Forms of life, forms of mind

There are many ways to write scientific manuscripts. The following is my personal process, and what I encourage most of my students to do.

There are remarkable people out there who can write linearly. They sit down, start at the beginning, and write a draft from beginning to end. They edit the draft, but the initial structure is solid so that all of the edits are a progressive improvement and eventually converges on a great final product. I am not one of those people. I have to write top-down (or as computer scientists would say, “breadth first“) – strictly following an outline format. So for example, for a Perspective piece or essay, I decide on the content by writing down a set of main ideas, ordering them, and then go on to the next level of the outline by putting down the sub-sections that are needed to develop each of the main sections. Rinse, repeat.

This cures a problem that plagues a lot of early stage academics: long hours spent writing and rewriting drafts which do not converge but float and morph laterally. The pieces shift and get repeatedly re-done because the detailed writing ends up in a corner that forces revision of large-scale structure. Some people dread writing and put it off until the last minute because they anticipate the chaotic process, or they write something that is not logical and clear and does not get their message across. Fortunately there is a method I have found to be very helpful to get over this issue.

The biggest idea here is to clearly separate the creative work from the mechanical work. I personally find it very difficult, in the same piece of text, to try to make executive level decisions about what I’m trying to say and at the same time wrestle with wording (or even add numbers/figures etc.). Thus, I like those processes to be entirely separate. The first phase is (to me) the most fun: the creative process of organizing ideas, deciding what story I’m going to tell and in what order. Thinking about how to convert my mind-map of relevant ideas into a linear structure suitable for a written piece. The key is that I don’t write any actual sentences and I don’t worry about wording until the very end of the outline – the bottom layer, after all the ideas are hierarchically in place. At that point, everything becomes purely mechanical. There are no more decisions to be made – it’s just “turn the crank” as you obediently follow the outline and put in the sentences around the detailed concepts indicated.

This is powerful, and effectively drains the most common barriers to writing. By not worrying about the wording or details at first, you get to focus entirely on the logic and the journey that you plan to take the reader on. This is the time to reorganize your own thoughts, to prioritize, to make decisions about what concepts are in and which are for later, and how everything relates to each other. That part is not scary because it’s a joy to play with ideas and how they relate and you don’t (at this stage) have to worry at all about how you will say anything (details or wording).

Once the outline is complete, you are relieved of the need to have deep thoughts: the rest is absolutely mechanical – nothing to decide, just go in and fill the details according to each bullet point. Trust the outline (or edit it if needed) – if the logic is sound, you don’t need to worry about it in phase 2 – just put sentences and numbers around each bottom rung of the outline. This makes the rest of the process benign as well – what could be simpler than mechanically filling in an existing structure? By segregating these two phases of the work, you diffuse the biggest barrier to continuous improvement in the draft – the chaos of having to make decisions and wordsmith at the same time. The first phase is like the prepatterning stages of embryonic development or large-scale appendage regeneration, where biochemical and bioelectrical scaffolds are set up indicating the large-scale logic of what is to come. The second phase is like wound healing, where purely local decisions are sufficient to allow cells to fill in gaps in a structure that is already set.

For primary data papers, I do it this way, in order:

