Slashdot Mirror
Why Most Published Research Findings Are False
Hugh Pickens writes "Researchers have found that the winner's curse may apply to the publication of scientific papers and that incorrect findings are more likely to end up in print than correct findings. Dr John Ioannidis bases his argument about incorrect research partly on a study of 49 papers on the effectiveness of medical interventions published in leading journals that had been cited by more than 1,000 other scientists, and his finding that, within only a few years, almost a third of the papers had been refuted by other studies. Ioannidis argues that scientific research is so difficult — the sample sizes must be big and the analysis rigorous — that most research may end up being wrong, and the 'hotter' the field, the greater the competition is, and the more likely that published research in top journals could be wrong. Another study earlier this year found that among the studies submitted to the FDA about the effectiveness of antidepressants, almost all of those with positive results were published, whereas very few of those with negative results saw print, although negative results are potentially just as informative as positive (if less exciting)."
Title is wrong. It says that the FDA is corrupt. And that published papers take around 3years to get peer reviewed where the bad ones are removed. What a blatent attack on science generally. Sure paper publishing needs to be reviewed but 'most published research is false' is an outright LIE. 'Most published research' includes all of our basis of scientific knowledge. If most of our theories on biology were wrong really we realistically wouldnt have been able to move forwards into working with genes if we didnt know what a cell did.
Basic idea: high-profile journals want papers that are new and exciting. This means that scientists have an incentive to 1) rush their work, 2) choose fields that are popular, and 3) claim that their papers solve more than they actually do. This leads to sloppy, dishonest papers.
I'm not going to judge this paper - I haven't read it thoroughly - but to pair a title like "Why most published research findings are false" to a pretty well-known problem seems itself like an example of problem 3!
The idea behind this is that pharmaceutical studies are difficult and expensive to perform, so they rarely get challenged for some time, and when they do, the challengers are more often in more obscure journals: so when people go to cite statistics and findings, they don't notice the fact that what they're quoting has been invalidated. Climate change has been studied over and over again and subjected to extensive analysis by many minds for many years, particularly because so many people have questioned it. To suggest what you are saying now is a little behind the times.
Bio questions? Ask me to start a Q&A journal. Computer analogies available for most topics!
My prediction for this thread:
Several people will post about how this validates the TINY, TINY, TINY, number of scientists and LARGE number of completely uneducated "opinion" formers and MASSIVE number of people who think that "belief" is the same as fact.
What they will miss:
This article talks about how things are put out there then invalidated by SUBSEQUENT PUBLISHED RESEARCH, not about how there is a great conspiracy around something being "right" and everyone shouting down those who dare to disagree. Global Warming is something that has consistently been found to be happening and while certain bits have been revised due to subsequent research, most of that research has found that previously incorrect models were in fact too optimistic in their view.
This article doesn't strengthen your misguided, and uneducated, belief that Global Warming isn't real. When even the Republican candidate says its real then its time to let go and become part of the solution.
An Eye for an Eye will make the whole world blind - Gandhi
I fail to see how you can draw any conclusions about the reliability of atmospheric physics papers from a study of biomedical research papers.
I came here for a good argument
Many are wrong. This is pretty obvious, and it's also why science works. Eventually, the wrong ones will be replaced with something less wrong, and so on.
SJW n. One who posts facts.
Peer review can both help and hinder - there's the reputation effect of guest authorship where having a well-known, senior, academic's name on the paper helps it through no matter how absurd the findings.
Then there are reviewers who review papers they do not have the expertise to review. And to be frank I've seen some pretty bloody ludicrous comments from supposedly expert reviewers - the sort of stuff 1st year students wouldn't make.
But I do think that the majority of researchers are dilligent and beleive in what they submit. And lets face it - if it is an emerging area and you have a neat result that either refutes someone else's grand theory or is just really novel you're going to want to see that in print. It is because we seek to replicate research that findings are later falsified. This isn't evidence that the system is broke it is pricesly how it should work. It is the work that can't be falsified that stands the test of time and contributes to our knowledge.
