Guidelines for Writing Academic Papers

A crash course in how to motivate and organize an academic paper.

Introduction

This page contains tips for writing conference and journal papers. I’ve found them helpful and would like to share them with you. If you read no further than this, remember:

• Have a story. A paper does not simply list results. It should introduce the reader to an open question in the field, and then describe what work has already been done to address this question. It should then explain the methods used in this work to address this question, such that a reader could (potentially) recreate the work themselves. It should then simply show the results without editorializing. Finally, in the discussion, it should interpret the results, explain how it contributes to answering the question from the beginning of the paper, how it relates to other work that has been done, and suggest ways to improve shortcomings in the study, or future work to more completely address the question. If your paper does not answer the question “So what?”, then it is not complete.

• Show don’t tell. It will be easier for you and your audience if you rely on pictures rather than text, and figures rather than verbal explanations. I often make a list of figures that I want in the paper before even collecting data. I make sure that seeing those figures, and nothing else, would already tell a story. This has two benefits:

  • A lazy reader would still get the point just by looking at the figures.

  • I know what data to collect to tell my intended story, rather than just collecting whatever I can get my hands on.

• Strike a balance between specificity and brevity. Good papers are short, clear and concise. In addition, conference papers almost always have a page limit. If a method is already explained somewhere else, cite it, don’t explain it again. If you want to refer to a result presented somewhere else, cite the original study, don’t write a page describing those results. Removing these other components will enable you to spend comparatively more time and space on the novel parts of your study.

  • An increasingly common way to address this issue is to write a brief manuscript with a substantial “supplementary document”, which may include additional figures, mathematical derivations, and other highly-technical pieces of information or data that would distract from the message of the paper, were it included in the main body text. Check the "guide for authors" of the journal you plan to submit to and read their policy on whether supplementary documents are allowed, and if so, how to format them.

Organization

Academic papers usually have the following components:

• Title

  • A statement that summarizes the main finding of the work. It should be brief, specific, and not sensationalist or arrogant. Here are some bad examples:

    • A robot model.

    • A Newtonian dynamical model of a six-legged robot modeled after the praying mantis can walk when controlled by a network built from nonspiking neuron models and conductance-based synapse models.

    • This robot solves all the world’s problems.

• Abstract

  • Answer the question, “What’s the point of this paper,” often with a limit of 200 words or so. This is a no-nonsense summary that lists the research question, what was done to try to answer this question, and what the findings were. If you don’t have a lot of experience writing abstracts, you may wish to write this last. Read the abstracts of your favorite papers to get a sense of what is included in and omitted from the abstract. I like to think of the abstract as how a robot would describe the paper.

• Introduction

  • Like the abstract, the introduction can be difficult to write if you don’t have a lot of experience writing them. The introduction should:

    • Motivate the work, by identifying a research problem or real-world issue to address (1 paragraph).

    • Describe the state of the art, that is, what is known related to this problem. This paragraph should cite many studies.

      • Consider using the phrase "to the authors' knowledge, no other study addresses..." to ensure that if a study does exist somewhere, your paper will not be immediately rejected (and you aren't snubbing the authors of that paper).

    • Describe methods that people have used to approach this problem, possibly comparing or contrasting the approach you plan to take yourself. Again, this paragraph should cite many studies.

      • Knowing the scope of whom to cite can be hard! Try to think about your audience, which is dictated by the conference or journal you are submitting to. At a biology conference, you don’t need to explain why animals are cool, but you may need to explain why models and robots are helpful to understanding animals. At a robotics conference, you don’t need to explain that some robots walk, but you may need to explain how the insect nervous system is organized.

    • Summarize the paper that you are going to write, “In this paper, we show…” (1 paragraph). People sometimes say you should “tell them what you’re going to say, then say it, then tell them what you told them.” I think this is a good approach. Consider this paragraph of the introduction the first part of that approach.

      • This is also the first part of the paper that is really about your work. It is considered informal to use the singular first person, “I made a robot.” Instead, you can use the plural first person, “We made a robot,” or the passive voice, “A robot was made.” I find the passive voice difficult to read, so I stick to the plural first person.

• Methods

  • Clearly and concisely explain:

    • The system you’re using (a robot, a test rig, control hardware, control software, etc.).

    • The mathematical bits underlying your work (mathematical neuron model, dynamics of a mechanical model, tuning techniques, data analysis techniques, etc.).

    • The experiments you ran (systems used to collect data, what assumptions you made, how you restricted the robot, how long the robot ran for, etc.).

