R help - Aug 2005 - Exponential, Weibull and log-logistic distributions in glm()

Dear R-users!

I would like to fit exponential, Weibull and log-logistic via glm() like
functions. Does anyone know a way to do this? Bellow is a bit longer 
description of my problem.

Hm, could family() be adjusted/improved/added to allow for these distributions?
SAS procedure GENMOD alows to specify deviance and variance functions to
help in such cases. I have not tried that option and I do not know how
does it work. I do not expect that others will do the job, although it 
would not harm ;)

Thanks in advance!

I fully understand that all mentioned distributiones are available through
"survival" functions in various packages but I don't have data
that
correspond fully to survival data. I have data on some cell parameters,
which show cell vitality. These data are highly skewed and I have
therefore modeled them via log-normal and gamma in glm(). Based
on deviance residuals, gamma seems to fit better. By the way, are there any
other means of chosing the right disribution for GLM?

However, some references from field of this data suggested that Weibull might
be better and I would like to try this, as well as with exponential and 
log-logistic. Of course, author in that paper did not say how they performed
the analyses, but I guess they took each group (several group were compared) 
separately and estimated parameters for distribution, say Weibull, via maximum 
likelhood. I do not like this approach, since not all data are used in one run, 
and would like to use model to get parameter estimates and perform inference.
With Weibull I am aware that it does not fit in exponential family, unless one
is read to specify a value/estimate for one parameter.

Any comment are welcome!

Lep pozdrav / With regards,
    Gregor Gorjanc

----------------------------------------------------------------------
University of Ljubljana
Biotechnical Faculty        URI: http://www.bfro.uni-lj.si/MR/ggorjan
Zootechnical Department     mail: gregor.gorjanc <at> bfro.uni-lj.si
Groblje 3                   tel: +386 (0)1 72 17 861
SI-1230 Domzale             fax: +386 (0)1 72 17 888
Slovenia, Europe
----------------------------------------------------------------------
"One must learn by doing the thing; for though you think you know it,
 you have no certainty until you try." Sophocles ~ 450 B.C.

On Wed, 10 Aug 2005, Gorjanc Gregor wrote:
> Dear R-users!
>
> I would like to fit exponential, Weibull and log-logistic via glm() like
> functions. Does anyone know a way to do this? Bellow is a bit longer
> description of my problem.
I think you want to use survreg().  It will still work when there is no 
censoring.


Adding these families to glm() would be difficult. They are really not 
generalized linear models in any of the useful senses: not exponential 
families, don't have estimating functions linear in the response 
variable, not fitted by iteratively reweighted least squares.


 	-thomas

>> Dear R-users!
>>
>> I would like to fit exponential, Weibull and log-logistic via glm()
like
>> functions. Does anyone know a way to do this? Bellow is a bit longer
>> description of my problem.
>
> I think you want to use survreg().  It will still work when there is no 
> censoring.
>
>
> Adding these families to glm() would be difficult. They are really not 
> generalized linear models in any of the useful senses: not exponential 
> families, don't have estimating functions linear in the response 
> variable, not fitted by iteratively reweighted least squares.
Thank you very much for the response. I didn't want to get into survival
since this is really not my field, but it seems that providing those 
distributions in glm() is not so easy as I thought. Based on your hint I 
tried with survreg and estimates seems reasonable.

best, Gregor