Deep Dive
Shownotes Transcript
Feel your max with Brooks Running and the all-new Ghost Max 2. They're the shoes you deserve, designed to streamline your stride and help protect your body. Treat yourself to feel-good landings on an ultra-high stack of super-comfy nitrogen-infused cushion that takes the edge off every step, every day. The Brooks Ghost Max 2. You know, technically, they're a form of self-care. Brooks, let's run there. Head to brooksrunning.com to learn more.
This episode is brought to you by Shopify. Whether you're selling a little or a lot, Shopify helps you do your thing, however you cha-ching. From the launch your online shop stage, all the way to the we just hit a million orders stage. No matter what stage you're in, Shopify's there to help you grow. Sign up for a $1 per month trial period at shopify.com slash special offer, all lowercase. That's shopify.com slash special offer.
There were a lot of stories about questionable science in the news this summer. Coming up, a Harvard scholar is accused of fabricating findings in a study about honesty. There was this story in June, and then in July. The president of Stanford University is stepping down over concerns about the integrity of his research. An independent review cleared him of research misconduct, but found, quote, serious flaws in five scientific papers that he authored.
Also in July, the journal Nature published a piece about problems with randomized controlled trials, so the gold standard of scientific research, basically. This one researcher found flawed data in over 40% of the medical trials he'd studied. So all these stories in some way mentioned scientific misconduct, this thing where researchers fudge data, say, or misrepresent things.
And it made me wonder, how common is misconduct in science? So scientific misconduct is more common than you likely think it is. Much more common. This is Stephanie M. Lee. She writes for The Chronicle of Higher Education. And she investigates the problem parts of science. So the fraud and the misconduct.
And she says that over the course of her career, she's come across some numbers that paint a picture of this issue. Like one number from this blog that catalogs papers that get retracted for problems. This blog called Retraction Watch. So Retraction Watch says that nearly 5,000 retractions are happening every year now. Not all those retractions are done because of fraud, obviously. They're genuine just errors in there. But a lot of them...
are fraud. To be clear, that's 5,000 out of around 2.5 million papers that are published every year.
So in some ways, that might not seem too bad. But then Stephanie told me some other reasons to suspect that more fraud might be out there. There was a 2009 survey, for example, where researchers asked scientists about their research practices. And they asked specifically whether scientists had ever messed with their data in some way. About 2% admitted to have fabricated, falsified, or modified data at least once, which is not good.
It's kind of bad. Like, if that scales up from this small survey, 2% of scientists is a lot of scientists, like tens of thousands of scientists. And even if scientists in the survey didn't admit to actually falsifying data... Up to a third admitted doing other questionable research practices that were not making up numbers per se, but are in a gray area of research practices that are frowned upon, though maybe not explicitly...
Overall, these numbers are imperfect measures of the misconduct in science, right? Like, these scientists might not all have answered the survey honestly, or Retraction Watch can only track the papers that actually get retracted. So we will probably never know exactly how much scientific misconduct exists out there in the published literature.
But when Stephanie digs into the cases of scientific misconduct that we do know about, she sees a bigger picture here. Some major flaws in the system of science as a whole, like cracks in our defenses that misconduct slips through. She actually broke a pretty big story this summer that kind of shows those flaws. And I think that walking through it helps us understand what's going wrong with science and
but also could help us understand how to prevent misconduct from being published in the first place. This is Unexplainable. I'm Bird Pinkerton. And this week on the show, what could we do about scientific misconduct? ♪
Should I start by introducing the paper? Our story starts with a paper, a paper published in 2012 that got a lot of coverage at the time. Francesca Gino, a professor at Harvard Business School, says we are signing forms in the wrong place. Using mock tax returns, she tested whether a small change in the form's layout, moving the signature line from the bottom to the top, might produce a big change in people's behavior.
It did. That's it. Instead of signing at the bottom, we should sign at the top. It's that simple. This 2012 paper actually featured three studies run by a bunch of different co-authors from prestigious schools.
But they all boiled down to kind of one conclusion. Citing at the top of a form rather than at the bottom of the form would lead to more honest reporting throughout the form. An important thing to know about this paper is that it went through the sort of normal scientific checks. So specifically, it went through peer review, which is where scientists
Other scientists in the field look over a paper. They kind of like look through a draft to make sure it adds new information to the field. They check the study design to make sure that's sound. They look for problems in the logic of the study, right? Stuff the author might have forgotten to address. And they don't usually look at like the raw data or check the statistics or redo experiments. But the process can still take several weeks.