If there are people who think that falsifying published research is somehow a bad thing - that is shows there's a problem in research standards - the they really really need to go back to school and read some Karl Popper.
How true that is.
The significant other is quitting grad school as soon as she gets her Master's in Neruoscience(she's in the PhD/Master Program). She can't stand the constant pressure of publishing nor the need constantly justify grant writing. She's not the best researcher, but the pressure is enough to drive her to not caring anymore. She'll get her consolation prize and get on with her life.
Maybe she's just not cut out for academia, though it's losing out on the great potential she has.
import system.cool.Sig;
Scientific research is just that -- research. If it were as easy as doing a couple of experiments, revealing the "truth" and moving on to the next thing, we'd all be living around Alpha Centauri by now. But science is hard and therefore a lot of conclusions are naturally going to be wrong. If that weren't the case then we wouldn't even need any scientific journals -- all we'd need would be newspapers.
Remember the whole "theory of evolution" issue that the creationists keep harping on? "They call it a theory so it must not really be true?" We all know that evolution is just about as "true" as any science gets -- and yet surely there are some portions of the current body of knowledge about evolution that will one day be falsified by later research. That's not a bad thing.
Notable research that has since been thought to be flawed or insufficient: Newtonian physics. Niels Bohr's model of the atom. Gregor Mendel's research into genetics. Einstein's theory of general relativity. Koch's postulates for determining disease causation. Quantum mechanics. And so on.
Breakfast served all day!
Even after reading the article, I'm still not sure if the authors are saying:
A) Given that research has been published, it is more likely to be false than not; or
B) Given that research is false, it is more likely to be published than is the case for true research.
I mean, it says:
So, (Wrong Articles)/(Total Articles) = >=0.5, right?
But the only figures I can find in the same article are:
So.. "most" is now "less than one third"?
I'm somewhat alarmed that The Economist lets people who don't seem to grasp basic statistics write their articles.
Antiquis temporibus, nati tibi similes in rupibus ventosissimis exponebantur ad necem.
Yes I agree - this paper is not about "Most Published Papers" in Science. It is about published papers in the area of therapy effectiveness. Especially those where we do not have a good model. Thus of course about half should be wrong I would guess, as established by later studies. This is statistics in action. When you are looking for high correlations and selecting for the positive, you will will get false ones. As long as this paper's authors could find LATER PUBLISHED RESEARCH showing this stuff was wrong, that is the scientific method. In fact if, say, 98 and 4/100's of papers were shown to be right later, I would smell something in the woodpile.
The meat of the article is the bias about reporting negative results. This is not a secret.
In regard to something like climate research, really it doesn't apply. but if you take the premise, it would generally bolster the 1000's (+) of papers over the years that show consistent effects and generally put more shadow on the couple showing otherwise. It would mean the papers the doubters bring up are wrong with even more percentage, since these are the papers with no validated mechanisms and generally many defects which immediately get pointed out. That is we would expect some wrong or null correlations to pop-up, given this paper and shouldn't put much support to isolated work that is not buttoned down to the max.
How long until some researcher releases a study showing that Dr. Ioannidis' research findings are themselves wrong?
Who needs a study? Simply reading the article shows that he has fallen precisely into the trap that he is complaining about i.e. overstating his results. He forgets one very simple point: not all science is medicine/biology.
As a particle physicist I would strongly disagree with his conclusions, at least as applied to experimental particle physics. It is certainly true that some papers turn out to be wrong but this is rare and usually ends up as a 'big thing' in the field. Outside my field I'd be very surprised if the majority of physics or even chemistry papers turn out to be wrong (but I certainly not a chemist so this is just my impression).
As for medicine I can certainly see that they have a problem. Afterall how many times have we been told "don't eat X/do Y it is bad for you" only later to find out that actually it isn't half as bad as they thought and may even have benefits? Just because a lot of medical research is often flawed does not mean that all of science has the problem on the same scale.
