R help - May 2006 - quasi glm start values

Hi,
I'm tying to fit a glm using quasi with variance=mu(1-mu) (to try and  
get "well behaved" residuals). I get an error message:
"cannot find valid starting values, please specify some".
I have tried using start= and specifying coefficients from a previous  
fit (with different error structure) but get the same message.
What would be valid starting values?

I am using a 64 bit build R on a MacG5.

Thanks



Lucy Crooks
Theoretical Biology
Universitätstrasse 16
ETH Zentrum, CHN 75.1
CH-8092 Zurich

Tel.: 0041 44 632 8326
Fax: 0041 44 632 1271

http://www.eco.ethz.ch/


	[[alternative HTML version deleted]]

On Thu, 4 May 2006, Lucy Crooks wrote:
> Hi,
> I'm tying to fit a glm using quasi with variance=mu(1-mu) (to try and
> get "well behaved" residuals). I get an error message:
> "cannot find valid starting values, please specify some".
> I have tried using start= and specifying coefficients from a previous
> fit (with different error structure) but get the same message.
> What would be valid starting values?
What link are you using?  You need 0 < mu < 1 (the actual function used is
                validmu <- function(mu) all(mu>0) && all(mu<1)
plus any restriction from the unspecified link).
> I am using a 64 bit build R on a MacG5.
The version of R is more important here.

-- 
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