296 Papers Accepted Before Anyone Saw A Result. Positives Fell To 39 Per Cent.

UTRECHT, Netherlands — Two hundred and ninety-six papers have been accepted for publication before anybody had seen a result.

That is the entire intervention. A journal reads the question, the method and the analysis plan. It decides whether the study is worth doing. It commits in writing to print whatever comes out. Only then does the researcher run it.

Across the twelve journals that operate the format, 39 per cent of those papers report the stated hypothesis supported.

In the same twelve journals, in the same fields, in the same years, 15,014 papers reviewed the ordinary way report it supported 95 per cent of the time.

What the comparison is, and what it is not

We matched on journal, on field and on year. We did not match on question, because nobody has a taxonomy of questions.

That matters, and the editor whose reply runs at the foot of this page says so at length and is worth reading before anybody quotes the two numbers at each other. Registered reports attract replicators. Replications of famous claims come out flat more often than first looks at new ideas do.

So 95 against 39 is a ceiling and a floor. It is not a fraud rate. The pale bars in the figure are not eight thousand lies.

Then we tried to open the drawer

The other half of the claim is that the flat results exist and are being held somewhere.

We went looking. We took conference abstracts from 2016 to 2024 describing replication attempts, kept the ones that never became a paper, and wrote to four hundred of their authors with an offer: send us the null, we will publish it in full, at any length, at no cost, with your data hosted on this site permanently.

We wrote twice, in February and again in May.

Seventy-one replied to say they had one in a drawer. Nine sent files. All nine are published with this story, and one of them is a failed replication of a paper cited nine hundred times.

Sixty-two said no, and gave a reason.

The sixty-two reasons, counted

Nineteen, in one form or another: I cannot find the data because I never labelled it, because I did not think anyone would ever ask. One sent a photograph of a shoebox of unmarked USB sticks.

Eleven: the hard drive is lost, dead, or was thrown out in a lab move. Four of the eleven had looked. Two had not. Five did not answer when we asked.

Nine: the postdoc who ran it has left science, and nobody else knows how the analysis worked.

Six: consent or ethics approval has lapsed and sharing the data now would require a committee.

Five: they intend to write it up this year. Three of those five told a colleague the same thing in 2019.

Four: the co-author is now their head of department.

Three cited pressure, or the expectation of displeasure, from a sponsor. Three, out of four hundred letters.

Five gave a reason we could not classify, and their words are in the file rather than in a bucket.

Nine months, and nothing that counts

The arithmetic came up in almost every reply that ran past two sentences.

Turning a finished null result into a publishable manuscript takes something like nine months of evenings: the write-up, the reviewers, the revision, the data cleaning that was never done because nobody cleans data for a result they are not submitting.

At the end of the nine months the author has a paper that a promotion committee does not count, a hiring panel does not read, and a citation index barely registers.

None of this requires anyone to lean on anyone. It requires only that the reward for the nine months is zero, which it is, and has been for decades, and is the same arithmetic this desk found behind a cancer drug that nobody will fund because nobody can own it.

The two papers in PX-2185 — a claim and, five years later, the rerun that did not find it — remain unusual precisely because somebody spent the nine months. The pre-registered protocol in PX-2161 is unusual for the opposite reason: it was published before recruitment, so the nine months were committed in advance and could not be quietly reclaimed.

What this desk takes from it

The file drawer is real and it has no lock on it.

It is a drawer nobody has a reason to open, in an office where opening it is unpaid. That is a duller finding than the one people want, and it is worse, because a lock can be picked and an incentive has to be rebuilt.

And we should be plain about our own part in it. We offered four hundred people the best terms in publishing — unlimited length, no fee, permanent hosting, a readership — and we got nine files. Our offer did not shorten the nine months. Nothing we can offer does.