The whole prior thing is ridiculous. I don't freaking have a prior, so I have to make one up? Seems like so much hokum.
I'm out on the Bayesians

I'm on your side op. Just don't understand why theory people are obssessed with bayesian models. There are many other useful alternatives.
I was talking to a coworker who loves Bayesian stuff. I pretended to agree with all his nonsense in order to avoid an argument.

The whole prior thing is ridiculous. I don't freaking have a prior, so I have to make one up? Seems like so much hokum.
If you are willing to make up a likelihood, then it does not take a lot more to also make up a (possibly diffuse) prior. In economics at least, one often has a pretty good idea what is a "reasonable" range for parameters. For instance discount factors should be close to 1, autoregressive parameters should be between 1 and 1 if the model is to be stable, etc. The Bayesian viewpoint is natural in such contexts.
Also in practice the prior may be determined by using the data itself, not just by "introspection".

Look at Wooldridge's tweets on this to see how clueless most Bayesians are about the important issues.
https://twitter.com/jmwooldridge/status/1367799492829995009
They can't get the basic point that modeling heteroskedasticity or serial correlation is not a substitute for robust standard errors or clustering. Or that OLS, fixed effects, and so on can be studied under very few assumptions. The typical Bayesian thinks the "solution" is to just model everything.
Frequentists need to up their game in explaining this to Bayesians. As Wooldridge points out, it's not the prior that's the biggest concern. It's all the other assumptions that come with it that seem very difficult to relax in a Bayesian setting.
My goal with OLS: Obtain the best mean squared error approximation to E(yx) without extra assumptions. Provide a confidence interval.
Now, to the Bayesian: Give me your best shot.

Look at Wooldridge's tweets on this to see how clueless most Bayesians are about the important issues.
https://twitter.com/jmwooldridge/status/1367799492829995009
They can't get the basic point that modeling heteroskedasticity or serial correlation is not a substitute for robust standard errors or clustering. Or that OLS, fixed effects, and so on can be studied under very few assumptions. The typical Bayesian thinks the "solution" is to just model everything.
Frequentists need to up their game in explaining this to Bayesians. As Wooldridge points out, it's not the prior that's the biggest concern. It's all the other assumptions that come with it that seem very difficult to relax in a Bayesian setting.
My goal with OLS: Obtain the best mean squared error approximation to E(yx) without extra assumptions. Provide a confidence interval.
Now, to the Bayesian: Give me your best shot.Clustering is overly pessimistic and tends to underrepresent the quantity of data contained in a study.