How to write a dissertation: structure, supervision, failure modes

How to write a dissertation: why the trouble arrives late

Most advice about how to write a dissertation is advice about habits. This brief is about structure, because the failures that sink a thesis are structural and they are committed early.

Doctoral candidates who get into difficulty rarely do so because they stopped working. The pattern is more specific and much crueller: the work proceeds, the reading accumulates, the data arrives, and then somewhere in the second half a chapter refuses to be written. What surfaces at that point is almost always a decision taken in the first six months whose consequences were invisible until now.

This is why motivational framing misreads the situation. Advice about discipline and writing habits addresses the symptom (a candidate who cannot write) while the cause is that the chapter as specified cannot be written, because the question and the method do not meet, or because the literature chapter never established the gap the findings are supposed to fill. A candidate in that position is not procrastinating. They are correctly detecting that something upstream is broken.

So the question of how to write a dissertation is really four questions about structure, asked early. Four failure modes account for most of it. Each is diagnosable in the first year with a question you can ask in ten minutes, and each is expensive to fix in the third.

Failure 1 · a question the method cannot answer

The commonest structural fault is a mismatch between what the research question asks and what the chosen design can establish. It is easy to commit because both halves are individually respectable: an interesting question, a competent method. The mismatch only becomes visible when the analysis is done and the candidate discovers that the data support a claim adjacent to the one they promised.

Typical shapes it takes:

• A question asking why something happens, answered with a design that can only establish whether it happens.
• A question about change over time, answered with data collected once.
• A question about a population, answered with a sample selected for accessibility, where the limitation is real and the question was never narrowed to match.
• A question about effect, answered with a design that cannot exclude the obvious alternative explanation.

The early test. Write the sentence you expect to be able to write at the end: the finding, in one sentence, with its qualifiers. Then ask whether the design as planned could produce that sentence, and what result would force you to write its opposite. If no possible result would force the opposite, the question is not being tested by the study. If the sentence needs a word the method cannot supply (because, increased, generally), either the method or the sentence has to change. Ten minutes in month four saves a rewritten analysis chapter in month thirty.

Failure 2 · a literature review that is a list

A literature chapter has one structural job: to establish that a specific gap exists, that it matters, and that this study is positioned to address it. A list of what other people found, however comprehensive and however well organised, does not do that job. It reads as diligence and functions as filler, and the candidate does not find out until the discussion chapter, where there is nothing to argue against.

The diagnostic is unkind and reliable: if the chapter would survive unchanged had the research question been different, it is not a review. A real review is shaped by the question it is clearing ground for. Its sections are the sub-claims required to make the gap credible, not the themes the literature happens to fall into, and certainly not the chronology of who published when.

The second symptom is the citation that does no work. If a source is named and its content plays no part in the argument of the paragraph, the sentence exists to demonstrate reading rather than to advance a claim. Examiners see this immediately, and it costs credibility precisely where the candidate was trying to build it.

The fix, when it is caught early: write the gap as a numbered argument first: three or four propositions that together make the case that something is unknown and worth knowing, and then let the literature enter only where it supports or complicates one of those propositions. Everything that has no proposition to attach to goes in an appendix or nowhere.

Failure 3 · a findings chapter that argues

Findings and discussion do different work, and merging them is the failure examiners comment on most often. A findings chapter establishes what was observed, with enough transparency that a sceptical reader could reach a different interpretation. A discussion chapter argues for one interpretation over its rivals. When the two are fused, the reader cannot separate the evidence from the reading of it, and the whole claim becomes unauditable.

The tell is a findings section in which every result arrives already interpreted — where the numbers or the quotations appear as illustrations of a point that has already been made rather than as material the reader can weigh. In qualitative work the equivalent tell is the quotation selected to prove a theme that was named before the data were read.

Splitting them also solves a writing problem. A findings chapter that only reports is dramatically easier to draft, because it does not require the argument to be finished. Many candidates who cannot write their findings are in fact stuck on the discussion, and have not noticed that these are two tasks.

The early test. Take one page of draft findings and delete every sentence containing suggests, indicates, demonstrates or because. If what remains is too thin to stand as a report of what happened, the chapter is a discussion in disguise.

