Oliver D. Reithmaier
Question for all the #BayesianStatistics and #epistemology folks: can a Bayesian falsify things? Or is falsification just a different updating function for knowledge?
Reason for saying this: falsification to me is kind of a Schumpeterian version of epistemology. You tear down a building (theory) to get a new, better one. Bayesianism on the other hand is more akin to a formative evaluation.
Aug 26, 2024 at 2:48pmWeb
Show threadclimbertobbyAug 26, 2024

@odr_k4tana

The book Statistical Inference as Severe Testing: How to Get Beyond the Statistics Wars, by Deborah G. Mayo seems like what you are looking for.

There's also a shorter essay on the topic, by the same author, but I forgot the title.

Show threadUlrike HahnAug 26, 2024

@odr_k4tana the Bayesian framework allows you to revise your degree of belief in the truth/falsity of a claim like "Joe Biden is the democratic presidential nominee". Evidence in favour will increase that belief, evidence against will decrease it.
It doesn't matter whether you happen to be more interested in its truth or its falsity.

but maybe I'm missing your point?

Show threadOliver D. ReithmaierAug 26, 2024
@UlrikeHahn well, I know how Bayes works 😅
My question comes from the angle where people create things like "Bayesian P-values" that (for me) come from a different epistemological interpretation which Bayes cannot answer. My question is basically am I right in saying this or whether that's just something my brain made up.
Show threadOliver D. ReithmaierAug 26, 2024
@UlrikeHahn in other words: is Freq not extremely alien to how things work in reality? I know there's conditions to meet for frequentist testing, but are they ever met? Aand is it efficient to do this? I feel like constructing theories to then test is a bit of extra mileage added onto an otherwise straight path.
Show threadUlrike HahnAug 26, 2024

@odr_k4tana Oliver, I had typed out a response that somehow failed to include your handle...(I've now edited it in, but doubt that works..)

it's here https://fediscience.org/@UlrikeHahn/113029731632614410

Ulrike Hahn (@UlrikeHahn@fediscience.org)

@odr_k4tana@infosec.exchange still not sure I understand completely what you are saying, but the epistemological difference feels like this: goal Bayes - evaluate truth or falsity of theory, to do that make assumptions about diagnosticity of data goal Freq - establish non-random nature of data pattern (phenomenon) and make assumptions about how that relates to theory ??

FediScience.org
Show threadOliver D. ReithmaierAug 27, 2024
@UlrikeHahn thank you!
Show threadVirginicusAug 26, 2024
@odr_k4tana If the odds are less than 20 to one in favor of the proposition, that’s how most people interpret flunking a p-value test.