To write a research question, start with a topic, read until you can see what is still unknown or disputed, and turn that gap into one question that names who or what you will study, where and when, and what you will measure or interpret. Then test it: a good question can be answered with evidence you can actually get, and its answer isn't a fact you could look up. Expect several drafts, because the wording you settle on decides the method.
This guide is the method, with one question built from scratch, a check from clinical research (FINER) that works well beyond it, and a framing tool for health questions (PICO). For finished questions in a dozen fields, and eight weak ones repaired, see the research question examples.
One question, three drafts
Much of the work is revision. Here is one humanities question taken through three drafts, with what each change makes answerable:
| Draft | Question | What it makes answerable |
|---|---|---|
| DraftRough | QuestionHow do critics write about immigrant food? | What it makes answerableNothing yet: which critics, where and when are all open, so no one could collect the evidence |
| DraftBounded | QuestionHow did restaurant reviews in one city's two daily newspapers describe immigrant-owned restaurants between 1990 and 2020? | What it makes answerableAdds an archive and a period, so the evidence becomes a finite set of reviews you can gather and read |
| DraftFinal | QuestionHow did restaurant reviews in one city's two daily newspapers describe immigrant-owned restaurants between 1990 and 2020, and did the words reviewers used for price and authenticity change over those three decades? | What it makes answerableAdds two features to track and a comparison over time, so anyone can see what would count as an answer |
The steps below are how you get from the first row to the third, and the volunteer firefighter question further down shows them one at a time.
Eight steps to a research question
George Mason University's Writing Center describes the core of the process in five steps: choose a general topic, do some preliminary research, consider your audience, start asking questions, and evaluate the question. The eight below split the middle into the parts that take the longest.
- Choose a topic you can live with
A thesis question stays with you for months, a dissertation question for years. Pick a subject you are curious about and that has sources, data or people you can reach.
- Read to map what is known
Start with two or three recent review articles, then the studies they lean on most. Keep notes in two columns: what researchers agree on, and where they disagree or stop. The limitations paragraphs of recent studies are where authors admit what they couldn't answer, and a well-built literature review ends by naming exactly that.
- Find the gap
A gap is the specific thing the reading leaves open. Gaps often fall into one of five kinds, and each points to a question shape:
- An untested group or place. Studied in cities, never in rural counties. “Does the pattern hold for…?”
- Conflicting findings. Two good studies disagree. “Under what conditions does…?”
- An old finding in new conditions. Measured before a policy, a technology or a demographic shift changed things. “Is it still true that…?”
- An unused method. Everyone surveyed; nobody interviewed or observed. “How do the people involved describe…?”
- A practice with no evidence behind it. Everyone does it, nobody has checked it. “Does [practice] change [outcome]?”
- Write ten rough questions
Don't judge them yet. Mix how, what, why and does questions, big and small. Ten drafts give you something to choose from; one draft gives you something to defend.
- Pick one and add boundaries
Choose the question you most want answered, then add four boundaries: who (the population), where (the setting), when (the period) and which outcome. Each one cuts the question down to something evidence can settle.
- Match the wording to a design
Decide what evidence would answer the question, and check that the wording asks for it. A question about how common something is needs counting; a question about cause needs a comparison; a question about experience needs people's own accounts. The guide to qualitative vs quantitative research covers the choice.
- Test it
Run the question through FINER (below) if it is for a study, or the Writing Center's five qualities if it is for a paper. Expect to revise; a first draft often fails at least one test.
- Write what comes next
A quantitative question gets a hypothesis that predicts its answer (see these hypothesis examples). A thesis gets two to four sub-questions, one per chapter or strand. An argumentative paper gets a working thesis, the answer you expect to defend, which the guide on how to write a thesis statement covers.
