[R-sig-eco] Question about GLM post hoc and chi square

[Martin Weiser]

Hi,
just a minor comment below.
Bob O'Hara píše v So 14. 06. 2014 v 09:45 +0200:
> On 06/14/2014 03:05 AM, Luis Fernando Garca wrote:
> > Dear all,
> >
> > I am making an analysis using a GLM using three explanatory variables and a
> > response variable. I need to obtain a table similar to this one,
> > http://postimg.org/image/5sau79wlt/r
> >
> >   nevertheless, I have not been able to do it. I am having a hard time
> > specially getting the chi square values. I would like to know how to obatin
> > them.
> Use anova(). The deviance follows a chi-squared distribution (usually - 
> if you have overdispersion it gets a bit more complicated).
> > I also would like to know what function could help me to make ad hoc
> > comparisons for single variables and interactions.
> These comparisons are called contrasts. There is a contrasts() function 
> in R, and also a contrast package (which, I'm guessing will be of more 
> use). Googling "R contrast" might help too - there seems to be plenty of 
> material, so hopefully one or two results will be exactly what you want. 
> Contrasts can get esoteric, so if you can find some a page with code 
> that gives you the comparisons you want, that should help a lot.
> 
> Good luck!
> 
> Bob
> 
> > If any of you knows how to do both estimations, I would really appreciate
> > it.
> >
> > All the best!!!
> >
> > This is my script
> > a=read.table("ricis3.txt",header=T)
> > attach(a)
> > model7=glm(Count~Sex+Time+Behaviour+Sex*Time+Sex*Behaviour+Time+Behaviour*Sex,family=poisson)
> > summary(model7)
It seem so me that your model is misspecified: if expanded and
reordered, your model would look like:
Count~Sex+Sex+Sex+Sex+Time+Time+Time+Behaviour+Behaviour+Behaviour
+Sex:Time+Sex:Behaviour+Sex:Behaviour

So: note the ":" and "*" difference, see help(formula)

Some less related tip: step-by-step work with deviance (anova) tables
and contrasts is done in Crawley: The R book.

Even more distant things, ignore if familiar with glms:
a,if residual deviance of your model >> degrees of freedom, you have
overdispersion (see summary(your.model)). Try family=quasipoisson and
testing with F distribution instead of Chi (test="F" in the anova
command) then (or glm.nb from the MASS package).
b, if numerical problems occur, try to standardise your predictors:
subtract each predictor's mean from the actual values (so
mean(new.predictor)==0 (not exactly, machine precision)), divide the
rest by the standard deviation of the original predictor (so most of the
values fall within -3,+3), see help(scale). Beware, you loose your
original scale with this. 

HTH.
Best,
Martin

> > _______________________________________________
> > R-sig-ecology mailing list
> > R-sig-ecology at r-project.org
> > https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
> 
> 
> _______________________________________________
> R-sig-ecology mailing list
> R-sig-ecology at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-ecology




-- 

------------------------------
Pokud je tento e-mail součástí obchodního jednání, Přírodovědecká fakulta 
Univerzity Karlovy v Praze:
a) si vyhrazuje právo jednání kdykoliv ukončit a to i bez uvedení důvodu,
b) stanovuje, že smlouva musí mít písemnou formu,
c) vylučuje přijetí nabídky s dodatkem či odchylkou,
d) stanovuje, že smlouva je uzavřena teprve výslovným dosažením shody na 
všech náležitostech smlouvy.