Thank you @jessegrabowski and @BjornHartmann for your answers.
I was currently evaluating the results at n_experiments = 1000 and given jessegrabowski’s answer above I felt confident that everything was working as planned. I think that I previously had to high expectations on how fast p and n should zero in on the true values.
However, following BjornHartmann’s suggestion to continue increasing n_experiments to 10000, I also get some weird results. Here is the traceplot (I ran this with chains=10):
It seems that each chain is homing in on a different solution giving weird results. Here are the overall plots of p,n:
p
n
So it seems that somewhere in the interval n_experiments=1000 - 10000, the model starts to deviate from the true values.
Following @junpenglao presentations on likelihood functions i have plotted the following likelihood function using the trace of p and n
Am i somehow “messing” up the likelihood function because of the large number of observed and getting the gradient to large ?
I am feeling rather confused at the moment, anyone has an explanation for this ?
best regards