1. Yes, you are looking a infinite coefficients. The default convergence
criteria for phreg is tighter than that for coxph() (more so than necessary,
in my opinion -- but computers are fast).
2. Within each strata (a case-control group), it suffices to have the time
value for the controls be larger than that for the case. Beyond that
constraint, you can use any values for "time" that you please, including
different choices in each strata. The documentation for coxph() says to use a
constant because
a. it was the easiest explanation to give to users
b. it rather vividly shows that "time" is in essence a dummy value.
Terry T.
Terry M. Therneau, Ph.D. (507) 284-3694
Head, Section of Biostatistics (507) 284-9542 FAX
Mayo Clinic therneau.terry@mayo.edu
Rochester, Minn 55905
|