Sönke H. Matthewes

Deutschland braucht dringend ein #Bildungsverlaufsregister um gute Forschung zu ermöglichen & den bildungspolitischen Blindflug zu beenden. Details in unserem (K. Schneider, @IngoIsphording, @CK_Spiess, F. Hertweck und ich) neuen VfS Standpunktpapier (cf. Kapitel Bildungsdaten).

@KolemanStrumpf @instrumenthull @J2dubyas What is this about? I’m not following.

@instrumenthull Honestly wondering how was data misused here? The NYT used a line of the file. Or is issue how it was obtained, in which case any paper using data with dubious origins should be criticized (@J2dubyas comment relating to a paper that hacked potentially PID)?

I don't think "academics" should hide behind pseudonyms and misuse data

@florianederer @wagonomics „this setup“ refers to EB5 or the new proposal? And could you elaborate on why you think one or the other is better? Thanks!

@wagonomics It's $1.8 million now and it's quite burdensome with recurring investments. Last year there were over 12,000 of these visas, but I think this setup is much better.

"Give me your wealthy, your rich, your huddled billionaires yearning to breathe free" (This is a good idea.)

Much more in the paper ofc. Check it out here in case you're interested. It's open access. sciencedirect.com/science/articl…

CS don't raise Abitur attainment likely bc they don't provide a better peer environment than lower tracks bc the academic-track still cream-skims the best students. Despite their name, CS fail to deliver a truly comprehensive alternative within an otherwise tracked system. [n/n]

Camilla Borgna and I have a new paper forthcoming in Economics of Education Review. We study how marginal increases in the flexibility of Germany’s between-school tracked school system have affected student attainment. A short 🧵on what we do & find. [1/n] sciencedirect.com/science/articl…

@Prashant_Garg_ Cool! When I asked it to only use uploaded papers, it indeed dropped the imaginary ones and gave me two real ones. I just tried without the explicit instruction in a new CClara window, but atm it says that it "can’t do more advanced data analysis right now." Will try again later!

@reflecmec I've added some additional instructions to ensure at different processes of it's retrieval, it's reminded of the strict reliance on uploaded data. I think it now behaves better. Thanks for the feedback! chatgpt.com/share/67868f35…

🚨Thrilled to share our new paper "Causal Claims in Economics"! 🚨 @fetzert and I analysed over 44,000 economics papers using AI to create a knowledge graph of economics and map out causal relationships. Here's what we found 🧵👇

@Prashant_Garg_ Maybe I can even post the results here? chatgpt.com/g/g-QE4aPgEVQ-…

@Prashant_Garg_ but none of them are real! (At least judging by a quick Google search - also never heard of any of them. Paper IDs are: 1023, 1145, 1307, 1489). What's going on here? Thank you! [2/2]

@RomanFolw @lakens Ok I agree that we shouldn’t teach this. But it is true that if they don’t overlap then the difference is significant. It’s just when they do then we can’t directly tell, right?? (Or am I missing something here?)

@lakens Another urban myth, which I have no idea about it's genesis, says that if you eyeball (non-)overlap of two 95CIs for group-mean A and for group-mean B, you can actually identify whether the mean difference is (non-)statistically significant with an alpha level of .05.

Confidence intervals are confusing intervals. The reason is because they tell us something about what happens across many confidence intervals, not in any single confidence interval. In the long run, 95% of 95% CI contain a population parameter. 1/n