• Results

  • Focus on figures in the results. A good results section may be five or six figures but only two or three paragraphs of text (although there is no set formula for success). You should provide only enough information for the reader to understand the data. You should not interpret the data for the reader. The results section should make sense for the rest of time. The interpretation of that data, however, may change, so keep it out of the results.

    • Also recall, this is the “tell them what you’re telling them” part.

• Discussion

  • The discussion is where you summarize and interpret results, relate results to other studies, and speculate about where the work will lead. A formula that I like to use is:

    • Summarize your paper, that is, “tell them what you told them” (1 paragraph).

    • Interpret the results, explaining what you think your results mean, and just as importantly, what they don’t mean (1-2 paragraphs).

    • Relate your work to related studies. “Bob did this, but I did this. I believe that my work more completely captures a particular phenomenon;” or, “Suzy did this experiment in an animal, and my results match hers. Therefore, it is possible that the nervous system is performing the same computation as my model.” Note that both of these are pretty “soft” statements. You should not:

      • Speak ill of somebody else’s work or personality.

      • Say anything with absolute certainty. Note the use of words like “believe” and “possible” above. You can make statements, but people will interpret what you write literally, so unless you have incontrovertible proof that what you are saying is true, qualify it. Technically science cannot prove anything, it can only rule out possibilities (i.e. reject hypotheses). Therefore, any "affirmative" statements must be qualified.

• Acknowledgements

  • Acknowledgements should be very brief.

    • They should list any and all funding agencies and grant numbers that supported the work. Such statements are often used to justify continued funding, so this is important.

    • If an organization paid for you to travel to the conference (such as your department, your university, the conference organizers, etc.), you should acknowledge that.

    • Authors should not be thanked in the Acknowledgements.

    • Non-authors who helped by reading the paper before submission, or having a particularly helpful or enlightening conversation, should be thanked in the Acknowledgements.

    • In a paper, you do not thank your mother, your partner, etc. Your shout-outs should be strictly professional.

• References

  • Cite all related works to the study. Specifically,

    • Cite works necessary to explain the state of the art in the Introduction.

    • Cite works that use similar or the same methods as you (often, you yourself or your colleagues).

    • Cite works that are worth comparing and contrasting to your own, in the discussion.

    • By the time you have cited these works, you may have cited 10-20 papers. Conference papers often have strict limits on how many papers you may cite (usually 10-20), so this may be sufficient.

      • If you end up with 3 references, then you need to broaden the scope of your introduction.

      • If you end up with 50 references, then you should probably reduce the scope of your introduction, or try to cite review papers instead of long lists of individual studies.

    • Journals rarely place limits on how many works you may cite. In that case, you want to be as thorough as possible. For a journal paper, 30-100 references is usually appropriate.

Tenses

One should make sure the tenses of the paper are clear and consistent. Possibilities, in order of decreasing clarity, are:

• Everything in the present tense.

• Everything in the past tense.

• Introduction in the future tense, and discussion in the past tense.

• Any other possibility (stick to the first three).

Layout

A conference or journal will almost always provide you a template for your paper. This way, the formatting is uniform throughout whatever book they publish the papers in. For MS Word, they will usually provide a file you can edit, with the proper fonts and headings. For LaTeX, they will usually provide a .cls file with the formatting details, and then a .tex file that you can compile to see what it looks like.

Increasingly, journals allow or even encourage you to submit papers with “no” formatting. This means

• Size 12 Times New Roman font, 1 column, left aligned/justified, easy to read spacing (maybe 1.15 or 1.5).

• No fancy journal logos or fonts or colors.

• References listed in easy-to-review (Author, Year) format.

• Lines numbered.

• All figures after the text. No in-line figures.

I like this method because it is easy for you the author to create, and easy for the reviewer to review (they can say “on line 123, you said…”, without referring to columns). I insist on no formatting for journal manuscripts.

Figures

It’s all about those figures, baby! Just like posters, vector formats like .eps and .pdf are great because you can scale them without losing resolution. However, MS Office handles them very poorly (it downsamples them into bitmaps, but badly), so you have two options for putting figures in papers:

• Using LaTeX:

  • Export .pdf files from whatever image editor you use, and then LaTeX will directly incorporate them into your document.

  • Make .eps files, and then use the package epstopdf to enable LaTeX to convert them into .pdf.

