Greetings,
I'm currently doing a bivariate analysis of CD33 expression which I input as categorical variable of 1 (positive expression) and 0 (negative expression). Previously, I calculated the hazard ratio of this variable using STATA and the result was 0.3 (0.1-0.9) with significant P Value. When I enter this variable for bivariate analysis as covariate in SPSS, I received the same hazard ratio value. However, when I input the variable as categorical covariate in the bivariate analysis, I instead received hazard ratio above 1. This doesn't make any sense to me as the kaplan-meier curve clearly showed that CD33 positive expression was associated with higher survival. So I want to ask, should I input it as categorical covariate in SPSS or just leave it as covariate only?
Many thanks for the help.
Hello Kevin,
Just to make life interesting, whoever coded the categorical variable handling for several of the SPSS subprograms (you're using Cox regression, the same thing happens in logistic) decided it would be fun to reverse the categories for binary variables (probably something to do with the default declaration of category). So, hazard ratios and odds ratios end up being the inverse (= 1/original value) of what you'd expect if you were to input the same variable as a continuous covariate.
So you can: (a) just take the inverse of the ratio as output by SPSS; (b) re-run the analysis, entering the variable as a continuous covariate (dichotomous variables only, of course!); (c) recode the variable opposite to the way you had it originally and run the same analysis (categorical covariate); or (d) try some other software package.
Good luck with your work.
Hello Kevin,
Just to make life interesting, whoever coded the categorical variable handling for several of the SPSS subprograms (you're using Cox regression, the same thing happens in logistic) decided it would be fun to reverse the categories for binary variables (probably something to do with the default declaration of category). So, hazard ratios and odds ratios end up being the inverse (= 1/original value) of what you'd expect if you were to input the same variable as a continuous covariate.
So you can: (a) just take the inverse of the ratio as output by SPSS; (b) re-run the analysis, entering the variable as a continuous covariate (dichotomous variables only, of course!); (c) recode the variable opposite to the way you had it originally and run the same analysis (categorical covariate); or (d) try some other software package.
Good luck with your work.
David Morse
Many thanks for the reply and the amazing explanation provided. I'm really grateful for the answer you provided.
Kevin
From David Morse : SPSS folks "decided it would be fun to reverse the categories for binary variables." Yes, those folks know how to have fun!
BTW this is one reason I left SPSS then went to SAS and then to R.Even if you tell beginners that SPSS does it backwards to the rest of the world they get confused. On the rare occasions that I have to deal with SPSS in these situations, I'll get confused more often than not. I feel pretty much the same about the use of the word reference to denote the 0 value in regression for ANOVA models.. Try to teach a beginner that one as well. Another nail in the coffin which is well deserved IMO.Thanks to David Morse for a lucid and truthful explanation. Best to all, D. Booth
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS