In comments on our recent post, The so-called “lucky golf ball”: The Association for Psychological Science promotes junk science while ignoring the careful, serious work of replication, Jim asked why so many of these ridiculous and unreplicated results kept coming up in the field of psychology. I shared my hypothesis: one reason that psychology has all these crappy claims that stay around even after failed replication is that this sort of research does not have active opposition.
The crappy research in question was promoted by the APS recently, but it had been published in the society’s journal in 2010, back when junk science was a norm.
But, even back then, there were people within the field who were calling out the problems. Two prominent critics back then were Uri Simonsohn and Greg Francis, two psychology researchers who disagreed on the details but from my perspective were making similar points.
The good news is that Greg Francis showed up in a recent comment thread with this story, commenting on my remark that a lot of bad research in psychology stayed afloat because it did not have active opposition. Here’s Francis:
I would say there is active opposition to the (admittedly small) opposition. In 2012, I submitted a commentary to Psychological Sciences on the Damish et al. (2010) paper when I realized the results seemed “too good to be true”. The editor rejected the commentary based on the feedback from “a very disitnguished professor of experimental design and statistical methods” who wrote (among other things), “I would not be at all surprised if there is publication bias involved. If I had run a study on superstition and the results were null, I would not likely submit it for publication.”
On the face of it, this is kind of amazing: flat-out admitting the problem but not wanting to do anything about it! As Francis says, it’s opposition to opposition.
More generally, there are people in academia who take an anti-anti-junk science stand. They’re not exactly in favor of junk science—if you pressed them on it they would accept that open data is better than not, that non-replication tells us something, that accurate measurement is a good idea, etc.—but what really bugs them is when people are anti-junk science.
We’ve seen this over the years, with prominent academics dissing research critics as being “second-string,” “terrorists,” “Stasi,” etc etc etc.
Regarding lack of opposition, commenter Jack wrote:
Keep in mind that most of [Brian “Pizzagate”] Wansink’s papers are in obscure, minor journals that are rarely read and cited. He gets by on quantity, volume. Also . . . Wansink’s papers are usually on “harmless” topics, uncontroversial and very specific. It would be different if the papers were on “big topics”, topics that everyone has a strong opinion about, and on which many people work using the same data or similar data. For example if you use CRSP financial securities data and make a dubious claim about what explains expected returns, you can be sure 1000 researchers will call you out on it.
It’s like the difference between fishing in a small pond no one but you knows about, vs fishing in the Grand Banks in the Atlantic.
To which I replied:
Ahhhh, but here’s the paradox: In a scientific context, Wansink’s work is obscure. Yet in the news media, he was not obscure at all, being featured in the New York Times, the New Yorker, NPR, etc., as well as Marginal Revolution, Freakonomics, etc. And he was respected enough to have received millions of dollars of government dollars and was appointed to a post in the government. And he was a superstar at Cornell, a well-regarded university. His work had real policy impact. So not such a small pond at all.
So, lots of publicity and influence but not much opposition. Lots of people where happy to promote Wansink’s junk science within the academic fields of psychology and business management and within the news media, but not much resistance. And similarly with other purveyors of junk science.
P.S. Greg Francis in his comment adds:
I had not noticed at the time, but I later realized that the means reported in Experiment 4 of the Damish et al. paper fail a GRIM test. The measure of performance is the number of correctly identified words, so the sum of scores across participants of each condition must be an integer value. Damish et al. do not report their sample sizes for each of two conditions, just the total sample size (n1+n2=29). The reported mean for participants with the lucky charm (M1=45.84) and the mean for participants without their lucky charm (M2=30.56) cannot simultaneously be produced by any combination of n1 and n2 sample sizes that together add up to a total sample size of 29. For example, if n1=14, then the sum of ratings would be n1*M1=641.76, which presumably rounds up to 642 (it has to be an integer because it is a count of correctly identified words). But 642/14=45.857, which does not match the reported 45.84. Rounding down does not help either because 641/14=45.785, which does not match the reported mean. There is no way to get M1=45.84 from n1=14 participants. For other combinations of n1 and n2, you can get one of the means to make sense, but never both simultaneously.
I’m reminded of Clarke’s Law: Any sufficiently crappy research is indistinguishable from fraud. I don’t know if the numbers in the article in question were made up, or rounded and unrounded too many times, or mistyped, or maybe Francis messed up in his calculations—I’m guessing the most likely possibility is that the authors messed up in some small way in their analysis, including certain data in some comparisons but not others—but it really doesn’t matter, except for historical reasons, to help understand how things went so wrong for so long in that field.
A comment on one part of the paradox: why the media and government agencies are willing to recognize quantity over quality. I think part of the fault lies with academics itself. So much research is so obscure and complex, that it is often rendered useless or at least sidelined. Of course, the world is complex and statistics is hard, so any carefully done study will likely be complex. But I think academics have been guilty of not adequately distinguishing between technical issues that don’t really matter and important ones that do. So, a “high quality” piece of research may spend pages discussing a detail that only a highly trained specialist cares about, while the “low quality” research gets paid more attention because it is more understandable. I’m not defending that situation (because it leads to an overall degradation in quality), but I do think some of the responsibility lies with the disciplines themselves. The failure to identify truly important details lends itself to this quantity over quality phenomenon.
A concrete example: there was a recent discussion on this blog that concerned the topic of conformal prediction. I exchanged some emails with someone who understood it better than me. My simplistic explanation turned out to be correct, but I was told at one point that there was an important detail that I needed to divide by (n-1) rather than by (n). When I said that seemed unimportant to me, this person told me (paraphrasing here) “yes, it only matters for the proof of the theorem.” I would submit that sort of detail as the kind that feeds the tendency of the media, the public, and even government officials to discount high quality research.
Anyone should be able to find the answer to a few simple questions.
Have these results been replicated by other groups? If so, are those groups more incentivized to find the same or other results?
Is there a quantitative theory put forward? If so, how surprising were the predictions relative to other possibilities? How accurate were they?
The details of the proofs, etc only become interesting to outsiders when the answers to the above are satisfactory.
I think there’s a deeper problem here: we don’t have good intermediate information infrastructure. There are a lot of different versions of this problem, but it basically works like this:
Consider a domain of information on which you lack knowledge. From a state of ignorance, our current society presents you with two options for improving your understanding: you can either sample the domain randomly, picking up isolated facts and claims without any kind of context to evaluate them, or you can do a holistic study of the entire field. If you’re lucky, a structure exists to help you do the holistic study in an organized way, but often you’re almost entirely on your own, especially if you’re lacking in time, money, or both. And this is assuming you’re aware of your own ignorance in the first place!
I don’t know that we as a society have ever been very good at dealing with this sort of problem, but it’s gotten much worse since computing technology has massively increased the volume of information individuals have access to with very little attention paid to making this sudden flood navigable. Our best tools for searching and categorizing still aren’t much better than “do active word association and pray”.
Well, people *should* be able to rely on argument from authority/consensus heuristics for most issues.
Unfortunately the ultimate source of info has been slowly replaced with pseudoscience (test null hypothesis instead of your hypothesis, peer review instead of peer replication). This environment of confusion is then being gamed for profit and power.
I would answer this question differently. In my opinion, the reason the media and the government are happy to recognize quantity over quality is that they are unwilling to recognize the answer “we don’t know”. Any claim to know beats any claim not to know regardless of the relative quality of the two claims. And when the quality answer is “we don’t know”, that just means that the recognized answer will be of low quality.
I’m not really sure that I see a paradox here… Never ending soup bowls seems like way better material for a news outlet to pick up than say CRSP financial securities data. Everyone can read and relate to the former, but the latter just sounds technical and boring. Like Dale says above, much of research (good and bad) is by necessity complex, detailed, and very niche, so doing clever seeming studies with interesting results on a ‘harmless’ topic that the general public finds interesting would seem to be a great way to gain publicity. I don’t see why doing good science, in and of itself, would be the most efficient way to gain publicity.
Doesn’t a lot of this junk make interesting little tidbits for news outlets to pick up? It’s all about clicking and reading, right?
Andrew:
You talk about the “scientist hero” meme and how that drives crap research. It’s good characterization. But the full essence of the phenomenon seems broader and deeper in society than just the idea of a scientist doing something.
IMO there’s a “Mystery Machine” element in the phenomenon as well: an idea that a few “meddling kids” with no knowledge of anything can walk in the door and figure out cold fusion in half an hour. “Aha! Velma found X and Scooby accidentally found Y while he was tripping over Shaggy and we discovered Cold Fusion! Now the World Is Saved”
It’s not *just* a “scientist” thing. There’s a concept that you don’t have to know anything or even do any work to uncover amazing truths and save the world.
jd writes, “I’m not really sure that I see a paradox here… Never ending soup bowls seems like way better material for a news outlet to pick up than say CRSP financial securities data.”
I think the paradoxical part is if you accept Andrew’s hypothesis that junk science proliferates in fields without much opposition, then why wouldn’t there be alot of opposition to Wansink, for example. He had a lot of influence. Wouldn’t another researcher want to challenge those results and get that influence for himself. I think the answer maybe that in his field, no one would want to challenge the research because it may undermine the entire field. There are areas of psychology and other social science where there just isn’t evidence for any of big claims not because it is all a fraud, but because it is just impossible or impractical to gather enough data or to seperate out the noise from the data. So, if you have a Wansink superstar, you don’t want to explore whether he is a fraud because he is bringing money to your field, and you know that if anyone looks too hard, none of the studies are on solid ground. But, once the superstar starts getting attention beyond his field eventually some one looks.
I don’t find that a convincing argument. There is a tragedy of the commons element here – even if it is in the interest of the entire field to not tarnish a star, the individual incentives to do so might still be strong. I think that is particularly true in the competitive academic realm where a minor career can be enhanced by critiquing a star’s work.
“a minor career can be enhanced by critiquing a star’s work.”
Judging by what I’ve seen on this blog critiquing a star’s work is a big mistake in social sciences. I’ve also heard many anecdotes from researchers in physical sciences that “star” researchers go out of their way to attack opponents directly and/or block publications that negate or contradict the “star’s” work.
From what I know if you’re going to do anything that contradicts or undermines or otherwise tarnishes the work of a well-known researcher, you better have plenty of your own special mojo or a lot of friends in high places.
I had a long comment about this that I had to delete because I think it was too personally identifying, but I will build on this and try to split the difference a bit. I think within a lot of fields, people only respond to critiques that speak the idiom of the field. If you want to destroy a widely accepted convention, you are going way out on a limb and you better be sure you are right, because people will be out to get you.
I have some training in a weird little corner of statistical analysis that historically was not that common in the field in which I currently work. Any time I raise certain kinds of criticisms that would be considered obvious in my little corner, they are essentially ignored. In so many words: “Everyone (in the field in which we work) does x, it’s very hard to write about this subject without doing x unless you have huge amounts of money available for collecting large amounts of your own data, so we are going to continue to ignore the problems with doing x.”
+1 It brings to mind the old Upton Sinclair quote, “It is difficult to get a man to understand something when his salary depends upon his not understanding it.” I my response to Dale’s concern about incentives is that we aren’t really at the level where incentives matter. The researchers have to look around and consider the possibility that everything they are doing may be BS. Instead, they quitely bury that suspicion. How many times during the debates of null hypothesis testing has the response been, “this is standard practice” until it got through to people that standard practice may be wrong.
100% agree, as a quant in education research who came in at the very end of the all-qualitative era. You’d feel bad for these people, stuck as they are in a system too big to change, except they’re the ones reviewing journal articles, funding proposals, and dissertations. The problem isn’t top-down, it’s bottom-up. Fortunately, that means new generations can make a big difference with the changing of the guard. Although, in research, the guard tends not to change until they’re well into their 70’s or 80’s. I mean, I’m a radical and I’m 45!
“Any time I raise certain kinds of criticisms that would be considered obvious in my little corner, they are essentially ignored. ”
This begs the question: does the work even mean anything?
To put in another way, why is it that getting better results isn’t important to the people you work with? If a bad model was used to calculate the trajectory of an interplanetary landing craft, the failure would be obvious. There would be repercussions. But in most of social science, no one cares because the failure isn’t even detectable. It has no negative effect on the “results” of anything because it’s meaningless in the first place. Which raises the question: why is anyone bothering to do the work in the first place? Does it have any use at all? :)
I’ve noticed three factors that make certain subfields of psychology more likely to produce and tolerate of iffy work. First, psych editors don’t like replications. Psychologists are encouraged instead to conduct “conceptual replications,” which assume the last study with p<.05 is irrefutable, and so test a related idea in a new setting. Such designs tend to be bespoke and idiosyncratic, with hypotheses that sometimes verge on unfalsifiable, making them vulnerable to overinterpretation and ungeneralizability.
