Ask HN: Criticisms of Bayesian statistics? | Hacker News Reader

Ask HN: Criticisms of Bayesian statistics?

1 pointmuraiki4 comments

In tech circles, it seems that Bayesian statistics is often favored over classical frequentist statistics. In my study of both Bayesian and frequentist statistics, it seems that the results of a Bayesian analysis are generally more intuitive, such as when comparing Bayesian credible intervals to frequentist confidence intervals. It also seems like Bayesian analysis avoids what I think is one of the most serious problems in analysis, the multiple comparisons problem. It's been easy for me to find any number of Bayesian critiques of frequentist stats, but I have rarely seen frequentist defenses against Bayesian stats. This may simply be because I mostly read technology related sites as opposed to more general statistics oriented sites. As such, I would really appreciate hearing some frequentist critiques of Bayesian stats. I feel like the situation can't be as cut and dry as one being better than the other in all things, so I would like to acquire a more balanced perspective by hearing about the other side. Thanks!

~bayesian logicism

https://plato.stanford.edu/entries/logic-inductive/ :

> It is now generally held that the core idea of Bayesian logicism is fatally flawed—that syntactic logical structure cannot be the sole determiner of the degree to which premises inductively support conclusions. [...]

Read an old stats text. One reason why Bayesian statistics was less popular in the past was that people didn't know how to do it, except in special cases. Techniques have e loved since then.

Thank you for your response, but I'd like to understand modern perspectives in the frequentist vs Bayesian debate. I know that in the past, Bayesian methods were oftentimes impossible to use in non-trivial analyses because calculating the integral over the probability of the data was infeasible.

But now that we have efficient Markov chain Monte Carlo algorithms along with fast and inexpensive computers for sampling from posterior distributions, why isn't every statistician embracing Bayesian methods if they provide so many benefits over the classical approach?

I think most of the people doing research on statistics now are Bayesians, but I think many practitioners still use the old methods because that is what they learned in school and they are still being taught as a result of inertia.

I think the teaching of statistics has often been wrong minded in a number of ways. I am not a professional statistician, but I know enough about it to know what I don't know, so I have a strong preference for nonparametric methods because these are more foolproof than the typical methods that assume gaussian distributions.

Reply on news.ycombinator.com