Increasing in recent years. For a long time, one had to use mathematical shortcuts to make the techniques computationally feasible (for example conjugate priors

Increasing in recent years. For a long time, one had to use mathematical shortcuts to make the techniques computationally feasible (for example conjugate priors). Now, with better sampling algorithms (like NUTS) you don't have to make those compromises anymore. This means as a practitioner you get all the benefits (like built in, rich error distributions and good small sample properties) without the compromises.