[R] glm

Spencer Graves spencer.graves at pdf.com
Wed Jun 1 18:26:18 CEST 2005


	  I would not use gamma or gaussian for a proportion.  If the 
percentages were success or failure rate out of a known number of 
trials, then I'd try function "glm" with family = binomial or 
quasibinomial.  However, "percent of plant root length colonized" does 
not sound to me like "glm" would be appropriate.  The beta distribution 
seems to me to be the most plausibly distribution to use.

	  I therefore did a search for "beta regression using 
"www.r-project.org" -> search -> "R site search".  This revealed a 
package "betareg" available from CRAN.  Have you considered that?  If 
no, I think I'd try it.

	  spencer graves

Mahdi Osman wrote:
> Hi, list,
> 
> I have got a dataset on soil and plant. I would like to fit a "glm" to my
> data. My response variable is percentage data. That is percent of plant root
> length colonized by Arbuscular micorrhiza fungi. Because of the nature of my
> data, I am not quite sure whether gamma or gaussian distribution is suitable
> for this type of data. If I use gamma distribution, which link function is
> appropriate?
> 
> I appreciate your suggestions very much.
> 
> Thanks in advance!
> 
> 
> Mahdi
> 
>




More information about the R-help mailing list