When comparing the two approaches, the first thing that we need to note is that as we have already seen from the box plots which we have observed in data presentation, the assumption of constant variance which is required for classical GLM is violated. Therefore, using a classical GLM is not the right approach with this dataset. First a comparison was performed empirically as we know that we have several model assumptions broken in question 2. So we have concluded that there is no need to compare the classical glm with our final Generalized linear model by numerical measures.
Between the generalized linear models, we select the final Gamma model over the final Poisson model due to a better AIC score and also due to lower residual deviances. When comparing Poisson and Gamma model, we can see on the figure below(The Poisson model is represented by blue circles and Gamma by orange squares), that coefficients for both models have roughly same sign and therefore the effects of the explanatory variables will be in same „direction“ for both Poisson and Gamma model.
