If Most Research Publications are Just Wrong, is that a Disaster?

NIH Director Jay Bhattacharya thinks so, but is he right?

So this will come as a shock to folks who haven’t heard about this. It turns out that for some chunk, maybe a large chunk — there’s a lot of debate about exactly how much — of the published peer-reviewed scientific literature, even in top journals, when independent research teams look and try to answer the same question, do not find the same answer. That is, a large chunk of the scientific literature is not reliable.

And this happens in field after field after field….

This is a disaster. It’s a disaster for everybody.

NIH director Jay Bhattacharya, in a January 29th interview in The New York Times

I think the world of Jay Bhattacharya. I’ve been a guest on his Illusion of Consensus podcast before he became director of the National Institutes of Health, and he was kind enough to read and endorse my book, Rethinking Diabetes. We had met socially in 2023 and had bonded over our shared experience being vilified by the medical research establishment—Bhattacharya for co-authoring the Great Barrington Declaration during the midst of the Covid epidemic, and me for my journalism and books on nutrition and chronic disease.¹

I had been skeptical when Bhattacharya accepted the job of running NIH under Robert F. Kennedy Jr. I questioned Bhattacharya’s wisdom in taking the job when he had no experience running any organizations, let alone the most influential research institution on the planet, and under the auspices of a president and a head of Health and Human Services who would guarantee that many of the 20,000+ NIH employees would be antagonistic to virtually everything he tried to accomplish.

His initial press reflected this context, and the fact that Bhattacharya was all too human, lacking the superhuman attributes it might have taken to overcome it.

Now, though, he’s settling in. The media profiles and interviews focus, as they should, on what he hopes to accomplish rather than on what he and his bosses represent to establishment medicine. See, for instance, the interview with Ross Douthat (a conservative columnist, true) in the Times from which the above epigraph is taken, or this recent profile in Science, aptly titled “Man in the Middle.”

Bhattacharya’s solution to the dismaying situation he describes above—the large volume of unreliable science published year after year in the scientific literature—is to create a system, led by the NIH, that funds and catalogs attempts to replicate or reproduce key science. Here he is describing it in the Douthat interview:

Bhattacharya: You can’t replicate every single paper. It’s hundreds of thousands — like, millions — of ideas. And I also don’t want the government to decide which ideas ought to be replicated….

What you have to do is you have to crowdsource — have the scientific community identify what are the key ideas that need replication. If they turn out to be true, then they would send science one way. If it turns out to be false, it would send science in another way.

You do that by essentially using the normal process of the N.I.H. to seek grant applications from the scientific community to do replication. And that has a big effect on essentially creating a cadre of researchers who are honored by the scientific community, because if I give them N.I.H. grants, then that’s a marker of scientific success….

Then, second, you have a journal where you can actually publish your replication work. Also, your negative findings. I have a drawer full of hypotheses that failed. I should be able to publish them and put them in a journal somewhere. Again, something the N.I.H. can do…

Then, third, you make a set of metrics that track good scientific behavior. If someone comes to you and says, “Ross, I’m going to try to replicate your paper,” you’re going to view it as a threat, because the culture is wrong.

If someone comes to a scientist and says, oh, I want to replicate your paper, or your idea — that is actually an honor. And we can put metrics around that so that people at the scientist level get credit for that.

I think the N.I.H. can, and under my leadership, we’re working to try to do all three of those things.

All reasonable ideas, except I don’t think they will have a prayer of working, and by working, I mean increasing the signal-to-noise ratio in science so that more of what’s published is reliable and can be trusted.

I also think Bhattacharya may be misconstruing the fundamental nature of science, at least in this interview.

That a large chunk of the published literature is unreliable is not a disaster. It’s the nature of the scientific endeavor. Yes, I suspect, as Bhattacharya likely does, that it’s worse now than ever, but I don’t think there’s any way to escape this essential problem. And I don’t think Bhattacharya’s solution will solve the problem.

In short, the problem is that good science is, well, really hard, and irreproducibility simply comes with the job.²

The unreliability of the relevant research is the natural state of the front lines of science: a large proportion of observations and experimental results will simply be wrong, or, at best, correctly observed and measured but nonetheless misinterpreted or meaningless. It’s true (or at least so I think) that a large proportion of the research community no longer understands how extraordinarily difficult it is to do good science, so they make no attempt to do it, and that is a problem, maybe even a disastrous one, but that doesn’t change the reality.

A case study of what it takes to establish reliable knowledge

Before we get into quantifying the expected signal-to-noise ratio in the scientific literature, even in the best of all worlds, let me tell you my favorite story of functioning science to set the context. This is a story about what it takes to get the right answer in science. And, as is my wont, it’s an old one. I’ll keep it short(ish).