On Academic Writing

It is only until recently that I come to realize the importance of professional academic writing. After revising one of my own manuscripts over and over again with my advisors, and reviewing two papers for journals, I feel it is better if I write down some thoughts.

The purpose of writing a paper is like a presentation - to propose something, tell a good story, and convince people. The only difference is that it is not presented orally. You may not always get recorded by a video camera when giving presentations but your paper is there forever, which means extra care should be taken when telling your story.

DOs

My advisors have given me ample education on good and bad practices during my 3 year PhD training. I learned some by myself too. Here are a few.

  • In the introduction, some existing work are often reviewed. One should try to establish the connection between these works, and put them into groups. Start with the motivation of doing this work, then explain what have been done before, and point out what they have not been able to tackle. Then proceed to your method, briefly state how yours solves the problem.

  • In proposing your methods, make sure everything is clearly defined. Do not assume people will look over your paper and trace it out.

  • Care about reproducibility. Many journals used to ask for code and data being presented as supplemental materials. Nowadays, journals are accepting GitHub repositories. Carefully documenting your code will be a much appreciated practice.

  • If your method is based on a very famous one, like LASSO or spline, or you are using MCMC, there is no need to repeat the details in the main text. Adding a proper reference is enough.

  • Make sure the simulation study settings are in accordance with your theory.

  • If things like bias, MSE, are presented under different parameter settings, figures may be better choices than tables, as comparisons can be easily seen.

  • In real data application, explain what pre-processing have been done to the dataset.

  • Interpret the results. Just presenting the numbers/graphs without any comment will look confusing.

  • Pay attenton to typography as well. Although it is the method that mainly decides your paper’s quality, the way it is presented also has some impact. For example, use \log and \exp instead of using log and exp.

  • In producing figures, use .pdf or .eps instead of .png. View each figure from the reader’s perspective, and evaluate if this figure is informative, both itself and the axis labels.

  • When you are using numbers less than 10 in a sentence (in non-mathematical contexts), you should write the numbers out (instead of using their numerical values) i.e. write “four” rather than 4. (credit to EDS)

  • When the “u” sound is made like a “y” sound, as in uniform or university or unique or useful, you use “a”; if the “u’ sound is more like an “uh” sound, as in umbrella or umpire or undeniable, you use “an”. (credit to EDS)

  • Be careful about notations. It is easy to mix-use one character for different variables/quantities over the entire text if there are too many variables to denote. For example, the identity matrix and indicator function can easily confuse people.

DONT’s

  • Never use words like “actually” or “somehow”. These make you look not confident or objective.
  • Avoid using “we can”. As a Chinese who uses English as the second language, I sometimes do a direct translation from Chinese to English, but “we can” do not read objective enough, and more like a subjective thought.
  • Do not use abbreviations in spoken English such as it’s or we’re, which look very unprofessional.
  • Do not write incomplete sentences, such as ‘Analyze the results via … method.’
Yishu Xue
Yishu Xue
Data Scientist / Coder / Novice Sprinter / Gym Enthusiast

The night is dark and full of terrors.