how you did it?
No, Rana Hamza Shakil a three-category variable requires only two dummies -- just as a two-category variable requires only one dummy. The regression intercept will represent the score for the "omitted" category (i.e., 0 on all the other dummies) and each of the remaining dummies will give the difference between that variable and the omitted category.
Entering three dummies for a three-category variables calls for a degree of freedom that you don't have. Neither SPSS nor R can handle this, but I do not know how they report this error.
- Badges
- Science topic