David Freedman in The Atlantic writes,
We think of the scientific process as being objective, rigorous, and even ruthless in separating out what is true from what we merely wish to be true, but in fact it’s easy to manipulate results, even unintentionally or unconsciously. “At every step in the process, there is room to distort results, a way to make a stronger claim or to select what is going to be concluded,” says Ioannidis. “There is an intellectual conflict of interest that pressures researchers to find whatever it is that is most likely to get them funded.”
The article is long, but worth reading in its entirety. I cannot excerpt enough.
the 49 articles themselves were the most widely cited articles in these journals. These were articles that helped lead to the widespread popularity of treatments such as the use of hormone-replacement therapy for menopausal women, vitamin E to reduce the risk of heart disease, coronary stents to ward off heart attacks, and daily low-dose aspirin to control blood pressure and prevent heart attacks and strokes. … Of the 49 articles, 45 claimed to have uncovered effective interventions. Thirty-four of these claims had been retested, and 14 of these, or 41 percent, had been convincingly shown to be wrong or significantly exaggerated…
Hansonian medicine, anyone?
“Often the claims made by studies are so extravagant that you can immediately cross them out without needing to know much about the specific problems with the studies,” Ioannidis says.
The $300,000 kindergarten teacher, anyone?
But of course it’s that very extravagance of claim (one large randomized controlled trial even proved that secret prayer by unknown parties can save the lives of heart-surgery patients, while another proved that secret prayer can harm them) that helps gets these findings into journals and then into our treatments and lifestyles, especially when the claim builds on impressive-sounding evidence. “Even when the evidence shows that a particular research idea is wrong, if you have thousands of scientists who have invested their careers in it, they’ll continue to publish papers on it,” he says. “It’s like an epidemic, in the sense that they’re infected with these wrong ideas, and they’re spreading it to other researchers through journals.”
Rational expectations macroeconomics, anyone?
Other meta-research experts have confirmed that similar issues distort research in all fields of science, from physics to economics (where the highly regarded economists J. Bradford DeLong and Kevin Lang once showed how a remarkably consistent paucity of strong evidence in published economics studies made it unlikely that any of them were right).
READER COMMENTS
david
Oct 15 2010 at 1:27pm
Don’t fall into the Austrian trap of “economics is complex and your econometrics is worthless, so clearly we should fall back to the default position of Austrian Capital Theory, which, being self-evidently the default, needs no econometric proof”.
david (not henderson)
Oct 15 2010 at 1:51pm
Gee, if such a thing can happen in the medical sciences, do you think it’s possible that these sorts of problems might also characterize other areas of the physical sciences (he asks innocently)?
rhhardin
Oct 15 2010 at 2:36pm
Actual science runs on curiosity, not credentials.
In that respect, the refutations of science cited are actually its workings. (“Hmm, that’s odd. I wonder why?”)
It’s global warming that tries to discredit curiosity, not the odd medical statistical study.
Joe Cushing
Oct 15 2010 at 6:34pm
” if you have thousands of scientists who have invested their careers in it, they’ll continue to publish papers on it,”
Hmmm. Can we think of any area of science where one of the primary arguments that the claims must be correct is that there are thousands of scientists that support those claims? Can we think of an area of science where there have been many millions if not billions of dollars spent trying to prove these claims, where a movies and global speeches have been made (full of falsehoods) trying to convince people that the thousands of scientists are right. Can we think of a set of scientific claims, that if true, would require the largest re-organization of the global economy in world history? Said re-organization would of course require an unprecedented growth in world government power. Can anyone think of a claim that has more at stake than any other claim made in history?
Can anyone think of any claims like this that are floating out there?
Jim Manzi
Oct 15 2010 at 6:38pm
Arnold,
Thanks for the pointer. It’s a fascinating article.
3 quick observations:
1. Unless I missed it, the article never defines what it is for a finding from a study to be “wrong”, other than to say it is the probability that it will later be “convincingly refuted”. Does this mean if we replicate it with identical methodology, we don’t get the same answer? Does it mean that if we quasi-replicate it with what the meta-researchers believe to be the “correct” methodology, we don’t get the same answer? Or is it something else? How do we know the refutation is “convincing” unless we can specify this?
