Update: Richard Van Noorden at Nature runs the numbers on this paper, and comes to the same conclusions: you can't use it to say that US papers are more prone to fraud. . .
There's a new paper in the Journal of Medical Ethics looking at fraudulent publications. The author went back and studied the papers in the PubMed database that have been retracted during the last ten years (there are 788 of them), looking for trends. And so far, the press coverage that I've seen of his conclusions seems to be missing the point.
One thing that stands out is that retracted papers tend to come from higher-profile journals. That makes sense to me, because that's (a) where they're more likely to be noticed, and (b) where the editors are more likely to care. As you move down the list, people just seem to start shrugging their shoulders. Has a paper ever been retracted from Bioorganic and Medicinal Chemistry Letters, for example? I can't think of any, and readers are invited to try to find one themselves.
Another conclusion is that retracted papers tend to come from serial offenders. Over half the retracted papers had a first author who had retracted something else. I think that probably represents cases where someone had published a body of fraudulent work which all got exposed at once, but if that's not a serial offender, I don't know what is.
But a third conclusion is the one getting the headline writers going. The paper examines the retractions, looking for whether they were withdrawn for error or for fraud. The US is notable for having more retractions in the latter category, as compared to the rest of the world: one third of its 260 retracted papers are attributed to fraud, while the rest of the world comes in between 20% and 25%. So you get things like "US Scientists More Likely To Commit Fraud". But no, for that conclusion to be valid, you'd want to know how many fraudulent papers were published as a percentage of the whole output.
Even that wouldn't tell you the whole story. Remember, we're looking at papers that have actually been retracted. Most published fraudulent research never gets to that point. Lower-end journals have a terrible problem with plagiarized, derivative junk. Piles of it gets sent to them, then too much of it gets into print, and it just sits there, with no one ever paying any attention. Well, years later, some poor person might try to reproduce a prep, find that it doesn't work, sigh, and try something else, but otherwise. . .
No, in the same way that the prominent journals are over-represented, I think that the US might be a bit over-represented because slightly more fraud gets caught. More US papers appear in prominent journals than average, and fewer appear in the absolute bottom-rung journals. The United States accounts for at least one-third of the total scientific paper output, so those 260 papers out of 788 are exactly what you'd expect if all other things were equal. But other things aren't equal. We may have more prominent (and more harmful) frauds here, but I'd be willing to bet that as a proportion of the whole, we have fewer of them.
One last note: the figures in this paper seem to conflict with an earlier analysis, which seems to have found fewer retracted papers in PubMed, and a higher proportion of them tainted by fraud. Who's right?