I've published stats papers on Bayesian methodology, consider myself a kind of Bayesianist, and stances like the link essay, and many of the posts here, make me sad and a bit frustrated. As you're saying, it's become an ideological, not functional, debate. Different inferential paradigms have their benefits and costs, and Bayesian methods have their own problems.
Whenever you make an inference, it's a gamble, in a sort of literal sense. The choice of frequentist versus Bayesian methods basically is a bet about whether the bias due to your prior is small enough to offset the variance due to the lack of one. If the bias due to a "misguided" prior is high enough, you'll end up making a worse inference than not using a prior.
You can, of course, use an optimally conservative prior, a reference prior, but for many cases what you're left with is a uniform prior, and therefore, frequentist inference. It's not always the same as a uniform prior, but in practice is often the same. I think the philosophical arguments against frequentist lines of thoughts are often unresolvable and involve strawmen characterizations of frequentism often derived from misperceptions of young adults learning any sorts of stats for the first time.
There's also a very strong argument against strong priors in competitive situations, such as in achievement tests for admissions or some such thing. Imagine making priors for your test score based on your demographic background, for instance. Technically this might be ok from a subjective Bayesian perspective but I think almost no one would agree this is acceptable, and the reasons why are telling about statistical inference more generally.
Many of the other arguments equally apply to Bayesian methodology too. The worst problems of NHST are not actually about the tests, they're about how they are used and interpreted, and would creep up if everyone was using credibility intervals too.
I kind of think Bayesian methodology should be taught more at an earlier stage, but it would be irresponsible in my mind to not teach frequentist methodology at the same time.
Whenever you make an inference, it's a gamble, in a sort of literal sense. The choice of frequentist versus Bayesian methods basically is a bet about whether the bias due to your prior is small enough to offset the variance due to the lack of one. If the bias due to a "misguided" prior is high enough, you'll end up making a worse inference than not using a prior.
You can, of course, use an optimally conservative prior, a reference prior, but for many cases what you're left with is a uniform prior, and therefore, frequentist inference. It's not always the same as a uniform prior, but in practice is often the same. I think the philosophical arguments against frequentist lines of thoughts are often unresolvable and involve strawmen characterizations of frequentism often derived from misperceptions of young adults learning any sorts of stats for the first time.
There's also a very strong argument against strong priors in competitive situations, such as in achievement tests for admissions or some such thing. Imagine making priors for your test score based on your demographic background, for instance. Technically this might be ok from a subjective Bayesian perspective but I think almost no one would agree this is acceptable, and the reasons why are telling about statistical inference more generally.
Many of the other arguments equally apply to Bayesian methodology too. The worst problems of NHST are not actually about the tests, they're about how they are used and interpreted, and would creep up if everyone was using credibility intervals too.
I kind of think Bayesian methodology should be taught more at an earlier stage, but it would be irresponsible in my mind to not teach frequentist methodology at the same time.