This is a territory map rather than a list of questions. For each subject it names the corners a school student can genuinely work in, the corners that are closed by equipment or policy, and what the data looks like in each. Pick the territory first; the question comes from the data once you are inside it.
Most topic lists hand you fifty subjects and leave you to find out which ones are actually workable. Half are not, usually for reasons nobody mentions until you have wasted a term.
This maps the territory instead. For each subject: the corner you can work in, the corner that is closed, and what the data looks like.
Before the map: what closes a territory
Three things, and only one of them is about you.
Policy. UNC's laboratory policy bars under-eighteens from BSL-2 agents, radioactive materials and animal work, requires constant direct supervision, and requires safety paperwork filed weeks ahead. Most research universities match it. This closes more territory than anything else and it is not negotiable.
Access. Some data exists but is behind an institutional login, a formal application or a fee. UK Biobank's full dataset is an application, not a download.
Scale. Some questions need more observations than you can obtain. Surveying your own year group gives you perhaps ninety people, which will not support a claim about adolescents.
What does not close a territory is difficulty. A sixteen-year-old can read a codebook.
The map
Economics
Open: Almost all of it. Labour markets, prices, public policy evaluation, development, behavioural questions using existing surveys.
Closed: Nothing meaningful, which is why this is the most reliable subject on the list.
What the data looks like: IPUMS for microdata, World Bank for cross-country indicators, ONS for the UK. Mostly spreadsheet-sized at aggregate level; microdata needs a bit of code.
The characteristic mistake: picking a question so large it is a dissertation topic. "The causes of inflation" is a career. "Whether the 2018 Sugar Levy changed sugar content or price" is a term.
Political science
Open: Voting behaviour, turnout, legislative output, party systems, media and opinion.
Closed: Anything requiring elite interviews — politicians do not reply to school students either.
What the data looks like: ANES for US survey data, Electoral Commission and the House of Commons Library for the UK, published election returns everywhere. Well documented and designed to be used.
The characteristic mistake: writing an opinion piece with citations attached. A political science project needs a relationship you can test, not a position you can defend.
Public health and epidemiology
Open: Mortality and morbidity patterns, policy evaluation, inequality, environmental health.
Closed: Anything involving patients, samples or identifiable records.
What the data looks like: CDC WONDER for the US, ONS and NHS England for the UK, WHO for cross-country. Query interfaces that export cleanly, so often no coding at all.
The characteristic mistake: confusing correlation with a claim about causes. Fast-food outlet density tracks childhood obesity. So does deprivation. Separating them is the project.
Psychology
Open: Anything using existing survey waves — adolescent behaviour, attitudes, educational outcomes, wellbeing.
Closed: Experiments on human subjects without institutional ethics approval, which you will not have.
What the data looks like: Monitoring the Future, the General Social Survey, PISA. Large, repeated, and built for secondary analysis.
The characteristic mistake: running an unsupervised survey on classmates. It feels like real research and the sample will not carry the conclusion.
Biology and chemistry
Open: The computational corners. Bioinformatics using public sequence databases. Epidemiology. Ecology using citizen-science records. Meta-analysis of published trials. Label-claim verification against published measurements.
Closed: Essentially all bench work, for the policy reasons above.
What the data looks like: Public sequence repositories, BTO survey data, Nature's Calendar phenology records, published trial registries.
The characteristic mistake: a kitchen-table demonstration dressed as research. Growing cress under different lights answers a question already in a textbook.
Computer science
Open: All of it. You own the apparatus.
Closed: Nothing, though anything needing serious compute will constrain you.
What the data looks like: The GitHub API, the Wikipedia API, public corpora, or data you log yourself. Often you build the dataset, which is itself a contribution.
The characteristic mistake: building a tool rather than answering a question. A working app is an achievement and not a finding. "Can a simple classifier distinguish X, and where does it fail" is a finding.
Linguistics
Open: Corpus work, language change over time, translation comparison, readability, sociolinguistic variation in published text.
Closed: Fieldwork with speakers, for ethics reasons.
What the data looks like: Google Books Ngrams, subtitle corpora, newspaper archives, social text. Enormous and barely touched by students.
The characteristic mistake: a claim about how people speak, evidenced from how people write.
History
Open: Anything with digitised sources — newspaper archives, parliamentary records, digitised collections, published statistics.
Closed: Archives requiring in-person access and a reader's ticket you cannot get.
What the data looks like: Newspaper databases, Hansard, digitised institutional collections. Often text rather than numbers, which suits students who dislike statistics.
