[R] ANCOVA/glm missing/ignored interaction combinations

[Mark Difford]

Hi Lara,

>> And I cant for the life of me work out why category one (semio1) is being
>> ignored, missing 
>> etc.

Nothing is being ignored Lara --- but you are ignoring the fact that your
factors have been coded using the default contrasts in R, viz so-called
treatment or Dunnett contrasts. That is, your intercept is semio1 and the
other coefficients are calculated off it, hence the name treatment
contrasts.

In general this type of coding is recommended for GLM models (but this can
raise deep divisions, as some don't consider them to be proper contrasts
because they are not orthogonal). Perhaps you should read up on how dummy
variables are coded.

To really understand this you will have to move away from ?contrasts as a
source. It might also help you to do the following on a simpler model:

?dummy.coef
dummy.coef(saved.glm/lm.model)

This shows how it is worked out.

HTH, Mark.


lara harrup (IAH-P) wrote:
> 
> Hi
> 
> I am using R version 2.7.2. on a windows XP OS and have a question
> concerning an analysis of covariance with count data I am trying to do,
> I will give details of a scaled down version of the analysis (as I have
> more covariates and need to take account of over-dispersion etc etc) but
> as I am sure it is only a simple problem but I just can't see how to fix
> it.
> 
> I have a data set with count data as the response (total) and two
> continuous covariates and one categorical explanatory variable (semio).
> When I run the following lines, after loading the data and assigning
> 'semio' as a factor, taking into account that I want to consider two way
> interactions:
> 
>> model.a<-glm(total~(temperature+humidity+semio)^2,family=poisson)
>> summary(model.a)
> 
> I get the output below. But not all interactions are listed there are 4
> semio categories, 1,2,3 and 4 but only 2,3,and 4 are listed in the
> summary (semio2,semio3,semio4). And I cant for the life of me work out
> why category one (semio1) is being ignored, missing etc.
> 
> Any help or suggestions would be most appreciated. Thanks in advance
> 
> Lara
> lara.harrup at bbsrc.ac.uk
> 
> Call:
> glm(formula = total ~ (temperature + humidity + semio)^2, family =
> poisson)
> 
> Deviance Residuals: 
>     Min       1Q   Median       3Q      Max  
> -22.212   -5.132   -2.484    3.200   18.793  
> 
> Coefficients:
>                       Estimate Std. Error z value Pr(>|z|)    
> (Intercept)          23.848754   2.621291   9.098  < 2e-16 ***
> temperature          -1.038284   0.150465  -6.901 5.18e-12 ***
> humidity             -0.264912   0.033928  -7.808 5.81e-15 ***
> semio2               22.852664   1.291806  17.690  < 2e-16 ***
> semio3                3.699901   1.349007   2.743   0.0061 ** 
> semio4               -1.851163   1.585997  -1.167   0.2431    
> temperature:humidity  0.012983   0.001983   6.545 5.94e-11 ***
> temperature:semio2   -0.870430   0.037602 -23.149  < 2e-16 ***
> temperature:semio3   -0.060846   0.038677  -1.573   0.1157    
> temperature:semio4    0.055288   0.046137   1.198   0.2308    
> humidity:semio2      -0.114718   0.013369  -8.581  < 2e-16 ***
> humidity:semio3      -0.031692   0.013794  -2.298   0.0216 *  
> humidity:semio4       0.008425   0.016020   0.526   0.5990    
> ---
> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 
> 
> (Dispersion parameter for poisson family taken to be 1)
> 
>     Null deviance: 10423.5  on 47  degrees of freedom Residual deviance:
> 2902.2  on 35  degrees of freedom
> AIC: 3086.4
> 
> Number of Fisher Scoring iterations: 7
> 
> ______________________________________________
> 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.
> 
> 

-- 
View this message in context: http://www.nabble.com/ANCOVA-glm-missing-ignored-interaction-combinations-tp19285259p19286859.html
Sent from the R help mailing list archive at Nabble.com.