that graph is an example of "telling on yourself" https://twitter.com/MCBazacoPhD/status/1279067353926586368

you want a hotter take? my guess is that the average econometrics course doesn't even give you the statistical vocabulary to understand why this is not ok. the problem is the signal to noise ratio. the data-generating process is clearly dominated by something not in the model.

well who cares, you might think, if you drank the econometrics kool-aid. "i get my partial effect, that's all i care about". but in cases like this the partial effect is gonna be highly non-robust. *whispers* I have a paper with some smart folks coming soon showing this formally.