Stop deliberating and run six tests: can you reach the data today, can you name the confounder, would any result surprise you, does someone disagree, can it be answered in your word count, and will you still care in four months. Score both out of six. The one that wins on paper is almost always the right one, and the one you are emotionally attached to is usually the one to drop.
Most students do not have no question. They have two, and they spend three weeks deciding — which is three weeks neither question gets worked on.
Here is how to settle it in an hour.
The six tests
Score each question out of six. Write the scores down, because the act of writing them is what stops you quietly favouring the one you arrived with.
1. Can you open the data today?
Not "does data exist" — can you have a file open in the next thirty minutes.
This is the test that eliminates most questions and it is the one students skip. Some sources need a registration that takes a day. Some need an institutional login you do not have. Some, like UK Biobank's full dataset, are a formal application rather than a download.
If the answer is no for one question and yes for the other, you may be finished already. Our list of free datasets covers twenty-two sources with notes on which need registration.
2. Can you name the confounder?
For each question, say out loud what else could explain the relationship you expect.
If you cannot think of anything, you do not understand the question well enough to research it — that is a fail, not a pass. If the confounder plainly explains everything, that is also a fail.
The sweet spot is a confounder that is real but controllable. This is the same judgement Think.Check.Submit. asks authors to exercise about venues: establish the thing that would change your answer before you commit. "Does X still predict Y once we account for Z" is a better question than "does X predict Y," and it is the shape most published papers actually take.
3. Would any result surprise you?
If every plausible outcome confirms what you already think, the question is closed, and four months will confirm it slowly.
This is the test students fail most often without noticing, because a question you already know the answer to feels comfortable to choose.
4. Does somebody disagree?
Find two sources taking different positions on your question or something adjacent to it.
That disagreement is where your analysis lives. A question answerable from a single textbook has nowhere to go, and in the UK qualifications it also costs you marks directly — AQA allocates 20% of EPQ marks to AO2, "research, critically select, organise and use information," where the operative phrase is critically select.
5. Can it be answered in your word count?
Four thousand words for an Extended Essay or an Advanced Higher project. Around 5,000 for an EPQ. Four to five thousand for AP Research.
Estimate honestly: if the literature review alone would take 1,500 words before you have said anything of your own, the question is too big.
6. Will you still care in four months?
Not "is it interesting" — will you open the file on a Tuesday in February when nobody is asking you to.
Most projects fail here rather than intellectually. One Stanford lab that publishes its policy takes a high school intern "only once every 4 to 5 years" — so for most students the project has to sustain itself on interest alone. Nothing external requires a draft, and a question chosen to look impressive runs out of fuel long before one chosen because you actually want to know.
How to score
| Question A | Question B | |
|---|---|---|
| Data open today | ||
| Confounder nameable and controllable | ||
| A result could surprise you | ||
| Someone disagrees | ||
| Fits the word count | ||
| You will still care in February | ||
| Total | /6 | /6 |
A question scoring 5 or 6 is ready. Four is workable. Three or below should be dropped or narrowed before you commit.
If both score 5, pick the one with better data access and stop. Equal scores mean the decision does not matter much, and continuing to deliberate costs more than choosing wrong.
The question students always keep
The one they are attached to.
Attachment is the single commonest reason a student carries an unanswerable question into February. You picked it because it interested you; interest is not the same as answerability, and the person who chose a question is the worst-placed person to see why it fails.
Two things help.
Write three candidates, not two. Three forces you to select rather than defend. With two, you tend to argue for the one you already prefer.
Show both to someone who does not care. A teacher with a doctorate, a graduate student, anyone with subject knowledge and no stake. Ask them which they would try to answer, not which is more interesting. This is precisely what a research mentor does in the first session, and it is free from anyone willing to spend ten minutes.
If you genuinely love the lower-scoring question, write it down and keep it. The constraint is the format and the timeline, not the idea — it may be a good undergraduate project.
Narrowing, if neither works
Four moves, in order of how much they cost you.
Add a control. "Does X predict Y" becomes "does X still predict Y once Z is accounted for." Cheapest and usually improves the question.
Restrict the population. All adolescents becomes US 12th graders in one survey year. Instantly makes the data finite.
Restrict the period. Twenty years becomes the three years around one policy change.
Switch from the outcome to the mechanism. If you cannot measure whether something works, you can often measure how it is implemented, which is less crowded territory anyway.
What does not work is choosing a broader question in the hope it will be easier. Broad questions are harder, because there is no point at which you are finished.
Writing the winner properly
Once you have chosen, write the question so it names its own terms.
"Does self-reported daily screen time predict self-reported sleep duration among US 12th graders in the 2023 Monitoring the Future survey, controlling for sex and region?"
Unlovely, and every clause is doing work. Anyone reading it knows what you did and what you did not do.
Underneath it, write the sentence saying what evidence would change your answer. If you cannot write that second sentence, go back and score again.
The full method is in how to build a question worth publishing, and research topics by subject maps which territories are workable at all.
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 six tests above are what the first session of a mentorship consists of. We have written them out because they work whether or not anyone is paid, and because a teacher with a doctorate can run them with you in twenty minutes.
The hour to spend tonight
Write your two questions. Score both out of six. Pick the higher one.
If they tie, take the one with better data and start. An hour spent scoring saves the three weeks you were about to spend deliberating.
Frequently asked questions
How do I choose between two research questions?
Score both against six tests — data access, confounder, surprise, disagreement, scope and durability — rather than deliberating. Scoring takes an hour and produces a decision; deliberating takes weeks and usually produces the question you were attached to at the start, which is rarely the better one.
What if both questions score the same?
Pick the one with better data access and move on. Two questions scoring equally means the decision does not matter much, and the cost of continuing to deliberate is far higher than the cost of choosing wrong. You can switch in week three; you cannot recover a month.
What if I love one question and the other scores better?
Take the one that scores better. Attachment is the commonest reason students carry an unanswerable question into February. If it helps, note the question you loved and keep it for a longer project later — the constraint is the format, not the idea.
How narrow is too narrow?
Almost never a problem. Nearly every failing project was too broad. A question whose answer is a single number is thin, but widening a narrow question takes an afternoon and narrowing a vast one usually means starting again.
Should I pick the question or the data first?
The data, almost always. A question invented in the abstract tends to need evidence that does not exist or is behind a login. A question that comes out of a dataset you already have open cannot have that problem.
When is it too late to switch?
Once you have written analysis, not before. Switching in week two costs you two weeks. Carrying an unanswerable question to week ten costs you the project, and that trade is not close.