What will happen once we start trying to integrate more with ML and CS and realize more of our results are like this.
Kasy and Sautmann Econometrica 2021 proven wrong
-
Oh, I feel bad for K-S. Must have been brutal to read the comment on their ecma.
Also Young's recent and heavily cited QJE.
Econ research, what's wrong?!Must have been brutal to miss out on a job because a competitor candidate had a forthcoming ECTA that turned out to be flawed.
-
It's been almost a month since this note is on arxiv. Probably they contacted the authors before. What happened? Nothing. It's a shame.
I mean, not even a refutal.
It's not possible to have a refutal.
The reason these authors were interested in this paper in the first place is because it claims to solve something that's not been solved by any stat person and a long standing open question. The proofs are entirely wrong and do not give any new leads to solving the problem.
-
It's been almost a month since this note is on arxiv. Probably they contacted the authors before. What happened? Nothing. It's a shame.
I mean, not even a refutal.
It's not possible to have a refutal.
The reason these authors were interested in this paper in the first place is because it claims to solve something that's not been solved by any stat person and a long standing open question. The proofs are entirely wrong and do not give any new leads to solving the problem.How do we know the proofs are wrong? Have you checked them? I have not (nor I am planning to). The stuff in ArXiv seems to be too short.
-
This case is interesting because while it is very common for proofs to have mistakes, even in famous papers, it is quite unusual for the theorem itself to be wrong. Usually referees get a good idea of whether the theorems are true, and once they believe them don't check the details.
Not for famous open problems
-
This case is interesting because while it is very common for proofs to have mistakes, even in famous papers, it is quite unusual for the theorem itself to be wrong. Usually referees get a good idea of whether the theorems are true, and once they believe them don't check the details.
Not for famous open problems
I think for bandits, it is better to publish in CS or ML outlets (COLT etc).
If you have a solution to an open problem in ML and CS, you will only find qualified referees in that area.
The set of econometricians that works on bandits is SA and MK. I don't know who else. Very hard to find referees.
Dudes, if you have a worthy result, don't send to Econometrica.
-
Read the paper yourself. If Kasy and Sautmann could do something about this they would have already. Kasy and the people behind this fiasco will simply try to bury it. Once upon a time Kasy had the decency to tell people not to cite his false Restud paper, but that warning had long disappeared from his website.
-
Read the paper yourself. If Kasy and Sautmann could do something about this they would have already. Kasy and the people behind this fiasco will simply try to bury it. Once upon a time Kasy had the decency to tell people not to cite his false Restud paper, but that warning had long disappeared from his website.
What makes you think Kasy won't acknowledge the error? In stats and math, it is common to see errata and corrections published. I think only in Economics, people don't publish corrections or acknowldgements of errors.
-
It's been almost a month since this note is on arxiv. Probably they contacted the authors before. What happened? Nothing. It's a shame.
Maybe they are taking some time to look at the critique, see if their pros can be patched, etc. No need to rush.
(Although I suppose a tweet or website note along the lines of "we're thankful for the criticism and we're looking at it" would be a decent thing to do )