s-news
[Top] [All Lists]

Re: error message in coxph with frailty

To: <s-news@lists.biostat.wustl.edu>
Subject: Re: error message in coxph with frailty
From: Mr J Gomez <ecsfi@csv.warwick.ac.uk>
Date: Wed, 21 Nov 2001 20:48:40 +0000 (GMT)
In-reply-to: <Pine.SOL.4.30.0111131830560.17895-100000@mimosa.csv.warwick.ac.uk>
Thank you very much to Dr. Therneau for the answer to my question. It
follows:

"In your case: don't worry about the message.

  The program has nested iterations.  During the early guesses at theta,
the
warning says that it did not get complete convergence for some ancillary
parameters.  However, the solutions were good enough to advance -- by
iteration
4 "theta" is close enough to the true solution that this is not longer an
issue.  At iteration 10 for theta it considered itself to be done.

  I've recieved comments in general on this issue -- some advise that
coxph
not print out "informative" warning messages at all, but only in the case
of errors or near errors, just because they bother so many people."

On Tue, 13 Nov 2001, Mr J Gomez wrote:

> Dear all,
> I do not know how to interpret this error message. I estimate a Cox model
> with coxph and I get the following:
> > coxph(formula = Surv(ANOI, ANOF, CENSURA) ~ log(SIZE) + INTER + PROFI +
> PROXIC +
> +     PROXIE + DENSI + MGROW + DEPHA + NUMSB + CMOBC + NMARK + (NMARK^2)
> +
> +     MMKC + (MMKC^2) + frailty(NCAJA), data = datos, na.action =
> +     na.omit, eps = 0.0001, iter.max = 100, method = "efron")
> Call:
> coxph(formula = Surv(ANOI, ANOF, CENSURA) ~ log(SIZE) + INTER + PROFI +
> PROXIC
> +
>       PROXIE + DENSI + MGROW + DEPHA + NUMSB + CMOBC + NMARK + (NMARK^2)
> +
>       MMKC + (MMKC^2) + frailty(NCAJA), data = datos, na.action =
> na.omit,
>       eps = 0.0001, iter.max = 100, method = "efron")
>
>                      coef se(coef)      se2  Chisq   DF        p
>      log(SIZE)   1.398600 0.160569 0.145692  75.87  1.0 0.0e+000
>          INTER   1.922725 1.704403 1.584675   1.27  1.0 2.6e-001
>          PROFI   0.403346 0.207586 0.192813   3.78  1.0 5.2e-002
>         PROXIC   2.407926 0.205021 0.203010 137.94  1.0 0.0e+000
>         PROXIE   1.607162 0.243733 0.241092  43.48  1.0 4.3e-011
>          DENSI   0.000121 0.000558 0.000556   0.05  1.0 8.3e-001
>          MGROW  -0.001793 0.025365 0.025237   0.00  1.0 9.4e-001
>          DEPHA  -0.163160 0.480386 0.477317   0.12  1.0 7.3e-001
>          NUMSB  -0.113098 0.025728 0.025555  19.32  1.0 1.1e-005
>          CMOBC -13.069912 3.944259 3.172306  10.98  1.0 9.2e-004
>          NMARK  -0.068676 0.053151 0.048440   1.67  1.0 2.0e-001
>     I(NMARK^2)   0.001272 0.000980 0.000895   1.69  1.0 1.9e-001
>           MMKC   0.278581 0.066723 0.066406  17.43  1.0 3.0e-005
>      I(MMKC^2)  -0.012808 0.003747 0.003718  11.69  1.0 6.3e-004
> frailty(NCAJA)                               63.00 23.2 1.5e-005
>
> Iterations: 10 outer, 317 Newton-Raphson
>      Variance of random effect= 0.514   EM likelihood = -1045.7
> Degrees of freedom for terms=  0.8  0.9  0.9  1.0  1.0  1.0  1.0  1.0  1.0
> 0.6
>   0.8  0.8  1.0  1.0 23.2
> Likelihood ratio test=965  on 35.91 df, p=0  n= 34529
> Warning messages:
>   Inner loop failed to coverge for iterations 1 2 3 in: coxpenal.fit(X, Y,
> stra
> ts, offset, init = init, iter.max = iter.max, outer.max  ...
>
>
> What does this message mean? I have tried to estimate the model in
> different ways and the message disappears when I exclude some variables
> (it works if, for example, I do not include the quadratic structure in
> the NMARK variable). But I would need the full model. I have also rescaled
> some of the variables and this does not solve the problem. Does the
> message mean that I cannot take these results as valid?
>
> Thanks
> Jaime Gomez.
>
> ---------------------------------------------------------------------
> This message was distributed by s-news@lists.biostat.wustl.edu.  To
> unsubscribe send e-mail to s-news-request@lists.biostat.wustl.edu with
> the BODY of the message:  unsubscribe s-news
>


<Prev in Thread] Current Thread [Next in Thread>