I think the methods would be the most important element of peer review and commitment to publish. If you have a good method, the hypothesis is almost irrelevant--it's important to understand what variable is under test, but predictions of what result will be found are somewhat arbitrary.
I would love the idea (for data-driven science) of a peer review before execution of the work. Write up a one- or 2-pager, review methodology and accept/reject based on that.
Once the work has been done, there would mostly be editorial comments.
Of course, that ignores the analysis section, which is somewhat important.
But still, a vast improvement over today's way, where you may end up spending a lot of time only to get a vague reject.
It's unclear to me precisely what you mean, and we may be agreeing here.
But the 'variable under test' is definitely part of the hypothesis, and a study without a hypothesis is, not useless, but much less likely to produce a useful result.
If the (pre-registered) hypothesis was that green jelly beans cause acne, then this is at least an interesting result.
If you run this experiment with no particular hypothesis, and then decide on that basis that green jelly beans cause acne, this is just a setup for a later failure to replicate.
At best. At worst, no one bothers checking your results, the company stops selling green jelly beans due to bad publicity, and people who enjoy the green ones are deprived of them for no good reason.
I think we probably agree, but we're talking about slightly different things.
In the proposed experiment, the hypothesis is, "Green jelly beans cause acne.", and the variable under test is "whether or not green jelly beans cause acne". The hypothesis is basically a statement of what variable is under test AND a prediction of what the results of the test will be.
What I'm saying is that the prediction part of the hypothesis is basically a statement of bias. It's useful in noting what an interesting result would be and what the biases of the researcher might be, but it arguably is counterproductive in that it takes the focus off of observing the variable with an open mind.
If we simply preregister the variable under test, that gives us everything we actually need from the hypothesis, and avoids the problem you describe: instead of testing 20 different variables and only publishing on the one that yields a low-P result, we test the single variable and that's it.
That's a narrow view of publishable results. There's a ton of really useful science done by simply filling in a curve with measured values. There's no hypothesis required. Think eutectic curves in metallurgy and things like that. ...But the methedology being sound is critical.
