[R] alternative to logistic regression

[roger koenker]

Perhaps I could just add that my experience with step halving in glm --
fitting models with somewhat non-standard links for binary response
[as reported in Rnews:  http://cran.r-project.org/doc/Rnews/ 
Rnews_2006-4.pdf]
was excellent.  Once some scaling issues for the links were resolved,  
we had
no problems whatsoever with the glm fitting.

url:    www.econ.uiuc.edu/~roger            Roger Koenker
email    rkoenker at uiuc.edu            Department of Economics
vox:     217-333-4558                University of Illinois
fax:       217-244-6678                Champaign, IL 61820


On Nov 16, 2007, at 10:50 AM, Prof Brian Ripley wrote:

> On Fri, 16 Nov 2007, Terry Therneau wrote:
>
>> Brian Ripley wrote
>> "Hmm ... I think you are generalizing from another R-like system."
>>
>>  Actually no, I was reading R source code that has no comments,  
>> and assumed
>> incorrectly that 1. the variable "okLinks" was the list of allowed  
>> links and
>> that 2. the error message further in the code
>>  	"... available links are 'logit', 'probit', 'cloglog',  
>> 'cauchit', 'log'
>> was accurate.
>
> It is:  those are the available unquoted names.  That error message  
> does
> not actually occur in the code: it is constructed at run time, and  
> is only
> used unquoted link names.  So if I do
>
>> identity <- NULL
>> binomial(identity)
> Error in binomial(identity) :
>    link "identity" not available for binomial family; available  
> links are
> .logit., .probit., .cloglog., .cauchit., .log.
>
> I get left and right directional quotes which my mailer does not know
> about and has substituted dots.  Those are not character strings,  
> and we
> try to make them typographically distinct (and never use single  
> quotes to
> mean a character string).
>
>
>>  Nevertheless, I should have tested my suggestion more thoroughly  
>> before I
>> posted it.  The step halving that I now see in glm.fit is a  
>> welcome addition,
>> but does not completely abrogate the fitting issues that can arise.
>
> It does mean that the comments about NA log-likelihoods were wrong.
>
>>    Terry T.
>>
>>
>>
>
> -- 
> Brian D. Ripley,                  ripley at stats.ox.ac.uk
> Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
> University of Oxford,             Tel:  +44 1865 272861 (self)
> 1 South Parks Road,                     +44 1865 272866 (PA)
> Oxford OX1 3TG, UK                Fax:  +44 1865 272595
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting- 
> guide.html
> and provide commented, minimal, self-contained, reproducible code.