I am currently trying to calculate a one-sample logrank-test (Gail & Ware
1979, Woolson 1981) for a test of equality between the survivor function of
a sample and that of the corresponding population.
I have searched the stata-list archives and the User Group Meetings and
came up with 2 results - which are both not satisfactory:
1.) I found a presentation by Pere�Joan Ventura on a User Group Meeting in
Spain. There is mentioned a Do-File onesam.do for this purpose, which I
cannot find anywhere in the net.
2.) I found a reply to an earlier request: Enzo Coviello mentions
equivalence to a test for intercept=0 in a poisson model with offset term
log(n_expected) (based on articles by Berry (1983) and a technical report
by Therneau and Offord (1999)).
This is more tricky: I tried it and the resulting p-values are plausible.
However one thing about them contradicts my intuition of this test:
As an approximation I can calculate a normal (two-sample) logrank-test for
equality of the survivor function of my sample and of the population. This
approximation should - in my opinion - overestimate the p-value (i.e.
underestimate the value of the test-statistic), because it treats the
population survivor function as a sample survivor function and thus
introduces uncertainty that does not exist. So this approximation should be
more conservative than the one-sample test.
However, when I follow Enzo Voviellos advice, I get p-values that are a
little bit larger than that of the approximation. I find this disturbing as
I think it should be the other way around. So I do not really trust the
results.
Therefore I would like to know, if anyone can point me to a
stata-implementation of this test.
