As someone who
has done science and who writes about science, I know the importance of
accuracy in reporting events or results. Sometimes, as a writer, one gets
things from sources that may not be reputable or have not done enough due
diligence on their own sources. Thus, there is always the risk of that. I don’t
always double-check sources and may make mistakes at times. There is always a
risk in science itself that results may be misinterpreted, and facts
unknowingly conveyed incorrectly. Science is self-correcting by nature, always
seeking to refine understanding, typically in light of new data. There is also
the possibility of deliberate misrepresentation of science for various reasons.
Freelance science journalist
Dalmeet Singh Chawla has written about factchecking scientific journals for
Undark Magazine and Chemical & Engineering News. In Undark Magazine, Chawla
tells the story of a doctoral student at the Australian National University who
found that guidelines and standards for concentrations of lithium, potassium,
sodium, calcium, and magnesium in drinking water that were often cited as
coming from the WHO and EPA, in fact, did not come
from the WHO and EPA since they did not produce those guidelines and standards.
This is a case of incorrect information being passed on and repeatedly cited
without being true. Chala notes that the researchers found twenty papers that
cited the false numbers. The above case precedes the proliferation of
generative AI, so that is not a likely culprit. However, in some other cases,
AI is a culprit:
“In today’s world, the immediate culprit that comes to
mind is generative AI, which is widely known to make up citations when it
hallucinates. Some researchers are using AI to draft, edit, or academic papers,
which could lead to corruption in the scientific publishing process.”
Chawla also mentions the
messy and shady corruption that is paper mills, which I wrote about last year
as fraudulent scientific research. In some cases,
the factual errors can be attributed to what I was describing in the first
paragraph, such as using values without checking primary
sources, a practice that has been referred to as “cold citing.” Chawla argues
that scientific journals:
“…should hire dedicated paid fact-checkers whose remit
it is to rigorously check all claims — including references — made in academic
papers before they go live. That would be an additional round of quality
control in addition to peer review.”
Chawla thinks that many
scientific journals are profitable enough to pay factcheckers just as many
media sources do:
“In a world marred with disinformation, misinformation,
and information overload, science could and should lead the way. Science should
pride itself in providing rigorously checked factual information that has been
manually scrutinized by a human, who uses automated tools at their disposal.
This approach is especially crucial amid fears that fake science could be
becoming harder to spot.”
In C&E News, Chawla
writes about Leslie McIntosh, vice president of research integrity and security
at the scholarly analytics firm Digital Science. McIntosh believes data-driven
approaches are crucial to the scientific sleuthing she does to find fraudulent
science, and she hopes to formalize the process to some extent. Chawla calls
people like McIntosh ‘research integrity practitioners.’ Some of these
practitioners think that standards for sleuthing should be established, while
others believe it is the responsibility of the whole scientific establishment
to weed out fraud. McIntosh explains why she sleuths in addition to her
full-time job:
“Science is a pillar of democracy which needs to be
defended and strengthened.”
McIntosh favors data-driven
approaches to spot patterns across a large number of studies and calls the
practice ‘forensic scientometrics.’
“McIntosh and colleagues recently reported on a pattern
of suspicious activity that implicated hundreds of researchers. In a study
published in February on arXiv, where researchers post papers that haven’t been
peer-reviewed, the team found that more than 120 papers list the name of a
fictitious organization called the Pharmakon Neuroscience Research Network as a
funder or affiliation for at least one author. These papers were coauthored by
more than 300 authors working at 230 institutions, primarily between 2019 and
2022.”
Such fakeries should incense
all of us. Some of these compromised (fake) researchers are drawing funding
from institutions like the NIH and NSF. Some authors are unaware that they may
be involved in research that has fraudulent elements. Small groups of research
integrity sleuths are meeting up, have some funding, and are working toward
developing standards. They are working to combat the paper mills.
McIntosh notes:
“This is not just for publicity. The declaration
highlights forensic scientometrics as a distinct field that could and should
attract funding, she says, noting that most sleuths don’t get paid for flagging
nefarious activities or faulty papers.”
Chawla writes in C&E
News:
“McIntosh is also in the process of putting together a
code of ethics that lays out standards to which sleuths should adhere when
digging into papers. She says she’s developing the code in part because she
worries that sleuthing is being weaponized to serve political agendas. For
instance, a few people deliberately highlight only papers authored by
researchers of certain ethnicities, backgrounds, or genders.”
“Another aim of the sleuthing community is to shield the
scholarly literature from interference by governments, McIntosh says. “I don’t
think that it is beyond certain countries to play a long game in manipulating
what goes into the scientific literature,” she says. “I think we underestimate
science and the power of it in our society if we think that other people aren’t
trying to also control and direct where information is going.”
