How to Write a Research Question

A strong research question is the foundation of any study. Learn to sharpen vague topics into answerable questions using PICO, FINER, and proven frameworks.

Key takeaways

A topic is not a question—use the ladder method to narrow from broad interest to a specific, testable question.
Every strong research question has four components: a clear concept, a defined population, a specified relationship or comparison, and answerable-with-evidence form.
Use PICO for health research or PEO and SPIDER for observational and qualitative work; FINER is your screening checklist after drafting.
Your question determines your method—descriptive, comparative, causal, and exploratory questions require different study designs.

A topic is not a research question

Most students start with a topic: "I'm interested in study breaks." That is not a question. It is a landing zone for your curiosity. A research question is specific, testable, and answerable with the data you can actually collect.

Think of moving from topic to question as climbing a ladder. You start broad and narrow step by step.

Fig. 1 — The topic-to-question ladder narrows your focus step by step. Each rung specifies one more element: the population (you, the student), the intervention (break duration and frequency), and the outcome (exam recall).

Here's how each rung works:

Broad topic: "Study breaks" could mean anything—coffee breaks, weekend breaks, stretching, napping, reading Twitter.
Narrowed topic: "How do study breaks affect memory?" You've picked a mechanism (memory) and a type of break (study breaks), but you still haven't specified who, what dose, or what comparison.
Working question: "Do short study breaks improve exam recall?" You've named the outcome (exam recall) and hinted at the comparison (with breaks vs. without), but you haven't said how long or how often.
Research question: "Does a 10-minute break every 30 minutes of studying improve recall of exam material compared to no breaks?" Now every element is defined: the population (you, a student), the intervention (10-minute breaks every 30 min), the comparison (no breaks), and the outcome (recall of exam material).

The four things every good research question has

Before you apply a framework, ask yourself: does my question have these four elements?

A single, clear concept of interest. What is the main thing you want to understand? In the breaks example, it's the effect of break frequency on memory. Not "study tips in general" or "productivity"—one concept.
A defined population. Who are you studying? Undergraduate students? All students? Nurses? Medical students preparing for board exams? Your population shapes which studies are relevant and whether you can recruit participants.
A specified relationship or comparison. Are you comparing two groups (breaks vs. no breaks)? Describing one group (how often do students take breaks)? Linking two variables (does break duration correlate with retention)? Your question must make the comparison explicit.
Answerable-with-evidence form. Can you measure it? Can you collect data that will answer the question? "Is studying bad for happiness?" is vague and hard to measure. "Does studying more than 6 hours daily correlate with lower mood scores on the PHQ-9 scale in first-year university students?" is answerable.

PICO: The structure borrowed from health research

The PICO framework originated in clinical and health research and has become the standard for systematic reviews and meta-analyses. It works for any study where you are testing an intervention or exposure.

Letter	Meaning	Study Breaks Example
P	Population (or Patient)	Undergraduate students (ages 18–22) preparing for final exams
I	Intervention (or Exposure)	A 10-minute break every 30 minutes of studying
C	Comparison	No breaks (continuous studying for 2 hours)
O	Outcome	Recall accuracy on a practice exam (% of correct answers)

Two common extensions exist:

PICOS adds a fifth element: S (Study design)—the type of research you plan to conduct. Are you running a randomized controlled trial, an observational cohort study, or a cross-sectional survey? Specifying this in advance prevents you from asking a question that demands an RCT but then settling for observational data.

PICOT adds T (Time): How long will the intervention run, and when will you measure the outcome? In the breaks example, PICOT specifies: "during a single 2-hour study session" (T = one session) or "over a 4-week semester" (T = ongoing)? The time frame affects what you can measure.

