Daryl Bem

Why do we need “publish or perish”?

pnrj ethics, personal, public policy, science and academia , , , , , , , ,

June 23 JDN 2458658

This question may seem a bit self-serving, coming from a grad student who is struggling to get his first paper published in a peer-reviewed journal. But given the deep structural flaws in the academic publishing system, I think it’s worth taking a step back to ask just what peer-reviewed journals are supposed to be accomplishing.

The argument is often made that research journals are a way of sharing knowledge. If this is their goal, they have utterly and totally failed. Most papers are read by only a handful of people. When scientists want to learn about the research their colleagues are doing, they don’t read papers; they go to conferences to listen to presentations and look at posters. The way papers are written, they are often all but incomprehensible to anyone outside a very narrow subfield. When published by proprietary journals, papers are often hidden behind paywalls and accessible only through universities. As a knowledge-sharing mechanism, the peer-reviewed journal is a complete failure.

But academic publishing serves another function, which in practice is its only real function: Peer-reviewed publications are a method of evaluation. They are a way of deciding which researchers are good enough to be hired, get tenure, and receive grants. Having peer-reviewed publications—particularly in “top journals”, however that is defined within a given field—is a key metric that universities and grant agencies use to decide which researchers are worth spending on. Indeed, in some cases it seems to be utterly decisive.

We should be honest about this: This is an absolutely necessary function. It is uncomfortable to think about the fact that we must exclude a large proportion of competent, qualified people from being hired or getting tenure in academia, but given the large number of candidates and the small amounts of funding available, this is inevitable. We can’t hire everyone who would probably be good enough. We can only hire a few, and it makes sense to want those few to be the best. (Also, don’t fret too much: Even if you don’t make it into academia, getting a PhD is still a profitable investment. Economists and natural scientists do the best, unsurprisingly; but even humanities PhDs are still generally worth it. Median annual earnings of $77,000 is nothing to sneeze at: US median household income is only about $60,000. Humanities graduates only seem poor in relation to STEM or professional graduates; they’re still rich compared to everyone else.)

But I think it’s worth asking whether the peer review system is actually selecting the best researchers, or even the best research. Note that these are not the same question: The best research done in graduate school might not necessarily reflect the best long-run career trajectory for a researcher. A lot of very important, very difficult questions in science are just not the sort of thing you can get a convincing answer to in a couple of years, and so someone who wants to work on the really big problems may actually have a harder time getting published in graduate school or as a junior faculty member, even though ultimately work on the big problems is what’s most important for society. But I’m sure there’s a positive correlation overall: The kind of person who is going to do better research later is probably, other things equal, going to do better research right now.

Yet even accepting the fact that all we have to go on in assessing what you’ll eventually do is what you have already done, it’s not clear that the process of publishing in a peer-reviewed journal is a particularly good method of assessing the quality of research. Some really terrible research has gotten published in journals—I’m gonna pick on Daryl Bem, because he’s the worst—and a lot of really good research never made it into journals and is languishing on old computer hard drives. (The term “file drawer problem” is about 40 years obsolete; though to be fair, it was in fact coined about 40 years ago.)

That by itself doesn’t actually prove that journals are a bad mechanism. Even a good mechanism, applied to a difficult problem, is going to make some errors. But there are a lot of things about academic publishing, at least as currently constituted, that obviously don’t seem like a good mechanism, such as for-profit publishers, unpaid reviewiers, lack of double-blinded review, and above all, the obsession with “statistical significance” that leads to p-hacking.

Each of these problems I’ve listed has a simple fix (though whether the powers that be actually are willing to implement it is a different question: Questions of policy are often much easier to solve than problems of politics). But maybe we should ask whether the system is even worth fixing, or if it should simply be replaced entirely.

While we’re at it, let’s talk about the academic tenure system, because the peer-review system is largely an evaluation mechanism for the academic tenure system. Publishing in top journals is what decides whether you get tenure. The problem with “Publish or perish” isn’t the “publish”; it’s the “perish”. Do we even need an academic tenure system?

The usual argument for academic tenure concerns academic freedom: Tenured professors have job security, so they can afford to say things that may be controversial or embarrassing to the university. But the way the tenure system works is that you only have this job security after going through a long and painful gauntlet of job insecurity. You have to spend several years prostrating yourself to the elders of your field before you can get inducted into their ranks and finally be secure.

Of course, job insecurity is the norm, particularly in the United States: Most employment in the US is “at-will”, meaning essentially that your employer can fire you for any reason at any time. There are specifically illegal reasons for firing (like gender, race, and religion); but it’s extremely hard to prove wrongful termination when all the employer needs to say is, “They didn’t do a good job” or “They weren’t a team player”. So I can understand how it must feel strange for a private-sector worker who could be fired at any time to see academics complain about the rigors of the tenure system.