Some of these sleuths, like
Elizabeth Bik, examine papers one at a time, looking for fraudulent research
groups. Bik thinks that there needs to be a society for sleuths and funding for
sleuths who can receive legal threats or get sued. The goal, of course, is to
expose and filter out fraudulent papers.
Lawsuits against publishers
of fraudulent papers are becoming more common, and a recent one involved the
Dana-Farber Cancer Institute agreeing to pay a total of $15 million, most of
which will go to the NIH, which funded the research that was compromised.
Sleuths can even become whistleblowers and be rewarded through the US False
Claims Act.
Statistical manipulation and
image manipulation are common fraudulent techniques. Sleuths found that a
landmark 2006 paper on Alzheimer’s research was fraudulent, a paper later cited
by thousands of other papers. More corrections and retractions of such papers
are needed.
A September 2024 article in
Undark Magazine by Jessica Wapner gives some modern history of scientific
fraud:
“Scientific fraud has existed for as long as people have
stood to benefit from it. In the early 1980s, Harvard Medical School heart
researcher John Darsee faked data in animal research on heart attack
treatments. Beginning in the early 1990s, Japanese researcher Yoshitaka Fujii,
an anesthesiologist, fabricated more than 170 papers. And Massachusetts
anesthesiologist Scott Reuben fabricated data in at least 21 studies dating
back to the 1990s, several of which highlighted the benefits of pain
medications made by Pfizer, which had supported much of his research.”
Again, sleuths note that the
journal publishers are very profitable and have little to no incentive to
expose their own publications as purveyors of fake research. They say
publishing is favored over scrutinizing the data that is published. The sleuths
also noted that universities are often reluctant to investigate allegations of
misconduct by researchers they employ. There is a disincentive since if they
expose false research, they risk having future research defunded. With the
possibility of getting sued for defamation, sleuths can be wary at times to
call out certain researchers without very strong proof. Of course, whether
certain researchers knowingly or unknowingly passed on fraudulent research is a
factor as well. It also takes a lot of time and effort to investigate
thoroughly.
“Tim Kersjes, who leads the resolutions team within the
publisher’s {Springer Nature} research integrity department, acknowledged that
investigations can take a long time. “Ideally a concern comes in, we
investigate it, and we can retract two weeks later,” he said. “But in practice
that’s impossible.” Kersjes said that authors don’t always respond to emails or
send explanations that require further review. Marcus noted that the peer
review process isn’t geared toward catching images for signs of tampering or
other types of misconduct. Jackson, at the Journal of Clinical Investigation,
said that the fact that some scientists are willing to fake their data caught
publishers off guard.”
It is also a sticky issue
whether a paper that contains some elements of fraud should simply be corrected
or retracted. This mainly depends on whether the fraudulent elements lead to a
change of conclusions or not.
“In an email that Michael Stacey, head of communications
for journals with Springer Nature Group asked to be attributed to a
spokesperson, he wrote: “Our investigations follow an established process,
which involves consultation with the authors and, where appropriate, seeking
independent advice from peer reviewers and other external experts. Other
factors, such as awaiting the outcome of institutional investigations, where
appropriate, can also impact the length of time an investigation takes.”
Juraj Vladika and Florian
Matthes of the Department of Computer Science, Technical University of Munich,
Garching, Germany, published a paper in the Findings of the Association for
Computational Linguistics: ACL in 2023 that sought to better define scientific
factchecking and its challenges. The abstract and figure below are from that
paper, followed by an explanation of some specific challenges of scientific
factchecking vs. general factchecking.
They noted that challenges of
scientific factchecking include evidence quality, reasoning and explainability,
dataset size, external knowledge, multimodality and multilinguality, and
human-centered factchecking. For evidence quality, scrutinizing the evidence in
detail and with the most recent findings in the field of concern is needed.
Sleuths need to analyze the reasoning utilized to arrive at
conclusions and whether it is really explainable. Small datasets can be
difficult to analyze and should ideally be combined with larger existing
datasets. The complexity of scientific knowledge makes it “suitable for
representation with structures like Knowledge Graphs (KGs) that model world
knowledge in the form of entities and relations between them.” Images and
fake videos are a common means of spreading misinformation. Scientific
factchecking needs to occur in multiple languages as well. Regarding
human-centered factchecking, they note:
“Making the process of NLP-based fact-checking more
human-centered is a promising future direction that will make it more reliable,
trustworthy, and easier for wide-scale adoption.”
