It's also clear that the data fabrication happened in Excel (based on styling differences that wouldn't pass through other formats) and the Excel metadata has Ariely as the creator of the file. If the data had been fabricated by the insurance company as an Excel file, they'd show up as the creators. If it had been fabricated by them as a e.g. a CSV and Ariely just imported it to Excel, it wouldn't show the inconsistent styling.
It's easy to imagine reasons. A lazy assistant who's supposed to go through 13,488 filled-out forms decides to count 1,000 of them and then fabricate the rest so they can spend the week watching soccer online instead? Or an employee lost the data on which form results were associated with which condition, and they might get fired or not get promoted if caught, so they made it up? And so forth.
And metadata as you've described it doesn't sem like a smoking gun either. It's easy enough to imagine Ariely created an otherwise empty Excel sheet in the column format he wanted them to insert data into, e-mailed it to them, and got it back with their results. Or he just copied their Excel results into a spreadsheet of his own that already had other tabs, or similar.
I'm not saying Ariely didn't do it or did do it, I'm just wondering if there's an actual smoking gun here, or if it's accusations but there isn't enough evidence to actually know.
He also was the only one handling it, so if the data is bad then the responsibility falls to him regardless, both as the handler, as well as the head of the investigation. He either deliberately faked it, or got garbage data and didn't realize it when, now, plenty of others have.
Clearly he had control over the file, but I wouldn't put too much weight on the metadata when there's plenty of other evidence.
> If the data had been fabricated by the insurance company as an Excel file, they'd show up as the creators.
Digital forensics does not work this way. It is very rare when many digital evidence line up and imply that an actual person did or did not do something in the digital realm. A user account that shows up in an Excel file is absolutely insufficient to tie a person to the keyboard.
> it wouldn't show the inconsistent styling.
This is also inconsistent how digital forensics works. It can imply that something changed, but nothing about who, when, and why. It could have been an incorrectly propagated filter, or coloring, template, or functions.
> This is also inconsistent how digital forensics works. It can imply that something changed, but nothing about who, when, and why. It could have been an incorrectly propagated filter, or coloring, template, or functions.
No, sorry, I don't agree. It shows the "when" of the data fabrication, i.e. after the data was imported to Excel not before. The nature of the styling inconsistencies shows the "why", since it shows that the data was duplicated and one half of the duplicates was manipulated. Half the data was in one font, half in the other, and the two sets of data can be matched with each other. It also shows the "who" in combination with the metadata, due to the chain of reasoning in my previous message, since it shows that the data fabrication happened after the import to Excel and we know Ariely is the person who did that import.
> it is impossible to tell from the data who fabricated it. But because the fourth author has made it clear to us that he was the only author in touch with the insurance company, there are three logical possibilities: the fourth author himself, someone in the fourth author's lab, or someone at the insurance company.
So you seem to be much more confident about it only being one possibility, when the link explicity provides three possibilities.
One of those logical possibilities can be excluded entirely thanks to everyone involved agreeing that Ariely was the only person in contact with the company and handling the data ingestion. It could not have been anybody else in the lab.
The other theory is beyond absurd. Not only did the insurance company fabricate the data, but they happened to fabricate results that showed the effect Ariely needed for a high profile publication. Theories like your "slacking off data entry clerk" just don't make sense there.
Now, maybe if Ariely could provide even a shred of evidence for this being the data he received, I could change my mind. But he can't produce the file he received from the company. He can't tell who at the company sent it to him. The trail just runs cold at Ariely.
He'd already done lab experiments that showed the effect he thought existed, and it's easy to imagine an associate who actually fabricated the data already knew what he was looking to confirm.
Again, I'm not saying I think he's guilty or not. I'm just saying that from everything I've read so far, it seems plausible it was sloppiness and not checking data for signs of manipulation. That this sloppiness isn't "beyond absurd", but rather more like the classic "never attribute to malice that which is adequately explained by stupidity" (or by an overworked/overambitious academic).
Based on Bazerman's response, actually, he hadn't:
> [The fradulent paper] came together in a merger of two prior non-published empirical efforts. Mazar-Ariely independently provided the data for Study 3, while Shu-Gino-Bazerman had written a paper containing two laboratory experiments (Studies 1 and 2). The Shu-Gino-Bazerman group knew of the Mazar-Ariely data from multiple public presentations by Ariely.
Mazar's response is basically "oh god, I'm ashamed this happened. I played no role in the data collection, and I don't know who did. But thank you for all the hard work you've done in exposing this fraud, and I should have done better at questioning the data."
Gino's response (for context, she has been implicated in other fabricated data studies) is basically "I'm ashamed. I didn't collect the data, and I trusted the people who did collect it. But thank you for all the hard work you've done in exposing this fraud, I should have done better, and this is what I'm doing in the future to do better."
Ariely's response is "I'm the only one of the authors who had the data. Also, it wasn't me."
Bazerman's response is "Thank you for exposing this fraud. Let me explain how the study came to be, and the history of my own doubts about the quality of this study [which includes two episodes which seem to be red flags]. I should have pushed harder on those doubts [he was in the minority who wanted to retract the paper after nonreplication in the follow up]."
Ariely's response to me feels like raising a preemptive defense against being thrown under the bus. Neither Mazar nor Gino implicate anybody (only disclaim responsibility themselves), whereas Bazerman suggests that it would have had to have been Ariely or Mazar. (There's an interesting bit where Bazerman has an unspecified issue with the [now known to be faked] data prepublication, where one coauthor has unconvincing responses but a later coauthor was able to provide plausible explanations, complete with reference to the underlying data file--but both coauthors are unnamed). Given that, I just don't find the other possibilities all that plausible (and in the case of the middle possibility, it's not exactly exonerative anyways).
Specific example: In the car insurance paper, which was covered by Data Colada, there was agreement among all the four authors that Ariely himself was the only person to touch the data. That data (IIRC) was completely fabricated.
I’m happy to edit this comment and my original one if those details are wrong. I’ll give Ariely a presumption of innocence which he does not deserve and indeed which I suspect he abuses in research integrity investigations.
(Edit) data colada link: http://datacolada.org/98
At the end under “replies” Dan says himself that he was the person in touch with the insurer. It really beggars belief to imagine that there was someone at the insurer who fabricated a dataset which perfectly tells the BS story of this paper.
Also note that another of the four authors (Gino) will soon be fired from HBS for fabricating data in a series of studies.
It is just remarkable to see how people with enough knowledge of datascience to perpetrate this fraud for so long would also make errors as rudimentary as producing a faked 'random' data set in a uniform distribution. And with zero cars reporting over 50k miles driven. (edit: and so many others...this is worth reading just to see it all, thanks for linking).
These frauds really are not as smart as they think they are. But I guess selling best-selling books about stuff you made up could make you believe you're a genius.
You would imagine a con artist of his caliber would be more careful than that, but still ...
They've been doing it for a long time and it's always worked, either because they've always been this sloppy but everyone else reviewing was sloppier or gave the benefit of the doubt for reasons, or that they've put less and less effort into it over time because it hasn't been necessary.
Also the simpler the fraud is in execution the more likely you might be able to explain it away as a simple mistake or lack of familiarity with certain "tradesperson" tools adjacent to the skills you have framed in latin blackletter on sheepskin.
Which points towards either him fabricating it himself, or asking somebody else to give him results without going through normal procedures for registering/conducting the experiment and then naming someone else as responsible for it...
I do think there are cases where academics will have unwittingly published falsified or partly false data and sometimes even garnered fame and citations for it x, and no doubt many more where the academics spot issues with the data they've received and bin it, but this doesn't look like one of them.