Hello, I want to perform a GEE analysis for my repeatedly measured outcome. But first in order to specify what variables to include in the analysis, I was wondering if I can first perform chi-square tests (categorical DV & IV) at each timeline then the statistically significant ones to be included in the GEE model?
thank you!
Hello Nivine,
Is there some compelling reason to exclude variables from your IV set from inclusion in a model a priori? (I couldn't discern one from your query.)
If your goal is to include only variables that show bivariate relationship with the DV, then, yes, you can do this. However, if your goal is to identify the best ensemble of IVs for your DV, this approach will not necessarily accomplish that. As well, potential instances of suppression effect would go undiscovered.
Perhaps if you could elaborate a bit on what your research question is, and what variables (and sample size) you're working with, someone could likely give you a more focused recommendation.
Good luck with your work.
If you run bivariate analysis before your full model analysis you will not have the full picture of the relationship between predictors and outcome. Actually , you may exclude significant variables and also include not significant variables due to suppression and moderation effect. Currently, what I recommend is to get the opposite route and carry out a full model and then check which variables you'll keep in the model. I hope this helps
- Badges
- Science topic
- Similar topics
- Citations