Should I start writing a paper early or after all research is finished?

Question

I am working on research that will lead to a paper. The research isn't finished but I've finished enough that I have a good idea of what the basic idea of what the paper will say and look like. Is it better to start writing the paper now and make revisions as my research progresses or is it better to finish the research, have firm conclusions already in place, and then start writing?

Answer 1

The concept of "finished" is problematic when it comes to research. I think that the same quote applies as for art: research is never finished, it is only abandoned.

Less poetically and more pragmatically, it is often only in the process of writing that certain critical aspects of the work become apparent. When a person is in the midst of working on a project, they tend to get very close to the material and begin to take as clear and obvious things that are very much not so for others who are not so deeply involved. Writing one's work up in a scientific paper forces one to step back and build those gone-implicit arguments from the ground up (or at least it does if you are writing well).

This often leads to discovering unexpected problems, which lead to new literature searches, new theorems, new experiments, and even whole new perspectives. I have had nearly the entirety of a paper change out from under me as we wrote it and revised it, and the work became much better as a result.

So, to return to your question, of when to begin writing up a paper. My advice and experience is this: begin writing when you think you have achieved the key results that you want to build the paper around. As you begin to do so, you will likely discover gaps that need to be filled in, which will shift how you write the paper, etc. When the process converges, you know you've got a good paper on your hands, and it is ready to submit into the tender mercies of your dreaded peers.

Don't let yourself move forward with the research, though, to try to achieve the next key result. It's fun and exciting to do new things, but you must also have the discipline to cross the Is, dot the Ts, and observe the little things that need to be corrected and might otherwise escape your notice.

Answer 2

To sharpen jakebeal's point a bit: my primary specific recommendation is that you not spend any significant amount of time polishing the paper until you're confident that very nearly the sum total of its contents are collected in front of you, literally or figuratively. A more-or-less-messy pile of scratch can be enough to facilitate the process of thinking through one's lines of argumentation, depending on one's personality and modes of thought, while taking a comparatively small amount of time away from continuing the necessary research/experimentation.

Just like it's often a terrible waste of time to plan most experiments or lines of research too far ahead, it's also typically a terrible waste of time to refine a manuscript too far ahead. You may find you've spent a couple dozen hours wordsmithing text that never finds its way onto an editor's desk.

Answer 3

For me, writing a paper is a process that is not unlike how an author writes a book. I am constantly thinking about the "story" while I am doing the research. While working on a research project, I will suddenly think of some nice manner of presentation, phrase or even a single word that capture nicely some aspect of the work and I write these down in a raw manuscript file. Then, as the project advances to a more mature state where I know the majority of the results I will jot down a very rough outline. The actual hardcore writing then consists of putting everything together.

So in short, I suggest to start jotting ideas about writing as early as possible, but don't worry waste time on organizing or polishing these notes.

Answer 4

It depends – on your content or type of research as well as on your approach to writing.

The two approaches to (scientific) writing I would like to distinguish are:

1. Start with writing a quick draft and then revise and restructure it many times.
2. Start writing with a clear structure in mind and try to optimise every sentence from the beginning.

In my experience, neither approach is generally better, but for most people, one approach is better suited than the other. If you are the person who prefers approach 1, you might start writing as soon as you finished an aspect of your paper; if you prefer approach 2, this may be a waste of time, depending on the content (see below). While there is a grey zone between the two approaches, I have not met anybody yet whose approach lies in it.

The types of content I would like to distinguish are:

• Modular papers: There are several chunks of work that have little interdependencies to each other. If you would practice extreme salami publication, you would publish each one as a single paper, with no paper building up upon an unpublished one. So while some of these papers would cite others, there would be no loops in the citation graph.
• Interdependent papers: There is no structure like the above. For example the results of experiment A lead to experiment B, whose results in turn inspire to repeat experiment A with other settings and so on.

Obviously, modular papers are much more suited for early writing.

To give an example from personal experience, I am the sort of person who prefers the second approch to writing and I wrote most of my papers so far after all the work was finished. Nontheless, I recently wrote a paper in a totally different style. However, this paper was a method paper, which I knew to be modular. I did things in the following order:

1. Encounter a lack of a method during research.
2. Have an idea for a method.
3. Look, whether somebody had the idea already or there is a better method.
4. Devise the core method.
5. Find central conjecture required for core method.
6. Prove conjecture.
7. Write down core method and conjecture (I started this step the very next day).
8. Perform theoretical runtime analysis of method.
9. Write down runtime analysis.
10. Apply method to artificial data to test its performance.
11. Write down results.
12. Devise artificial test case to compare method with best existing method and perform the comparison.
13. Write down results.
14. Apply method and existing method to real-life problem from step 1.
15. Write down results.
16. Write abstract, introduction and conclusion.

At no point in the process did I need to perform revisions to already written stuff other than adding a sentence for explanation or renaming a variable. While I am very happy to have done it this way and this saved me a lot of time, I also know that this approach would not have worked at all for any of my other papers.