Most drafts submitted as a “research question” are still a topic wearing a question mark. “What are the effects of remote work on employee wellbeing?” reads like a question, but it names no population, no specific outcome measure, and no boundary on scope — it could not be answered by any single, feasible study design as written. This page is a worked bank of examples, organized by design type, each shown as a weak version and a stronger revision with a note on exactly what changed. For the step-by-step process of narrowing a topic into a question and evaluating it against FINER, see CASRAI’s guide to how to write a research question — this page assumes that process and shows its output.
What separates a good research question from a topic
A workable research question needs five things a topic does not have on its own:
- Specificity. A named population, variable, relationship, or intervention — not a general subject area.
- Answerability with obtainable data. The data needed to answer it must actually be collectible by the research team, within a real design, not hypothetically available in principle.
- A defined scope. A boundary on population, setting, and timeframe tight enough that one study could plausibly close the question.
- Non-triviality. An answer that isn’t already self-evident or a matter of definition.
- A connection to a gap. Some reason — contested evidence, an untested population, a methodological limitation in prior work — that the answer is actually needed.
The examples below apply those five tests directly, design type by design type.
Quantitative research question examples
For a deeper treatment of this category — including how wording shifts across descriptive, correlational, quasi-experimental and true experimental designs — see quantitative research question examples by study design.
Quantitative questions ask “how much,” “how many,” or “is there a relationship” using measurable variables. This section also covers what is commonly searched as research questions for quantitative research — the four sub-types below are the standard taxonomy.
Descriptive
Weak: What is the state of nurse burnout?
Strong: What proportion of registered nurses at acute-care hospitals in a given region report high emotional exhaustion on the Maslach Burnout Inventory?
What changed: a named population (registered nurses, acute-care setting), a named measurement instrument, and a quantifiable target (proportion), replacing a vague “state of.”
Correlational
Weak: Is social media related to sleep?
Strong: Among 13-17 year olds, is daily short-form video use associated with self-reported sleep latency?
What changed: both variables are now named and measurable, the population is bounded by age, and the relationship being tested (association, not causation) is explicit — a correlational design cannot claim more than that.
Causal / experimental
Weak: Does exercise help diabetes?
Strong: Among adults aged 45-65 with type 2 diabetes, does a supervised 12-week resistance-training program reduce HbA1c compared with usual-care activity advice, measured at 6 months?
What changed: intervention, comparison group, outcome measure, and time horizon are all specified — this is a PICOT-structured question, covered in more detail below.
Comparative
Weak: Do different teaching methods work better?
Strong: Do first-year undergraduate statistics students taught with flipped-classroom instruction achieve higher final-exam scores than students taught with traditional lecture, controlling for prior GPA?
What changed: the compared groups are named, the outcome is a specific measured score, and a confound the design needs to account for (prior GPA) is already flagged.
Qualitative research question examples
Qualitative questions ask “how” or “why,” oriented toward meaning, process, or lived experience rather than measurement. See CASRAI’s guide to qualitative research methods for the underlying methodologies these questions map to, and qualitative data for how the resulting answers are coded and analyzed.
Phenomenological
Weak: What do cancer patients feel?
Strong: How do adults within the first year of a stage III colorectal cancer diagnosis describe their lived experience of decision-making around treatment?
What changed: a bounded population and timeframe, and a specific phenomenon (decision-making experience) rather than an open-ended emotional state.
Grounded theory
Weak: How do teams collaborate?
Strong: What process do newly formed cross-functional product teams use to establish shared norms during their first 90 days?
What changed: the question now points toward generating a process/theory (grounded theory’s signature output) rather than a description, with a bounded population and timeframe. See CASRAI’s grounded theory entry for the methodology this question type is built for.
Ethnographic
Weak: What is hospital culture like?
Strong: How do norms around interdisciplinary communication develop among staff on a single hospital intensive care unit over a period of sustained observation?
What changed: a single bounded setting (one unit, not “hospitals”) and an explicit method signal (sustained observation) that only an ethnographic design can deliver.
Case study
Weak: Why do some nonprofits succeed?
Strong: How did one regional nonprofit organization adapt its service-delivery model during a specific funding disruption?
What changed: “some nonprofits” (an unbounded population no single study could cover) became one case, bounded by a specific organization and event.
Narrative
Weak: What are immigrants’ experiences?
Strong: How do first-generation immigrant small-business owners in one city narrate their pathway from arrival to business ownership?
What changed: population narrowed by generation, occupation, and location, and the question now asks specifically for a narrated account, which is what a narrative design is built to elicit.
