Overwhelming evidence of research fraud and he still earns 190k/year www.ontariosunshinelist.com/people/steve...

On the upside, I got to talk to the guy that blew the whistle on him. He did it while he was a postdoc! He had every reason to believe this was career suicide, so it was just incredibly brave of him. Good news is he landed well and now has a good position.

I had not because I’m trying to minimize the amount of manual work I have to put into this, but you’re right that it’s a neat idea

I suppose I could always drill the holes asymmetrically and leave the axis unlabelled, to convey trying to publish a result where scrutiny is asymmetric, but ideally I'd like to convey just the minimal point about selection on significance (not sign) first.

the main problem is that while this vividly shows the missing estimates, I don't think it conveys the problem of bias when the distributions is not centered on zero. I could literally add numeric labels to the bottom, but this feels intuition-breaking

I don't mean to be rude but I don't think this conversation will be productive. We both lack enough context about the other person's probably complicated beliefs, and I don't want to spend the time to build that context right now. sorry.

I don't mean to dodge a sincere question, but I think the question is badly formed. replace them *doing what*? I do get meaning from my work and I really like it, but most work is drudgery and I think we can find lots of meaning outside of much (possibly all) of it.

that has not been my experience

I very much think that “evidence based policy people” should be interested in this, but because it is a fairly general thing and not targeted to a specific domain it’s a bit hard to pitch. You’d think this is a strength, but IME it’s not.

I don't think so (though I'm not super familiar), and the public validation part seems pretty important.

We'll never earn it back and we'd rather not spend our own money on this (though tbh I've often thought about if I could just buy out my own teaching from my own salary—Deans don't like this conversation). Our goal is to release everything publicly for free.

Main upshot is we should end up with a tool where you give it an article and it gives you rows in a spreadsheet recording results, methods, research context, etc, for basically free. It's so cheap that we plan to do it on all of PS from 2010 forward and release the results publicly.

We can do a variant of this right now for something like 0.05 USD/paper and costs will probably fall. We're looking for money to buy out a programmer profs time so he can devote more time to structuring the info we give the model to boost performance. This seems to matter a lot.

We're building & validating a way to extract info from academic articles, starting in political sci (because it's a very hard case). The end result of this is a pipeline that goes from article to machine readable data on: key outcome, key independent variable, coefficient, standard error, &much more

no. It's of political science. but our data extraction task is very likely general (e.g. if we can get LLMs to reliably figure out the key hypothesis, DV and IV, and pull the right coefficient and SE from tables in PS then they can probably do it in econ, heath, etc)

yes, I said 9k (it's actually like 8,500 USD).
Canadian profs are so cheap you're being ripped off by not buying out our teaching time!

I’ll read it (again?) with interest. Thanks for sharing

Defeats the whole purpose

I probably should back off the “justification” frame, but I think it is an interesting finding that does offer a reason for us to push at least to 80%, and it’s funny that the max in my setup lines up with Cohen’s convention