[S] parentheses in interaction terms - details

["Isabelle Robichaud" <eir8049@umoncton.ca>]

 hello all,

Yesterday I posted a message saying I obtained  different results whether I put parentheses or not in  glm interaction terms.  Here are my outputs (terrdens  is a continuous variable, year an ordered factor with  five levels):

> intercept <- glm(resid.deviance~1)
> txy <- glm(resid.deviance~terrdens:year)

> anova(txy, intercept, test="F")
Analysis of Deviance Table

Response: resid.deviance

Terms     Resid.Df Resid.Dev  Df
1 terrdens:year    64     12.4338                        2             1    69     13.8501   -5  

 Deviance F Value    Pr(F) 
-1.41632  1.45804  0.215999

> txy2 <- glm(resid.deviance~(terrdens):(year))
> anova(txy2, intercept, test="F")

Terms      Resid.Df Resid.Dev  Df 
1 (terrdens):(year)   60    10.4744                      2                 1   69    13.8501    -9 

Deviance  F Value     Pr(F) 
-3.37571  2.14855  0.0387868

> txy3 <- glm(resid.deviance~terrdens*year)
> anova(txy3, intercept, test="F")

Terms      Resid.Df Resid.Dev  Df 
1 terrdens * year  60     10.4744                
2               1  69     13.8501    -9

Deviance  F Value    Pr(F) 
-3.37571  2.14855  0.0387868

As you can see, (terrdens):(year) is not equal to  terrdens:year, but rather to terrdens*year.

When I use anova() with only one model at a time, the  terms added sequentially (first to last) are:
for terrdens:year...NULL, terrdens:year  (as expected)
for (terrdens):(year)...NULL, year, terrdens:year (?)
for terrdens*year...NULL, terrdens, year, terrdens:year  (as expected)

It is the second model that puzzles me.  Any  suggestions?  Thank you very much.

Isabelle Robichaud

Departement de biologie
Universite de Moncton
Moncton NB Canada E1A 3E9
e-mail: eir8049@umoncton.ca    
----------------------------------------------------------------------- This message was distributed by s-news@wubios.wustl.edu.  To unsubscribe send e-mail to s-news-request@wubios.wustl.edu with the BODY of the message:  unsubscribe s-news