--- begin included message ---
Hello R users,
I am analizing survival data (mostly uncensored) and want to extract the
most out of it.
Since I have more than one factor, I?ve read that the survival regression
can help to test the interactions between factors, and then decide how to do
the comparisons using the Log-rank test (survdiff).
1- if I chose the Weibull distribution, does the output inform the goodness
of fit to it? perhaps in this part of the output...
Weibull distribution
Loglik(model)= -1302.8 Loglik(intercept only)= -1311
Chisq= 16.49 on 11 degrees of freedom, p= 0.12
Number of Newton-Raphson Iterations: 7
n= 873
2- one of my factors is "gender" (2 levels). With survreg, it appears
as
significant, but if I compare them with log-rank it turns not significant.
Are they comparing different things? or is it a test power issue?
------- end inclusion -------
1. To understand goodness of fit you need to look at the residuals in
multiple ways. (The same answer applies to ordinary linear regression.)
2. You have not given us enough information to answer the questions. If
the data is p=.049 vs p=.051, the the answers are in agreement even
though the artificial label of "significant" changes. The logrank
test
and survreg are not the same model. If the data is p=.02 vs p=.8, then
you have an error in the code.
Terry Therneau