But there are some important differences here: The academic job market is not nearly as competitive as the private sector job market. There simply aren’t that many prestigious universities, and within each university there are only a small number of positions to fill. As a result, universities have an enormous amount of power over their faculty, which is why they can get away with paying adjuncts salaries that amount to less than minimum wage. (People with graduate degrees! Making less than minimum wage!) At least in most private-sector labor markets in the US, the market is competitive enough that if you get fired, you can probably get hired again somewhere else. In academia that’s not so clear.

I think what bothers me the most about the tenure system is the hierarchical structure: There is a very sharp divide between those who have tenure, those who don’t have it but can get it (“tenure-track”), and those who can’t get it. The lines between professor, associate professor, assistant professor, lecturer, and adjunct are quite sharp. The higher up you are, the more job security you have, the more money you make, and generally the better your working conditions are overall. Much like what makes graduate school so stressful, there are a series of high-stakes checkpoints you need to get through in order to rise in the ranks. And several of those checkpoints are based largely, if not entirely, on publication in peer-reviewed journals.

In fact, we are probably stressing ourselves out more than we need to. I certainly did for my advancement to candidacy; I spent two weeks at such a high stress level I was getting migraines every single day (clearly on the wrong side of the Yerkes-Dodson curve), only to completely breeze through the exam.

I think I might need to put this up on a wall somewhere to remind myself:

Most grad students complete their degrees, and most assistant professors get tenure.

The real filters are admissions and hiring: Most applications to grad school are rejected (though probably most graduate students are ultimately accepted somewhere—I couldn’t find any good data on that in a quick search), and most PhD graduates do not get hired on the tenure track. But if you can make it through those two gauntlets, you can probably make it through the rest.

In our current system, publications are a way to filter people, because the number of people who want to become professors is much higher than the number of professor positions available. But as an economist, this raises a very big question: Why aren’t salaries falling?

You see, that’s how markets are supposed to work: When supply exceeds demand, the price is supposed to fall until the market clears. Lower salaries would both open up more slots at universities (you can hire more faculty with the same level of funding) and shift some candidates into other careers (if you can get paid a lot better elsewhere, academia may not seem so attractive). Eventually there should be a salary point at which demand equals supply. So why aren’t we reaching it?

Well, it comes back to that tenure system. We can’t lower the salaries of tenured faculty, not without a total upheaval of the current system. So instead what actually happens is that universities switch to using adjuncts, who have very low salaries indeed. If there were no tenure, would all faculty get paid like adjuncts? No, they wouldn’t—because universities would have all that money they’re currently paying to tenured faculty, and all the talent currently locked up in tenured positions would be on the market, driving up the prevailing salary. What would happen if we eliminated tenure is not that all salaries would fall to adjunct level; rather, salaries would all adjust to some intermediate level between what adjuncts currently make and what tenured professors currently make.

What would the new salary be, exactly? That would require a detailed model of the supply and demand elasticities, so I can’t tell you without starting a whole new research paper. But a back-of-the-envelope calculation would suggest something like the overall current median faculty salary. This suggests a median salary somewhere around $75,000. This is a lot less than some professors make, but it’s also a lot more than what adjuncts make, and it’s a pretty good living overall.

If the salary for professors fell, the pool of candidates would decrease, and we wouldn’t need such harsh filtering mechanisms. We might decide we don’t need a strict evaluation system at all, and since the knowledge-sharing function of journals is much better served by other means, we could probably get rid of them altogether.

Of course, who am I kidding? That’s not going to happen. The people who make these rules succeeded in the current system. They are the ones who stand to lose high salaries and job security under a reform policy. They like things just the way they are.

The sausage of statistics being made

pnrj education, statistics , , , , , , , , , , , , , , , , ,

 

Nov 11 JDN 2458434

“Laws, like sausages, cease to inspire respect in proportion as we know how they are made.”

~ John Godfrey Saxe, not Otto von Bismark

Statistics are a bit like laws and sausages. There are a lot of things in statistical practice that don’t align with statistical theory. The most obvious examples are the fact that many results in statistics are asymptotic: they only strictly apply for infinitely large samples, and in any finite sample they will be some sort of approximation (we often don’t even know how good an approximation).

But the problem runs deeper than this: The whole idea of a p-value was originally supposed to be used to assess one single hypothesis that is the only one you test in your entire study.

That’s frankly a ludicrous expectation: Why would you write a whole paper just to test one parameter?

This is why I don’t actually think this so-called multiple comparisons problem is a problem with researchers doing too many hypothesis tests; I think it’s a problem with statisticians being fundamentally unreasonable about what statistics is useful for. We have to do multiple comparisons, so you should be telling us how to do it correctly.

Statisticians have this beautiful pure mathematics that generates all these lovely asymptotic results… and then they stop, as if they were done. But we aren’t dealing with infinite or even “sufficiently large” samples; we need to know what happens when your sample is 100, not when your sample is 10^29. We can’t assume that our variables are independently identically distributed; we don’t know their distribution, and we’re pretty sure they’re going to be somewhat dependent.

Even in an experimental context where we can randomly and independently assign some treatments, we can’t do that with lots of variables that are likely to matter, like age, gender, nationality, or field of study. And applied econometricians are in an even tighter bind; they often can’t randomize anything. They have to rely upon “instrumental variables” that they hope are “close enough to randomized” relative to whatever they want to study.

