What's the big deal about it? Why the hype? Aren't we all already Bayesians anyway? I mean we don't treat all possible hypotheses as equally plausible so there's always some underlying 'prior distribution' even among frequentists.
Redpill me on Bayesianism
Jacob Parker
Easton Miller
Requires different tools and can potentially yield different outcomes.
What's the deal? Do you have a specific question?
William Ramirez
Could you elaborate a bit more on that? Sorry I'm a bit of a brainlet.
If Bayesian updating would lead to a posterior distribution with the same expectation as the maximum likelihood estimation of a parameter asymptotically then why bother using this method at all considering the much greater computational costs. Even Stan, the fastest Bayesian software, still takes ages to run and doesn't seem to get any better results. So why are Bayesians such a rabid adherents to the method when computationally cheaper and equally effective methods exist within the frequentist paradigm?
Jaxson Sullivan
>So why are Bayesians such a rabid adherents to the method when computationally cheaper and equally effective methods exist within the frequentist paradigm?
You might want to extend your scope a bit. Bayesian frameworks are applied in many areas. I work in remote sensing and Bayes' theorem allows bringing a bunch of things together that otherwise would be a massive faff and allows for quicker results compared to e.g. MCMC.
Gabriel Sanchez
It's a fad among comp-sci brainlets right now. You won't find any serious mathematician that is not a frequentist. Bayesianism is an unrigorous heuristic that appeals to engineers because it's "intuitive".
Bentley Cox
>So why are Bayesians such a rabid adherents to the method when computationally cheaper and equally effective methods exist within the frequentist paradigm?
They are not. The point of Bayesian statistics is that it provides the underlying math for frequentist methods. Bayesian statistics is not about computational methods at all; it's the background theory on which all computational methods must be based, and can explain things like when they work, why, and under what assumptions.
Sometimes this theory suggests computational tools unlike anything in frequentist statistics, and then we call those tools "Bayesian methods". But that doesn't actually have anything to do with Bayesian statistics; it's just basically a name for any methods that do not arrive from the surface-level understanding of frequentist statistics.
Xavier Green
This is bait.
Nathan Reyes
Why use a fat unattractive neckbeard as an OP
Tyler Wood
Yudkowsky is the poster child for the 'rationalist' Bayesian cult
Adam Allen
Landon Jenkins
Last I checked he was not a practicing statistician.
Just because someone screams the loudest doesn't mean they represent a majority viewpoint.
Blake Lewis
Read "Probability Theory: The Logic of Science"
Ryan White
>The phrase “Bayes’ Theorem” has caused a fair amount of confusion, by creating the nebulous sense that the Bayesian machinery is rigorously grounded in some single fundamental mathematical result, some deep inviolable idea like conservation of energy.
Evan Mitchell
Yeah this guy is bait. Look at any ivy league's Statistics department and 50% if the tenured faculty will be doing some form of research in Bayesian or non-parametric statistics. Furthermore look at any of their physics departments and you'll see most of the astronomers using it. It's useful when you are working in fields with small amounts of usable data. I personally am doing research into it's usefulness in calculating Nuclear EFTs as nuclear physicist.
The guy up there is right. It's not necessarily 'better' than frequentism. It's just sometimes more useful for the situation.
Grayson Walker
Justin Bailey
>stat.columbia.edu
OP here. I actually am at Columbia and recently started attending some of Gelman's seminars, hence the question. I don't really see anything special about 'Bayesian' methodology that hasn't already been incorporated in mainstream statistics, nor do I see any appreciable improvement in predictive results compared to explicitly non-Bayesian methods such as neural nets.
Levi Kelly
>nor do I see any appreciable improvement in predictive results compared to explicitly non-Bayesian methods such as neural nets
Neural networks are uninterpretable. Not all statistics is about prediction, sometimes your goal is to understand the system under consideration.
Thomas Reed
Correct me if I'm wrong, but I'm fairly sure you can derive every frequentist approach through bayesian maths.
Benjamin Moore
>Redpill me
gtfo pill-popping /pol/itard
Ryan Moore
This is true. Frequentism makes immediate constraints. Sometimes it's better not to. I don't really see why people feel the need to argue about this.
Grayson Rodriguez
>Statistics department .. physics department
>mathematicians
Go away brainlet.
Colton Parker
It's the truth.
Angel Rivera
yes, but you also sometimes get different results with frequentist vs bayesian analyses
Logan Nguyen
t. IQ-signalling role-player
Gavin Mitchell
lesswrong is a cult of libertarians pretending they are radical centrists. /thread
Jack Ross
Wtf does that have to do with anything.