How can I stratify individual categories for a variable when doing Cox models to see if a particular category within the variable is a significant predictor in SPSS? For example, the variable can be race, and I would like to see the influence of each (Asian, Caucasian, etc.) on survival. Would I need to separate each race into a different variable/column, code it as a binary value, and choose one to be a reference? Would I need to do this for all the possible combinations though? Thank you!
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
This is an issue in all regressions, not just Cox. With a multi-level categorical predictor, you can't ask whether an individual level is a "significant" predictor on its own. You can examine the contribution of the entire categorical predictor (whether any levels are associated with differences in outcome), and you can examine differences among specific levels with respect to associations with outcome.
Would I need to separate each race into a different variable/column, code it as a binary value, and choose one to be a reference?
That's how the standard "treatment" or "dummy" coding of a categorical predictor works "under the hood." You don't have to do it yourself, and you don't have to re-run the model using each level individually as a reference, as the information about the differences between levels and their significance is present in the model's regression coefficients and the coefficient covariance matrix. I don't use SPSS, but it almost certainly has tools for doing pairwise comparisons among levels of a categorical predictor once the model is fit. It might go by a name something like "estimated marginal means." In R the emmeans package is a widely used tools for such post-model analysis.
