This is a continuation of a long topic, addressed by post number rather than by page. Start at post 1 · go to the accepted answer.
Author track record is a weak signal and it is a signal. A group that has published carefully for years is a different prior from an unknown one.
The arithmetic in post #89 is right; the assumption feeding it is the part to check.
Two things can be true about preprint at once: the mechanism is plausible and the evidence for the size of the effect is thin. Most of the argument here is people defending the first against attacks on the second.
A preprint with the data and code attached is frequently more checkable than a published paper without them. Reviewed and reproducible are different properties.
If that is already documented somewhere, ignore me and link it.
Everything in post #93 holds. The case it does not cover is the one I have.
Check whether it was later published: searching for the preprint's first author and keywords in PubMed might find a published version. The published version is the stronger citation if it exists.
If this contradicts something upthread, the upthread version may well be the better one.
Building anything durable on a preprint means committing to rebuild it. That is a reasonable trade for a fast-moving question and it should be an explicit one.
On post #98 — agreed on the reasoning, with one qualification.
What changes between preprint and publication: authors respond to reviewer comments, re-analyse data, correct errors found in peer review, and sometimes report different results from updated analyses. The published version is not the same as the preprint.
It took me longer than it should have to see that.
Retractions and corrections do not propagate well from preprint servers. A citation that was fine when made can quietly become wrong.
I have been on both sides of the preprint argument in this category within eighteen months, which should tell you how strong the evidence for either side is.
Post #101 is the version of this I will quote in future. One addition.
Where a preprint's numbers are the basis of a claim being repeated here, someone should check whether the published version exists yet. That check is nearly always overdue.
I am aware this is the third time this month I have made this point.
Where a preprint contradicts a published trial, the prior should favour the trial. Where it extends one, the preprint can be genuinely useful.
It is the sort of thing that seems obvious in retrospect and was not at the time.
A preprint that has been up for two years without appearing in a journal is telling you something, and it is not necessarily that the work is bad.
Everything in post #105 holds. The case it does not cover is the one I have.
Screening on a preprint server is minimal and is not review. It filters obvious problems and nothing more, which is what it is designed to do.
I looked this up rather than remembered it, which is the right order.
Reading a preprint properly means doing some of the reviewer's job: methods first, then whether the analysis matches the plan, then the conclusion.
Not the whole picture, but the part of it I can speak to.
Building on post #108 rather than restating it.
Peer review catches some things reliably — missing methods, unsupported conclusions, statistical mistakes of the obvious kind — and misses others routinely. Knowing which is which makes preprint reading easier.
I have deliberately not rounded that, because the rounding is where the argument starts.
The useful distinction on preprint is between what was measured and what was inferred from it. Both end up in the same sentence and only one of them has error bars.
If you cite a preprint in a maintained document here, the convention is to label it in the sentence and to note the check date. The document review will catch the update.
Answering the question post #110 raises rather than the one it answers.
Always label it as a preprint in any discussion. A preprint is not a published trial and it is not settled knowledge. Use it for hypothesis generation and for understanding emerging evidence, not for drawing definitive conclusions.
I would treat the number as indicative rather than as a measurement.
Taking post #112 at face value and following it one step further.
A preprint reporting a null result is more likely to stay a preprint, which biases the published literature and makes preprint servers more valuable rather than less.
I am confident about the direction and much less about the magnitude.
Post #114 and I disagree about the size of the effect, not about the direction.
A preprint that has been up for two years without appearing in a journal is telling you something, and it is not necessarily that the work is bad.
A qualification I should have led with rather than closed on.
Two claims get bundled together under preprint and they need separating. The descriptive one — this is what was observed — is usually well supported. The causal one — this is why — usually is not.
Almost every disagreement in threads like this one dissolves once you say which of the two you are making.
Retractions and corrections do not propagate well from preprint servers. A citation that was fine when made can quietly become wrong.
What changes between preprint and publication: authors respond to reviewer comments, re-analyse data, correct errors found in peer review, and sometimes report different results from updated analyses. The published version is not the same as the preprint.