Mixed-methods research question examples
A mixed-methods study typically needs three questions, not one: a quantitative question, a qualitative question, and an integration question that states how the two strands will be combined. The integration question is the piece most drafts leave out.
Quantitative strand: To what extent do first-generation college students report lower rates of faculty mentorship access than continuing-generation students?
Qualitative strand: How do first-generation college students describe barriers to seeking faculty mentorship?
Integration question: In what ways do the qualitative accounts of mentorship barriers help explain the quantitative gap in reported mentorship access between first-generation and continuing-generation students?
What changed: the integration question is what makes this a mixed-methods design rather than two unrelated studies bundled together — it explicitly names which strand explains or elaborates on the other.
Systematic review / evidence synthesis question examples
Weak: What does the literature say about telehealth?
Strong (PICO-structured): Among adults with hypertension (P), does telehealth-delivered blood pressure monitoring (I), compared with in-person clinic monitoring (C), improve blood pressure control (O)?
What changed: the review question is now structured to define eligible studies — the PICO elements double as inclusion/exclusion criteria for the search strategy, which an unstructured “what does the literature say” question cannot do.
Frameworks, applied to one topic
The same broad topic — access to mental health support among university students — produces a different, equally valid question depending on which framework structures it:
Run through PICO (quantitative / clinical-style)
- P — Undergraduate students at four-year universities
- I — Access to a campus-based counseling service
- C — No access / waitlisted students
- O — Self-reported anxiety symptom scores
Resulting question: Among undergraduate students at four-year universities, does access to campus-based counseling services, compared with waitlisted access, reduce self-reported anxiety symptom scores?
Run through SPIDER (qualitative / mixed evidence synthesis)
- S (Sample) — Undergraduate students who have sought campus mental health support
- PI (Phenomenon of Interest) — Experiences of barriers to accessing that support
- D (Design) — Qualitative interview or survey studies
- E (Evaluation) — Reported facilitators and barriers
- R (Research type) — Qualitative or mixed-methods
Resulting question: How do undergraduate students who have sought campus mental health support describe the barriers and facilitators they experienced in accessing it?
The same subject area, run through two frameworks, produces a measurable-outcome question and a meaning-oriented question — neither is more correct; the framework should follow the design the research actually needs, not the reverse.
PICo and PEO, briefly
PICo (Population, Interest, Context) is a qualitative-synthesis variant of PICO: interest replaces intervention, since qualitative work is usually not evaluating a treatment. Example: among first-time mothers (Population), experiences of postnatal anxiety (Interest), in low-income urban settings (Context).
PEO (Population, Exposure, Outcome) is used for observational and etiological questions where there is no intervention being tested, only a naturally occurring exposure. Example: among adults living within one mile of a major roadway (Population), long-term exposure to traffic-related air pollution (Exposure), incidence of childhood asthma diagnosis (Outcome).
FINER as the evaluation test, not a drafting framework
Unlike PICO, PICo, SPIDER, and PEO — which help build a question — FINER (Feasible, Interesting, Novel, Ethical, Relevant) is a check applied after a candidate question exists, to decide whether it’s actually usable. See CASRAI’s research question guide for the full FINER breakdown; every worked example on this page was written to already pass it.
SMART, adapted from objective-setting
SMART (Specific, Measurable, Achievable, Relevant, Time-bound) originates in objective-setting rather than research methodology, but the same five checks map cleanly onto a research question: is the population and variable specific, is the outcome measurable, is data collection achievable with available resources, is the answer relevant to the field or a decision, and is the study boundable within a defined time period.
Research question vs. hypothesis vs. aim vs. objective
These four terms are routinely used interchangeably in early drafts, but they are distinct parts of a proposal that serve different jobs:
- Research question — the thing the study is asking, in interrogative form. Every study has one.
- Hypothesis — a specific, testable, falsifiable prediction of the answer, stated as a claim rather than a question. Not every study needs one.
- Aim — a broader statement of purpose (“to examine,” “to determine”), often the framing sentence a proposal opens with, one level above the question.
- Objective — a concrete, often numbered, operational step toward the aim (“to recruit 200 participants and administer …”), the most granular of the four.
Which designs need a hypothesis and which do not follows directly from the type of question being asked: causal, comparative, and most correlational quantitative designs are hypothesis-driven, because they are testing a specific predicted relationship. Descriptive quantitative work (prevalence, characteristics) is often question-only, with no hypothesis to test. Exploratory and most qualitative designs — phenomenological, grounded theory, ethnographic, narrative — generally should not carry a formal hypothesis at all: stating a predicted finding in advance runs against the inductive, meaning-generating logic those designs are built on, and can bias data collection and coding toward confirming it.