A question built step by step
Here is one question taken through the steps, for an invented master's thesis in public policy. Watch what each step adds.
| Step | The question at this stage |
|---|---|
| Step1. Topic | The question at this stageVolunteer fire departments in rural areas |
| Step3. Gap (from the reading) | The question at this stageSuppose the studies you find count how many members departments have, but none follows new recruits to see who leaves, when, and why. |
| Step4. Rough question | The question at this stageWhy do volunteer firefighters quit? |
| Step5. With boundaries | The question at this stageAmong people who joined volunteer fire departments in three rural counties in western Pennsylvania between 2018 and 2023, who left within two years, and why? |
| Step6. Matched to a design | The question at this stageMain question: How many recruits who joined between 2018 and 2023 left within two years, and how do those who left explain the decision? Sub-question 1, from department rosters: what share left, and at what point? Sub-question 2, from interviews: how do former recruits describe the decision to leave? |
Feasible: departments keep rosters, and 20 to 25 interviews fit a two-semester thesis. Interesting: fire chiefs and county boards want the answer. Novel: it follows recruits instead of counting members. Ethical: former members can decline, and names stay out of the thesis. Relevant: it tells departments where in the first two years to spend their retention effort.
Two things happened between rows 4 and 6. “Why” became two questions with different evidence, a count from rosters and accounts from interviews, which makes this a mixed-methods study. And the boundaries made it finishable: three counties, five years, one group of people. The media question at the top of the page went through the same process with texts as its evidence.
Check it with FINER
FINER is a five-part check on whether a draft question is worth a study. It doesn't help you word the question; use it once you have one, as with the firefighter question above. It comes from Designing Clinical Research by Stephen Hulley and colleagues, and Patricia Farrugia and coauthors set it out in a 2010 article in the Canadian Journal of Surgery, “Research questions, hypotheses and objectives.” It was written for clinical research, but the five questions apply just as well in education, social science and policy.
| Criterion | What it covers | Ask yourself | If it fails |
|---|---|---|---|
| CriterionFeasible | What it coversEnough participants, the skills, affordable in time and money, a manageable scope | Ask yourselfCan I get the data and finish in the time I have? | If it failsNarrow the population, shorten the period or use existing data |
| CriterionInteresting | What it coversThe answer matters to you, your colleagues and the wider community | Ask yourselfWho besides me wants to know? | If it failsTie the question to a decision someone has to make |
| CriterionNovel | What it coversIt confirms, overturns or extends what earlier studies found | Ask yourselfWhat will we know afterward that we don't now? | If it failsChange the group, the setting or the method |
| CriterionEthical | What it coversAn ethics board would approve the study | Ask yourselfCould I get approval to collect this? | If it failsUse anonymized or public data, or change the question |
| CriterionRelevant | What it coversIt matters to the science, to practice or policy, or to future research | Ask yourselfWhat changes if the answer is yes? If it is no? | If it failsName the field, policy or next study that would use the answer |
For a class paper or an argument, a lighter check works. George Mason's Writing Center lists five qualities: clear, focused, concise, complex and arguable. Its complex test is the useful one for students: a question that a simple yes or no can settle, without analysis, isn't ready.
Frame health questions with PICO
Where FINER checks a question, PICO helps you build one. Clinical and nursing questions are often written to a framework of named parts, and the framework doubles as the list of search terms for the literature review. PICO comes from a 1995 ACP Journal Club article by W. Scott Richardson and colleagues, “The well-built clinical question: a key to evidence-based decisions.” Oxford's Centre for Evidence-Based Medicine describes its four parts as the patient or problem, the intervention (which can also be a cause or a prognostic factor), a comparison if there is one, and the outcomes.
| Framework | Stands for | Use it for | Source |
|---|---|---|---|
| FrameworkPICO | Stands forPatient or population, intervention, comparison, outcome | Use it forQuestions about treatments and other interventions | SourceRichardson and colleagues, ACP Journal Club, 1995 |
| FrameworkPICOT | Stands forPICO plus time, the period over which the outcome is measured | Use it forOutcomes that need a time frame, such as 30-day readmission | SourceUsed, for example, by Farrugia and colleagues, 2010 |
| FrameworkPECO | Stands forPICO with an exposure in place of the intervention | Use it forThings people are exposed to rather than given: pollution, noise, shift work | SourceMorgan and colleagues, Environment International, 2018 |
| FrameworkSPIDER | Stands forSample, phenomenon of interest, design, evaluation, research type | Use it forSearching for qualitative and mixed-methods studies | SourceCooke, Smith and Booth, Qualitative Health Research, 2012 |
PICOT: In adults with type 2 diabetes (P), does a weekly group walking program (I), compared with usual care (C), lower HbA1c (O) after six months (T)?
