R help - Dec 2007 - problem with coxph

Hi everyone,

I encountered a problem using the coxph function for the conditional 
logistic regression. I am trying to do some simulations and I really don’t 
understand a mistake which happened maybe only 1 time among more than 
1,000 simulations.

What appeared on the screen is the following:

Error in fitter(X,Y,strats,offset,init,control,weights=weights,:
NA/NaN/Inf in a foreign function (arg 6)
Warning message:
Ran out of iterations and did not converge in: 
fiiter(X,Y,strats,offset,init,control,weights=weights,

I don’t know if the warning message could have caused the error or vice 
versa.

Does anyone have just an idea of what kind of error it can be? I don’t put 
all the code because it is very long. But I guess that this error message 
may have an "answer" whatever the code is?

Thanks so much in advance!

Willy

	[[alternative HTML version deleted]]

On Fri, 14 Dec 2007, willy.wynant.chum at ssss.gouv.qc.ca wrote:
> Hi everyone,
>
> I encountered a problem using the coxph function for the conditional
> logistic regression. I am trying to do some simulations and I really
don???t
> understand a mistake which happened maybe only 1 time among more than
> 1,000 simulations.
>
> What appeared on the screen is the following:
>
> Error in fitter(X,Y,strats,offset,init,control,weights=weights,:
> NA/NaN/Inf in a foreign function (arg 6)
> Warning message:
> Ran out of iterations and did not converge in:
> fiiter(X,Y,strats,offset,init,control,weights=weights,
>
> I don???t know if the warning message could have caused the error or vice
> versa.
>
> Does anyone have just an idea of what kind of error it can be? I don???t
put
I suspect that the data set in that run has 'issues'.  These might
include
having a near dependency among the regressors - at least when weighting as 
to form the likelihood. Perhaps, (some of) the data are separable or 
nearly so.

To get at this I would wrap the call in 'try' or one of its relatives
and
save out the data when the error is triggered to see what gives.

Another possibility is to run interaactively and set

 	options( error=recover )

to try to take things apart.

HTH,

Chuck
> all the code because it is very long. But I guess that this error message
> may have an "answer" whatever the code is?
>
> Thanks so much in advance!
>
> Willy
>
> 	[[alternative HTML version deleted]]
>
>
Charles C. Berry                            (858) 534-2098
                                             Dept of Family/Preventive Medicine
E mailto:cberry at tajo.ucsd.edu	            UC San Diego
http://famprevmed.ucsd.edu/faculty/cberry/  La Jolla, San Diego 92093-0901