In practice, you only need to use the non-parameteric prior when you have a very large deviation from the parametric prior. In most cases, the results are nearly identical within a 1-2% difference in the expression values. In your case, I would recommend just trying it and seeing if it makes a difference. If it does, trust the non-parametric. Evan On Jun 13, 2013, at 6:32 AM, Arne M?ller wrote: > Hello, > > I've a bias of the delta.hat parameter when looking at the prior density plots (par.prior = T). The empirical data has a long tail and therefore deviates from the normal data. Would it make sense to use par.prior=F in this case? When is the nonparametric estimate recommended? > > thanks a lot, > > Arne > >