So, Dr. Ioannidis either show us some data from chemistry, maths and physics or stop complaining that all of science has a problem on this scale. From where I stand your evidence points to a problem with bioscience/medical research only.
When did "almost a third..." become "most?"
So, a paper claiming that small sample sizes lead to wrong conclusions is based on analysing a sample of only 49 other papers? The mind boggles with the self-applicability ...
> I fail to see how you can draw any conclusions about the reliability of atmospheric physics papers from a study of biomedical research papers.
Biomedical research is a lot more amendable to verification and falsification, thus an argument can be made that errors are getting corrected. Global Warming is faith based, it's predictions aren't made in anything resembling a controlled scientific environment and the only way to test it's predictions is to do nothing for twenty years and see if the disasters predicted come to pass. Now consider that rerunning a medical test and the origional paper wrong will get a researcher rewarded while writing anything whatsoever questioning human caused global warming gets a researcher labeled a whore of the oil companies and the argument that the science on GW might be at least as flawed as these biomedical papers grows.
Democrat delenda est
It does indeed. Thirty years ago an assistant professor could get tenure by publishing one good paper per year in an archival journal. Nowadays an assistant professor is expected to publish four or more journal papers per year. This leads to the well-known academic concept of the "MPU", i.e. the minimum publishable unit, or "just how many papers can I squeeze out of this one good idea?". This also leads to the backwards situation where a senior professor sitting on a Promotion & Tenure Committee may have fewer published papers (and fewer awarded research dollars) over his entire career than the assistant professor whose tenure he is voting on. Believe me when I say that the hypocrisy of this double standard is not lost on the junior faculty.
There's no doubt in my mind that the signal-to-noise ratio in archival journal papers has plummeted in the past two decades. 90% of all journal papers are superfluous, repetitive, or lacking in any significant advancement of the art, and I'll plainly admit that includes my own papers. Everyone in academia realizes what's going on, and knows it isn't good for the students or the faculty, but unfortunately that's the way the beans get counted in the academic world.
"Global warming does not fall under this. It has been researched, and retested, and re-challenged numerous times.'
On this point you are walking on loose sands. What do you mean retested? You cannot test Global Warming. You can only make observations about it and then form your own opinion based on those facts. You cannot create 2 identical planets ,mess with their CO2 levels and then compare the results. All your data is based on your measurements and conclusions you draw from it. That is where the controversy lies. In order to test the Global Warming theory you need 2 carbon copies of 1900 earth and have only one use massive amounts of carbon fuels and the other next to none. Only then will you truely approach(you cannot mimick everything) how the system works.
And let us not talk about history because that is even more a topic of disputes.
Knowledge is power. Knowledge shared is power lost.
It's important to identify a problem no matter what, but are any of these biases fixable? I would argue that some of them, specifically the bias toward positive results, is not fixable and is inherent to how science works.
To quote one of the articles "...negative results are potentially just as informative as positive results, if not as exciting." But negative results often require much much more verification than positive results, if they can be verified at all, and are limited in how much they can tell you. The antidepressant studies mentioned, a negative result, that the antidepressants did nothing, only tells you that in the patients tested, the doses tested did not give you a noticeable positive result. Publishing a negative result on that would have very limited conclusions. The next year, they could find that doubling the dose was actually effective, making the writeup of the earlier negative result pointless and even more trivial. A waste of time, plus then you've published saying your own product doesn't work.
Negative results get even more pointless in other fields. If someone does a mutagenesis screen for a particular defect in C. elegans, and doesn't find any mutants affecting that, it could be noteworthy, indicating that any genes affecting that process were so vital that when you took one away you didn't get a worm at all, or it could just be luck that genes affecting the process were never mutated, or the researcher didn't do it correctly, or all genes involved were redundant, or some combination. What conclusions could you draw from that? It would be a negative result that would be nigh impossible to tell anything from. Without any positive hits, you could go to the trouble of making sure you did it correctly, but you're not going to make sure every gene got hit at least once, that would be impossible.
