How to write a thesis, step by step

Deze pagina in het Nederlands

The nine steps

Every step ends in something another person can read. A step producing only a feeling of progress is not a step, which is why “do the research” sits so late.

• Pick a subject you can get data about. The interesting question you cannot get access to costs more weeks than the dull one you can.
• Reduce it to one research question, in one sentence, with a verb you can test. If it needs a semicolon, it is two questions.
• Break it into three to five sub-questions. Each becomes a chapter or a section; nothing else does.
• Write the plan — problem, question, sub-questions, method, planning — and get it approved before you collect anything.
• Set the document up once: citation style, page setup, a contents block, a word goal per section.
• Collect sources as you read. A reference reconstructed from a browser history is one you get wrong.
• Do the research, writing the theory chapter alongside it. Waiting rooms and transcription are the only free hours you get.
• Draft in the order the argument needs. Method and results first: easiest to write, hardest to fake.
• Revise in three passes — structure, paragraphs, sentences — never two in one sitting.

Chapter outline, and a word goal per section

A proportion, not a template. Results and discussion are together about two-fifths of a thesis, and the introduction is the shortest chapter people spend the longest on.

Word goals per section

A whole-document target tells you nothing in week 10: you satisfy it by writing 4,000 words of theory and none of method. So goals sit on sections too. The outline panel keeps a target per section, keyed to that section, and shows each as words / target with a hairline meter; the header carries the overall figure as a percentage. The useful number is not the total but which section is furthest behind its own.

Twenty weeks, week by week

Weeks are the unit because a supervision cycle is a fortnight. The two blocks people cut when they fall behind are the last two, which is exactly backwards.

The four settings that decide week 19

Citations, in four styles

Four are supported, named in the code rather than approximated: APA 7, MLA 9, Chicago and Harvard. All four are author–date or author–page, which is why switching is a setting and not an edit — the in-text marker is derived from the source, so reordering the bibliography never renumbers your prose. Two authors get an ampersand, three or more collapse to et al., and the list sorts by family name, then year.

You add a source by pasting what you have: a DOI, arXiv id, ISBN, BibTeX record, URL or raw citation is detected and parsed, and anything the parser inferred is flagged uncertain. The limit worth knowing: the shipped resolver never touches the network, so a bare URL yields a title guessed from the slug — and says so.

Footnotes and endnotes

Notes are numbered in document order, computed without a DOM, so the screen, the export and the assistant cannot drift apart. On screen they always sit at the end: a browser cannot print at the foot of a page, because float: footnote is in the CSS spec and in no shipping engine. The foot-or-end choice is therefore about the exported file — and one honest gap is that the Word writer emits real footnotes either way, so a programme demanding endnotes needs one conversion in Word.

A contents page that cannot go stale

It is derived from the document's headings on every render and nothing is stored. That is the whole feature: a stored contents page can disagree with the document, which is why every word processor storing one also has an “update field” ritual. It is a block rather than a fixed strip, so it sits where the document wants it.

Cross-references that renumber themselves

Two independent sequences run through the document, Figure n and Table n; pictures and charts share the figure sequence. Nothing is stored, so a figure inserted halfway renumbers everything after it — and every reference to it — in the same keystroke. A reference whose target you deleted renders [missing reference] rather than leaving a sentence that reads “as shown in , the effect”.

Handing it to a supervisor who wants Word

Supervisors review in Word with track changes on, and most writing tools treat that as an export problem. It is a round-trip problem. The .docx written here is real OOXML, not Word-flavoured HTML with the extension changed: footnotes Word lays out itself, w:ins and w:del tracked changes it can review, a header and footer with a real PAGE field, and a contents field, TOC \o "1-3" \h, with the current entries baked in as its result — so it reads correctly on open and updates on F9.

The way back matters more. Reading a .docx brings in headings, lists, tables and footnotes, and your supervisor's tracked changes arrive as proposed changes rather than flattened text — accept or reject them one at a time, or all at once across blocks you have not opened. Page setup covers A4, Letter and Legal, portrait or landscape, margins in millimetres, and page numbers bottom-right, bottom-centre or top-right.

Three limits. Headers, footers and page numbers do not work in a browser print at all — CSS margin boxes are unimplemented in every shipping engine — so they go into the Word export only. PDF export is window.print() into a clean window, not a generated file. And equations reach Word as LaTeX source in a monospaced run, because LaTeX to OMML is a compiler rather than a function. If your faculty runs on Office, if your university hands you Microsoft 365 has the rest.

Where Word is still better

A thesis is the one document where the incumbent is genuinely hard to beat. The rows go both ways.

Word wins on what the print industry standardised thirty years ago, and loses on what has to stay correct while you edit. If your results chapter does arithmetic, formulas for your results chapter covers the functions worth learning, and writing in a library with no wifi covers the week your connection is not part of the plan.

Questions people ask