Part 1 | Foundations — Why a therapist becomes someone who works with data
Chapter 3. The whole map — question, collect, analyse, communicate
Key points
- Think of data work as a journey through four steps that loop
- AI accelerates every step, but each step needs a judgement about what to entrust and what to check
- The first and last steps — the question and the communication — are where human work stays densest
Data has a journey
Say "data analysis" and most people picture only the moment of calculation in a statistics package. In reality there is a long road either side. This guide treats data work as a journey through four steps.
- Question — what do you want to know? Form a hypothesis and design what to measure.
- Collection — gather the data you need, quantitatively and qualitatively.
- Analysis — clean it, read the relationships, and scrutinise causation.
- Translation — turn the result into a form that lands, and deliver it into research, practice and publishing.
These four are not a one-off straight line. The response to what you communicated generates the next question. It is a circle that keeps turning — as Snow drew the map, stopped the pump, confirmed the effect, and moved on to the next question.
AI accelerates every step
Generative AI, the protagonist of this guide, accelerates all four.
| Step | What can be entrusted to AI | What a human carries |
|---|---|---|
| Question | Putting the hypothesis into words, organising prior work, a first draft of the design | Deciding what you actually want to know |
| Collection | Survey design, refining items, transcription | Ethics and consent; building relationships in the field |
| Analysis | Cleaning, calculation, charting, suggesting methods | Whether the method is valid; interpreting results; judging causation |
| Translation | Generating figures, reports and slides; drafting | What claim to make; to whom, and what, to deliver |
Notice that the middle of the journey — the labour of collection, the calculation of analysis — is easiest to hand over, while the two ends — the first question and the final claim — are where human work stays densest.
Coding and Building Systems used the image of a dining room and a kitchen. In data work, hand the kitchen of calculation to AI, and keep the dining-room judgements — what to serve, how to deliver it — with the human. That division is what lets speed and correctness hold together.
AI becomes a research partner
Foundations called AI intellectual training wheels. Working with Your Own Files brought it closer to a collaborator; Coding and Building Systems to a partner you build with.
Through this guide it evolves once more. The AI walking the data journey with you is, in effect,
a research partner sitting beside you, working around the clock
It organises prior work overnight, takes on tedious data cleaning, runs hundreds of lines of calculation in an instant, draws the charts, drafts the manuscript. An excellent research partner. But — the name on the paper, and the responsibility for the result, are yours. That distance is the spine running to the end of this guide.
A prompt to try
You are an experienced researcher and a teacher of data work. I want to look into (a theme you are currently curious about — e.g. the relationship between exercise habits and cognitive function in older adults). Break this into the four steps — question, collection, analysis, communication — and set out concretely what to do at each, as a roadmap for a beginner. For each step, separate what can be entrusted to AI from what I must judge myself.
Chapter 3 summary
- Data work is a looping journey: question → collect → analyse → communicate
- AI accelerates every step, but each needs a judgement about entrusting and checking
- AI becomes a research partner; responsibility for the result stays human