Yeah that's valid. I should have posted those two graphs together. Unfortunately there's no "amend" option and corrections never travel as far.

I'm not sure it changes the interpretation all that much—it makes sense that both legit and illegit accommodations would be requested after a slump.

The most marketable skill right now is having good judgment regarding how and when to use AI.

Using AI to generate your writing assignments is having bad judgment.

every student taking the easy path here is putting a little more alpha on the table for their peers who don't

just register for your own protection. it will be fine

We also published the entire database with metadata on 80 million Indian court cases from 2010–2018: devdatalab.org/judicial-data

If you want to learn more, we made an interactive demo of how publication bias works:
devdatalab.org/publication-...

And you can learn more about our paper on judicial bias here: devdatalab.org/judicial-bias

4/

If there's publication bias, studies look asymmetric on the graph — because noisier studies need bigger effects to be statistically significant.

The left graph is a simulation of publication bias. The black points in the right graph are past studies on ingroup bias. 3/

In the absence of publication bias, estimates should form a funnel centered around the true effect.

Statistically significant estimates are outside the solid lines — but point mass should be similar just on either side of the line. 2/