PECO: In children aged 5 to 12 (P), is living within 200 meters of a major road (E), compared with living farther away (C), associated with more emergency visits for asthma (O)?
Notice the verbs. The PICOT question asks whether the program lowers a measure, which needs a comparison group and ideally random assignment. The PECO question asks only about an association, because nobody can assign children to live by a highway. PICO fits interventions and exposures measured on an outcome, as with the walking program. It would not help with the firefighter or media questions on this page, which ask about reasons and language; for a question about meaning or experience, use SPIDER or no framework at all rather than forcing one.
Words that choose your method
Some words promise evidence you may not be able to collect. Swap them before the proposal goes to your supervisor.
| Instead of | Write | Why |
|---|---|---|
| Instead ofaffect, impact | Write“is associated with,” or “compared with… does X change Y” | Why“Affect” claims a cause. Say whether your design can show one. |
| Instead ofeffective, successful, better | WriteThe measure itself: missed appointments, pass rates, minutes | WhyNobody can measure “effective” until you say effective at what. |
| Instead ofpeople, society, students | WriteA named group, place and period | WhyBoundaries are what make a question answerable. |
| Instead ofthe best way to | Write“Which of [two options] does more for [outcome]?” | Why“Best” would mean comparing every possible way. |
| Instead ofshould (in a study) | Write“What happened when…?” or “How do… decide?” | WhyData describe; the should goes in the discussion. |
In an IMRaD paper the final question closes the introduction, right after the gap. In a proposal or dissertation it follows the problem and the purpose statement; these purpose statement examples show the whole chain, and the research proposal template shows where each piece goes.
Mistakes that cost months
- Starting from the method. “I want to run a survey” is a plan looking for a question. Decide what you want to know first, then whether a survey can tell you.
- Asking what you already believe. If only one answer seems possible, the question is a conclusion in disguise. Ask it so that a surprising answer could come back.
- Skipping the search. A question someone answered in 2019 isn't novel unless you change the group, the setting or the method, or set out to replicate it. Search for it in the words a published study would use before you commit. You can also ask the Strokes Assistant to look up whether a published study has already asked it; it browses the web and answers beside your draft, but read what it finds before you decide the gap is real.
- Sizing it for a career. A semester question names one group in one place. A dissertation question can be larger, but its sub-questions still have to fit the time you have.
- Hiding two questions behind “and.” Two causes or two outcomes in one sentence can be two studies. If the parts need different evidence, split them into sub-questions, as the firefighter question does, or keep the one you can answer best.
- Changing the question quietly. Questions change while you read and plan, and that's normal. Changing it after the results and presenting the new one as the original misleads readers.
Frequently asked questions
What is a PICO question?
A clinical question built from four parts: the patient or population, the intervention, the comparison and the outcome. For example: in adults with type 2 diabetes, does a weekly walking group, compared with usual care, lower blood sugar? PICOT adds a time frame. The format comes from a 1995 ACP Journal Club article by W. Scott Richardson and colleagues. It suits questions about interventions and exposures, not questions about experience or meaning.
How long should a research question be?
Usually one sentence, occasionally two. If it needs a semicolon or a second “and,” it may be two questions. George Mason's Writing Center asks for questions expressed in as few words as possible, but a research question still has to name the population, the setting and the outcome, so concise doesn't have to mean short.
Can a research question change during a project?
Yes, and many do while you read and plan. Changing it before you collect data is normal; keep each version and note why. Changing it after you have seen the results, and presenting the new question as the original, misleads readers. If the data suggest a better question, report it as a finding or as a question for future work.
