Publishing stats papers in general science journals

I got an email request to write a blog post about how I go about trying to publish statistics papers in general purpose journals like Nature, Science, PNAS, etc. One recent example of such a paper is our work on the problem of post-prediction inference that appeared in PNAS. But we have a few others over the years in these journals.

I'll start by saying that I don't think I have any special secret sauce. I'm going to say the things that we usually think about when targeting these journals but your mileage will absolutely vary.

First there are two types of work I have had success in getting into these journals: (1) opinion pieces and (2) scientific work. In general my experience is it is much easier to get statistics ideas into these journals as opinion pieces rather than research. Some journals like PNAS have an entire subset of the editorial staff who are statisticians, but many general science journals do not. If there isn't an editor with explicit expertise in statistics, statistical research is going to be very hard to publish in that journal. There isn't anything nefarious going on here. As a former editor, if a paper is way outside my expertise I won't be able to tell if critical referee comments are serious flaws or simple disagreements. This sets the bar at getting research published in the journal squarely at the level of your paper basically having no major issues come up in review...this....rarely happens for me. Opinion pieces, on the other hand tend to be a bit easier to do since you can avoid diving into complex statistical discussions and focus on high level issues. This is comprehensible to most technical editors at these journals who are really good at jumping between fields at the high level.

A few specific thoughts when I'm trying to write one or the other type of piece are:

Opinion Pieces

I've had the best luck with these when I'm responding to a popular trend, a recent scientific debate, or a paper that just got published in the journal.
I tend to have the most luck when the topic somehow touches on areas where I already have some recognized expertise.
I try to keep them short and truly opinionated, since those tend to play the best for me at these journals.
I realized the best workflow for these types of pieces is to develop a relationship with an editor - at a conference, through cold emailing, on social media - and then just run the idea by them. They can give you a pretty good idea of whether it is interesting or not.
Opinion pieces tend to be much more of a collaboration between you and the editorial staff member assigned to your piece. That has a couple of implications
1. Don't over-optimize the content before you submit. It will get edited a lot.
2. Try not to overwhelm the editors with tons of ideas, only bring something to them you feel really strongly about.
3. Be ready to do a comprehensive revision with the editor acting almost like a collaborator rather than a pure form reviewer.
Be timely. If you see a fresh issue developing, write something up quickly and go for it.

Research Pieces

Here again, I've had my best luck with these journals when they have a specific statistical editor or statistical review track.
If you go to a general purpose journal without a statistics editor, you'll need to downplay the statistical work and really focus on the scientific side. Sometimes this is a good tradeoff for the higher visibility, but sometimes its worth it to avoid these general purpose journals to be able to focus on the method.
When writing for a general interest journal it helps if your method is general interest. One way we have had some luck in showing this is making sure you show applications of your method in wildly different fields. Our most recent piece showed applications of post-prediction inference in genomics and in verbal autopsy analysis.
Use a cartoon to illustrate what your method is doing as the first figure. Since it is a general audience, they will appreciate a cartoon to explain the high level concept.
Alternatively consider starting with a simple simulated or toy example to make the concepts super clean and clear, then move into the real examples.
Make sure that the real data results are clear and show both a statistical and scientific improvement.
Most of these journals require you to format the papers as "results first" and then methods later. Making the statistical model the result only works if you have a statistical editor.
Have someone who isn't a statistician review the paper - ask them to explain the method to you and why you are using it. If they can't, you need to revise until they can.

That's about all I can think of off the top of my head. I'll say that for every one of these I've published I've probably had 3-5 rejected. So I guess my last piece of advice is my usual academic advice. Always be submitting.