So this is the vetting that this paper went through. And then after peer review, it was published in a reputable place. The Proceedings of the National Academy of Sciences, which is considered to be a very prestigious journal. A recent article from The New Yorker actually shows that there were a few questions about drafts of the paper. But there was no obvious reason for journalists not to cover this finding or for people not to trust it.
And so people took the results seriously. The company that I buy my renter's insurance from brought one of the co-authors on as a consultant to tell them how to reduce fraud in their forms. Governments got interested. A government in Canada reportedly spent thousands of dollars trying to change its tax forms. It seemed like an example of science going pretty well, right? Like a research finding making a difference in the world, getting cited hundreds of times.
Except it turns out that this paper was actually pretty flawed. The wheels on the proverbial bus start to fall off in 2020. So by 2020, there had been a couple of attempts to replicate the findings of the study. Basically, people trying to see if they could get the same results again. Those attempts failed. And then the original co-authors tried to replicate the paper. And this new attempt also failed. They couldn't recreate the results.
Which does happen. Papers don't replicate all the time for reasons that don't have anything to do with deliberate fraud. There are many other reasons why a paper doesn't replicate. But this failed replication did catch some people's attention because it came with some extra information. In addition to posting their failed replication report, they also posted the original data from the studies when they were done back in 2012.
When you read a paper, you can't always access the raw data that the scientists were working with. Even peer reviewers don't always have access to the raw data. Instead, a reader will see the results in like some tables and figures of data that have already been analyzed in some way. But after this failed replication, the original data was available for anyone to see. And that data eventually came to the attention of a group called Data Collada.
DataQuad has three business school professors, Leif Nelson at UC Berkeley, Joe Simmons at the University of Pennsylvania, and Iri Simonson at Asade Business School in Spain. These guys are kind of like...
vigilante data sleuths. Since 2013, they have run this blog with a weird punny name, Data Collada, and they dig into the data of papers looking for problems, basically evidence of flawed analysis or even fraud. They basically look for results or statistical analyses that seem anomalous, and they look
take whatever data is public or provided by the author, and then they analyze it according to what they see fit. So when the authors of the 2012 paper sort of made their data public for the replication process, the Data Collada team decided to analyze it much more closely than peer reviewers normally would. They basically did like a CSI Excel on one part of the paper. And when Data Collada looked at the data in the study, they found some weird stuff.
Half of the data was in one font in the Excel spreadsheet, and the other half of the data was in another font. That seemed fishy, like maybe some data had been added at a different time. But then the data itself was also weird. Like when they plotted it out on a graph, the data collada folks thought it looked more like it had been generated by a random number generator than actually collected from people. Because it didn't look like the standard bell shape that you would expect from this sort of data was accurate.
A lot of people in the middle and then fewer people at the extremes. All in all, their post suggested that there was a serious problem with this study, that some of the data in it might have been fabricated or doctored in some way. So in 2021, they published their concerns about this one experiment.
That revelation got out into the news. So you had articles with headlines like... An influential study of dishonesty was dishonest. And... Is this psychology's most ironic research fraud? But Data Colada... They did not stop there. Over the next several years, they continued to examine this paper and also started looking at other research done by one of the paper's co-authors. So this is Francesca Gino, the Harvard Business School professor.
They wound up looking at four studies, all co-authored by Dr. Gino, and these were all peer-reviewed, all in good journals...
But they found issues with the data in all of them, evidence that they could have been tampered with. And they told Harvard about their concerns in the fall of 2021. The folks at Harvard did their own review. They wrote up a report that was over a thousand pages long. And they found evidence of doctored data and reached out to the journals that had published these studies. So at this point, all four have been retracted.
And they also put Dr. Gino on administrative leave, which means that she was stripped of a salary, her research role there of teaching at Harvard, and it was effective immediately as of mid-June.
