Show newer
Ruben Arslan

RT @emollick
🤯🤯Well this is something else.

GPT-4 passes basically every exam. And doesn't just pass...
The Bar Exam: 90%
LSAT: 88%
GRE Quantitative: 80%, Verbal: 99%
Every AP, the SAT...

0
Ruben Arslan

RT @StuartJRitchie
Pretty much *the No.1* thing you learn NOT to do in science/statistics is to look at graphs that have similar lines and then imply that one thing must be causing the other.

But now loads of very smart people are doing just this, and everyone is nodding along! What's going on!?

0
Ruben Arslan

New blog post on the Killingsworth, Kahneman, Mellers paper and censored mixed effects location-scale models.
TLDR: sampling error, censoring, and heterogeneous within-subject variance could all bias the KKM estimates. Better models exist (with brms)

rubenarslan.github.io/posts/20

1+
Ruben Arslan

Ah yes, the good old "we tried chopping up the data many ways to find a specific three-way-interaction we had in mind". You love to see it in 2023.

1+
Ruben Arslan

RT @ehudkar
@dingding_peng This is cursed.
so I redid the facet titles

0
Ruben Arslan

RT @annemscheel
Helpful quality content from @dingding_peng as usual, featuring The Law of @lakens' Guidelines (footnote 13) twitter.com/the100ci/status/16

0
Ruben Arslan

RT @dingding_peng
@rlmcelreath I'm terribly sorry Richard but I saw two of your tweets in close succession and then this just sort of happened.

0
Ruben Arslan

RT @Sam_Dumitriu
Rishi Sunak just announced up to £50m in funding for new “focused research organisations.”

FROs are a new way of doing science. In effect, they apply the startup model to scientific research. 🧵

0
Ruben Arslan

RT @Sam_Dumitriu
What might a Biobank FRO look like?

Our paper sets out one idea — a biobank family study.

0
Ruben Arslan

RT @jrgptrs
Is Economics self-correcting? We have a new discussion paper, joint with N Fiala & @flneubauer. Quick answer is: No, rather not. We reviewed all replications published as comments in the AER. Plus, we surveyed the authors. A short thread 1/. bit.ly/3y0wMYp

0
Ruben Arslan

If anything, I see this episode as evidence that adversarial collaborations with a massive power differential don't work.
Like, income explains <1% in well-being. It's loglinear, i.e. diminishing returns. But there has to be a threshold too? How much cope do you need?
---
RT @rubenarslan
@AaronCharlton If I fit a spline on the 15% quantile, it doesn't seem like the threshold is 100k either, more like 200k. Not that this nonlinear fit i…
twitter.com/rubenarslan/status

0
Ruben Arslan

RT @rubenarslan
@AaronCharlton Also, I have to induce an artificial ceiling effect of at least the scale midpoint (50) to see the line flatten. Look at that distribution.

0
Ruben Arslan

RT @rubenarslan
@AaronCharlton I'm a simple man, I see someone claiming that they found just the right quantile regression after lots of tinkering, I run a LOO comparison between a location scale model with and without the moderator.

0
Ruben Arslan

RT @dingding_peng
I’ve seen a lot of confusing interpretations of statistical models, but reporting a quantile regression as if it were conditioning on the outcome probably takes the cake, but only for people who have taken only little cake.

0
Ruben Arslan

I thought there would be an effect, but the effect size surprised me! It's half as large as the IQ effect and if you compare the least to most attractive, the differences beat sex, migration background and parental education (yeah Table 2 fallacy, I know I know).

Show thread
1
Ruben Arslan

A lunch discussion on beauty premiums in school led me to dig this up. As I read it, one SD (.72) in attractiveness is associated with a .4 increase in grade point (from 6 to 1) in math exams and 1.3 points in arts/music/literature. 10 and 15 year olds.
degruyter.com/document/doi/10.

1+
Mastodon

a Schelling point for those who seek one

Trending now

Resources

Developers

What is Mastodon?

schelling.pt

More…

v3.5.3 · Privacy policy