A short book for Master's students — and the people who supervise them
Raphael Cohen Susewind
LSE Department of Methodology
I · The situation
Six months and ten thousand words — examined as if it were a year.
~6months — most of it shared with coursework, a job, a crisis you can't predict
~10kwords in which to ask a question worth asking and answer it credibly
No slow apprenticeship, no year in the field. The book is for thinking like a social scientist when there is little time to think slowly.
02
I · The premise
I stopped telling students what to do, and started asking them.
What it doesn't do
Teach you to run an interview
Teach you to fit a regression
Teach you to format a bibliography
What it does
Asks 23 questions, in four steps
Each answered in a single sentence
Because a dissertation is a chain of judgment calls
The first questions I inherited from Howard Becker; others from colleagues and my own students.
03
I · The architecture
Four steps, twenty-three questions, one sentence each.
IFind an intriguing puzzle & a manageable questionQ1–6
IIEmbed it in the literature — find shoulders to stand onQ7–11
IIIPlot a viable design — a path toward answersQ12–17
IVDo the research, make sense, step back, circleQ18–23
04
The 23 questions
Part I · Finding an intriguing puzzle and asking a manageable research question
1What is your topic of research and which incident or reading inspired you?
2What exactly puzzles you about this topic, incident or reading?
3Which different kinds of questions does this puzzle put on the agenda?
4Which of these questions do you want to answer as your research question?
5What might be a few imagined answers to your question?
6Can readers disagree with these answers, even if they agree with your facts?
05
The 23 questions
Part II · Embedding the puzzle in the literature — finding shoulders to stand on
7What larger issue is your puzzle a case of, and why should your sceptical reader care that it is?
8How have other scholars posed and answered your question?
9What convinces you about their approach, what not, and why?
10Which one study / paper / book do you want to imitate and why?
11How would you deviate from and tweak this study for your own purposes?
06
The 23 questions
Part III · Plotting a viable research design — a path towards answers
12What is the main pattern of difference you want to explain or understand?
13What could be several hypothetical explanations for (or understandings of) this difference?
14What would make your answer right rather than its rivals, and what could rule the others out?
15What kind of material do you need to convince readers of either explanation / understanding?
16Where and from whom are you most likely to obtain that material, and what will access and ethics require?
17What do you need to do to that material to figure out whether it supports your argument?
07
The 23 questions
Part IV · Doing the research, making sense, stepping back, and going in circles
18What, in one sentence, is your answer?
19What would be a good contrasting solution, answer and explanation / understanding?
20Which of your material does not fit your argument, and what do you do about that?
21What are good reasons to believe you? Are they clearly stated?
22What are good reasons not to believe you? Are they clearly stated as well?
23Which new questions, puzzles, differences did you encounter on the way?
08
II · The rhythm
It looks like a staircase, but behaves like a roundabout.
Answer Question 4, arrive at Question 13 — and find Q4 has quietly changed underneath you.
So you go back, adjust, come round again.
That is not failure. It is how research is actually done.
09
II · The one rule
One rule: every question wants a one-sentence answer.
“Vagueness is comfortable, and it survives beautifully in paragraphs; it dies in a single sentence.”
Answer all twenty-three questions in twenty-three good sentences and you have a dissertation; the rest is elaboration.
the sentences accumulate → the introduction & abstract
10
II · The pedagogy — writing
The writing isn't the record of the thinking. It is the thinking.
Each chapter ends in a short, timed, ungraded “Try it” — a low-stakes genre chosen because performing it rehearses that chapter's thinking move.
A postcard (Q1)forces concrete recall of the incident
A hostile margin note (Q6)instantiates the resisting examiner
A referee report (Q22)self-assessment in the discipline's genre
the diary becomes the quarry — snack, don't binge. Lineage: Emig (1977), Elbow (1973), Bean & Melzer (2021).
11
II · The pedagogy — diagnosis
A stuck student is usually at a threshold, not short of information.
Threshold concepts are troublesome, transformative, irreversible portals. The questions map onto them — so stuckness becomes diagnosable.
Q6the argument threshold — can they say what a reader would resist?
Q13the theorising threshold — deriving expectations, not quoting theory
Q17the analysis threshold — doing things to material, not displaying it
Q21–22knowledge creation — a claim owned, warranted, and priced
Meyer & Land (2003); Kiley & Wisker (2009) — and the diagnostic runs both ways: a student who sails through too fast is the one to probe.
12
II · The examples
The same question lands three ways — so three students carry it.
A
Amara
Public policy · interviews
Watched her council declare a climate emergency — then, six weeks later, approve a multi-storey car park.
T
Tomasz
Sociology · secondary survey
Doomscrolled the loneliness statistics and stopped: the loneliest in Britain are not the old, but the young — people like him.
S
Siobhán
Education · insider study
In the staffroom as the head called the phone ban “transformational” — catching the eye of a colleague who'd just let it slide.
13
III · Generative AI — the threat
Generative AI improves the dissertation while hollowing out the learning it certifies.
The product ↑
better essays, more fluent prose, cleaner grammar
The learning ↓
no greater knowledge gain or transfer — “metacognitive laziness” (Fan et al. 2024)
And markers cannot reliably tell AI-assisted work apart (Kofinas et al. 2025). For an output-marked dissertation, “examined, not defended” is a vulnerability. [early, mixed evidence]
14
III · Generative AI — the turn
As a ghostwriter, AI replaces the learning; as a sparring partner, it sharpens it.
Ghostwriter ✗
Answers for you
Produces the sentences — and skips the education that was the only point of the exercise.
Sparring partner ✓
Argues with you, never for you
Widens the rivals, plays the hostile examiner, reverse-outlines a draft — you still render the verdict, and verify.
“AI may argue with you, it may not answer for you, and whatever it hands you, you must always verify.”
The 23 questions demand the student's own sentence before any model could — which is exactly the mode the evidence rewards (Lee et al. 2024).
15
III · For us, the supervisors
Read as assessment design, the apparatus is a quiet integrity architecture.
Sentences as a standing agenda
bring the stack to every meeting; drift and progress visible at a glance
The diary read as fieldnotes
with interest, without a pen — never graded, or it dies
A season of 4–6 meetings
one shape, so neither side invents an agenda from a blank page
The role tilts toward three things the book already trains: scaffold · witness · verify. A student who has narrated their decisions across four meetings is hard to swap for a generated artefact.
16
23
The invitation
What would your tradition push back on?
I'd value your scepticism more than your assent.Tell me where this breaks in your discipline — and, if it's useful, borrow it, adapt it, or test the questions with your own students this year.