Harvard Business School declined to comment on the situation, and Dr. Gino did not respond to my emails asking for her side of the story. But she is suing Harvard and Data Colada for defamation and gender discrimination, and she's asking for $25 million. She says, and I'll just read from her website, she says,
There is one thing I know for sure. I did not commit academic fraud. I did not manipulate data to produce a particular result. I did not falsify data to bolster any result. I did not commit the offense I am accused of, period. Gino outlines a few defenses on her website. She argues that Harvard didn't follow its usual processes in investigating her and that Data Collada didn't give her an opportunity to respond to their concerns.
She and others have taken issue with Data Cloud as methodology, and she's raised some other points. She says that generally that what makes it hard for her to prove her innocence is that a lot of the records for these studies were collected on paper, and these paper records were then transferred manually into electronic spreadsheets. Errors may have occurred in that
that process. This story is still unfolding. There will likely be more rebuttals and counter-rebuttals on the fine details of what happened here and who did what. And just to be clear, for at least that first 2012 dishonesty paper, one of the other co-authors has come under a lot of scrutiny for his involvement. So people are not just looking at Dr. Gino.
But whether or not a single academic was responsible for the anomalies in these papers, I think the more important story here is what all this tells us about science as a whole. Scientific institutions, like peer review and journal oversight, they aren't really set up to detect signs of misconduct. The main reason that issues with the data in these papers were detected was because three researchers with a blog decided to look closer.
Without Jada Kolata, we would not be having this conversation that we're having today. Their long hours of research triggered Harvard to look into this, for these papers to now start being retracted, for all this to happen. So I wondered, is this how science is supposed to work?
And Stephanie says that it is not unusual for sort of like freelance and hobbyist data sleuths to be one of the last best defenses we have against misconduct in science.
But she also says that this is not some kind of formal system. Many data sleuths, like Data Collada, are not paid for this work. This is a labor of love for them, as far as I understand. They are business school professors, all of them, and they produce their own research. But they have an interest in making sure that their field stays honest as well. So this is something that they do basically on the side. Like any vigilantes, the people at Data Collada are not universally lauded.
Some people have accused them of chasing clout, sort of saying that they are unobjective or biased towards finding mistakes. That said, there's also a case that doing this work is risky. Like, it is hard to call out your professional peers. Some data sleuths actually operate anonymously. And as we mentioned, in this particular case, Data Claw does thanks for all their work,
is a lawsuit. The lawyer they've spoken to estimate that their defense could cost anywhere between $50,000 and $600,000, depending on how far the lawsuit gets. And their employers have so far only agreed to pay part of the legal fees. This whole story made me wonder, one, why people put themselves at risk for science in this way. And two, is there a better way to look for misconduct in science?
Data Colada wouldn't speak with us, but after the break, we do speak with another data sleuth about why she does this and also about how the system ought to change. This episode is brought to you by Shopify. Whether you're selling a little or a lot.
Shopify helps you do your thing, however you cha-ching. From the launch your online shop stage, all the way to the we just hit a million orders stage. No matter what stage you're in, Shopify's there to help you grow. Sign up for a $1 per month trial period at shopify.com slash special offer, all lowercase. That's shopify.com slash special offer. If you like data, get caught.
Elizabeth Bick used to be a normal microbiologist doing normal microbiology sorts of things. And then one day, a few years ago, she was looking at some scientific papers. And specifically, she was looking at the photos in them.
And she noticed that someone had used the same photo twice in two different papers. So it was the same photo, except it was upside down. They had rotated it. And in biology papers, photos aren't just like illustrations, right? They can be part of the results of an experiment. So they can show cells doing things, for example, or blots that indicate the presence of proteins in a tissue, right?
And scientists use these photos to draw conclusions that can have important consequences if the photos are in a paper about diseases, say, or like the write-up of a drug trial. It's not normal to just kind of copy-paste a photo and use it as a result for two different experiments. And with photos of humans, an upside-down photo is pretty obvious. But with scientific photos, these blots are sort of close-ups of cells. That kind of thing is less easy to see.
Unless you're someone like Elizabeth. I think I'm suited because I've always seen duplicated bathroom tiles. So, you know, if you have bathroom tiles, they sometimes have a pattern, right? Like it looks like marble, but it's fake. Or floor planks, laminate flooring, obviously are not real wood. They're photos of wood. And so I would always spot, oh, this plank or this tile is the same as that one. So...