Second, a lot of hypotheses in psychology can be as complex as those in, say, neuroscience, since they test predictions about the minutiae of human thought, memory, and behavior. Except, the literature contains a lot of theory-building-by-analogy, and most of the evidence is indirect. But, their procedures can often be iterated (relatively) cheaply and quickly. With the above-noted pressure for novel results, and the statistical significance filter, it's easy for the literature to set off down a road to nowhere and hard to get it back on track.
Third, if you don't have a solid, replicable, logically-consistent base for your work, the next best thing is a good story. We have analogies, just-so stories, common sense, and counterintuitive explanations that are just so crazy they they have to be true! All the better if you can, through a process of chained associations, tie your niche findings to a real-life phenomenon of which the existence and relevance are undeniable. That lends your work credibility AND topicality, which, as we all know, equals publishability.
TL;DR: Many lines of research depend on extended syllogisms from that one study no one ever replicated, and small p-values (i.e., luck). Everybody knows this is a house of cards, so it's best to avoid criticizing anyone who can even tangentially justify their design and conclusions by connecting them with published results. They might, after all, be your results. To coin a phrase, if we do not all publish together, we will surely all perish separately.
+1 You mention the pressure for novel results. I often wonder if in psychology or social psychology if novelty is an especially bad goal. We afterall have evolved to be social animals. We are really good at understanding other humans and figuring out how to please them, upset them, manipulate them, sale them, etc. We’re good at social psychology. But, academics want to say something new and exciting. “Sex sells” isn’t novel. If the power pose person just told us that you can pump yourself up before giving a speech, and that will help sometimes that wouldn’t have been novel. We know that those things help sometimes. But put some crazy quantative dressing on the claim and it becomes novel and ridiculous. Unlike physics or chemistry, our intuitions about how humans think and react are much better because we’re humans. We ought to assume that psychological claims about how average humans work that surprise us are probably false.
+1, this blog has convinced me that one’s belief as to (pretty much) any new claim in psychology should be … one’s prior. It should take enormous persuasive power to move you.
Not for claims that fall way outside our experience like serious mental illness, but yes for the psych of normal behavior.
“We afterall have evolved to be social animals….We’re good at social psychology. ”
And in fact it’s not just us: many mammals are social creatures that live in groups and take their cues on how to manage in the group without even speech. Mammalian brains were social before they were analytical, before language, before opposable thumbs.
About the CRSP data: There’s certainly tons of data mining in that field that doesn’t get called out. True, it doesn’t get into the news media, but it gets to the thousands of investment firms who are trying to generate abnormal returns and selling their strategies to clients. Pretty similar to the dissemination of junk science in psychology, except here, it costs people real money (and makes money for the snake-oil salespeople). Here’s a paper that finds very low replicability of published asset pricing anomalies: https://academic.oup.com/rfs/article-abstract/33/5/2019/5236964 (though a previous entry on this blog also covered a paper that finds a much higher replication rate: https://statmodeling.stat.columbia.edu/2021/11/05/is-there-a-replication-crisis-in-finance/)
I disagree that the data fails a GRIM test. If group 1 contains 13 people with a total rating of 596, and group 2 contains 16 people with a total rating of 489, then the group 1 mean is 45.84615… and the group 2 mean is 30.56250… . It seems defensible to report those as the truncated values 45.84 and 30.56?
I’d prefer to reserve the claim “fails a GRIM test” for when you can’t arrive at the reported data by any reasonable means. Rounding numbers down instead of “to nearest” is a reasonable means.
Your quantitative observation is correct. I have mixed feelings about your interpretation. Is it reasonable to suppose that the authors truncated values rather than rounded to nearest values? I think not because, outside of some researchers in mathematical psychology, my impression is that research psychologists think rounding _means_ reporting the nearest value. If these authors were using some other rounding approach (and there are many!), it seems they would need to explain/justify that in their manuscript.
