Skip to content

question about chi values GLM

3 messages · Luis Fernando García, Peter Dalgaard, Brian Ripley

Original
Luis Fernando García
# 
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. I have used the function ANOVA, but it shows me the deviance but not
the Chi-Square values, can be used these values?

I also would like to know what function could help me to make ad hoc
comparisons for single variables and interactions.

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)
anova(model7,test="Chi")
-------------- next part --------------
Sex	Time	Behaviour	Count
Male	Night	Exploring	189
Male	Night	Interacting	11
Male	Night	Feeding	170
Male	Night	Mating	13
Male	Night	Resting	240
Male	Day	Exploring	58
Male	Day	Interacting	1
Male	Day	Feeding	12
Male	Day	Mating	3
Male	Day	Resting	399
Female	Night	Exploring	187
Female	Night	Interacting	10
Female	Night	Feeding	95
Female	Night	Mating	8
Female	Night	Resting	175
Female	Day	Exploring	45
Female	Day	Interacting	6
Female	Day	Feeding	10
Female	Day	Mating	4
Female	Day	Resting	406
Immature	Night	Exploring	186
Immature	Night	Interacting	15
Immature	Night	Feeding	175
Immature	Night	Resting	200
Immature	Day	Exploring	68
Immature	Day	Interacting	8
Immature	Day	Feeding	6
Peter Dalgaard
# 
The column labeled "Deviance" pretty much _is_ the chi-square, specifically the likelihood ratio test statistic, which has an asymptotic chi-square distribution. (Using test="Rao" gives you the alternative Rao efficient score test, which in your case doesn't make much of a difference.) 

Notice though, that those displays are sequential and it is not clear that the one in the image you attach is made in the same way (or in a sensible way for that matter).  In particular, you have highly significant interaction terms, in which case the main effects tests are mostly irrelevant. You may need to consult a textbook on Poisson modelling or generalized linear modelling -- the discussion is a bit too long to be fitted into a mailing list.

-pd
On 14 Jun 2014, at 10:01 , Luis Fernando Garc?a <luysgarcia at gmail.com> wrote:

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Brian Ripley
    

    

      
      #
    

  


  

      
On 14/06/2014 09:45, peter dalgaard wrote:

            

      
      
      
    

        
We do not know what is meant by the entries in those tables.  Many years 
ago (before even the term 'deviance' was coined) H.O. Lancaster and 
others partitioned chi-squared tests, with a similar table to an 
analysis of deviance but different numerical quantities.  Although 
Peter's guess is the most likely one, it is not the only possible one.

As an overall test of fit in a Poisson log-linear model there is a 
likelihood-ratio aka deviance test (sometimes called a G^2 test or 
G-test) and a Chi-squared test (which is different).

The 'df' in the table make no sense.  For three binary variables, Season 
x sex has more df than Season or Sex, and Time x Sex has fewer .... 
Even if we had the reference (and the OP really should have given us 
that to allow him to reproduce it under 'fair use' copyright law) I 
guess we would be little the wiser.