Retraction Watch links to this news article by James Heathers, who writes:
It’s not that, under such conditions [coronavirus], a few bad studies were bound to slip through the net. Rather, there is no net. Peer review, especially when conducted at pandemic speed, does not exert the rather boring scientific scrutiny needed to identify the problems [of data quality] described above. . . .
Yet it has not yet sunk in to the public consciousness that our system for building biomedical knowledge largely ignores any evidence of widespread misconduct. In other words, the literature on ivermectin may be quite bad—and in being so, it may also be quite unremarkable.
If this is the case, how does medical science manage to navigate all the bad research? How have we not returned to the ages of leeches and bloodletting?
The secret, again, is simple: Much research is simply ignored by other scientists because it either looks “off” or is published in the wrong place. A huge gray literature exists in parallel to reliable clinical research, including work published in low-quality or outright predatory journals that will publish almost anything for money. Likewise, the authors of fabricated or heavily distorted papers tend to have modest ambitions: The point is to get their work in print and added to their CV, not to make waves. We often say these studies are designed to be “written but not read.” . . .
In a pandemic, when the stakes are highest, the somewhat porous boundary between these publication worlds has all but disappeared. There is no gray literature now: Everything is a magnet for immediate attention and misunderstanding. . . . An instantly forgettable preprint, which would once have been read by only a few pedantic experts, can now be widely shared among hundreds of thousands on social media. . . .
Just one thing. I like Heathers’s article, but it may be taken to imply that the problem here is with scientific lowlifes, various nobodies at Nowhere U. who are flooding the junk journals with junk papers which then get picked up by the malicious and the credulous—the knaves and the fools—and enter our public discourse. But it’s not just the lowlifes. We’ve seen shaky studies coming out of top institutions. Remember the Stanford crew? And, much worse, the Surgisphere study, which was laundered through Harvard and appeared in Lancet and New England Journal of Medicine? You can’t get much more blue-chip than that.
So, no, it’s not just the “gray literature” that’s causing problems. It’s a bit too late for a Halloween reference, but . . . some of the calls are coming from inside the house.
For covid, this is all new. But a few years ago we were talking about junk science in social psychology, and a few years before that we were talking about junk science in neuroimaging, and a few years before that we were talking about junk science in evolutionary biology. Before that maybe we were ok, or maybe the junk science was just happening under our radar.
OK, that’s all throat-clearing. Here’s the problem I want to talk about. How can we do science, and science communication, in an era of too much science? It’s the spam problem, right?
Until now we’ve much handled the problem by a mix of ignoring it—doing our work and not worrying about the bad stuff—or fighting hopeless battle after battle. As Heathers points out, it can take a huge amount of work to get a paper retracted, even before accounting for the efforts taken by authors who want to protect their ill-gotten reputations. That sleep guy is still around, so is the voodoo guy, the gremlins guy, and most of the rest of them. Junk science is hydra-headed, and the only good reason for us to keep hacking at the new heads that pop up is that the process of inquiry gives us some insight into what’s going wrong.
So what to do?
I’m still thinking that the effort the scientific community puts into pre-publication review could be more usefully spend on post-publication review. It’s the usual argument: pre-publication review is done on every paper; post-publication review could focus on the papers that people care about.
But Heathers points out that pre-publication review typically doesn’t address data quality problems; instead it’s mostly about ensuring the paper has sufficient novelty, that it appropriately cites the literature, and that its results are presented clearly. That’s all fine, but none of these has much to do with the paper’s scientific claims being true or even being consistent with its data.
Science reformers have come up with lots of ideas, but I’n not sure they’re clear on the big picture of how to make this system work.
It’s also not clear how much of this is just a result of political polarization. Back in the 70s there were lots of people who believed in fake cancer cures, astrology, etc., but aside from the occasional feature article, this stuff was pretty much sidelined. Even though various pseudosciences maybe were believed by something close to half of Americans, they weren’t really taken seriously by many people who mattered. It was sort of a gray world, as Heathers might put it, that was left on its own. But now you get vaccine denialists on the partisan news media. Imagine if we had the science environment of 2021 but the media environment of 1975. Maybe then ABC News would hire James Heathers and Nick Brown as special correspondents?
