I have mentioned before that I have some issues with the most vocal proponents of Bayesian analysis. Nothing against the approach as such - I have used it myself often enough. Indeed Bayesian phylogenetics and Bayesian analysis of population structure are fairly standard in my line of work.
But the problem is that every so often you run into people who think that it is the alpha and omega of scientific methodology, that it is hugely superior to every other approach, and/or that every other approach just sucks, or who simply try to apply it to everything, even where it is very difficult to apply.
That last one is often a problem because there is a tendency to conflate an intuitive use of Bayesian logic - something that we all daily do even if unconsciously, and that is consequently so trivial that it does not need to be hyped - and an explicit use of the Bayes formula to calculate actual, numeric probabilities - which is much harder to do because the probabilities and especially the priors are often pulled out of some dark place. Chris Hallquist just did a good post mentioning that distinction.
Really I get the feeling that many dedicated Bayesians, especially phylogeneticists just a bit younger than myself, exhibit a kind of missionary zeal. They may have the feeling that they are kind of a new wave, transforming their area of science for the better, throwing out what needs to be thrown out.
And this is where it hit me: That is exactly the impression that cladists must have given colleagues using old-established approaches when they swept biological systematics in the 1980ies and 1990ies. And I can understand; there must be something really insufferable about somebody telling you that what you have done the last thirty years is bad, and that you have to adopt their shiny new technique or you aren't a good scientist.
Still, and that may only be my bias, but to me there is significant difference: The cladists were right. It really does make more sense to classify by relatedness, and it does not make sense to classify inconsistently. You need to apply the same criterion everywhere because otherwise your classification will be uninformative and non-predictive, in other words useless.
On the other hand, while Bayesians have a shiny new approach they do not have the one principle that has to be applied everywhere. Indeed their approach cannot possibly apply to every question a scientist may ask (just as cladism does not apply to, say, geology). It is a powerful tool but it is only one of several powerful tools we have. When use it for cases where it is the best tool, and we use parsimony, for example, for cases where parsimony is the best tool.
Let's just say that the claim of Bayesian analysis being superior to all other tools, or demands we apply it to everything, seem a bit extreme.