  • I highly recommend using overleaf.com to write your paper in LaTeX. It is a browser-based editor and compiler. It simplifies finding all of the right libraries. If you use Overleaf, I highly recommend you export .pdf files of your figures, and then upload those. Overleaf will not compile your paper if it takes too long, and forcing it to convert image files into .pdf will drastically increase the compilation time.

• Using MS Word:

  • Make figures however you would like, then export them to .png or .jpg files with the program you are using to make the figures. Such images should be at 600 DPI or higher. When you insert them into MS Word, you will likely open a Pandora’s Box of annoying auto-formatting. In my experience, there are two possible ways to improve (but not solve) MS Word’s image-handling issues:

    • Add the figures to the bottom of the manuscript, one figure and its caption per page. Use the “page break” option to keep them separate. This is how “no formatting” manuscripts should be organized.

    • Add figures where you want them into the text but do this after writing the entire body text. This can be difficult, because seeing the figures may help you write, and putting the figures in the text will help you gauge the length of the paper. But it will prevent figures from shifting around, losing their captions, and sitting on top of one another.

      • This method may be improved by having two documents, one with text only, and the other with figures and captions. This way you can look at both as you write. When the writing is done, you can pop the figures at the end of the text. See “no formatting” above.

    • Put the figures inside of other MS Word objects.

      • A popular choice is to put them inside a Table. Then, you can add the caption below the figure but inside the Table, keeping them together.

      • Alternatively, you can use the mythical Frame object, which is hidden by default. To access it, click the down arrow next to the undo and redo buttons on the top of the window. Click “More Commands”. View “All Commands,” and sift down until you see “Format Frame.” This will add a circle icon next to the redo button, which will enable you to click and drag a “frame” into your text. Think of it as a Table, but without any of the Table’s specific properties. It will just be a box you can put stuff in, add a border to, and define alignment and text-wrapping. I find these easier to wrangle than a naked figure with a caption.

Here are a couple other platform-nonspecific tips for figures:

• Colors are a great way to distinguish different objects or data sets in figures. However, conference proceedings are almost always printed in grayscale (but color online). Therefore, you should make sure that your figures are still clear, even without color. One way to do this is to use colorblind-friendly color schemes. For instance, Matlab’s default color scheme “Parula” plots everything on a blue-green-yellow continuum. This way, it is both colorblind and grayscale friendly. Other options are to use different line types (solid, thick, thin, dotted, dashed, etc.) to outline objects, or to use data markers in your plots (circle, square, up triangle, down triangle, etc.).

  • You can also use websites that check for colorblind accessibility, or Adobe Illustrator has a built-in tool.

• The text on figures needs to be legible. It is really easy to make a figure that looks nice in the editor, but then has impossibly small text when inserted into the document. Journals usually specify that text in figures is no smaller than the figure caption text (which is usually the smallest in the document). Ensure this requirement is met by making your figure the proper size in the editor, and then restricting yourself to using the correct font size. This may seem obvious, but it’s really easy to keep adding stuff to a figure until it ends up 20 by 30 inches, and then you need to shrink it down to fit in the paper.

  • When you keep the figure the proper size, nine times out of ten, the text feels restrictively large on the figure, but ten times out of ten, the figure looks much better with legible text.

• Figures need captions. At the very least, captions should serve as legends, explaining what each line or item is in a figure (necessary but insufficient to understand the figure). You may wish to include more explanation of the results, such that a figure and its caption may stand completely alone (necessary and sufficient to understand the figure). Which one you choose is up to your personal style. However, what you should not do, is

  • Copy and paste the caption into the body of the Results section. That is redundant. If you refer to the figure, you can assume that the reader will read the caption.

  • Include legend-like information in the body of the text, but not in the caption (“The green line is…”). All legend information must be in the figure itself, or in the caption.

• I highly recommend using a vector graphics program, such as Adobe Illustrator or Coreldraw. The advantage of these programs is that one can scale text and figures however they wish without losing resolution. I find this particularly useful for the following development loop:

  • Collect data in Matlab.

  • Export figures as .eps (encapsulated postscript, a vector image format).

  • Use a vector image editor to scale, recolor, and change line weight as appropriate.

  • Export to .pdf (LaTeX) or .png (Office) to insert into your manuscript.

Takeaway

The goal of publishing or presenting your results is to show someone else how to do what you did and share your conclusions based on your results. Recording your work is a wonderful way to contribute to the human tradition of science. Furthermore, recording your work will help you build upon what you have done in the past, and will help you train mentees who wish to work with you. Learning how to present your work logically and clearly is a critical skill for all engineers and scientists.