Factchecking in Scientific Reporting
Knight Science Journalism at
MIT runs the KSJ Fact-Checking Project. They give ten common ways in which
mistakes are made in scientific reporting. 1) Correlation ≠
Causation – this is common and needs to be called out when it
happens. 2) Numbers and Units – these simply need to be
double-checked for accuracy, since they can be misleading at times and be
entered incorrectly. 3) Absolute Risk ≠ Relative Risk –
they explain this as follows, using a study that coffee increases cancer risk
by 25%, but that 25% is relative, and the absolute risk increase is very small
and likely negligible:
“Risk communicates how likely it is that a certain
harmful event will happen. For instance, an epidemiological study, also called
an observational study, may show the likelihood that a particular material
causes cancer, while a medical study on a new drug may show how often the drug
reduces the risk of a disease.”
“But studies often report these likelihoods in terms of
relative risk, which compares two test groups. A reporter may confuse the
relative risk as absolute, or the likelihood of that event happening in any
scenario.”
4) Single Study Syndrome – if reporting cites
a single study, especially when there are many other study conclusions that
refute the conclusions in the single study. I have noticed this used by
environmentalists, overly citing a single study, or more often, a single or small
group of researchers who share their own biases. 5) Statistical
Significance and P-values – these are based on probabilities.
Statistics need to be significant to be important and real, not just possible.
Data is often manipulated to look as if it is statistically significant when it
is probably not. 6) Size Matters – they explain
this as follows:
“If a sample size is too small, for example, it won’t
reflect the larger population. If a story cites a study with a small sample
size, the story should give context: What the sample size actually was and why
it may not mean much more broadly.”
7) False
Balance – journalists often seek to be balanced in their reporting,
by giving both sides of debates a voice. However, when one party’s view is much
more plausible than the other party’s view, it conveys a false sense that it is
difficult to determine who is winning the debate when in reality the likely
winner is quite clear. We see this a lot, unfortunately. 8) Mice Aren’t
Humans – this is simply the caveat that studies with mice or other
animals do not necessarily mean that similar studies with humans would give the
same result, as we know from several situations where results differ for
humans. 9) Consider the Source – the reputation or legitimacy
of sources needs to be considered in factchecking determinations. Scientific
journals are best peer-reviewed. 10) Don’t Believe the Hype –
claims need to be contextualized, especially claims that are bold and go
against previous established understanding. As the following graphic shows,
skepticism should be the default when investigating claims.
Factchecking Scientific Claims by Political Partisans
Factcheck.org, a project of The Annenberg Public Policy Center, has a subsection called SciCheck that focuses exclusively on false and misleading scientific claims that are made by partisans to influence public policy. It was launched in January 2015. This site is now dominated by spurious claims made by Trump. Consider his claims about the reflecting pool, among many others. One cannot deny that he is often quite loose with the truth. Remember the “alternative facts” idea during his first term. Trump’s cabinet members, several of whom have little qualifications for the positions they hold, are also often fodder for factcheckers. RFK Jr., in particular, is associated with wild claims about vaccines, pesticides, and genetic engineering dangers, and several other topics. Of course, other politicians of both parties should be scrutinized as well for scientific claims that may not be true. We need to hold all politicians to the fire for any claims they make. Trump, in particular, has made some very wild claims, many of which have been proven to be untrue. Thus, most people take his claims with a grain of salt and ignore them as hubris. He has certainly gotten away with a lot. Other people would have been annihilated for saying some of the things he has said.
References:
Scientific
sleuths come in from the cold: Research integrity investigators are starting to
organize, but the field, and the people, remain idiosyncratic. Dalmeet Singh
Chawla. Chemical & Engineering News (C&EN). March 11, 2026. Scientific sleuths come in from the
cold
Opinion:
Scientific Journals Need Dedicated Fact-Checkers: An additional layer of
quality control could help academic publishers weed out problematic content
before it propagates. Dalmeet Singh Chawla. Undark Magazine. April 9, 2026. Scientific Journals Need Dedicated
Fact-Checkers
The
Rise of the Science Sleuths: When an Alzheimer’s paper came under scrutiny,
correcting the scientific record meant battling much bigger problems. Jessica
Wapner. Undark Magazine. September 11, 2024. The Rise of the Science Sleuths
Fact-Checking
in Science Reporting. In Fact-Checking 101.KSJ Fact-Checking Project. Massachusetts
Institute of Technology. Fact-Checking in Science Reporting -
KSJ Fact Checking Project
Scientific
Fact-Checking: A Survey of Resources and Approaches. Juraj Vladika and Florian
Matthes. Findings of the Association for Computational Linguistics: ACL. July, 2023.
[2305.16859] Scientific
Fact-Checking: A Survey of Resources and Approaches
SciCheck.
FactCheck.org. SciCheck Archives - FactCheck.org
No comments:
Post a Comment