Post-hoc test after using general linear model (GLM)

Question

I want to see if there's a significant difference between 3 groups (klasse) and their agressive behaviour (by counting the number of agressive interactions). The situation is not relevant, but it must be taken into account. I compared the agressive interactions between the groups with a general linear model and I used the family poisson for the count data.

This is the data input in Rstudio:

data.frame(agressiekrab=c(8,5,1,10,6,12,4,17,1,1,5,9,11,2,3,0,21,17,4,1,10,4,14,15,22,8,19,0,6,16,4,10), klasse=c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,3,3,3,3,3,3,3,3,3,3), situatie=c(1,1,1,1,1,1,2,2,2,2,2,3,3,3,3,3,1,1,1,2,2,2,1,1,1,1,1,2,2,2,2,2))

structure(list(agressiekrab = c(8, 5, 1, 10, 6, 12, 4, 17, 1, 
1, 5, 9, 11, 2, 3, 0, 21, 17, 4, 1, 10, 4, 14, 15, 22, 8, 19, 
0, 6, 16, 4, 10), klasse = c(1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 
1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3
), situatie = c(1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 3, 3, 3, 3, 
3, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2)), class = "data.frame", row.names = c(NA, 
-32L))

It's non-parametric data, so not normally distributed This is the code that I used in Rstudio:

agmodel1<-glm(agressiekrab~klasse*situatie,poisson) 
summary(agmodel1)

The result is this:

Call:
glm(formula = agressiekrab ~ klasse * situatie, family = poisson)

Deviance Residuals: 
   Min      1Q  Median      3Q     Max  
-3.688  -1.623  -0.462   1.306   3.718  

Coefficients:
                Estimate Std. Error z value Pr(>|z|)    
(Intercept)       1.5198     0.3812   3.987 6.68e-05 ***
klasse            0.7223     0.1827   3.953 7.71e-05 ***
situatie          0.0905     0.2112   0.428  0.66836    
klasse:situatie  -0.3251     0.1169  -2.781  0.00542 ** 
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for poisson family taken to be 1)

    Null deviance: 171.19  on 31  degrees of freedom
Residual deviance: 120.51  on 28  degrees of freedom
AIC: 240.09

Number of Fisher Scoring iterations: 5

Now i would like to use a post-hoc test to see where the difference is between de groups (klasse). I tried the pairwise wilcox test, but it doesn't give me two-by-two comparisons. I don't know how to interpret the outcome. I also tried the TukeyHSD test, but that didn't work either (and I don't think it's right because of the non-parametric data)

pairwise.wilcox.test(agressiekrab,klasse*situatie,P.adj="Bonj")
   Pairwise comparisons using Wilcoxon rank sum test 
   data:  agressiekrab and klasse * situatie £
      1 2 3 4
    2 1 - - -  
    3 1 1 - -
    4 1 1 1 -
    6 1 1 1 1 
  P value adjustment method: holm

The tukey method gives an error:

TukeyHSD(aov(agressiekrab~klasse*situatie))    
  Error in TukeyHSD.aov(aov(agressiekrab ~ klasse * situatie)) : 
  no factors in the fitted model

Answer 1

TukeyHSD(aov(agressiekrab~factor(klasse)*factor(situatie)))