The relatively rigid format of research papers is often a pain point in scientific writing: What goes in the introduction vs. the discussion? How much detail should be in the methods section? When should I mention limitations of my analysis and why? How do I do this without it coming off as stiff and formulaic?
After awhile, the inflexibility of the form may actually reveal itself to be freeing, because it provides a template you can use to ensure that your ideas are clearly organized and communicated in a way that as broad an audience as possible can read, extend, and potentially replicate. The key thing to remember is that the sections of a research paper each have a distinct role to play, but also should cohere and interact. Put another way: Your paper is an extended argument about the relationship between a question (Introduction), the way you decided to answer it (Methods), what came out of it (Results), what it all means (Discussion), and what it implies for what is next (Conclusion).
If you start with a clear idea of what you want to accomplish with each of these sections, it will free you up to focus on their content and (hopefully) make your writing process less painful, more productive, and more successful.
Some questions to ask yourself:
- What is the major idea/contribution of the paper? If you have more than one, you either need to reconsider what your main idea is or think about splitting into two papers. The reason to do this is simple: you have not much space and not much attention from other people and you want to make the most of what you have. Sticking to 'one idea per paper' is in general a good practice to ensure that your message comes through as clearly and impactfully as possible.
- What are your personal/professional goals in writing this paper? Your career stage and level of engagement with the topic can and should dictate the way you approach a paper. Is this the publication you want to showcase when looking for your next job? Is this part of building a research program on your own by testing out new ideas?
- Who is your intended audience? This could also be described as "what do you want to accomplish with your paper?" Is the goal to showcase some kind of incremental improvement to existing approaches? Push back hard against a mistaken idea? Contribute to policy or public conversation? All of them are valid; you should just have an idea of who you are trying to reach and why?
- What is the right outlet for this work? Do you plan to submit to a broad field journal (AJE, Epidemiology, AJPH), a speciality within a field (e.g. Environmental Health Perspectives, Clinical Infectious Diseases), a methods-focused journal (Statistics in Medicine, Statistical Science), or a general audience journal (PNAS, Nature, Science). This will dictate your style, organization and length. If you are thinking of going for a high-risk/high-reward outlet, think about whether that serves your goals in 1 & 2, also think about if you want to take the time to go through multiple submissions or if you'd rather go for something that is more of a sure thing and keep moving. It is easy to forget that submitting and resubmitting papers is a lot of work and that a paper in a good but less-prestigious journal can be as or more helpful to you and the field than a super high-profile one.
- What is the problem your analysis is meant to address or solve? Use the first paragraph or two of the introduction to outline the scope and importance of the problem you're addressing.
- What else has been tried? This is the part where you get into the approaches that have been taken to this problem and the results that have come out of this.
- No need to be negative in this part: This isn't about why everything else is awful, but just what has already been accomplished, what we've learned from work done before by others (and potentially ourselves).
- Nevertheless... This is where you can identify the gap left by the previous work. For example may fit into one of these example categories:
- Previous analyses got at an important question but not with the kind of data you are using, which may be more: detailed, contextually relevant, etc.
- Previous work used a methodological approach that was not able to get at some important dimension of the problem, over/under-stated variability in outcomes, etc.
- Earlier work didn't address the broader context of the problem, i.e. too narrowly focused on a specific dataset or place and less on understanding the processes that cut across contexts.
- In this paper we will... For the love of everything holy, please tell me what you are going to do before you do it. This is the point of departure for your reader on the journey that will be reading and metabolizing your paper: give them a map!
- Use this part to mention the data you will use, i.e. where it comes from and what outcomes you are focusing in on.
- Also describe the methods in brief - don't get into a ton of depth, but just mention what you're going to do and provide a 1-sentence justification.
- Introduce the methods and data in broad terms. Use the first paragraph of the methods section to do a bit more 'signposting', i.e. giving the reader a sense of the reasoning behind the methodological approach and for using a particular dataset to address your particular question.
- This applies even to methods/stats papers: In most cases, the introduction of the entire paper should be largely focused on the epidemiological or public health problem at hand. If it is a more methods-focused or statistical paper, that may not entirely be the case. But even then, the introduction should be motivating the scientific importance of the method rather than getting into the nuts and bolts of the approach.
This section can and should be a bit more formulaic than the ones that came before. It really is as much 'just the facts ma'am' as possible, but the trick is highlighting the facts abut the data that are most relevant to what you're trying to do. The motivating question in putting this section together - which includes figures, tables and written description - should be: "What does the reader need to know about the data to understand the results?"
- What can go into a figure? Visualizing data is almost always preferable to talking about it or presenting summary statistics in a table. But it requires care: Think about what view(s) of the data is most important for a descriptive 'Figure 1' that provides an introduction to the data and the problem. Remember that the point is not to make the reader as much of an expert about the data as you are by the time you write the paper, but be familiar enough with the data to understand both the motivation for the analysis and the results when they come along.
- Do I need a table? Really, do you need a table? Ok fine, if you do: Keep it short and sweet - present relevant information that is hard to get into a single figure (e.g. cramming the number of people in a study, the proportion in different age groups, attack rates etc. into a figure could be difficult and annoying b/c of the different scales).