Anyhow, to fail a GRIM test simply raises doubts about the validity of the reported results, which I think is appropriate here. The cause of the failure could be trivial (such as a typo), but the authors should address the issue; just to set the record straight.
I would tend to agree that many people (research psychologists or not) believe that “rounding” means reporting to the nearest value at some particular decimal position.
But I would also say that the same people often report numbers by truncation. The use of the term “rounding” for truncation is my wording, not theirs. They would just tell you they’re reporting unrounded numbers.
Taking a particularly prominent example, the digits of pi are generally held to be fixed, eternal constants (which is the truth!), regardless of how many of them you are currently reporting. (Depends on your perspective!) Thus the fourth (post-decimal) digit of pi is 5, and pi to four places is 3.1415, despite the fact that the fifth place contains a 9.
Wow, so can we trust the Western media reporting on Ukraine?
Kien:
It’s all about the specifics. As far as I’ve seen, the general reporting on Ukraine (Russia starting a war, bombing civilians, Ukranians fighting back, etc.) doesn’t seem to be disputed in any serious way. I think that if the bulk of the reporting were untrustworthy, we’d see serious refutations of the claims of the reports. In the above post I talked about the value of active opposition. The reporting on Ukraine does have active opposition, it’s just that the active opposition seems more like propaganda than anything else.
Speaking more generally, in the above post I’m not arguing that we should distrust everything reported in the news media. What I’m saying is that it’s good to see multiple perspectives and then evaluate what we see.
My guess is kien thinks that the Ukrainians were the ones who killed civilians in Bucha, who hit that train station with a missle, that Ukrainians staged other (false flag) scenes of atrocities, that Zelensky is just a pawn in the hands of Nazi rightwing Ukrainians, and US necons, etc.
And western media has gone along with the conspiracy because…um because…er because they’re captured by the arms industry and the deep state. Yeah, that’s the ticket.
Joshua:
Recall the phrase, “Merchants of doubt.” It’s not required that someone present any evidence that the Ukrainians killed civilians in Bucha; all that’s needed is to sow a general mistrust in objective evidence. Just like way back when with the doubt on the cigarettes-cancer link. It’s tricky because I make the recommendation that people not blindly trust authority (just see my screeds against various publications in PNAS, JPSP, etc.), but that does not mean that I think nothing can be believed.
I have the impression that a lot of people think in terms of procedure rather than in terms of truth and evidence. For example, many leaders in academic psychology seemed to think it was improper to criticize papers that had been published in their journals. When the Freakonomics guys were climate-change-denial-friendly, it’s not like they had to believe in the ridiculous claims they were promoting, they just needed to believe that the deniers weren’t given a fair shake. Similarly with Ukraine: news deniers don’t need to actually believe whatever stories the Kremlin is pushing this week, they just have to feel that the news media is generally unfair to fascists, so if they can foster a general distrust of news, that will be good enough for them.
I spent most of my professional life working in agriculture, in both government and industry, interacting mainly with applied biologists and applied chemists of various flavours. I didn’t encounter any junk science and studies were easily replicated. If we did launch a product that didn’t work then farmers wouldn’t buy it, but we didn’t do that because we would have shown that it didn’t work in the development (or earlier) studies.
Now compare with nutrition. From the 40s onwards heart disease was on the increase; in the 50s Ancel Keys carried out a number of studies that appeared to show that fat in the diet was the culprit; In 1982 WHO published a report that reviewed the evidence linking early death in males from heart disease with the consumption of saturated fat. Then we had the healthy eating revolution featuring low fat, high sugar products. This turns out to have been a load of nonsense ( https://www.youtube.com/watch?v=qUQnf_ok63w&t=494s). How come nobody noticed. Well some did – e.g. “Eat Your Heart Out” by James le Fanu (1987), a GP in London and medical correspondent for the Sunday Telegraph. But most of us accepted the advice of the professionals and cut fat from our diets. Also, in the Early 70s John Yudkin (https://en.wikipedia.org/wiki/John_Yudkin) wrote “Pure, White and Deadly: How Sugar Is Killing Us and What We Can Do to Stop It “, but hardly anyone noticed. It appears that Ancel Keys had better PR.
So, why is it easy for a lay person to be able to tell when a pesticide does not work, but not be able to tell whether a bacon sandwich is dangerous?