Writing

• Each page of text should contain 3-5 paragraphs. And each paragraph should be concerned with one thing. Make sure the first sentence in the paragraph clearly outlines this. (See more below.)
• Your introduction should state your research question, preferably highlighted (in italics or bold font). See below
• The introduction is the most important section in your thesis, period. I have written a separate guide on how to write the introdcution which you can find here.

How to structure a paragraph of text

The first sentence should inform the reader what the paragraph is about as well as why it is logically placed where it is. Here are two examples that have the same contents but present them differently.

1. Don’t: Table Y shows a different set of results. These results are very similar to those in Table X. They are based on a different specification with fixed effects at the firm level. Such fixed effects control for time-invariant confounders, meaning that the results in Table Y are more robust than those in Table X. In this sense, by comparing the two tables, we may conclude that time-invariant confounders are unlikely to be a major issue.
2. Do: A natural critique of the results in Table X is whether there could be any omitted variables. To address this, Table Y presents the results from a specification with firm fixed effects. These fixed effects control for any time-constant omitted confounders that might otherwise bias my results. As is evident, the results are extremely similar, so I conclude that time-constant confounders cannot be a major issue.

• In 1, the first sentence gives no indication of the logic behind how table Y relates to table X. It is as if the paragraph is just a part of a long list: Table X shows this; Table Y shows that; Table Z shows this. It is exceedingly tiresome to read text like this since the reader is not given any guidance as to the thought process behind why the author has chosen those three tables, nor how they relate to each other. Those thoughts become clear only by reading the full paragraph and stepping back and thinking oh that’s why Table Y comes after Table X. Always give the reader this crucial guideline before and not after presenting the long convoluted analysis.
• What 2 does: The reader is able to skip the entire paragraph without losing track of the narrative arc. If all the thesis is written like this, it is possible to understand what happens by always reading only the first sentence. Then, if the reader is feeling curious or disagrees, she will choose when to wake up and engage and fully concentrate on the contents of a particular paragraph.

The research question

The single hardest thing in a good thesis is the research question. It is not possible to write this in its final form before after the thesis is fully written. But it very much is possible to write it in a broader form at the start of the work and then refine the question as you narrow in on what specifically you want to investigate.

It is very difficult but very important to work very hard on the research question. When you hand in, you should ideally have had 30 different versions of the research question as you experiment and zoom in. Tiny seemingly inoccuous differences in how you phrase your question can make or break a good question. A good question should

• Be focused around a tradeoff
  • E.g.: What’s the optimal fuel tax that balances consumer welfare against climate change?
  • A good question has rich arguments for prioritizing both sides of the equation
• Hint at the policy question (i.e. who the issue matters for?)
  • E.g. How can we design markets to mitigate the anti-competitive effects of pricing algorithms? (implicitly, the policy maker is the market designer)
• Motivate all of your thesis
  • If you have two key things you want to study, then your question must be sufficiently broad that it cannot be answered without addressing both things.
  • This will also force you to think carefully about why you need both things in your thesis and help you avoid the following thesis structuer: In the following, I will cover 7 unrelated regressions that I chose to run in order they occurred to me. All are interesting in their own regard.
• Be as precise (and thereby narrow) as possible