For observational and qualitative work, use these alternatives:

PEO (Population, Exposure, Outcome) is used when you are not testing an intervention but observing naturally occurring exposures. For example:

P: People who self-reported as "heavy caffeine users" (≥400 mg/day) | E: Caffeine consumption (observed, not assigned) | O: Self-reported sleep quality measured on the Pittsburgh Sleep Quality Index

SPIDER (Sample, Phenomenon of Interest, Design, Evaluation, Research type) is used for qualitative and mixed-methods research, where the research question is more exploratory:

S: Final-year nursing students | PI: The experience of burnout during clinical placements | D: In-depth interviews | E: Thematic analysis of interview transcripts | RT: Qualitative descriptive study

FINER: The sanity check

After you draft your research question, apply the FINER checklist. These are not drafting criteria; they are screening criteria you apply after you have something on paper. Each letter represents a yes-or-no gate that weeds out unfeasible, trivial, or unethical questions.

Letter	Criterion	Concrete Failure Mode (Undergraduate Example)
F	Feasible	Your question requires recruiting 400 nurses across three hospitals in six weeks with zero budget. You cannot deliver. ✗ Revise to: "Can we identify burnout in 20 nurses at our university hospital using a validated scale?"
I	Interesting	Your question is: "Do students with last names starting with A have different GPA than those starting with B?" (Yes, it's probably measurable, but no one cares.) ✗ Revise to: "Does peer tutoring improve GPA in students scoring below the 25th percentile at baseline?"
N	Novel	Your question: "Do antidepressants help depression?" has been answered 10,000 times. You are not adding new knowledge. ✗ Revise to: "Do antidepressants + digital cognitive behavioral therapy reduce depression faster than antidepressants alone in first-episode depression?"
E	Ethical	Your question requires randomly assigning students to a "no studying" group to test the effect of studying on exams. You cannot ethically deprive one group of education. ✗ Revise to: "Among students who study, does the timing of breaks correlate with exam scores?"
R	Relevant	Your question is answerable but has no connection to your field, your institution's mission, or any real-world problem. "Do different colored pencil erasers smell the same?" ✗ Ensure your question addresses a gap in practice, policy, or theory.

Question types decide your method

The structure of your question predetermines the type of study you will need and the analysis you will run. Here are the five main question types:

Question Type	Question Stem	Study Design	Analysis
Descriptive	What is the prevalence of X? What are the characteristics of Y?	Cross-sectional survey or observational cohort	Descriptive statistics, prevalence rates, proportions
Comparative	Do groups A and B differ on outcome Y?	Case–control, cohort, or RCT	t-test, ANOVA, chi-square, or Mann–Whitney U (see which statistical test should I use?)
Relational	Is X associated with Y? Does X predict Y?	Observational cohort or cross-sectional	Correlation, linear or logistic regression, Spearman's ρ
Causal	Does X cause Y? What is the effect of X on Y?	Randomized controlled trial (RCT) or quasi-experimental	ANCOVA, linear regression with treatment group as predictor, instrumental variables
Exploratory-Qualitative	What is the lived experience of X? How do people understand Y?	Interviews, focus groups, ethnography	Thematic analysis, phenomenological analysis, grounded theory

If your question asks "what is the prevalence," you need a descriptive study and descriptive statistics, not a statistical test. If your question asks "does X cause Y," you need an RCT (or a very strong observational design with sensitivity analyses), not just correlation. Mismatching question type to design is one of the most common errors in student research.

Start with your question type. Use the table above to identify the design, then consult our sample size calculator to determine how many participants you need.

From question to hypothesis

Once you have a research question, you may need to write a hypothesis—a prediction about the answer. Not all research requires a formal hypothesis (qualitative and purely descriptive studies often do not), but most quantitative studies do.

A hypothesis has two forms:

Null hypothesis (H₀): There is no relationship or difference. "Breaks have no effect on exam recall." This is what a statistical test assumes to be true, and you either reject it or fail to reject it.
Alternative hypothesis (H₁): There is a relationship or difference. "Breaks improve exam recall" or "Breaks either improve or harm exam recall."

The alternative hypothesis can be directional ("Breaks improve recall," H₁: μ_breaks > μ_{no breaks}) or non-directional ("Breaks change recall in some direction," H₁: μ_breaks ≠ μ_{no breaks}). Directional hypotheses allow one-tailed tests, which have more power but are only valid if you committed to the direction before seeing any data. If you are exploring, use a non-directional hypothesis and a two-tailed test.

