Everyone makes mistakes.

And  scientists are no exception. Researchers and publications  usually have safeguards to prevent shoddy or outright false  information from being published.

But sometimes, bad science  slips through the cracks. In the worst cases, the paper  will get retracted, meaning that the journal no longer stands  by that paper’s conclusions and says that it should not have  been published in the first place. And that’s a good thing, because it shows us that science is self-correcting.

When mistakes happen, someone  catches them and we adapt. So let’s look back on some of the most regrettable scientific papers of all time,  and why they were retracted. [♪INTRO] The first retracted paper on our list is an oldie, but it’s had real, dangerous staying power. This is the paper that started an international vaccine scare that’s still  happening decades later.

To be clear: vaccines do not cause autism. But some people still think they do. And that claim goes back to one paper published in 1998 and retracted in 2010.

Andrew Wakefield, then a doctor  at a major London hospital, led a research project looking into the measles, mumps, and rubella, or MMR, vaccine. Wakefield’s team published a study  claiming that the onset of autism was linked with an unspecified  gut disorder, and argued that this disease, and therefore the autism,  were caused by the MMR vaccine. The study made headlines, and received immediate criticism from the scientific community.

One immediate red flag was that his  study only looked at 12 children, and all of those children were  chosen because their parents already suspected that their vaccinations  were what gave those kids autism. The entire reason they were  included in a study trying to link autism and the vaccine  was the parents’ belief that they were connected,  which is just horrible logic. The paper also gets a bunch  of stuff completely wrong.

For instance, when they talk  about gut inflammation metrics, they present the normal values  for adults, not children. We’ve barely scratched the surface  of all the wrong in this paper, but despite all these flaws,  the idea spread faster than you can say “measles outbreak”. And the real twist came in 2011,  when the British Medical Journal published a 3-part investigation  by journalist Brian Deer.

In it, Deer presented evidence that  Wakefield had been paid to conduct this study by a group of people  planning to sue the vaccine makers, but needed a medically  justified reason for doing so. That is, of course, a major ethical violation. The 12 children were subjected to  all kinds of invasive procedures for research that only existed to  make money for a bunch of lawyers.

The paper was retracted in 2010,  Wakefield lost his medical license, and his conclusions have been  disproved time and time again. A study published in 2019 of  over 650,000 children showed that the MMR vaccine had absolutely no relationship to the likelihood of developing autism. And for the record, a full  course of the MMR vaccine is 97% effective against measles,  88% effective against mumps, and 97% effective against rubella.

But this story is sadly an  example of how hard it can be to correct misinformation once it’s out there, and the long-lasting harm that bad science can do. So maybe it’s comforting that this next problematic publication isn’t life or death. It involves a totally unrealistic  and kind of terrifying rat penis.

This paper was published on February 13, 2024, and it lasted three whole  days before it was retracted. It was a review paper  summarizing existing research about a molecular signalling  pathway called JAK/STAT that’s present in sperm-producing stem cells. The review focused on studies done  on rats, which is relevant, I swear.

Because JAK/STAT is involved  in everything from cancer to autoimmune disease, it’s worth knowing about. So what was the problem? It wasn’t anything about the text of the paper.

It was the images. They haunt my dreams. Turns out the authors used generative  AI to make all of their figures, which is how the world was introduced… to rat dck.

While it’s clear that these  atrocious images have no business being in a study on anything  other than nightmare fuel, the authors arguably followed  the journal’s policy. See, while some journals ban  AI images, not all of them do. In this case, the journal said  that as long as you were honest about using them, generative  AI figures weren’t forbidden.

This brings up a more worrying  possibility that other erroneous AI-generated figures might  be present in other papers, and might fly under the radar if  they’re more subtle than this guy. At least in this case, the  errors are clear just from looking at this monstrosity,  and these figures clearly mean nothing and are too flawed to keep. So… retracted.

Rest in peace, rat dck, we barely knew you. But erroneous images in papers can be sneakier than the rat dck, with worse consequences. And in this next case, it took 20 years to gather enough evidence to bring down the faked paper.

Alzheimer’s is the most common form of dementia, and there’s been a lot of  research into why it happens, and how to prevent it. A paper published in 1992 pointed the finger at amyloid beta molecules, which build up in the brain and form plaques around nerve cells. These plaques are made up of lots of amyloid proteins of all different sizes and shapes.