In still other cases, a negative result is often retrospectively found to be the fault of the researcher. Who wants to publish something that is basically telling your peers how dumb you are?
There's also that it requires a lot of extra work to make sure it's a negative rather than a null result. Usually when I hit a negative result, my inclination is to see if I did it wrong by repeating the experiment if possible, if it comes up negative again I usually take a different approach, if that also gets a negative result I re-evaluate. I don't ever do all the other supporting experiments that would be needed to convince a reviewer it's a real negative result. If I use an RNAi construct to knock down a gene, and it doesn't do what I'm expecting or anything else interesting, I don't verify the gene is actually knocked down, since that's more effort that would probably be a waste. I'm definitely taking a risk that it's a real result, but it's hard to prove a negative and there's also less motivation to do so.
The limited ability to make positive conclusions about negative results also limits where they could be published. There is a journal for negative results, but a publication there is not something I personally would put on a CV.
So while it is interesting that a bias against negative results may be throwing us off, it's not very usefull knowledge, because I don't see us able to do anything about it.
At the risk of being modded down to oblivion, I am still curious to how this effects popular theories like global warming.
Global warming is not popular, it is down right scary. If a group of scientists disprove that our use of hydrocarbons have a significant effect on global warming, these scientists will be extremely popular and probably share a Nobel Prize.
--- guns don't kill people, people with guns kill people ---
That's about as smart as saying that you can't test the theory of evolution because we didn't have two identical earths from 2 million years ago so we cant verify that species evolve. Most of science works using models and any basic chemistry lab can verify that increased levels of CO2 reflects more light of certain wavelengths which lead to higher temperatures. Any more professional biology lab can verify that organisms evolve by growing bacteria for a few days.
Football Odds
This latter group are far more prone to different interpretation by different people.
Engineering is the art of compromise.
So to test the lunar theory of tides, do we need make or capture a second moon? Or to test the theory of evolution, should we be thinking of terraforming Mars, seeding it with primitive life and watching for a billion years? And of course testing anything to do with stellar formation is going to be hellishly difficult, how are we going to afford 2 billion yottagrams of gas at todays prices?
No-one from Popper onwards has restricted testing to experiment only. Popper's own example in "Science as Falsification" is the Eddington solar eclipse observations performed to test General Relativity.
You test a theory or hypothesis by comparing the expectations of that theory with reality. This can be by either experiment or observation. If you reject GW because we can't "do the experiment", then you are rejecting most of cosmology, biology and geology along with it.
While Ioannidis, the author of the original work that was discussed here may be correct that a large fraction of a very specific type of clinical research findings, are incorrect, there is no reason to believe (from his published work) that "most published research findings are false". Most of the ones he looked at were not reproduced, but they had well understood limitations. My papers, I can assure you, are only incorrect about 10% of the time.
Nobody is arguing that the climate isn't changing, it's about the cause, the AGW premise is plausible and now even looks very likely, yet you have to consider the following the the Earth has warmed and cooled before without human intervention and establishing cause based on evolving and unproven computer modeling eliminating other causes, while we are still discovering potential causes is a bit of a stretch. Why would you think that Climatology is easier, cheaper and more certain than pharmacology and bio-medicine anyways? Perhaps you believe that Exxon-Mobile will just cease operating when the last drop of extractable crude comes out of the ground?
Apocalypse Cancelled, Sorry, No Ticket Refunds
Science in general isn't about "publishing what is right" but rather creating a network of accountability in the form of methods, ideas, data, procedures, etc. so others can try and reproduce and critique the results. Even if the published results are shown to be incorrect by other studies, this does not mean the system is broken. The scientific process is an iterative, self correcting, one. However, if after many years and many studies, a particular field fails to converge on an accepted baseline conclusions, there is a good chance something is wrong (you may even be doing pseudoscience).
i\hbar\dot{\psi}=\hat{H}\psi
Indeed in my field (a sub-area of computer science) people are usually highly skeptical of any supposedly important new result in the field that was first published in one of the highly prestigious but generalist journals, like Nature or Science. These often end up being, if not outright wrong, at the very least seriously over-extending their claims or the importance of their claims, in a way that would never get them published in a specialized journal filled with an editorial board who were actually experts in the specific area in question.
