I am doing a survival analysis with time independant and time
dependant variables. For the latter, I use the Time Program command to
test for univariate associations with the survival time. SPSS gives me
a warning for some of the variables: "A time dependent covariate is
missing in some risk sets and not missing in others. Cox Regression
cannot process such data. This command is not executed."
Since I have failures and censures, it's normal not to have complete
information for those subjects who are removed from the risk set.
For example, one of my variable is about homelessness in the 6 months
preceding the interview, 1 = yes and 0 = no. I know I can by-pass the
problem mentionned in the warning by setting some value to the missing
values. But if I set it to 0, will the model still makes sense (since
"0" is a category of th variable)? And if I set the variable to
something else (eg: 99), then SPSS creates a 3rd category to my
variable and generates 2 HR in the model.
What shall I do with that?
Thanks a lot!