But what we don’t fully  understand is why they happen. These plaques are more common  in people with Alzheimer's, but the idea was still pretty controversial, and the evidence connecting the two isn’t 100%. So when a 2006 paper identified  a specific amyloid molecule that could cause Alzheimer’s symptoms  in mice, that was a big deal.

The authors claimed to have  found a new protein called Aβ*56, which was a type of amyloid protein that built up in the brains of mice with Alzheimer’s. The authors suggested that this new protein is the reason these clusters build up in the first place. It’s like the thing in your shower drain that collects all the hair and  makes a nasty, gross clog.

And now that researchers had  a specific protein to target, they could finally explore how  to stop the plaques from forming. Or at least, they could try. Because after that 2006 team identified it, for some reason, nobody was  able to find Aβ*56 again.

Fast forward to 2022. A neuroscientist named Matthew  Schrag found evidence that the images claiming to show  Aβ*56 were heavily manipulated. The images were the results of  something called gel electrophoresis, where molecules get separated  by size into specific bands.

But in this case, it looks  like somebody copy-pasted some bands to make it look like these proteins were in the sample when they really weren’t. Does that mean the whole amyloid  beta hypothesis is debunked? Well, experts are still divided.  But it sure doesn’t look good.

The 2006 paper was retracted in 2024 with the agreement of all of the  co-authors, except for one. But that was only after it  was cited almost 2,500 times, meaning its impact on Alzheimer’s  research has been truly massive. How the field will recover remains to be seen, but hopefully with this one fraudulent paper gone, the future of Alzheimer’s  research will be a lot clearer.

We’ve got more terrible research for you, right after this ad break. This SciShow video is supported by Sendacake.com: the unique gifting experience for every occasion. Send A Cake sent us a sweet  flower shower box that plays music and shoots confetti  automatically when you open it.

A stream of flower confetti  bills rains down on you, and then you get to eat a mini bundt cake! What more could you want?! I mean, other than  to relive that experience over and over again.   And since these boxes can be reloaded and  reused, you can relive that magic moment.

If you know a mom anywhere in  the US, you can send them this   memorable Mother’s Day gift for 10% off  using the code SCISHOW at Sendacake.com. The bigger they are, the harder they fall. That’s true for this next  one, which is the story of a researcher with not one,  but eighteen retractions.

We’re talking about Brian Wansink, former head of the Food and  Brand Lab at Cornell University. He published dozens of papers about how to use psychology to change how people eat. You’ve probably heard of some of his ideas before.

Like, if you’ve ever been told not to shop hungry because you’ll buy more junk  food, that’s because of him. But that paper was retracted, so  feel free to forget that advice! This empire-toppling debunk started in 2016, when Brian Wansink wrote a blog post.

While we all have probably  posted some things we regret, this particular blog post really takes the cake. He was trying to give helpful advice to graduate students about  the “tenacity” and “hard work” it takes to get published in academia these days. But what a lot of people noticed was that his recommendations were essentially  just… do bad statistics.

See, Wansink’s lab would run their experiments and gather as much data as they possibly could, even if it wasn’t relevant to the study at hand. So, let’s say they were in a  restaurant collecting data to see if eating is affected by like,  napkin color, or something. They’d also take down lots of  other details, like how many people were in each group,  what they were drinking, their location in the restaurant, and more.

In his blog post, Wansink  recommended taking all of that data and reanalyzing it, even if it  turned out to be a totally different question than the original  experiment set out to examine. So if their results showed that  napkin color is insignificant, but that the location in the restaurant is, well, you’ve got a paper, baby! Dice up the data, re-run the  numbers, mix and match and saute and flambee until the computer spits out  a statistically significant result.

But while he presented this  as a research life hack, it’s already been around for a while. It’s called p-hacking, and it is a  big no-no in science and statistics. Here’s the thing about statistics.  There’s always a margin of error.

In fact, that’s essentially what a p-value is. In these analyses, you’re trying to see if your observed data can disprove a null hypothesis. In the napkin experiment, the null would be that napkins have nothing to do with eating choices.

So let’s say I crunch the numbers  and I get a p-value for the correlation between napkins  and food choice of p=0.45. That means there’s a 45% chance that my data are consistent with the null,  and in academic research, that means it’s not a  statistically significant result. But if I swap table location for napkins, I might get a p-value of 0.05, which most people would call  statistically significant.

