R help - Apr 2008 - GLM fitting in R and Statistica

Hi,
I have a problem concerning discrepances between R (which I use) and 
Statistica (which uses my supervisor). I can't say what is the origin 
of these differences but unfortunately my supervisor doesn't know that 
either.

Our response variable is number (or presence/absence) of parasites in 
rodents and explanatory variables are presence/absence of several 
alleles. The rodents were sampled in three sites and the sites differ 
in parasite frequency so we decided to include "site" as a factor.

The problem concerns calculations of factor variable. In Statistica 
output there is only one term, just "site", and in R there are two 
contrats "site A", "site B". I realized that I can obtain
similar
results in Statistica clicking on "Estimate" button instead of
"1LR"
(what does my supervisor which gives only log-likelihood and p of 
variables but doesn't estimate parameter).

But there is still one problem I can't explain. When we fit 
interaction terms (site x allele1, site x allele2 and so on) the 
results are completely different. I tried several different contrasts 
in R, such as contr.SAS, contr.treatment etc but I couldn't get 
nothing similar to Statistica output.

Have anybody any idea how to deal with that? Or how to explain why R 
results are different (and hopefully better)? I tried to argue that I 
did everything according to Crawley's R Book so probably the models 
are constructed correctly but my supervisor wasn't convinced...

Looking forward any suggestions,

Agnieszka

-- 
Agnieszka Kloch
Instytute of Environmental Sciences, Jagielonian University
ul. Gronostajowa 7, 30-387 Krakow, Poland, tel. (12) 664 51 51

> I have a problem concerning discrepances between R (which I use) and
> Statistica (which uses my supervisor).
		     ---------------

can this be submitted as a fortune?

any other programs around that use supervisors, it's always good to  
have a couple of
those around

ingmar
	[[alternative HTML version deleted]]

Can you get fitted values out of Statistica?  If the 2 models are
equivalent, just using different encodings of the categorical variables,
then the fitted values will be the same (within roundoff error) even
though the coefficient estimates may differ.

So as a first step you should compare the fitted values or predicted
values for new points and see if those are close or not.  

Also, make sure that Statistica is treating site as a categorical
variable (does it show 2 degrees of freedom?), if it is seeing it as a
linear variable rather than a categorical, then it would show 1 line and
everything would be messed up.

Hope this helps, If not, can you send code/output from both using a
sample dataset?

-- 
Gregory (Greg) L. Snow Ph.D.
Statistical Data Center
Intermountain Healthcare
greg.snow at imail.org
(801) 408-8111
 
 
> -----Original Message-----
> From: r-help-bounces at r-project.org 
> [mailto:r-help-bounces at r-project.org] On Behalf Of Agnieszka Kloch
> Sent: Wednesday, April 09, 2008 12:09 PM
> To: R help
> Subject: [R] GLM fitting in R and Statistica
> 
> Hi,
> I have a problem concerning discrepances between R (which I 
> use) and Statistica (which uses my supervisor). I can't say 
> what is the origin of these differences but unfortunately my 
> supervisor doesn't know that either.
> 
> Our response variable is number (or presence/absence) of 
> parasites in rodents and explanatory variables are 
> presence/absence of several alleles. The rodents were sampled 
> in three sites and the sites differ in parasite frequency so 
> we decided to include "site" as a factor.
> 
> The problem concerns calculations of factor variable. In 
> Statistica output there is only one term, just "site", and in 
> R there are two contrats "site A", "site B". I realized
that
> I can obtain similar results in Statistica clicking on 
> "Estimate" button instead of "1LR" 
> (what does my supervisor which gives only log-likelihood and 
> p of variables but doesn't estimate parameter).
> 
> But there is still one problem I can't explain. When we fit 
> interaction terms (site x allele1, site x allele2 and so on) 
> the results are completely different. I tried several 
> different contrasts in R, such as contr.SAS, contr.treatment 
> etc but I couldn't get nothing similar to Statistica output.
> 
> Have anybody any idea how to deal with that? Or how to 
> explain why R results are different (and hopefully better)? I 
> tried to argue that I did everything according to Crawley's R 
> Book so probably the models are constructed correctly but my 
> supervisor wasn't convinced...
> 
> Looking forward any suggestions,
> 
> Agnieszka
> 
> --
> Agnieszka Kloch
> Instytute of Environmental Sciences, Jagielonian University 
> ul. Gronostajowa 7, 30-387 Krakow, Poland, tel. (12) 664 51 51
> 
> ______________________________________________
> 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.
>