The upside down kind of copied blot in the second paper
made Elizabeth angry. After that first anger or being upset, I sort of got used to the idea. I just wanted to know how often does this happen? Can we publish this in a scientific way? So she took her ability to spot patterns and duplications, and she used it to do a systematic review of all the relevant images in a bunch of journals. So in the end, I looked for 20,000 papers. So to be clear, in your free time as a hobby, you looked at
20,000 papers. That amazes a lot of people. But I only looked at the images. I didn't read the papers.
So no big deal then. Like, after all, if she remembers it, she would only come home from her full-time job and do a few hours in the evenings. And then in the weekend, if I had the time off and no other obligations, I would spend perhaps a full day in total on it. So many hours for about two or three years. I did this, you know, every free hour that I had in my spare time. If you're asking yourself,
why anyone would do this. Elizabeth says that in order to get a sense of the scale of the issue and in order to get people to acknowledge it, she had to be this thorough. I felt like I'm seeing this problem
But I need to write this as a scientific paper, because if I just start, you know, yelling, this is a problem, who is going to believe me? It needs to be done in a scientific way. I need to have some idea of how many papers, what is the percentage of these papers. So she did this enormous research project. She got two other people, editors-in-chiefs of journals, to back up her findings and
And they wrote a paper with the results that was published in 2016. And I found that 4% of those papers, 4% of those 20,000, contained those duplications. For the less mathematically inclined, that's 800 of about 20,000 papers. So 800 papers that had images with duplications or photoshopping of some kind.
And Elizabeth says that for maybe half of these images, it could just be like an accident. But that's still hundreds of papers that seem to have real problems. And that frustrated me. It frustrated her so much that Elizabeth wound up becoming a full-time data sleuth. Like she now has a Patreon where people crowdfund her work. And she consults for people who think papers have data issues and mistakes.
So she is now actually getting financial compensation for her efforts to improve science and spot issues with data instead of doing this all in her free time after work or on the weekends. But despite these rewards, the risks for this work seem pretty real. So I personally have gotten...
angry emails or threats of lawsuits. And so that is a huge risk because most of us who criticize other papers for either for a living or for a hobby, we do this every
unprotected. And lawyers in the U.S. are very expensive. So, you know, I would not have the money to defend myself and neither do a lot of other people. Even if you work for a university, the university could say, well, you know, you do this work. This is not the work we pay you to do. So we're not going to protect you. Elizabeth has been doxxed before. So people have shared her personal information online to make it easier to target her offline.
She has been harassed on Twitter. And I couldn't help but wonder why, given the risks, she continues. I have thought about stopping, yes. I've never been actually involved in a real lawsuit. I've had some threats. But, I mean, maybe I'm too naive. I just thought, well, it's just the threat. You can just write me back.
you know, a letter saying you're going to sue me if you, unless you take down these blog posts. But I never have taken that perhaps as seriously as I should have. I've kept on going. And yeah, some people are going to be upset that I criticized their work. But in general, I feel a lot of scientists are supporting my work. So that helps enormously to know that other people appreciate it. So Elizabeth's continuing on, but this just does not seem sustainable.
As I mentioned, Data Colada didn't respond to my interview request, but in one of their blog posts, they wrote something that kind of summarizes the issue for me. Quote, addressing the problem of scientific fraud should not be left to a few anonymous and fed up and frightened whistleblowers and some fed up and frightened bloggers to root out.
The consequences of fraud are experienced collectively, so eliminating it should be a collective endeavor. And to me, this is also the big lesson of Stephanie Lee's reporting. Like, if we want to figure out how to reduce misconduct in science...
We can't rely on a system where people like the Data Colada team or Elizabeth Bick have to put their necks on the line. The ecosystem of science on its face seems to be set up in a way where there's multiple checks and balances, right? To make sure that things like this don't happen or they happen, you know, much less than they should. But when you examine each of those pieces separately,
and the incentives within each one, you start to see that the system is actually largely not incentivized to root out misconduct and fraud. So what are the problems with the system that we have? We'll start with peer review. This is the system we mentioned before, where papers get sent out to other people in the field.
That should be a way to evaluate the paper, but peer reviewers are not paid for their time and labor. There's no glory professionally from doing it. You just kind of feel like you're supposed to do some amount of it. So what do those conditions create? They don't create an environment where people are incentivized to spend hours and hours going over spreadsheets of data, checking fonts or looking at the distribution of car miles driven.