But remember, that means the chances of the null explaining the dataset  are 5%, which isn’t nothing. The more tests you do, the more likely  it is you’ll get a false positive. Readers of Wansink’s blog quickly pointed out these shady practices in the comments, which led people to dig even  further into his research methods.

Aside from the p-hacking,  they found inconsistencies in his datasets across multiple  papers, among other issues. While Wansink denies there was any fraud, he did step down from his position  at Cornell and retired in 2019, after 6 of his papers were retracted in one day. So in the end, he got his just desserts.

Our next example is actually a  great example of how a field can change their conduct in response  to super big challenges. In case that pun wasn’t clear,  we’re talking about superconductors. A conductor is a material that  electricity easily flows through.

Electricity moves more easily  through some materials than others, which is why we make wires out  of copper and not, like, wood. The holy grail in materials research  right now is a superconductor: something with zero resistance. And even better would be one  that worked at room temperature.

This could enable improvements  in everything from maglev trains to batteries, and could  even lead to magnet-powered nuclear fusion, and quantum computers. So far, we’ve only been able  to create superconductors either at low temperatures, or high pressures. Like, millions of times higher  than atmospheric pressure.

But in 2023, a research team claimed to have made superconducting  materials at room temperature. They claimed that they made  nitrogen-doped lutetium hydride, which means they added nitrogen  impurities to this crystal of lutetium and hydrogen, to make  it better at carrying a charge. Their data showed that the material  was acting like a superconductor at about 21 degrees Celsius and  just under 10,000 atmospheres of pressure, which is still a lot, but it’s way better than millions of  atmospheres it used to take.

But in November 2023, the paper was retracted after an investigation revealed  that the lab fabricated data, falsified the results, and  committed full-on plagiarism. And here’s the happy ending. In 2024, physicists held an entire conference just to discuss retractions  in superconductor research, and how to prevent more problematic work.

Physicists and folks from research  journals gathered and brainstormed some possible guidelines to make  sure research is reproducible. That’s exactly what it means for  science to be self-correcting, and I think that’s pretty super. And we can’t leave out the social sciences.

One huge question in the world  of political science is how you can change people’s minds  about divisive political issues. A 2014 paper by graduate researcher  Michael LaCour and his advisor, Donald Green - no relation -  set out to provide some answers. They wanted to know if door to door canvassing could convince people to change  their views on gay rights.

Both gay and straight canvassers held 20-minute conversations with strangers in Los Angeles County to try to get  their support for gay marriage. According to the paper, the  researchers conducted follow-up surveys 3 weeks, 6 weeks, and 9 weeks after the strangers’ first  conversation with the canvassers. They concluded that only the gay canvassers were able to convince people to change their minds and support gay marriage, even in the long-term.

This was a big result, and the  paper was published in Science. Naturally, some other  graduate students wanted to do a follow-up study to expand the research, so they asked the original  authors for some guidance. And as they started digging  deeper, things got fishy.

The survey company that LaCour said he worked with didn’t recognize him,  and didn’t have the ability to do the followup surveys he claimed to have run. This led the grad student  sleuths to suspect that the initial study lifted their  long-term results from something called the Cooperative Campaign Analysis Project, which used nearly identical  metrics in a national population. So, they brought their suspicions to Donald Green.

And when Green asked LaCour for his original data, LaCour said that he accidentally deleted it all. Believing this was a lie, Green wrote  to Science to ask for a retraction. But that’s not the end of the story.

Those intrepid grad student researchers ended up running their own version of the study. This time, they directed  canvassers to talk to people in South Florida about transgender  rights and transphobia. Like the original study  design, some of the canvassers were transgender and some were cisgender.

But this time, they really did  do all their follow-up surveys. They found that both transgender  and cisgender canvassers could reduce transphobia  through those conversations, and the effect persisted  long after that conversation. So even though the original  study was faked, the real, follow-up study showed something more hopeful: that anyone can help share  information and change minds through productive conversations,  even with strangers.

It’s tough to look back and  see the harm that some of these papers have done, and the silliness that occasionally gets through peer review. But these papers were retracted,  which means the system is working! The bad science gets called out  and cleared away, and that makes room for all sorts of fantastic, new  research that can bring us forward.

Let’s just try to avoid any images of  giant rat penises in the future, OK? [♪OUTRO]