On Writing

Each year I supervise several data-science master’s students, and each year I find myself repeating the same advises. Situation has worsen since students started (mis)using GPT models. I therefore have written this blog post to highlight few important, and often overlooked, aspects of thesis-writing. Many points apply also to writing in general.

On writing

“Easy reading is damn hard writing.” (Nathaniel Hawthorne). An inexperienced writer can easily take two working days to write (well) half of A4 page. Writing properly becomes even more challenging if you’re stressed because of a deadline. So, allocate ample time for the writing process. Working few hours each day, most students need about 3-4 weeks for writing.

Not all theses follow the same format. For the structure of your own thesis you can borrow guidance from other documents, such as a related academic paper or a well-graded thesis in the public domain.

Write for your audience (for your thesis committee in this case) rather than for yourself. Pretend to read what you wrote as if for the first time. It will help you identify incoherent sentences, and logic which is clear in your head, but not yet clear on paper.

It’s about the reader’s energy. Trivial, or vague, or too general sentences don’t add any informational content. So avoid those at all cost. Such sentences deplete the reader’s energy. By way of example, while syntactically correct the following sentence is a pathetic energy thief: “This underlines the critical challenge and ongoing research into methodologies aimed at skillfully navigating and optimizing the accuracy.”

On content

Growing up, we were compelled to write at length (”your answer should be at least such and such pages\lines..”). Of course we need exactly the reverse. The world today values brevity.

Write with courage. Be precise and concise. Avoid qualifiers such as “in a way”, “possibly”, and “somewhat”. Are you sure you need the word “almost”, the word “sometimes”? Get to the point quickly. What do you do? Why do you do? What do you find? Why do you find that? What is interesting? Why is it interesting?

Do not use the output from a GPT model as is. PLEASE!! it’s clear when you do, it creates antagonism, and most of all it’s tiring. Using a GPT model requires an initial understanding of your message. Use a GPT model to help you better your writing, never to replace it. In our context GPT models are for refinement, and not for creation.

Strike balance between complexity and brevity. Independent thinking is usually rewarded – but if you do something original, make sure you spend enough time on it until it’s crystal clear, rather than glossing through it and instead spending too much time on trivialities which can be relegated to an Appendix.

On presentation

Tables or charts should be only called for if they are better in delivering the point you want to make. Conversely, text is preferred when it is more effective than using a table or a chart. You should never struggle with questions like: “why is this chart here?” “what is the insight from this table?” If you pause before answering, consider dropping it.

Help the reader by providing long captions for tables and charts. Explain what do we see directly in the caption, rather than sending the reader searching the text for the point made in the table/chart.

Make sure tables and chart are presentable. For example, not 0.7739904 but 0.78. Don’t mix your code with your chart (what the hell is d_561..?). What is on the axis? What are the units here (e.g. pounds, dollar, euros, thousands, millions)? What about alignment? columns in a table, figures etc.

Include sources for all images. For images you created, use “Source: Author’s calculation.”

To place your research in the context of existing work, you must read a lot. Cite relevant papers in the main text, but refrain from citing sources outside the main narrative, even if you found those sources relevant for your work. And of course, sight what you cite.

Done? wait 24 hours. Re-read all of it again (yes, again, even though you are probably sick of it already). You will find spelling mistakes, incoherent phrases, and other details in need for corrections. Make the adjustments and submit thereafter.