Sorry for the late reply,
at the moment I am testing my stan model by generating artificial data that is made with the same likelihood function as in stan and find them with my stan model. So currently it is possible to determine if stan found the right values by comparing the generated parameters with the guessed.
The weird behavior I am facing at the moment is that by increasing the number of chains it can happen that my effective sample sizes drops dramatically and I don’t understand why this happens. As I understand it at the moment is that every chain works independently and contributes effective samples to the model based on autocorrelation of the parameters within a chain.
So if I have 4 chains (which all computing an accurate posterior) and get an n_eff value of 4000 and then I increase the amount of chains that run to 7 I suddenly get an n_eff value of 2 or 4.
This makes me wonder if there is some sort of cross validation between the chains that could ultimately cause this.
In case my model could help you can find a detailed description in this post here. The model is still pretty much the same except for applying the changes suggested in the post.
Thank you for your help!