- What do the data measure?
- Include outcomes, covariates, etc. If you have multiple sources (e.g. case data + census) introduce both.
- When/how were they collected? Who collected the data? This is what is sounds like: Give a sense of the provenence of the data. If they come from a larger/long-term study, refer to earlier analyses from the study. If a published study protocol exists, be sure to cite it.
- Any important caveats. Emphasis here is on important: What does the reader need to know about what the data don't show in order to make sense of the analysis. Don't spend time here getting into nuance - you can do that in the discussion.
As methods-y people, we often want to get into a lot of detail here (I will speak for myself at least!). But this is another moment where you really need to consider your audience and overarching goals carefully. You are probably doing a number of things in the paper that merit explanation in the methods section, but the ones you highlight in this section should fit the following criteria :
- What is the most important thing you're trying to do with your paper, and which methods are most crucial to understanding that This is another way of saying that you should focus on the most model or set of models in the paper that are most important to understanding your results and advancing your overall agenda with the paper. You may have sub-analyses that provide extra detail but don't need to be outlined in depth in the main part of the manuscript. In the worst case, you can always refer the reader to a supplement containing this information.
- Which methods are important enough to describe is a function of your audience. Think about what kind of paper this is and who you expect to read it; that will dictate what goes in this section. The more generic the audience is, the more likely you will have a lot of the methods in a supplement, whereas for a more specialist audience - either a sub-field in epidemiology or a more methodological journal - you will probably go into more depth.
- How does what you're doing methodologically relate to the major questions of your analysis? Remember that the methods section exists to give the reader the ability to understand evaluate the results you are presenting to them.So, if it's not going to be in the results in some way, it shouldn't be in the methods.
The point of the results section is to make the results of your analysis - parameter estimates, posterior predictions, model simulations - as clear as possible to the reader. The first part of that is figuring out the best way to communicate each relevant piece of information. I would boil this down into a set of simple-ish suggestions:
- If it can be conveyed visually, do it! Prefer figures over tables and in-text descriptions where you can. This is subject to limitations that force you to prioritize what goes into a figure and what doesn't: How many figures are allowed by the journal? Is it enough information to take up a whole figure or would briefly mentioning it in the text be a better use of space?
- Figures and tables should stand on their own. A reasonably informed reader should be able to get what is going on from looking at your figure and reading the legend, even if they haven't read the rest of the paper. This is not a hard-and-fast rule, but if you work towards it you will ensure that the figures convey as much information as possible.
- Each figure should make a clear point of its own. If two independent figures convey overlapping information, try to combine them either into a single panel or multi-panel figure. But each full figure should touch on a single idea/result.
- If you have to make a results table, keep it small and simple. Big, complex tables are where reader attention goes to die. If information is best conveyed by a table, be sure to include the minimal set of results needed to make sense of what is going on.
- Use the beginning of the results section to hit the highlights in the figures and tables. Imagine you are explaining the figures to someone: What is the most important thing you want them to get from the figure? Talk about that in the beginning of the results section. Whatever you do, do not recapitulate entire figures and tables. They are part of your results, you can and should refer to them, but they should be complementary to what you are writing here not duplicative or completely disjointed from that.
- Use the remaining text of the results section to provide information not in the figures. What else is important to know that isn't camptured by a figure or table? Is there a single estimate from a side analysis that fills in the story but doesn't warrant a figure or table on its own?
The discussion is where you get to be a bit more expansive and opinionated. As ever though, think about how each of the things you put in here will impact what you are trying to accomplish. I tend to think of the discussion as having its own subsections that roughly look like this:
-
First paragraph of discussion: Summarize, summarize, summarize. What did you accomplish? How did your results relate to the problem/hypotheses you laid out in the intro? If I hadn't read the paper at all but just opened it to the discussion and read this paragraph, I should be able to get what you did.
-
Second paragraph of discussion: Sell the product. This is it: time to make the affirmative case for what you did! Why is it important? Why was your approach well-suited to answering the question? What gaps have you worked towards closing that you highlighted in the intro? Again - no need to be negative about other work, just show how you have moved the ball down the field in some meaningful way.
-
Third paragraph: Limitations. Writing this part can be uncomfortable or scary sometimes b/c it seems like you are being asked to undermine your work. But a well-written limitations paragraph should add to your credibility by showing you have thought about what can and can't be done with the data available and the methods employed. Try to answer the question of "What question might someone else want to answer that my paper doesn't/can't address?" This is a different question from "What did I do wrong?". Things that are substantial enough to seem 'wrong' are things you need to fix before writing. Limitations are reasonable stopping points for what you are trying to do that demarcate the boundaries of your analysis.
-
Fourth paragraph: Conclusions. This is where we tie everything together: What does it all mean and why does it matter? What comes next? What is the big point you want to stick with the reader? Here is where you can get a bit more opinionated and editorialize a bit - but be careful not to go too far beyond the data you have or the results you have generated.