P.S. One of my co-bloggers wants us to have more posts with math and fewer posts about academic malfeasance. I told him, no problem: I post at a pretty constant rate of one per day so if he and the others want to add more math posts, they can make the ratio go as high as they want. Academic malfeasance falls in the Social Science category of Statistical Modeling, Causal Inference, and Social Science.
1) Amp-up – as opposed to the current direction of toning down – the scientific and mathematical rigor across the board at US schools. It’s OK if people flunk out. They can work on the GM assembly line or run eBay stores instead of write newspapers or become social scientists
2) Stop giving journalists a science pass. In a scientific society, we need journalists up to the job.
Could one of the preeminent statistical associations create a certification for journals that audited the data quality capabilities and data review processes of the journal? If the journal had the appropriate capabilities and processes in place, they would receive the certification. Seems to me that the issue of “junk science” is fundamentally an issue of the relevant gate keeper, journals in this case, not valuing the gold standard in statistical methods and/or having the correct processes in place to make sure the papers that they receive are reviewed appropriately. A certification process could help solve this problem.
The quality of the pandemic research has been pretty standard in my opinion.
As usual, the solution to low quality and fraud is regular direct replication.
Then once we have reluable results, the solution to poor interpretation is testing otherwise surprising predictions.
> The quality of the pandemic research has been pretty standard in my opinion.
What is described in the video I linked below isn’t “pretty standard,” IMO.
It’s not a completely new phenomenon, and it’s part of a contiuum, IMO, but I do think the landscape has significantly changed looking pre-pandemic to now. IMO, the pandemic has provided a mechanism to consolidate and channel existing trends (the politicization of scientific research, the explosion in podcasts and social-media influenced forms of informaiton promotion, etc), into a tighter and more influential process.
In terms of overall effect I’m not sure how it might compare to something like the promotion of misinformation when Johnson lied about Vietnam. I’ve read some reasonable arguments recently that question whether “misinformation” is more prevalent or influential than in the past. But I do think that within the domain of science and research in particular, for all the benefits of enlarging public access to more information, the potential means for promoting and weaponizing misnformation isn’t well-described as “pretty standard.”
Medical research has been primarily people generating piles of conflicting results and following fads until they fizzle out for awhile.
Eg, the low sodium diet to reduce blood pressure fad is starting to fizzle: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7832857/
Framingham study is just an observational one. It is similar to the studies of Yanomami people of Amazon whose intake of sodium is minimal, but they live without modern diseases.
I would take all that with a grain of salt (pun intended). Agree with most of medical research being way to noisy going in all directions though.
They should have checked the BP of an anthropologist who lived with the tribe and shared their diet.
This is an interesting video (by Greg Tucker-Kellog, who has commented here) to show, in pretty much real time, how the promotion of bad science landscape has changed with the advent of social media.
https://www.youtube.com/watch?v=oRkNU1FlQGE&feature=youtu.be
Basically the story – a young researcher did some research which obstensibly shows ivermectin to be extremely efficacious, particularly in comparison to (a more expensive treatment) remdesivir. He wrote it up with co-authors and submitted it to revview. The reviewers pointed out fatal flaws in the research (uncontrolled confounders) and after consideration, the authors decided to not continue with the investigation.
Subseuently, an abstract of the research was published for a conference proceeding.
That abstract was picked up by John Campbell (complete with plausible deniability promtion of conspiracy ideation with grandfatherly concern to his over 2 million Youtube subscribers), Pierre Kory (probably the leading advocate for ivermectin), Jordan Peterson (darling of the IDW crowd with a massive audience), the Epoch Times (an anti-China mouthpiece that has been a big source of anti-vax material), ivnmeta (a website that reviews literature on ivermectin from an advocacy perspective), etc.
The lead author has come out and explained, clearly, that his abstract should NOT be considered as substantive evidence for the efficacy of ivermectin, that he believes ivermecting ot be inefective.
Obviously, peer review and the existing process of research publications has significant problems. But really, I hope that people really step back a bit from the replicability “crisis” to consider what the full range alternatives looks like
…sorry, that should be ivmmeta….
