[R] What is the convergence criterion for binomial logit in glm?

[David Winsemius]

On Jan 22, 2013, at 3:59 PM, Dimitri Liakhovitski wrote:

> I already looked. This help file for "loglin (http://127.0.0.1:12583/library/stats/html/loglin.html) says:
> "The Iterative Proportional Fitting algorithm as presented in Haberman (1972) is used for fitting the model. At most iter iterations are performed, convergence is taken to occur when the maximum deviation between observed and fitted margins is less than eps." And the default eps is 0.1
>  
> So, is it then the convergence criterion used by glm when family=binomial("logit")?
> I just need to know for sure.

I assumed that you would follow the link on help(glm) to `glm.control` where the convergence criteria is described and can be altered. The link to that help page is at the end of the line that reads:

"control	
a list of parameters for controlling the fitting process. For glm.fit this is passed to glm.control."

The default epsilon is NOT 0.1

-- 
David.
>  
> Thanks for confirming!
> Dimitri
> 
> 
>  
> On Tue, Jan 22, 2013 at 6:37 PM, David Winsemius <dwinsemius at comcast.net> wrote:
> 
> On Jan 22, 2013, at 2:55 PM, Dimitri Liakhovitski wrote:
> 
> > Dear R-ers,
> >
> > I am running logistics regression using package "glm": glm(myDV ~ .,
> > data=mydata, family=binomial("logit"))
> >
> > I have a general question: in "glm" (binary logit) - what convergence
> > criterion is being used?
> 
> You should look at the help page for `glm` (and follow the obvious links.)
> >
> > --
> 
> David Winsemius
> Alameda, CA, USA
> 
> 
> 
> 
> -- 
> Dimitri Liakhovitski
> gfk.com

David Winsemius
Alameda, CA, USA