How many research questions should a study have?
There is no fixed number, but a well-scoped study is usually organized around one primary research question, sometimes supported by two or three secondary questions that break the primary question into components (a specific population subgroup, a secondary outcome, a moderating variable). A proposal built around five or six co-equal questions is a common sign that the topic hasn’t actually been narrowed yet — the fix is usually to promote one question to primary and either cut the rest or demote them to secondary.
Common failures, with fixes
- Double-barrelled. “Does telehealth improve access and satisfaction?” asks two questions at once, each of which might have a different answer. Fix: split into two questions, or pick the primary outcome and make the other secondary.
- Yes/no questions. “Is remote work good for employees?” invites a binary answer with no room for the actual variation a study would find. Fix: reframe around degree or relationship — “to what extent,” “how does … relate to.”
- Presupposing the answer. “How does social media harm adolescent mental health?” assumes the causal direction and valence the study is meant to test. Fix: neutral framing — “how is social media use associated with adolescent mental health outcomes?”
- Unmeasurable constructs. “Does the program improve wellbeing?” leaves “wellbeing” undefined and unmeasured. Fix: name the instrument or operational definition — a specific validated scale, a specific behavioral indicator. CASRAI’s guide to construct validity covers how to check that a chosen measure actually captures the construct it claims to.
- Unobtainable populations. A question requiring access to a population the research team has no realistic route to (an undocumented population, a closed patient registry, a competitor’s proprietary users) fails on feasibility regardless of how well-formed it otherwise is. Fix: identify the actual accessible population before finalizing the question, not after.
- Scope no single study could deliver. “What causes income inequality?” is a lifetime research program, not a study. Fix: narrow to one mechanism, one population, one time period the available design and resources can actually address.
The ethics and feasibility gate
A question can pass every drafting framework and still be unaskable in practice. This is the administrative reality CASRAI’s audience deals with directly, and it deserves separate attention from the wording checks above:
- Vulnerable populations. Questions involving children, prisoners, people with impaired decision-making capacity, or other populations requiring additional protections need a design an institutional review board (or equivalent research-ethics body) will actually approve — not just one that is scientifically ideal. A question requiring, for example, withholding standard care from a vulnerable group is not answerable by an ethical design, however well-formed the wording is.
- Obtainable consent. If the population the question requires cannot practically be consented — incapacitated patients without a legal representative pathway, deceased individuals’ records without the applicable waiver, minors without guardian access — the question is not feasible regardless of scientific merit, unless a specific regulatory exception applies and has actually been secured.
- Data access. A question that depends on proprietary, embargoed, or legally restricted data the research team has no confirmed access pathway to is not yet a usable question — it is a question conditional on a data use agreement, IRB approval, or licensing arrangement that has to be resolved first, not assumed.
The practical implication: run the ethics and feasibility gate before, not after, time is invested in a full proposal. A question that fails here needs to be reformulated — a different population, a different data source, a different design — not defended.
Frequently asked questions
What is an example of a good research question?
“Among adults aged 45-65 with type 2 diabetes, does a supervised 12-week resistance-training program reduce HbA1c compared with usual-care activity advice, measured at 6 months?” is a strong example: it names a bounded population, a specific intervention, an explicit comparison, a measurable outcome, and a timeframe.
What are examples of quantitative research questions?
Quantitative research questions fall into four types: descriptive (what proportion or how many), correlational (is X associated with Y), causal/experimental (does X cause a change in Y), and comparative (does group A differ from group B on Y). Worked examples of each are in the quantitative section above.
What is a sample qualitative research question?
“How do first-time mothers describe their experience of postnatal anxiety in the first three months after birth?” is a sample qualitative research question — it names a bounded population and asks for a described experience rather than a measured variable, appropriate to a phenomenological or narrative design.
Do all research questions need a hypothesis?
No. Causal, comparative, and most correlational quantitative designs are hypothesis-driven. Descriptive quantitative work and most qualitative designs proceed with a research question alone.
What is the difference between PICO and SPIDER?
PICO (Population, Intervention, Comparison, Outcome) structures quantitative and clinical questions around a measurable intervention effect. SPIDER (Sample, Phenomenon of Interest, Design, Evaluation, Research type) structures qualitative and mixed-methods evidence-synthesis questions around an experience or process rather than a treatment effect. The same topic run through both, shown above, produces different but equally valid questions.
For the process of narrowing a topic into a question and the full FINER and PICOT breakdown, see CASRAI’s guide to how to write a research question. For the design decisions a finalized question feeds into, see the rationale of a study, qualitative vs. quantitative research, and power analysis and sample size calculation.
