I'm working on a survival analysis project and, based on previous work of
my colleagues on similar projects, am trying to estimate an AFT model with
a generalized gamma distribution. When I execute the following code:
>flexsurvreg(timesurv ~ cycles + rcycle + cyctime, dist="gengamma")
I get the following error message:
Error in optim(optpars, minusloglik.flexsurv, t = Y[, "time"], dead =
Y[,
:
non-finite finite-difference value [4]
I get an almost identical error (a 3 in the brackets instead of a 4) if I
use the dist = "weibull". But have no problems when I run survreg
with
Weibull distribution on the same data.
I've seen some messages about similar errors for other functions, but
don't
think the responses apply (at least not in a what I understand). I tried
setting initial values based on the survreg output values and that didn't
help. Also, I checked and none of the variables are particularly
well-correlated.
Any idea how to deal with this?
Thanks,
--
Brett Close
[[alternative HTML version deleted]]