1. Decide on the story, write the outline of how you plan to tell it and what will be included. Write the Abstract that summarizes what problem was addressed, how, and what was found.
2. Make the Figures. Start with a Figure title for each. These figure titles are the bones of the paper – each one is one large “Result”. I like titles that make a claim and have a verb in them (with the exception of Figure 1, which is often a descriptive schematic of a process or system). Each figure should contain the graphical elements needed to convince the reader of the figure’s title claim. Once you have the figure titles, that’s a good time to check in with your PI or your collaborators, before you go filling in the content, to make sure that the outline of the figures is solid.
3. Write the Results. Each subsection of the Results section has a subtitle, which tightly matches the Figure titles. It’s not that each Result subsection maps to only 1 figure necessarily, but they are at the same level and should match. My preferred formula for the Results is simple and designed to make it easy to fill in without getting stuck. The framework is that each Results subsection goes like this:
   • In order to discover/determine X, we took approach Y
   • We performed experiments Z, which showed that Q (see Figure N, statistics).
   • Therefore, we conclude that W.
4. Write the Methods – a full list of all the techniques you used with enough details for others to repeat the work. Especially mention whatever was tricky or likely to trip up others or keep them from getting the same results. Also write the Acknowledgements – anyone that helped, and funding sources.
5. Discussion: briefly recap the main results. Cite a summary figure if appropriate, which brings it all together into a model. Mention anything that is interesting, puzzling, or not obvious from the raw data. Interpret for the reader what all this means and why it’s important. Tie it in to existing work in the field – what matches, doesn’t match, extends, or overturns other work that has been done on this quetion. Add subsections for “Limitations of this study” (what was not done? what was not certain? what big issues remain open?), and “Future work” (what can be done next, and what is now possible given the answers in this paper). I like to add a Conclusion that ends on a positive statement that inspires others to be excited about this work, its overall field, and the possibilities of future research on it.
6. Introduction – I write it last. This is the most difficult part; it needs to be skillfully done, to frame the problem in the way you think as the most useful and let the reader know what the knowledge gaps are and why they matter. Give plenty of references to other good work on the relevant topic. The Introduction’s job is to make the reader need to keep reading – it sets up the problem as too fascinating to not continue engaging with the rest of the paper, and frames the work in the perspective which you want the reader to adopt on this problem. The content might be like this:
   • describe the overall field of science being addressed here
   • what is known – overview the relevant structure of past work and set the scene for how to think about it
   • what is not known – knowledge gaps, and opportunities
   • presage the content in the paper by saying which of the gaps you plan to address and how
   • a teaser sentence or two saying (to some level of detail) what you found and what the main claims will be that are shown in this paper: “Here, we XYZ”.

Introductions should, conceptually, be a funnel shape – leading from the broadest concepts to the specific details. The Discussion goes in the reverse order (from the specifics to the broader impacts of the ideas). Abstracts are a full hourglass, with the general->specific->general content, all simply said in ~250 words.

A point regarding references. Be generous with citations – cite all relevant prior work and make sure that the really engaged reader can track it backwards to the giants on whose shoulders you stood to do the work. If you’re dealing with a journal that limits citations (a terrible practice), at least make sure that the preprint you post before submission to a journal has all the references. Use a reference manager; I like Endnote and my personal library can be downloaded here. By the time you are writing a paper, you should have done enough reading and thinking that you know what needs to be cited; but it’s a good idea to let someone familiar with the field to read your draft and make sure you didn’t miss something or someone.

One other thing: while writing, do not worry about word counts (length limits) – thinking of cutting while writing inhibits the creative process. Say everything you want to say (assuming the outline is sound). Then later, go back and remove whatever is not essential; ask yourself about each paragraph: is this serving the mission of taking the reader where you want them to go? What work is it doing for the main theme? What will the effect be on the readers? And cut mercilessly as needed. Remember though that preprints have no length limit and there you can say everything you want to say.

Finally, the current paradigm of a “linear paper” as the only way to tell a story has to get radically expanded as we go forward. At some point there should be software and publication platforms where non-linear mind-maps and other structures can be published, allowing a much richer information conduit by means of which writers and their readers can interact and exchange ideas. Until then, here’s a suggestion for content that is really non-linear. Suppose your paper has pieces A,B,C,D where the reader needs to know each piece to appreciate the other pieces – like an interlocking puzzle that only makes sense as a whole. Which should they read first? Think of it as a spiral. Don’t try to say everything in a linear way, but do it in small circles and then bigger circles. For example, order it like this: a,b,c,d, A, B, C, D. That is, make a few short and simple statements about all of the parts, then really launch in. That way, when they encounter A, they’ve already seen b,c,d which helps to understand A even if it wasn’t the details of B,C,D yet. You can do multiple circles of increasing size – take the reader through the spiral of ideas saying each time just enough of the key concept they need to keep moving through the structure.

And of course remember that none of the above is written in stone – if you have a process that works well for you, use it – there are many ways to do lovely papers. I’ll do a separate post later on how I decide what to write, when, and for whom. That is a much bigger question, highly personalized, and all I can say is what I myself do.

---

Feature image by DALL-E.