Dear all,
thanks to Jim I was referred to the SAS-L list, from which I got two
responses to my question, both mentioning the same paper: Callaert H.
Comparing statistical software packages: The case of the logrank test in
StatXact. American Statistician 2003 Aug; 57(3): 214-217.
The paper perfectly addresses my question. Briefly, the difference is due to
differences in the test statistic used when ties are apparent (StatXact uses
different scores than the most commonly used logrank scores). For details
and a possible implementation of an exact logrank test with the "standard"
scores using StatXact, refer to the above mentioned paper.
Thanks again to the three respondents from the two lists!
Best regards, Michael
-----Ursprüngliche Nachricht-----
Von: A UK-based worldwide e-mail broadcast system mailing list
[mailto:[log in to unmask]] Im Auftrag von Michael Meyners
Gesendet: Donnerstag, 3. Februar 2005 11:55
An: [log in to unmask]
Betreff: QUERY: Logrank scores -- SAS vs PROC StatXact
Dear all,
I want to compare the (partially censored) survival times of two groups
using the logrank test. To my surprise, I get different asymptotic p values
using PROC LIFETEST from SAS or PROC TWOSAMPL from PROC StatXact. It seems
as if the scores used are different between the two procedures, as the rank
statistics slightly differ. Furthermore, LIFETEST obviously uses a
Chi-Squared approximation, while TWOSAMPL rather uses a normal
approximation. I wouldn't expect this to make much difference, though.
Did anybody come across this before and can offer any explanation about the
computational differences? Which scoring is to be preferred? I am willing to
use the exact version of the test only if the scores used are reasonable.
Of coures I'll post a summary of responses.
Thanks in advance for any suggestions,
Michael
*****************************************************
Dr. Michael Meyners
Boehringer Ingelheim Pharma GmbH & Co. KG
Medical Data Services
Birkendorfer Strasse 65
88397 Biberach an der Riss
Germany
Tel.: +49 (0)7351 54 4219
Fax: +49 (0)7351 83 4219
E-Mail: [log in to unmask]
*****************************************************
|