Small concentrations

Topic: Categorical data and the Dirichlet distribution · more questions on this topic
‹ Why not a multinomial prior? The simplex CLT ›

For K=3 compare \operatorname{Dirichlet}(1,1,1), \operatorname{Dirichlet}(5,5,5), and \operatorname{Dirichlet}(0.1,0.1,0.1). Give their common centre, compare their total concentrations, and describe where a typical draw lies. Why is “all categories are near zero” impossible?

This exercise has no worked answer.