I am working on a regression task where the response is count data… due to several factors i have to scale the data such that the response is divided by an factor thus leading to an continous response.
Although it should be fine? i am having a hard time interpreting this posterior predictive plot and it seems quite off:
can someone give me an hint as from where the bumps in the observed comes from… i cant find an reasonable explanation for this looking at e.g histograms of the response.
And secondly, is my approach of scaling the count data fine or should i tweak that in some sort of way?
