9 thoughts on “Response to criticisms of Bayesian statistics”
> “Trust, but verify,” recommending Bayesian methods with frequentist evaluations, but I usually don’t get around to that. My actual practice is closer to “Trust, but check”: assume a model, fit it, then check its fit to existing and future data.
Wondering if computational advances of the past changes this much – easier to get around to?
Following a course I took with Michael Betancourt in 2019, I always do such checks now.
This paper is great, Andrew, somehow I missed it. It will be a great go-to reference for a talk I am about to give today in Hamburg on the role of replicability in BDA.
I have done a couple intro Bayes talks for non-statistician Data Science folks based on Betancourt’s work on Principled Bayesian Workflow and it seems to make sense to them.
They can “see” how Bayesian models repeatedly work in fake universes where you know the the true parameter values, some fake data was generated by a known fake data generating model and obtain the MCMC posterior sample. Also makes clear the downside of using small noisy data sets and or poorly thought out priors.
They can “see” how Bayesian models repeatedly work in fake universes
what are you talking about? there’s this thing called bayes theorem. it is an algebraic consequence of the axioms of probability theory. is is an algebraic formula that has enormous relevance for modeling human behavior and for developing mathematical computer algorithms and forecasting models in economics and elsewhere.
Perhaps where you know the the true parameter values, some fake data was generated by a known fake data generating model and obtain the MCMC posterior sample.
I do not mean to be rude but this makes no sense and I do not think you are in a position to trash talk the reverend.
I’ve got Firefox 70.0.1 on Linux and it displays in browser. So this may be an option you can configure somewhere which you clicked ages ago and it’s dutifully remembering…
> “Trust, but verify,” recommending Bayesian methods with frequentist evaluations, but I usually don’t get around to that. My actual practice is closer to “Trust, but check”: assume a model, fit it, then check its fit to existing and future data.
Wondering if computational advances of the past changes this much – easier to get around to?
Following a course I took with Michael Betancourt in 2019, I always do such checks now.
This paper is great, Andrew, somehow I missed it. It will be a great go-to reference for a talk I am about to give today in Hamburg on the role of replicability in BDA.
I meant the other 2019, i.e., 2018.
I have done a couple intro Bayes talks for non-statistician Data Science folks based on Betancourt’s work on Principled Bayesian Workflow and it seems to make sense to them.
They can “see” how Bayesian models repeatedly work in fake universes where you know the the true parameter values, some fake data was generated by a known fake data generating model and obtain the MCMC posterior sample. Also makes clear the downside of using small noisy data sets and or poorly thought out priors.
One was recorded and is available here https://www.youtube.com/watch?v=I7AVP9BCm1g and also discussed in this post https://statmodeling.stat.columbia.edu/2019/10/15/the-virtue-of-fake-universes-a-purposeful-and-safe-way-to-explain-empirical-inference/
They can “see” how Bayesian models repeatedly work in fake universes
what are you talking about? there’s this thing called bayes theorem. it is an algebraic consequence of the axioms of probability theory. is is an algebraic formula that has enormous relevance for modeling human behavior and for developing mathematical computer algorithms and forecasting models in economics and elsewhere.
Perhaps where you know the the true parameter values, some fake data was generated by a known fake data generating model and obtain the MCMC posterior sample.
I do not mean to be rude but this makes no sense and I do not think you are in a position to trash talk the reverend.
I am not able to view that file.
Really? It’s just a pdf. That’s odd.
Some time ago, Firefox’s behavior changed, it started auto-downloading pdf’s when you click on them. Earlier they would show in the browser.
I’ve got Firefox 70.0.1 on Linux and it displays in browser. So this may be an option you can configure somewhere which you clicked ages ago and it’s dutifully remembering…