Zeta Of 1's Avatar

Zeta Of 1

@zetaof1.bsky.social

Economist, Enthusiast of Chess, Math, Music and similar items. Data Dude. Some call me the space cowboy, some call me the gangster of love. Sober 10 years. Known to rearrange thoughts in other people's brains.

1,265 Followers  |  600 Following  |  5,380 Posts  |  Joined: 01.07.2023
Posts Following

Posts by Zeta Of 1 (@zetaof1.bsky.social)

I also wonder about this, if the frequentist interpretation is "in 95% of cases the true value is contained", why would we, as *f r e q u e n t i s t s*, not call that "chance" if you already label frequencies as probabilities or ar least good proxies for it?!?

04.03.2026 07:09 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Go agentic. Make Claude build it by itself and then have it register when baby wakes up and automatically activate cradle rocking at night.

04.03.2026 06:57 β€” πŸ‘ 3    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

People think we actually have an algorithm to compute the line of best fit in a data cloud, in reality it's all just Davey who's just really good & fast at drawing good-looking lines

03.03.2026 12:11 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

On a very unrelated tangent, I am somewhat convinced that at some point in the next decades some great figure in Econ research will pivot to Bayesianism and then everyone else follows along and people will say "How silly that none of us considered doing that before". That's how econ research works.

03.03.2026 11:11 β€” πŸ‘ 6    πŸ” 1    πŸ’¬ 0    πŸ“Œ 0

least strength x to disregard our results"? 2/2

03.03.2026 11:08 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

I wonder, is there any established name for what I would call the "hater-prior", i.e. that mildest prior you need to have that is so strong that your results (e.g. HDPI) do include a null-effect? Would be an interesting way to argue "In order to not believe our results you need a prior of at 1/

03.03.2026 11:08 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Reading that table of contents: I'm buying this. I'm buying this so hard.

03.03.2026 11:05 β€” πŸ‘ 3    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

your study and still likely end up with the desired results. Or am I missing anything? 3/3

03.03.2026 10:49 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

perform some simple simulations at home in secret ex ante on different scenarios on how data might show up to determine how informative your prior needs to be in order to most probably achieve a satisfying result. So you can "fabricate the data" ex ante just by simulating and then register 2/

03.03.2026 10:49 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Don't your priors bias your results into the direction which you deem your prior expectation? Therefore, whatever you want your outcome to be close to ex ante you can set your prior to and you can be sure that you will at least have somewhat biased everything in that respect to. You can even 1/

03.03.2026 10:49 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 2    πŸ“Œ 0
Proof

Proof

Here also visually on paper

03.03.2026 10:40 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Edit: an earlier version mistakenly wrote that you add k observations with a single lambda per line, but it should be √(lambda). I've fixed that mistake.

03.03.2026 10:33 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

mean value in all others still leads to mean value in the outcome, i.e. your outcome will be robust against small changes, biasing the OLS coefficient per regressor against it towards zero because you add a positive change that does not lead to any change in the outcome variable. 4/4

03.03.2026 10:32 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

with k "fake"observations that are a single √(lambda) per line and rest of the regressors 0 that lead to an outcome of 0. Thus you've basically added fake data to bias your problem, saying that, if we speak of standardized data here, having a single value of mean + √(lambda) in one regressor and 3/

03.03.2026 10:31 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Z equals X and √(lambda)*I stacked, i.e. Z = (X √(lambda*I).T. put differently, you add √(lambda)*I as new observations at the bottom of X.
y' is a stack of y and k observations of just zero.
Finally, verify that b_hat = (Z.T @ Z)^(-1) @ Z.T @ y'.
Effectively, you have reduced this to OLS 2/

03.03.2026 10:30 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Sure! Observe that the analytic solution for the Ridge Regression is b_hat = (X.T@X + lambda*I)^(-1) @ X.T @ y, where @ is matrix multiplication, lambda is ridge penalty, and I is identity matrix with dimensions of k x k, k being number of regressors (including constant).
Now define as follows: 1/

03.03.2026 10:25 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Absolutely. That's why I find it so weird. Effectively, doing a frequentist Ridge regression is also the same as data fabrication. Do we throw that out of the window now as well?

03.03.2026 09:49 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 2    πŸ“Œ 0

by definition the same. 2/2

03.03.2026 09:46 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Without reading this: it is kind of trivial though? Bayesian Models usually tend to reach a sort of "compromise" between the prior and the data. Effectively, the prior adds like additional data from a frequentist perspective. Of course that is then indistinguishable from data fabrication. It is 1/

03.03.2026 09:46 β€” πŸ‘ 3    πŸ” 0    πŸ’¬ 3    πŸ“Œ 0

Sehr freundlich, sehr zuverlΓ€ssig, verkauft Zwerghahn Seidenhuhn. So wie alle Julia Rohrers.

03.03.2026 09:22 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
Preview
Zwerghahn seidenhuhn Seidenhuhn Hahn vom letzten FrΓΌhjahr,Zwerghahn seidenhuhn in Baden-WΓΌrttemberg - Dunningen

I think this person sells some, maybe that can help you?

www.kleinanzeigen.de/s-anzeige/zw...

03.03.2026 09:19 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

All I ever see is AI going for the boring low-hanging fruits but what about the things of real actual value like automatically changing reels that say "girlfriend" to "wife" so that I can send them to my wife

03.03.2026 09:17 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

I find the idea quite funny that you make increasing, or at least proportional, additional revenue from adding arbitrary amounts of adding vending machines. Just put 5000 vending machines in one place and get R I C H

03.03.2026 07:49 β€” πŸ‘ 6    πŸ” 1    πŸ’¬ 2    πŸ“Œ 0

Aside from the quite obvious simplicity of this, I always laugh when I hear "7 figures in revenue". Making revenue is easy. Making profit is hard.

03.03.2026 07:21 β€” πŸ‘ 5    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Ah I see. Are you within Germany? Then either via ARD Mediathek (free) or Wow (subscription). Outside Germany - depends on your country

03.03.2026 07:19 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

OTT?

02.03.2026 23:02 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

I used to think that Dark was the best thing german cinema of recent years had to offer but lord did Babylon Berlin prove me wrong
Hands down best german show I have seen in recent years. Maybe the best german production I've ever seen

02.03.2026 22:48 β€” πŸ‘ 4    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

The entire city or just this place? Because this is actually not bad looking. Or are we rather talking about the overall looks of Monnem? That I am not so sure on...

02.03.2026 13:06 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

🀒🀒🀒🀒

02.03.2026 09:37 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

It's more about the name than the concept, but don't worry, cannot eat this again now either

02.03.2026 09:37 β€” πŸ‘ 3    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0