default prior for general likelihoods

[stephens999]

@mengyin for general likelihoods does it make sense to have the "null-biased" prior as default? Might be better to make it uniform? What is your experience?

[mengyin]

Yes I always use the "uniform" prior instead of the "null-biased" prior. It makes more sense to make the default prior "uniform".

But "null-biased prior" might still be useful in some contexts? Say when we compare singlecell data of two replicates and try finding "outlying genes", we estimate p:=count1/(count1+count2), the mode should be around 0.5 and we expect most genes to come from the "null component" (mode pointmass component).