Failure 4 · an unnegotiated supervision rhythm

The fourth failure is the only one that is not about the text, and it is the one that most reliably turns the other three into crises. Candidate and supervisor operate on unstated and incompatible assumptions: the candidate believes a submitted chapter is being read closely and awaiting judgement, the supervisor believes they are receiving a sketch for orientation. Months pass in which each behaves reasonably given their own model, and nobody is wrong until a deadline makes the discrepancy visible.

The literature on doctoral supervision keeps arriving at the same conclusion: what matters is not the frequency of meetings but whether expectations are explicit. Four things need to be stated once, in writing, and revisited:

1. What a submission is for. Comment on structure, on argument, or on prose — these require different reading and very different turnaround.
2. Turnaround time, stated as a number of weeks, including what happens in fieldwork and vacation periods.
3. Who decides what. Which choices are the candidate's to make and report, and which need agreement before work proceeds.
4. What a warning looks like. How the supervisor will say "this is not on track" in a way the candidate will recognise as that, rather than as a suggestion.

None of this requires a difficult conversation in month one. All of it requires a difficult conversation in month twenty-four if it is skipped.

Dissertation structure: what each chapter is for

Dissertation structure varies by discipline and institution while the underlying functions stay stable. Stated as functions rather than titles, so they survive a different template:

Read as a whole, that is the dissertation structure the examiner will assume. The table is also a coherence check. The claim in the conclusion, the gap in the review and the promise in the introduction should be three statements of one thing. Where they diverge, the divergence is the examiner's first question.

Writing early, and what it is for

Anyone asking how to write a dissertation is told to write from the start. That advice is right and it is usually justified badly. The point is not that early prose will be reusable; much of it will not be. The point is that writing is the only reliable instrument for detecting the four failures above. A question that cannot be answered by the method becomes obvious the moment you try to write the finding sentence. A review that is a list becomes obvious the moment you try to write the paragraph that names the gap.

Which suggests a specific practice rather than a general exhortation: at the end of each stage, write the one page that stage was supposed to make possible. After the design is settled, write the abstract as though the work were finished, including the finding. After the literature is read, write the gap as a numbered argument. After the data are collected, write the findings summary before any analysis is polished. Each of those pages is cheap, disposable, and diagnostic.

The timeline that actually happens

Doctoral schedules are written as though the phases were sequential and equal. They are neither, and the mismatch between the plan and the shape of the work is itself a source of the late panic. Three properties of real timelines are worth planning around.

Approval is a gate with a queue. Ethics review, access negotiations and, in some fields, equipment or archive scheduling sit between the design and the first data, and their duration is set by other people's calendars. A design that can only be executed one way is hostage to that queue; a design with a fallback is not. This is the practical reason to settle the method early, not the bureaucratic one.

Analysis expands and writing does not compress. The stage that reliably overruns is the one between having data and knowing what they say, because it is the stage where the first failure mode surfaces. The stage that cannot absorb the overrun is final writing, which has a hard floor: a chapter takes as long as it takes to think through, and no amount of pressure converts three months of argument into three weeks.

The last year is not for new work. In a schedule that holds, the final phase adds no data and no new literature; it makes an argument out of what exists. A candidate still collecting in the final months has usually been rescuing an earlier decision, and the rescue is what consumes the writing time.

The planning consequence is to schedule backwards from the submission date through the writing floor, and treat whatever is left as the true research window, rather than scheduling forwards and discovering the floor at the end.

Reading your own draft as an examiner would

Candidates re-read drafts for prose and examiners read them for structure, which is why a heavily polished chapter can still come back with fundamental comments. Four passes, each done separately, catch most of what an examiner will raise, and none of them is about sentences.

1. The promise pass. Read only the introduction and the conclusion, back to back, ignoring everything between. Do they describe the same study, at the same size? Divergence here is the single most common examiner opening.
2. The claim-support pass. Take the three strongest claims in the discussion and find the specific evidence for each in the findings. Any claim whose support you cannot point at has to be softened or removed.
3. The alternative-explanation pass. For your main finding, write the best rival explanation, in one honest paragraph. If the thesis does not address it anywhere, that paragraph is the question you will be asked.
4. The load-bearing citation pass. Mark every citation that the argument would collapse without. If there are very few, the chapter is decorated rather than grounded; if a load-bearing one is a secondary source, read the original before submission.

