Dear Adrian,
I’ve also tested the “horseshoe”-Variant now. So far I do not see any signs of divergence (edit: still true after lot of sampling), so maybe it’s the way to go.
EDIT: I missed on first glance that there seems to be no model parameter measuring the change due to altered condition on population level. It only returns “altered_subject”, but no “altered”. Hence the model seems to be hierarchical with regard to intercept, but not slope. Did I miss something?
EDIT: the following only applies for some of the model parameters. distributions are not wide in general. However, the general question remains.
However, since I increased prior SD again, the posteriors are very wide. This might have to do with the limited size of my data set.
General question:
Is there any way to estimate if my data is actually sufficient for a certain model complexity? Something in line with the thinking of Boettiger et al. (2012)? I often have the feeling that, though Hierarchical models are an excellent tool, they might be more demanding than plain hypothesis testing in terms of sample size on each level. This is rarely discussed and it would be great to quantify.
Regarding Junpeng’s suggestion on the difference between v3.0 and v3.1: is there currently a way in master to ignore the funnel? I see the general advantages of accounting for the funnel, yet in my case with outlier components in the data probability density, I think it’s reasonable to base step size only on the width of the major mass component.
Best,
Falk
Boettiger, C., Coop, G., and Ralph, P. (2012). Is your phylogeny informative? measuring the power of comparative methods. Evolution, 66(7):2240–2251.