The golden rule: Write and register your hypothesis before you collect or analyze data. Changing it after seeing the results (a practice called "p-hacking" or "HARKing"—Hypothesizing After Results are Known) is a form of research misconduct.

Eight weak questions rewritten

Here are real problems from student proposals, with diagnoses and rewrites:

Weak Question	Diagnosis	Rewritten
Does depression exist in college students?	Yes/no trivial	What is the prevalence of major depressive disorder in first-year college students at our university, measured by the PHQ-9 ≥10?
Are vitamins good?	Too broad, unmeasurable	Do daily multivitamin supplements reduce cold incidence in sedentary undergraduate students compared to placebo?
What is the effect of social media and sleep on academic performance?	Two questions in one	Pick one: Does daily social media use >2 hours predict lower sleep duration in undergraduates? OR Does sleep duration mediate the relationship between social media use and GPA?
How does motivation affect learning?	Unmeasurable construct	Do students with higher self-efficacy (measured by the New General Self-Efficacy Scale) achieve higher final exam scores?
Why do students prefer online learning?	Assumes the answer	Do students in hybrid courses report higher or lower satisfaction than those in fully in-person courses, and what factors predict satisfaction?
Can mindfulness improve health?	No population, no outcome	Does an 8-week mindfulness-based stress reduction program reduce cortisol levels and self-reported anxiety in first-year nursing students?
Is tutoring effective?	No comparison, too broad	Do students receiving peer tutoring within 48 hours of failing a quiz achieve higher scores on the next quiz than those receiving tutoring 1 week after?
How do teachers perceive student mental health?	Not answerable with available data	Can university teaching staff accurately identify depression in their students using brief screening questions compared to trained clinical interview?

A checklist to run before you submit your proposal

Print this list and tick off each item. If you cannot tick all ten, your question needs work.

☐ My question has named the population (who you are studying).
☐ My question specifies the main variable or intervention (what you are measuring or changing).
☐ My question includes a comparison or group (versus what, or in what context).
☐ My question specifies the outcome (the measurement you are taking).
☐ I can write a hypothesis or predicted direction for the answer.
☐ I have checked my question against FINER—it is feasible, interesting, novel, ethical, and relevant.
☐ My question maps to one of five types: descriptive, comparative, relational, causal, or exploratory.
☐ I can name the study design that matches my question type (RCT, cohort, survey, interviews, etc.).
☐ No one has already answered this exact question in the literature (or if they have, my version adds a novel population, intervention, or outcome).
☐ I can gather the data to answer this question within my time frame and budget constraints.

FAQ

How long should a research question be?

A research question should be one to three sentences. If it takes more than that to explain, it probably has too many moving parts. Aim for clarity and specificity, not brevity—a 2-sentence question that specifies P, I, C, and O is better than a 1-sentence question that is vague.

Can I have more than one research question?

Yes, but be strategic. A study can have one primary question and several secondary questions. Primary questions are the focus of your analysis and your power calculation; secondary questions are exploratory. For an undergraduate project, stick to one primary question. For a thesis, two to three related questions are typical. Avoid a laundry list—each question should fit the same study design and population.

What is the difference between a research question and a hypothesis?

A research question is a query: "Does X cause Y?" A hypothesis is a prediction: "X causes Y." The question comes first, drives your study design, and then you write directional or non-directional hypotheses to test. Some studies (especially qualitative and descriptive work) answer the research question without ever formally stating a hypothesis.

Do qualitative studies need a hypothesis?

No. Qualitative research is exploratory; it asks open-ended questions like "What is the experience of X?" or "How do people understand Y?" You may have an initial research question, but the findings emerge from the data, not from a pre-specified hypothesis. Qualitative research uses frameworks like SPIDER (Sample, Phenomenon of Interest, Design, Evaluation, Research type) instead of PICO.

When is it too late to change my question?

Before you collect any data, you can change it freely. Once you have data, changing the question is problematic because it risks p-hacking. If you find your original question is infeasible or not what you want to study, reframe it into a secondary analysis question or declare it as exploratory analysis (not a confirmatory test). Changing a question mid-analysis after seeing which comparisons are significant is research misconduct.