Done at the end of each chapter rather than at the end of the thesis, these four passes convert most late structural corrections into small early ones.

The examination, and what examiners look for

Examination formats differ — a closed viva, a public defence, a written report followed by an oral — and the questions converge. Examiners are establishing three things: that the contribution is real and correctly sized, that the candidate understands the limits of their own method, and that the work is theirs in the sense of being defensible under questioning rather than merely completed.

That is why the most dangerous answer in a defence is an overclaim. A candidate who says the study shows more than it does invites the examiner to dismantle the method; a candidate who states the limitation first, accurately, has removed the attack and demonstrated exactly the judgement being examined. Knowing what your study cannot show is not a weakness to be managed. It is a substantial part of what is being assessed.

Thesis or dissertation: the terms differ by country

How doctoral training is organised around this text — supervision, progression reviews, completion times — is treated in the higher education hub. The two words are used in opposite senses in different systems, which makes generic advice easy to misread. In much of the United States a dissertation is the doctoral work and a thesis is the master's; in the United Kingdom and large parts of Europe the reverse is common, and a thesis is the doctoral text. Some systems distinguish a monograph thesis from a compilation of published papers with a framing chapter, and the regulations for the two are genuinely different documents.

The practical consequence is small but worth stating: when you read guidance, check which system it comes from, and treat your own institution's regulations as the only binding text. Nothing on this page overrides them.

Works referred to

1. Writing a Graduate Thesis or Dissertation — Sense Publishers, 2016.
2. Authoring a PhD: How to Plan, Draft, Write and Finish a Doctoral Thesis or Dissertation — Patrick Dunleavy, 2003.
3. How to Write a Thesis — Rowena Murray, third edition 2011.
4. Helping Doctoral Students Write: Pedagogies for Supervision — Barbara Kamler and Pat Thomson, second edition 2014.
5. The Craft of Research — Wayne Booth, Gregory Colomb and Joseph Williams, fourth edition 2016.
6. Writing Your Journal Article in Twelve Weeks — Wendy Laura Belcher, second edition 2019.

Questions readers send

How long should a dissertation be?

The limit is set by your institution and discipline, and it is a ceiling rather than a target. Doctoral limits in the humanities and social sciences commonly sit near eighty to a hundred thousand words, with lower limits in the sciences where the argument is carried partly by figures and papers. Ask for the regulations and the last three passed examples in your department; those two documents tell you more than any general guide.

Should I write the introduction first?

Write a draft of it first and rewrite it last. Its early job is to force you to state the question, the claim and the contribution in a form you can look at; its final job is to promise exactly what the finished chapters deliver. Those are different documents that happen to sit in the same slot.

What is the difference between a literature review and a background chapter?

A background chapter tells the reader what is known. A literature review argues that a specific gap exists, that it matters, and that your method can address it. If your chapter would survive unchanged had you chosen a different research question, it is background rather than a review.

How much should a supervisor read, and how often?

Whatever the two of you agree explicitly, written down, in the first month. The specific arrangement matters much less than its existence: the common failure is not a bad rhythm but an unstated one, where the candidate believes chapters are being read closely and the supervisor believes they are seeing sketches.

Can a dissertation be a set of published papers?

In many systems yes, and it changes the structural problem rather than removing it. A paper-based thesis still needs a frame that states one overall contribution and explains how the papers make it, and that frame is where the same four failure modes reappear.

What actually happens if the results are negative or messy?

Nothing fatal, provided the question was worth asking and the method could have answered it. A clean negative result against a well-posed question is examinable; a messy result against a question the design could never settle is the problem. This is the first failure mode arriving in its usual disguise.

Where to go next

1. Learning theories The four families of learning theory compared on the same three questions: what each one can explain, what it cannot, and which teaching decisions it licenses.
2. Adult education What adult education research supports and what it only asserts: the six andragogical assumptions checked one at a time, plus the participation evidence.
3. Curriculum design Backward design, spiral sequencing and constructive alignment explained as a coverage-versus-depth trade, with the cost of each choice stated in teaching hours.