I’m coming around to the view that social media influencers are weaponising ignorance not just of science, but of science communication. Both Campbell and Mobeen eventual “apologised”, but begrudgingly, and shifted the blame to the authors for the abstract instead of accepting their own responsibility as scientific communicators/educators. They basically pretend not to know that an isolated abstract is no basis for anything https://youtu.be/biv6N25c9Zo
Did you know that having dogs in your house can lead to higher chances of getting covid?
“Study 1 shows that people in states with a higher percentage of dog (cat) owners search more promotion- (prevention-) focused words and report a higher COVID-19 transmission rate.”
https://journals.sagepub.com/doi/10.1177/00222429221078036
I believe the reason for a lot of bad covid science (by this I mean exaggeration or dichotomizing of effects) that bypasses the radar is general lack of familiarity with research methods in some fields of Biology. It’s quite easy to spot BS in a social Psych. paper because we can all relate to its methods (we all know what it means to survey a bunch of undergrads using a ‘validated’ ‘instrument’).
OTOH, reading and trying to interpret results from ELISA, X-ray crystallography, cryogenic microscopy, etc. requires some level of familiarity with those and many other methods, regardless of one’s level of statistical expertise.
Just look at many claims regarding ‘immunity’ (as if it was one thing). Geometric mean titer is used as a proxy for protection. Exaggerated claims of the booster benefits and undermining the effect of good masks or anything that is not a vaccine is rampant. Never mind that recent up and downs in hospitalizations have no connection to the vaccination levels or anything else. How about the whole concept of a ‘case’, the most meaningless term used in all this mess. To make one thing look semi-effective, it’s easiest to make all the other things look really ineffective. Not a bad strategy.
p.s. I hope I’m wrong about all of the above.
Lots of good points — especially, “OTOH, reading and trying to interpret results from ELISA, X-ray crystallography, cryogenic microscopy, etc. requires some level of familiarity with those and many other methods, regardless of one’s level of statistical expertise.”
My own experience is that both scientists (in their particular fields) and statistically involved people in other fields need to communicate more — and that both groups need to be pro-active in getting together. I was very fortunate that when I was first getting interested in statistics (my formal education and my research are as a mathematician), there was a biologist in my university who actively encouraged mathematicians to attend his faculty/research-student discussion group. As a result of my participating his discussion group, I ended up being on the Ph.D. dissertation committees of several of his students. It was very interesting for me, and I think my participation helped improve the work of some of his students.
I think there are a couple of big problems with the review and publishing process.
Separating the vetting (editorial decision) from the review by peers is definitely a positive development. What Andrew highlights here (and I would love to have references listed here) is the concern that the pre-publication review is of “lower quality”. I think we are witnessing the survivorship bias – in the conventional process the rejected papers do not get reviewed at all, so we don’t see at least half of the reviews. They are just not written. The second concern is that the lion’s share of reviews is actually done by a handful of scientists (Tennant & Ross-Hellauer, 2020). It’s not enough to just open the scope of scientific work which could be reviewed (by including preprints), it is also important to increase the pool of reviewers. The threat of retaliation by senior colleagues is still holding back many junior reviewers from entering the peer-review field. If the peer-review process is separated from the vetting process and if the identity of peer reviewers is protected from retaliation by colleagues, perhaps we could scale up the review process and address the quality dilemma by the increased number of reviews.
Second issue: post-publication review is a thing and should be done more often. It is partially addressed by the replication papers, but many replication failures could be avoided if journals adopted a simple and transparent replication policy. Peer Reviewer’s Openness Initiative (Moorey et al 2016) is one example of how it can be done. Journals that fail to demand their authors to publish data and code should be publicly reprimanded and put pressure on. Those who do not quality control the research are actively participating in degrading the quality and reputation of science.