In practice what we tend to do is… fudge it. We use the formal statistical methods, and then we step back and apply a series of informal norms to see if the result actually makes sense to us. This is why almost no psychologists were actually convinced by Daryl Bem’s precognition experiments, despite his standard experimental methodology and perfect p < 0.05 results; he couldn’t pass any of the informal tests, particularly the most basic one of not violating any known fundamental laws of physics. We knew he had somehow cherry-picked the data, even before looking at it; nothing else was possible.

This is actually part of where the “hierarchy of sciences” notion is useful: One of the norms is that you’re not allowed to break the rules of the sciences above you, but you can break the rules of the sciences below you. So psychology has to obey physics, but physics doesn’t have to obey psychology. I think this is also part of why there’s so much enmity between economists and anthropologists; really we should be on the same level, cognizant of each other’s rules, but economists want to be above anthropologists so we can ignore culture, and anthropologists want to be above economists so they can ignore incentives.

Another informal norm is the “robustness check”, in which the researcher runs a dozen different regressions approaching the same basic question from different angles. “What if we control for this? What if we interact those two variables? What if we use a different instrument?” In terms of statistical theory, this doesn’t actually make a lot of sense; the probability distributions f(y|x) of y conditional on x and f(y|x, z) of y conditional on x and z are not the same thing, and wouldn’t in general be closely tied, depending on the distribution f(x|z) of x conditional on z. But in practice, most real-world phenomena are going to continue to show up even as you run a bunch of different regressions, and so we can be more confident that something is a real phenomenon insofar as that happens. If an effect drops out when you switch out a couple of control variables, it may have been a statistical artifact. But if it keeps appearing no matter what you do to try to make it go away, then it’s probably a real thing.

Because of the powerful career incentives toward publication and the strange obsession among journals with a p-value less than 0.05, another norm has emerged: Don’t actually trust p-values that are close to 0.05. The vast majority of the time, a p-value of 0.047 was the result of publication bias. Now if you see a p-value of 0.001, maybe then you can trust it—but you’re still relying on a lot of assumptions even then. I’ve seen some researchers argue that because of this, we should tighten our standards for publication to something like p < 0.01, but that’s missing the point; what we need to do is stop publishing based on p-values. If you tighten the threshold, you’re just going to get more rejected papers and then the few papers that do get published will now have even smaller p-values that are still utterly meaningless.

These informal norms protect us from the worst outcomes of bad research. But they are almost certainly not optimal. It’s all very vague and informal, and different researchers will often disagree vehemently over whether a given interpretation is valid. What we need are formal methods for solving these problems, so that we can have the objectivity and replicability that formal methods provide. Right now, our existing formal tools simply are not up to that task.

There are some things we may never be able to formalize: If we had a formal algorithm for coming up with good ideas, the AIs would already rule the world, and this would be either Terminator or The Culture depending on whether we designed the AIs correctly. But I think we should at least be able to formalize the basic question of “Is this statement likely to be true?” that is the fundamental motivation behind statistical hypothesis testing.

I think the answer is likely to be in a broad sense Bayesian, but Bayesians still have a lot of work left to do in order to give us really flexible, reliable statistical methods we can actually apply to the messy world of real data. In particular, tell us how to choose priors please! Prior selection is a fundamental make-or-break problem in Bayesian inference that has nonetheless been greatly neglected by most Bayesian statisticians. So, what do we do? We fall back on informal norms: Try maximum likelihood, which is like using a very flat prior. Try a normally-distributed prior. See if you can construct a prior from past data. If all those give the same thing, that’s a “robustness check” (see previous informal norm).

Informal norms are also inherently harder to teach and learn. I’ve seen a lot of other grad students flail wildly at statistics, not because they don’t know what a p-value means (though maybe that’s also sometimes true), but because they don’t really quite grok the informal underpinnings of good statistical inference. This can be very hard to explain to someone: They feel like they followed all the rules correctly, but you are saying their results are wrong, and now you can’t explain why.

In fact, some of the informal norms that are in wide use are clearly detrimental. In economics, norms have emerged that certain types of models are better simply because they are “more standard”, such as the dynamic stochastic general equilibrium models that can basically be fit to everything and have never actually usefully predicted anything. In fact, the best ones just predict what we already knew from Keynesian models. But without a formal norm for testing the validity of models, it’s been “DSGE or GTFO”. At present, it is considered “nonstandard” (read: “bad”) not to assume that your agents are either a single unitary “representative agent” or a continuum of infinitely-many agents—modeling the actual fact of finitely-many agents is just not done. Yet it’s hard for me to imagine any formal criterion that wouldn’t at least give you some points for correctly including the fact that there is more than one but less than infinity people in the world (obviously your model could still be bad in other ways).

I don’t know what these new statistical methods would look like. Maybe it’s as simple as formally justifying some of the norms we already use; maybe it’s as complicated as taking a fundamentally new approach to statistical inference. But we have to start somewhere.