Hi – OK, sure, normal assumptions can be violated. Personally I don’t find this a major concern because it’s relatively easy to catch, and if so one can either transform the summary statistics, or one can use another distribution (nothing in SL requires normal). But it’s true, the normal as default may be a bit less generally applicable than ABC. Thanks!

]]>I would think that using a normal approximation for the posterior is adding one level of approximation to the regular ABC…

]]>But this is something that arises primarily through the use of summary statistics, not through the method that is used to approximate the density p( s=S | theta ), which is the point in which ABC and SL differ.

So, the issue of summary statistics aside, to you see any additional problem that is create by synthetic likelihoods?

]]>