[R] number of iterations in a Tobit model
Terry Therneau
therneau at mayo.edu
Fri Jan 21 18:57:36 CET 2011
-- begin inclusion ---
I'm running a Tobit model but convergence can not be reached within 30
iterations. Is there anyway I can change the max number of iterations?
Thanks.
--- end inclusion ---
"Tobit" is simply a linear model with censored data. You don't say how
you are fitting this, but I'll assume you are using survreg
fit <- survreg(Surv(time, status) ~ x1 + x2, dist='gaussian',..)
with appropriate additional arguments to the Surv function if the data
is left or interval censored.
If survreg doesn't converge in 30 iterations it likely won't converge
in 100 or more. The Newton-Raphson algoritm has gotton lost. Data sets
with a very large fraction of censored observations can be numerically
challenging. help(survreg.control) will tell you all the necessary
details however.
Over the years I have accumulated a few data sets that were very
difficult maximizations for survreg, and led to further tuning of the
underlying algorithm. Yours would be the first new one in a while; if
you are willing to share it that would help me track down the issue.
You likely will need to use specific starting estimates.
If you are not using survreg, try it. Perhaps it is already robust
enough for your data.
Terry Therneau
More information about the R-help
mailing list