2. Assuming some rough agreement that “right” means something crudely like “competent researchers acting in good faith will consistently get this result in large randomized trials”, the difference between the success rate for large RCTs (90%) and non-randomized methods (20%!) seems pretty striking. 90% sounds like a pretty good batting average when somebody describes a clinical finding like “take this pill”, and being wrong 80% of the time means that its a lot worse than flipping a coin – I ought to pretty much ignore it.
3. It is interesting that examples that are cited for how findings go wrong (after a long portion of the article about researcher bias and so on) are really about causal density. One sub-aspect is pure density, note this passage:
“But even if a study managed to highlight a genuine health connection to some nutrient, you’re unlikely to benefit much from taking more of it, because we consume thousands of nutrients that act together as a sort of network, and changing intake of just one of them is bound to cause ripples throughout the network that are far too complex for these studies to detect, and that may be as likely to harm you as help you.”
And another is simply that the heath phenomenon exhibits behavior that takes a long time to fully manifest itself. Consider the sentences that follow those from above:
“ven if changing that one factor does bring on the claimed improvement, there’s still a good chance that it won’t do you much good in the long run, because these studies rarely go on long enough to track the decades-long course of disease and ultimately death. Instead, they track easily measurable health “markers” such as cholesterol levels, blood pressure, and blood-sugar levels, and meta-experts have shown that changes in these markers often don’t correlate as well with long-term health as we have been led to believe.”
In effect, the rhetoric of the article is to question the internal validity of the studies (or as per point 2, RCTs), when in fact these examples are really about generalizability, or external validity.
Best,
Jim Manzi
Dave
Oct 15 2010 at 9:14pm
“The $300,000 kindergarten teacher, anyone?”
How about the $300,000,000 ( yes,9 places) CEO. I’d much prefer to pay a teacher $300K than a financial officer that produces nothing of substance for his salary and perks!
Get real.
Jacob Oost
Oct 15 2010 at 11:31pm
Dave, I didn’t know you paid CEOs and were thus in a position to put a number on their value to you based on what you think their productivity is.
My mistake.
rpl
Oct 18 2010 at 1:01pm
The statistical rate of type I (i.e., false positive) errors normally quoted for these studies is 1-5%. If the actual rate of false positives is 2-10 times that, then that’s a strong indicator that there is some sort of bias at work. RCT may be the best methodology we have, but if we can find some way of wringing out that remaining bias, it would be even better.
And, of course, these findings are applicable to other areas of science as well. In some fields, RCT aren’t even possible in principle, so it’s useful to know what kind of bias results from alternate methodologies.
Jim Manzi
Oct 19 2010 at 6:29pm
rpl,
Thanks, and I agree about both further improvements and knowing bias for other methods.
The 5% level conventional test for significance is with respect to the specific study under consideration. There is a huge issue with generalizing this result to other circumstances. This problem gets worse in more complex areas touched on in the article, especially behavioral arenas.
Arnold linked to an article I did in City Journal about some of this a few weeks ago. Just google Jim manzi City Journal if interested.
Best,
Jim
[Arnold’s post is at http://www.econlib.org/archives/2010/08/james_manzi_on.html and Manzi’s City Journal article is at http://www.city-journal.org/2010/20_3_social-science.html –Econlib Ed.]
Oren Grad
Oct 19 2010 at 7:32pm
Jim Manzi: the Atlantic article necessarily presents an oversimplified account. Best to read Ioannidis’ papers if you want to understand what claims he is actually making, and on what basis.
A good starting point is the two papers mentioned in the article:
Why Most Published Research Findings Are False
Contradicted and Initially Stronger Effects in Highly Cited Clinical Research
Douglass Holmes
Oct 20 2010 at 1:36pm
What? Does David Freedman really think that scientists are human? Does he really think that scientists are likely to infuse their own biases into their research? Or that drug companies have a vested interest in making their products appear more effective?
I’m shocked, shocked.
Comments are closed.