The characteristic mistake: narrating events rather than arguing about them. A history project needs a disagreement between historians to enter.
Sociology and urban studies
Open: Neighbourhood-level analysis, housing, transport, inequality, local policy.
Closed: Ethnography, for the same ethics reasons as psychology.
What the data looks like: Local authority open data, DfE statistics, census output areas, transport records. Geographically granular, which makes for vivid findings.
The characteristic mistake: finding that deprived areas do worse on something. They do, on nearly everything. The project is what survives controlling for it.
The territories at a glance
| Subject | Open corner | Closed corner | Where the data is |
|---|---|---|---|
| Economics | Nearly all of it | — | IPUMS, World Bank, ONS |
| Political science | Voting, turnout, legislatures, media | Elite interviews | ANES, Commons Library |
| Public health | Patterns, policy evaluation, inequality | Patients, samples, records | CDC WONDER, ONS, WHO GHO |
| Psychology | Existing survey waves | Unapproved human experiments | Monitoring the Future, GSS |
| Biology / chemistry | Bioinformatics, epidemiology, meta-analysis | Essentially all bench work | Public sequence repositories, BTO |
| Computer science | All of it | — | GitHub API, self-logged data |
| Linguistics | Corpora, language change, readability | Speaker fieldwork | Google Books Ngrams |
| History | Digitised sources | Reader's-ticket archives | Newspaper databases, Hansard |
| Sociology | Neighbourhood analysis, housing, transport | Ethnography | Local authority open data, DfE |
Picking between territories
Three questions, in order.
Which of these do you want to study at university? Start there. A project in your intended subject gives you material for an interview in that subject, which is where it actually pays off.
If that territory is closed, what is its open neighbour? A would-be medic cannot work in a lab but can work in epidemiology, which is clearly medical and entirely reachable. A chemist can do cheminformatics. Keep the subject, drop the equipment.
Can you name a dataset in under five minutes? If not, the territory is probably still too broad. Narrow until the next step is obvious.
From territory to question
This page stops where the question begins, deliberately.
The move from "I am working in public health" to "I am testing whether X predicts Y in this dataset, controlling for Z" is a separate skill, and it is where most projects fail. Our guide to building a research question worth publishing covers that step; where to find a free dataset covers the sources named above; and if you are doing a formal qualification, EPQ ideas and Extended Essay ideas give worked questions in each.
Where we fit
The Oxford Centre for Advanced Research pairs students with PhD candidates and early-career researchers at Oxford, Cambridge, other leading UK universities and Ivy League institutions. The Oxford Scholar Programme is £2,000 — about $2,600 — for ten contact hours over ten to fourteen weeks; the Oxford Publication Fellowship is £3,400, roughly $4,400.
Most of what a mentor does early is exactly the territory-narrowing above: establishing which corner of your subject is reachable before you spend a month discovering it is not. We have written up what those first sessions contain so you can run the process yourself if you would rather not pay for it.
One thing to do today
Pick the territory closest to what you want to study. Open the dataset named beside it. Read until you find a variable you did not expect to exist.
That is where the question is, and it is an hour's work to get there.
Frequently asked questions
Is this a list of research questions?
No, deliberately. It maps where the workable territory is in each subject, because most students pick a field before they pick a question and the usual failure is choosing a field where nothing is reachable. Once you know your corner, the question comes out of the data — our guide to building a research question covers that step.
Which subject is easiest to research at school level?
Economics, political science and public health, because the data is public, documented for non-specialists and large enough that unchecked questions remain. Linguistics and computer science are close behind. The hard ones are wet-lab biology and chemistry, and the difficulty is institutional rather than intellectual.
Should I research what I want to study at university?
Usually yes, because it gives you something specific to be questioned on at interview in that subject. The exception is a field where nothing is reachable — a would-be medic may do better with epidemiology than with anything requiring a laboratory, and that is still clearly medical.
Can I do a topic that spans two subjects?
Yes, and the overlaps are often where the unclaimed questions are — health economics, computational linguistics, political psychology. The caution is that a formal qualification like the Extended Essay is marked against one subject's criteria, so pick which discipline's methods you are using.
What topics should I avoid?
Anything requiring equipment, animals or human subjects under supervision; anything so broad it names a field rather than a relationship; and anything where you cannot imagine a result that would surprise you. The third is the one students miss.
How specific does my topic need to be before I start?
Specific enough to name a dataset. 'Health inequality' is not a topic you can start from. 'The urban–rural life expectancy gap in England, using ONS data' is, because the next step is obvious: open the data.