These slides from my talk last year contain a lot more references and discussion of the open peer-review process: bit.ly/CEC-OPR
REFERENCE:
Morey RD, Chambers CD, Etchells PJ, Harris CR, Hoekstra R, Lakens D, Lewandowsky S, Morey CC, Newman DP, Schönbrodt FD, et al. The Peer Reviewers’ Openness Initiative: incentivizing open research practices through peer review. Royal Society Open Science. 3(1):150547. https://doi.org/10/gf4293
Tennant JP, Ross-Hellauer T. 2020. The limitations to our understanding of peer review. Res Integr Peer Rev. 5(1):6. https://doi.org/10/ggth5b
Good points.
Good comment, will check out your slides!
Andrew — your comments at the end were not good. They were great.
This topic, for how important it is, is nowhere near discussed enough. We have some scientists trying to contain the fire in an attempt to limit the spread. Others are fighting it and trying to reclaim lost ground with more robust open science, New Statistics, etc. That’s all great, but the fools in the other room playing with matches to pad their CVs need to be sanctioned — hard.
We live in a bizarre world where plagiarism can run your career before it starts, but almost nothing can nuke your career from orbit once a few journals print your crappy papers. We’ve lost our collective minds.
Andrew, it seems that if this is a spam problem we should cut the incentive to produce spam :) This means a serious rethink of how tenure and promotion work no?
Chris:
Maybe so. I’m not the first to note that the academic publishing system is really weird. It’s a world where authors (including me!) send articles to journals for free, and then the journals make money selling subscriptions. As a participant in this system, I understand the incentives. It’s not just tenure and promotion (which have not affected me for decades); it’s also that I have some idea that if I publish an article in a journal rather than just posting it on my website, the article will get more readers or be taken more seriously.
Standardized testing! :)
These days I would settle for a review of computational reproducibility on a lot of papers. That is, if someone can get the same numbers out of the dataset in a re-analysis of the exact procedures the authors claimed to use.
Heather’s and colleagues focus on basic mathematical consistencies in papers is appealing in the sense that that those sorts of problems should be uncontroversial (e.g., the P value for a t test of 1.1 isn’t 0.001 or whatever). So you can’t just spin it wlaway with words.
Stuff can be wrong for lots of reasons, but it feels like passing a computational reproducibility check should be a reasonable low bar to cross. We don’t even do that for almost any paper!
A requirement to submit both your data and code should be non-negotiable. I think many of us would be surprised, though, how much “and then I copy-pasted it into Excel and smeared the data around” happens.
Even in the case where I lack the language expertise to parse their research steps (let’s say they’re using a tool that both sucks and reads like you’re casting spells – like SAS), the bare minimum of “my analysis is reproducible” can at least cross that off the list of potential sources of error.
That’s just a no-go for a lot of fields. Thinking about epidemiology, sometimes the hoops we have to jump through regarding ethics and data security to just access the data as the main researcher are massive. There’s no way we’d be allowed to send it to a reviewer/journal on a non-secure network. You could submit some code, and fake data, but it wouldn’t be the same code and data (the code might well require a major rewrite depending on how it was written to interface with the secure environment, for example).
I’m not saying that it shouldn’t be done where it can, but having it as an absolute requirement isn’t realistic.
This new institute is working on computational reproducibility (in addition to robustness) for economics and political science:
https://i4replication.org/definitions.html
Why does this post remind me of how medical science treated handwashing as pseudoscience, because how can what you can’t see hurt you?
《Despite various publications of results where hand washing reduced mortality to below 1%, Semmelweis’s observations conflicted with the established scientific and medical opinions of the time and his ideas were rejected by the medical community. He could offer no acceptable scientific explanation for his findings, and some doctors were offended at the suggestion that they should wash their hands and mocked him for it.》(Wikipedia article on Ignaz Semmelweis)
Andrew wrote: 《Even though various pseudosciences maybe were believed by something close to half of Americans, they weren’t really taken seriously by many people who mattered.》
How many of us realize that animals have personalities, curiosity, intelligence, and rights, while wet labs continue experiments willy-nilly?
The problem is also one of science and science communication in an era of too much communication. By that I mean (mostly) social media, but I include platforms like YouTube, Medium, Substack and others that lend themselves to amplification and self-appointed authority. The most influential professional science and medical communicators are on those platforms, whether or not they’ve ever given any consideration to responsible professional standards for science communication.