I have a dataset with say about 50 variables related to lifestyle risk factors, and I want to measure the association between these lifestyle predictors (e.g. smoking, alcohol consumption, physical activity etc) and a dependent variable, say high cholesterol (three levels: Normal; Borderline High; High) using multinomial logistic regression.
My questions are:
Should I run the model for each variable separately and interpret the result (significance of association, relative risk ratios etc), or it is better to use all (or at least a group of) the variables?
If likelihood ratio chi-square test shows that the model does not fit significantly better than an empty model, how does it affect the interpretation of the result?