I am analysing fish consumption data in a 2 factor design (predator species, 2 levels + prey species, 2 levels). The consumption data (number of prey eaten) is not normally distributed so I have been trying to use GLMs to establish whether there are differences in consumption of the prey species by the 2 fish. Initially, I used a Poisson error distribution and switched to quasi-Poisson error when I found the data to be overdispersed. However, the quasi-Poisson GLM is returning no significance for any factor, despite there being no overlap of error bars when I create a barplot of the same data (indicating that significant differences are indeed present).

I have therefore tried an ANOVA on untransformed data as some articles in the literature claim that ANOVA is robust to deviations from normality. Would love to hear some thoughts on this to help me decide on how to continue?

Thank you!

PS. All analyses are being carried out in R.

More Kyle David Green's questions See All
Similar questions and discussions