HR (95% CL) is different depending on 3 groups versus 2 groups for same reference group in %NEWSURV

[svh · Accepted solution]

Isn't the hazard ratio different because the models themselves are different? The reference group is the same, but the variable has 3 levels in one analysis and 2 in another. If a group of participants is omitted from one of the models, there will be a new estimates of the model parameters.