This is only exacerbated by the fact that, because generalist publications know they don't have expertise in every specialized area on staff, they often ask the authors to suggest potential reviewers of their own papers. Of course, authors are likely to suggest reviewers who they think will like the paper, not the ones who would give it a grilling.
I think the interest of this particular study is not so much that a lot of science turns out to be wrong, but that a lot of the most prestigious publication venues turned out to be wrong more often.
10 PRINT CHR$(205.5+RND(1)); : GOTO 10
Afterall how many times have we been told "don't eat X/do Y it is bad for you" only later to find out that actually it isn't half as bad as they thought and may even have benefits? Just because a lot of medical research is often flawed does not mean that all of science has the problem on the same scale.
The problem here is that the popular press always report the very latest 'finding' in what is a complex field. Yet we should know that not only in medicine, but in virtually all experimental sciences, a single paper is not sufficient to establish some new profound truth.
Dr Ioannidis' largest problem is that he thinks he has identified a problem. There isn't one. This is how science is supposed to work! We publish methodologies so that the work can be replicated by other teams. Some findings survive futher scrutiny, some don't. The "hotter" the field, the less you are going to rely on the latest single study, no?
So he's found 1/3 of studies were refuted, but later work. Great, they were refuted, what's the problem? And how do we move from that to the conclusion that "most" scientific papers (even outside the hotter fields of bio-medical research) are wrong. And what about looking at outcomes? The advances of medicine even in my lifetime are astounding, this is hardly the result of a system that isn't working!
Better to be despised for too anxious apprehensions, than ruined by too confident a security. --Edmund Burke
You said, "So, Dr. Ioannidis either show us some data from chemistry, maths and physics or stop complaining that all of science has a problem on this scale."
I'm sympathetic to the direction you are going, but I don't agree completely.
The problem is due to being able to get extra money by exaggerating claims. The problem is in every area of science, in my experience. If there is no chance to get more money by exaggerating claims, then I agree, the problem seems minimal.
In computing, claims about "Artificial Intelligence" have been extremely exaggerated.
In physics, there are those who claim they may have found a method of cold nuclear fusion. Search for Sonofusion, for example, fusion that is caused by extremely intense ultrasonic sound. Some of those claims are exaggerated, or there are omissions of the limitations.
I think the more important thing to note here is the irony in the fact that published research has found that most published research is false.
I hate printers.
I can't see what is so surprising here. Basically when you do research, you are groping in darkness - after all, it wouldn't really be worth doing if the results were already known, would it? Approaching a new problem is a bit like looking at the notorious elephant through a keyhole; different people will have different guesses as to what it is and most will be wrong, until at some point enough observations are made and you can construct a more complete picture.
When you publish scientific articles you don't claim that "THIS IS THE TRUTH" - you are merely putting forward your opinion and then somebody else comes along and says "No, because ...". And even the articles with the "wrong" results are valuable, because they tell us that this particular interpretation is not the right one. It can take a lot of false turns before you find the right way through a maze, and in fact it tells us something about the generally high quality of research that we are not seeing about 90% wrong results.
"I had this idea, and I did a small little experiment to see if it was worth anything. Maybe it is."
If you just had an idea, and you did a small little experiment you would get knocked back for "methodological issues". You are better having an idea, then using a big simulation output and datamining (with a boutique distance metric) to try and bluster your the readers into thinking you had done something original. Just testing ideas with experimentation is soooo 1950s. Besides, it blows the budget and doesn't attract funding.