Also this week: automating ecology, data transformation vs. global warming, Simpson’s paradox vs. Covid vaccine efficacy, vaccine hesitancy (polio edition), the case for pandemic optimism, another retraction for Denon Start, and more.
Shu et al. 2012 PNAS, an influential psychology paper on how the right “nudge” can reduce dishonest reports of information, is…dishonest. Yes, really. The fakery–and that’s definitely what it is–was discovered when the raw data were included in the supplement of a 2020 PNAS paper by different authors. The 2020 paper failed to replicate the 2012 findings; now we know why! This case illustrates what I now think of as the main argument for mandatory data sharing: data sharing makes it harder to get away with faking data. Kudos to the anonymous authors of the linked piece for some excellent data forensics (I say this as someone with a bit of experience in data forensics myself…). Noticing the use of two different fonts in the data file, and using that clue to work out exactly how some of the data were faked, is an especially good bit of sleuthing. The linked post in turn links to responses from four of the authors of Shu et al. You should read them as well. And here’s first author Lisa Shu’s Twitter thread on the case. There is strong evidence pointing to the source of the fakery: it seems to be either co-author Dan Ariely (a very famous psychologist at Duke), or the insurance company from which Ariely says he obtained the data. In his statement, Ariely blames the insurance company, and indicates that he would welcome an investigation by Duke. Note that there are some apparent conflicts between Ariely’s statement, and the available evidence. We’ll have to wait and see if an explanation for those apparent conflicts is forthcoming. Hopefully there will be a quick and thorough investigation that will get to the bottom of all this. It would sure speed the investigation if Ariely would just name the insurance company…This isn’t the first time that questions have been raised about data that Ariely claims he got from a private company (and as an aside, that story illustrates one reason why Ariely might not want to name the insurance company…). Meanwhile, others have started doing deep dives into Ariely’s other papers, somewhat hindered by the fact that he ordinarily only provides summary statistics, not raw data. Anyway, as in all such cases, you feel terrible for the co-authors who’ve been burned for doing what we all do (and could hardly avoid doing): assuming that the people we’re working with are honest. A bit of further context: Ariely received an Expression of Concern for another paper of his just last month, due to a bunch of statistical discrepancies that couldn’t be fully explained because Ariely couldn’t provide the raw data. And finally, what would you bet that PNAS will retract Shu et al. 2012? That’s not a rhetorical question: you literally can bet money on it. UPDATE: BuzzFeed identified the insurer that purportedly provided the data. It’s The Hartford, which didn’t reply to multiple requests for comment. /end update
UPDATE #2: The Hartford confirmed that it partnered with Dan Ariely on a “small project” in 2007-8, but says it can’t locate any data, results, or other deliverables from the project. /end update #2
Start et al. 2019 Am Nat has been retracted at the request of all authors but Denon Start. The retraction notice is admirably detailed. Kudos to Ben Gilbert and Art Weis for doing the right thing here, after spending considerable effort getting to the bottom of the problems with the data and analyses. Sorry that the burden of correcting the scientific record fell on them, rather than on Denon Start where it belonged. Am Nat EiC Dan Bolnick has a tl;dr version of the retraction notice. Add this paper to the growing list of papers by Start that have been retracted, corrected, or subjected to expressions of concern (see here for a very incomplete list; there are PubPeer threads about 16 different papers of his). As I’ve noted previously, most EEB researchers seem to have stopped citing any of Denon Start’s papers.
This is old but I missed it at the time, and it’s newly relevant in light of some of the links above: maybe scientific funding agencies should allocate a small amount of their research budgets (say, 1%) to researchers who want to reproduce or double-check the work of others. What do you think?
Sociologist David Weakliem with a series of three blog posts on hesitancy to take the polio vaccine back when the vaccine was first introduced in the 1950s (part 1, part 2, part 3). Very interesting historical comparative context for covid vaccine hesitancy.
Timothy Keitt and Eric Abelson on automating ecology.
Simpson’s paradox vs. Covid vaccine efficacy. Good example for teaching Simpson’s paradox, and more broadly the importance of covariates.
Ooh, here’s a good example for my collection of statistical vignettes. Wolkovich et al. 2021 GCB show how failure to use a linearizing transformation makes it look like biological processes are becoming less sensitive to temperature change as the earth warms. Here’s a blog post with the story behind the paper. Fascinating (and depressing, and unsurprising) to learn that there was so much resistance to this paper from the reviewers.
Need a version of this for profs like me who’ve suddenly realized they need to start prepping to teach. 🙂
And finally, this is lovely:
Have a good weekend. 🙂
Bonus! I’ve linked to this a couple of times before, most recently in the comments on a post earlier this week. But it’s funny and timely so I’m going to link to it again: