They have 253 retracted papers and they're making a generalization about falsification in PubMed?
I'm pretty sure the retracted papers are just those that were falsified most stupidly and not anything like a representative sample of those falsified.
I read a pretty good account of the Schon scandal. From interviews with his colleagues at Bell Labs, it seems that one of the reasons he got away with it for so long is that he didn't act like a blowhard. People had a preformed image of fraudsters as arrogant self-promoters, and so Schon's apparent modesty threw them off.
I'm pretty sure the retracted papers are just those that were falsified most stupidly and not anything like a representative sample of those falsified.
The bad wig theory of fraud.
I think 2 is right. The real finding seems to be "bad liars are easier to catch."
The obvious next step is to take their metric and apply it to the non-retracted papers to see if there are any that ought to be looked at more carefully.
Where are they saying they found fraud associated with boasting / big talk? The abstract says the fraudulent papers had more jargon, lower readability (definition important there), and more citations. The press release also says what I'm not sure is a hypothesis or a finding, "a fraudulent author may use fewer positive emotion terms to curb praise for the data, for fear of triggering inquiry".
Looking at all of the evasively written paper loaded with jargon ought to be a quick job.
The methodology of the linked paper is sort of hilariously susceptible to shenanigans. I would -- yes! -- really like to see a replication of this. (For one thing, that's a tiny sample to be looking at such a broad set of measures. Corpus linguistics works best when you have thousands and thousands of exemplars. Also, statistical significance is really not going to tell you much in this instance. If they built a predictive model with independent training and test sets (preferably with a very very large negative set) that would be a lot more convincing)
I think it would be especially tricky here since a lot of the fraudulent studies we know about tend to be discovered when someone gets caught out once and then everyone goes back through their earlier stuff and goes "Oh, oh shit. This is all made up." And without a lot of data and different fraudulent researchers involved you're running the risk of something like "Statistically speaking, fraudulent scientific studies look like they were written by Diederik Stapel."
I really, really want this paper about fraudelent research to be fraudulent.
They could include themselves as a data point!
I suggested this strategy once to a researcher talking about the difficulty of doing studies on some kinds of bad behaviors in medical research/doctors generally where you both need people to answer honestly about stuff and also you're under an obligation to report people if they'd broken the rules. He said it sounded like the sort of thing a philosopher would think of, which made me proud.
That's not an uncommon problem in survey research, if you leave out the mandatory reporting part. Supposedly, you can ask questions like "Is your birthday in December or have you ever deliberately falsified research?" and then get the prevalence after adjusting by 1/12th. But I've never seen anybody actually do that.
I assume they're impressed by the elegance of my solution, and have decided to adopt it instead.
It's a shame outright fraudulent research gets all the attention when the really massive problem is on the 2nd and 3rd tiers of scientific dishonesty: real data, cherry picked or spuriously analysed to support the hypothesis.
16: One of the biggest recent scandals in a field adjacent to mine wasn't exactly provable fraud, but it involved not doing an obvious control, presumably because the researchers in question suspected that their Nature worthy result would go up in smoke if they looked at it too closely.
1.5 percent more jargon. "significant" but not very significant
I keep reading this post title as "fraudulent blowjobs," and wondering how such a thing might occur.
19: Like the end of "Revenge of the Nerds"?
They're all fraudulent if no blowing is involved.
OT: is there a modern, non-offensive phrase generally in use in replacement of "Indian-giver"?