[R] glm's for a logistic regression - no warnings?

[Xochitl CORMON]

<>< <>< <>< <><

Xochitl CORMON

Le 01/10/2013 17:29, Dimitri Liakhovitski a écrit :
> Thank you very much, Bert - it's very helpful.
> This post says that R issues a warning:
>
> Warning message:
> *glm.fit: fitted probabilities numerically 0 or 1 occurred
> *

Actually the warning message should be something like:
glm.fit: algorithm did not converge

The fist warning is not fatal contrary to the second one.. 
(https://stat.ethz.ch/pipermail/r-help/2012-March/307352.html)

> However, in my case there is no warning. How could I detect complete
> separation in my data? I need to be able to flag it in my function.

As said use the separation dectection function: separation.detection{brglm}

> Thank you very much!
> Dimitri
>
>
>
> On Tue, Oct 1, 2013 at 10:52 AM, Xochitl CORMON
> <Xochitl.Cormon at ifremer.fr <mailto:Xochitl.Cormon at ifremer.fr>> wrote:
>
>     Hi,
>
>     I did have warning messages about convergence issues using binomial
>     GLM with logit link with my data in the past....
>
>     Do you detect separation using the function separation.detection{brglm}?
>
>     Regards,
>
>     Xochitl C.
>
>
>     <>< <>< <>< <><
>
>     Xochitl CORMON
>     +33 (0)3 21 99 56 84 <tel:%2B33%20%280%293%2021%2099%2056%2084>
>
>     Doctorante en sciences halieutiques
>     PhD student in fishery sciences
>
>     <>< <>< <>< <><
>
>     IFREMER
>     Centre Manche Mer du Nord
>     150 quai Gambetta
>     62200 Boulogne-sur-Mer
>
>     <>< <>< <>< <><
>
>
>
>     Le 01/10/2013 16:41, Dimitri Liakhovitski a écrit :
>
>         I have this weird data set with 2 predictors and one dependent
>         variable -
>         attached.
>
>         predictor1 has all zeros except for one 1.
>         I am runnning a simple logistic regression:
>
>         temp<-read.csv("x data for reg224.csv")
>         myreg<- glm(dv~predictor1+predictor2,__data=temp,
>                        family=binomial("logit"))
>         myreg$coef2
>
>         Everything runs fine and I get the coefficients - and the fact
>         that there
>         is only one 1 on one of the predictors doesn't seem to cause any
>         problems.
>
>         However, when I run the same regression in SAS, I get warnings:
>            Model Convergence Status  Quasi-complete separation of data
>         points
>         detected.
>
>         Warning: The maximum likelihood estimate may not exist.
>         Warning: The LOGISTIC procedure continues in spite of the above
>         warning.
>         Results shown are based on the last maximum likelihood
>         iteration. Validity
>         of the model fit is questionable.
>
>         And the coefficients SAS produces are quite different from mine.
>
>         I know I'll probably get screamed at because it's not a pure R
>         question -
>         but any idea why R is not giving me any warnings in such a
>         situation?
>         Does it have no problems with ML estimation in this case?
>
>         Thanks a lot!
>
>
>
>
>
>         ________________________________________________
>         R-help at r-project.org <mailto:R-help at r-project.org> mailing list
>         https://stat.ethz.ch/mailman/__listinfo/r-help
>         <https://stat.ethz.ch/mailman/listinfo/r-help>
>         PLEASE do read the posting guide
>         http://www.R-project.org/__posting-guide.html
>         <http://www.R-project.org/posting-guide.html>
>         and provide commented, minimal, self-contained, reproducible code.
>
>
>     ________________________________________________
>     R-help at r-project.org <mailto:R-help at r-project.org> mailing list
>     https://stat.ethz.ch/mailman/__listinfo/r-help
>     <https://stat.ethz.ch/mailman/listinfo/r-help>
>     PLEASE do read the posting guide
>     http://www.R-project.org/__posting-guide.html
>     <http://www.R-project.org/posting-guide.html>
>     and provide commented, minimal, self-contained, reproducible code.
>
>
>
>
> --
> Dimitri Liakhovitski