So ideas could be like, what if we paid peer reviewers? Or what if peer reviewing could be on your CV, for example?
Or if journals also employed statistics experts on their staffs who were giving papers like a really deep number scrub before they were published, I think that would also be helpful. We can't assume that every peer reviewer in the world is going to have Elizabeth's ability to spot patterns in photos or, you know, be as good as Data Kolata at ferreting out problems in an Excel spreadsheet. Right.
So why not have people like Data Kolata and Elizabeth Bick on staff at journals looking at papers with all the protections and salaries that come along with that? Some journals have actually started doing this, including the journal Science. And one benefit is that you could hopefully have people find errors before a paper is published instead of afterwards. But we could also work on what happens after a paper is out in the world. So journals...
Once a paper is published, they're not really incentivized to look too closely at what they've published because retractions are kind of embarrassing and, you know, make them look like they weren't doing their job in the first place. We also don't really incentivize people to do replication work to check that things are true. Instead, researchers are mostly incentivized to make sure that their papers are exciting in some way. There are very strong incentives in science to publish a lot of research, but
with attention-grabbing results in a very short amount of time because that's the pathway to tenure, to getting funding, to rising to the top of your field. So maybe science needs to shift its priorities a little bit.
And then there are suggestions like making data more public or even building new institutions. What if there was an IRS for science? And what if there was some agency, some entity that was just...
mass auditing random papers all the time. And would just the threat of that agency maybe knocking on your door one day incentivize you, the scientist, to be more honest or scrupulous, more careful with your research? You'd have to figure out who would fund something like this, but it would still have more institutional backing than professors and freelancers working on their own. If we start changing the incentives
structure that would help the field overall steer in a better direction. There needs to be a wholesale reckoning with all the parts of the system, not just one or two individual actors, though they should be held accountable as well, of course. If we make these kinds of systemic changes, we still won't know exactly how much misconduct is out there, but we might at least know that we're doing a better job of preventing it.
But when I look at a case like this 2012 dishonesty study debacle or these other retracted papers that Stephanie reported on, it all makes me a little uneasy. There are so many incentives for messing with data. And if we're relying on vigilante data sleuths to catch those problems, I start to wonder what that says about the state of science. Like, I genuinely had a moment while working on this piece where I wondered...
Should I be worried about all science? Both Stephanie and Elizabeth were pretty quick to say no. The takeaway here is not all science is fraud. We need science to solve, you know, climate change, pollution, hunger,
Pandemics. We need science in order to solve these problems. The message should not be, don't trust any science and all scientists are corrupt and crooked and fraudulent. I do trust science in general, but... We've been trusting a lot by default and not really incentivizing enough verifying. So we need to do both. Leave room for the correcting and the errors that happen every day in science.
Trust but verify? Trust but verify is a better slogan. Stephanie M. Lee is a senior reporter for the Chronicle of Higher Education. She wrote this story, and you can find her ongoing coverage over at the Chronicle of Higher Ed. I highly recommend it.
Also, a note: Data Colada was first tipped off by a researcher named Zoe Ziani, who then worked with the Data Colada team. So you can read more about her involvement and the whole process on her blog. We'll link to it in the transcript.
Kelsey Piper also has some great pieces on Vox.com about this. Look up Is It Defamation to point out scientific research fraud to read more. And just in general, there's been a lot of in-depth reporting from various outlets on this subject if you want to read more. So Planet Money has a good piece. The New Yorker and The New York Times both have pieces.
This episode was produced by me, Bird, and by Brian Resnick. Brian edited the piece along with Meredith Hodnot, who also runs our show. Noam Hassenfeld did the music. Christian Ayala did the mixing and the sound design. Serena Solon checked the facts. And truly, I cannot shout out Serena Solon enough. Fact-checking is kind of a thankless task, so here's just an extra thanks for you, Serena. And
And finally, Manning Nguyen is coming into the office more, which makes me really happy. Special thanks to Jillian Robbins for her help as well.
This podcast and all of Vox is free in part because of gifts from our readers and listeners. You can go to vox.com slash give to give today if you would like to. And if you want to bring me endless joy, please leave us a review or write us an email with your thoughts or questions. We're at unexplainable at vox.com. And really, I can't emphasize enough how much I love hearing from you all. Unexplainable is part of the Vox Media Podcast Network. And we'll be back next week.