Formulating Research Questions

Overview

A research question converts an identified problem into a precise request for knowledge. It defines what the study seeks to determine and creates the interface between the problem and the methodological plan. Population, design, measurement, sample size, analysis and interpretation all depend on the question.

A strong question is focused enough to guide a study but not written so narrowly that it assumes unverified details or excludes meaningful inquiry. It identifies the relevant population or setting, the phenomenon, exposure, intervention, comparator, outcome or experience, and the kind of conclusion required. Clinical-research authors repeatedly emphasise that one well-developed primary question is more useful than a long list of loosely connected questions [1–3].

This lecture focuses on the logic of question formulation. Structured frameworks such as PICO, PECO, PICo, PCC and SPIDER are addressed in the next lecture.

Why this matters

An imprecise question allows almost any data collection to appear relevant. During analysis, investigators may search among many outcomes or subgroups because the primary aim was never established. This increases complexity, weakens interpretation and may encourage selective emphasis on favourable results.

A clear primary question determines the central outcome or phenomenon, key comparison, target population, timing and intended inference. General health-research guidance likewise places question formulation before the detailed design and data-collection plan [4]. It also helps participants, collaborators, ethics committees, funders and reviewers understand what the study is designed to accomplish.

Question, objective and hypothesis

A research question is interrogative: What proportion completes testing within 12 weeks?

An objective states the action the study will take: To estimate the proportion of eligible women completing testing within 12 weeks.

A hypothesis states a testable expectation: Women who receive a scheduled appointment before discharge are more likely to complete testing than women who do not.

These elements should express the same underlying inquiry in different forms. A hypothesis is not required for every question. Descriptive studies, early exploratory work and many qualitative studies may be guided by questions and objectives without a formal predictive hypothesis.

Types of research questions

Descriptive questions

These ask about amount, frequency, distribution, characteristics or experience.

• What proportion of eligible women complete postpartum glucose testing within 12 weeks?
• How do women describe the transition from obstetric to primary care after gestational diabetes?

Descriptive questions do not require a causal explanation. They still require a defined population, measurement or qualitative approach and time frame.

Comparative questions

These ask how an outcome differs between groups, interventions, periods or strategies.

• Is testing completion higher among women given an appointment before discharge than among those receiving usual referral?

The question should identify the direction of the contrast and whether the groups arise through assignment or observation.

Associational questions

These ask whether an exposure, characteristic or service factor is related to an outcome.

• Is distance from the testing facility associated with completion within 12 weeks?

An association does not automatically imply causation. The design and adjustment strategy determine what can be inferred.

Explanatory or causal questions

These ask whether changing an exposure or intervention would change an outcome, or through what mechanism an effect occurs.

• Among eligible women, does scheduling a test before discharge increase completion compared with routine advice?

Causal wording creates stronger design requirements, including attention to comparability, confounding, adherence and co-interventions.

Predictive questions

These ask how well information available at a specified time can estimate a future or currently unknown outcome.

• Can information available at discharge predict non-completion of testing within 12 weeks?

Prediction focuses on performance for individuals or groups, not whether predictors cause the outcome.

Exploratory and interpretive questions

These seek understanding of meanings, experiences, processes or contextual mechanisms.

• How do women and primary-care staff experience the transfer of responsibility for postpartum testing?

Such questions are often open, flexible and refined as understanding develops, while still requiring a clear phenomenon, context and participant group.

One problem, different questions

illustrative

Intended knowledge	Example question
Description	What proportion completes testing within 12 weeks?
Association	Which patient and service factors are associated with completion?
Causal evaluation	Does a scheduled appointment plus reminder increase completion compared with usual care?
Prediction	How accurately can discharge information identify women at high risk of non-completion?
Experience	How do women explain their decisions and ability to complete testing?
Implementation	What factors influence delivery and uptake of the follow-up pathway?

Characteristics of a strong research question

Clear

The wording has one main interpretation. Terms such as “effective”, “knowledge”, “outcome”, “risk” and “quality” are defined sufficiently to guide measurement.

Focused

The question is neither a broad programme of research nor a trivial fragment. It prioritises one central uncertainty and avoids unrelated populations or outcomes.

Answerable

A plausible design, population, data source and analytical approach exist. The required exposure and outcome can be measured with acceptable validity and timing.

Important

The answer could advance understanding, improve a decision, address inequity, reduce uncertainty or support subsequent research.

Ethical

The question can be investigated with a reasonable balance of scientific value, burden and risk. A scientifically unanswerable question is not rescued by a complete consent process.

Neutral

The wording should not presume the result. “Why does poor awareness cause non-attendance?” assumes both poor awareness and causation. A neutral question asks whether knowledge is associated with attendance or explores how women describe decision-making.

Primary and secondary questions

The primary question is the central question the study is designed to answer. It usually determines the main outcome, sample-size basis and principal interpretation. Secondary questions add important but subordinate information.

A study may legitimately have several questions, but they should form a coherent hierarchy. Too many “primary” questions create competing priorities, multiple outcomes and an impractical protocol. Exploratory questions should be labelled as such and should not be presented later as though they were pre-specified confirmatory aims.

DECISION GUIDE

Prioritise the primary question

Choose the question that:

1. addresses the most consequential gap;
2. can be answered by the planned population and design;
3. identifies the main outcome or phenomenon;
4. determines the central sample-size or information requirement;
5. would still justify the study if secondary questions were removed; and
6. supports a conclusion that the study is genuinely capable of making.

A practical formulation process

1. Restate the problem in one sentence. Identify the gap and why it matters.
2. Name the intended inference. Description, comparison, association, causation, prediction, exploration or implementation.
3. Define the population and context. Specify who or what system, where and during which period.
4. Specify the central phenomenon or variables. Exposure, intervention, comparison, outcome, experience or process.
5. Add timing where it changes meaning. Baseline, follow-up duration, diagnostic window or implementation phase.
6. Test feasibility and neutrality. Can the required information be obtained without assuming the answer?
7. Identify one primary question. Reclassify other useful questions as secondary, exploratory or future work.
8. Check alignment. Ensure the question can be translated into objectives, variables, data collection and analysis.

Worked example

WORKED EXAMPLE

Refining an unfocused question

Initial question: Why do women not attend follow-up after gestational diabetes?

Problems: “Follow-up” is undefined; the population and time frame are absent; the question assumes non-attendance; and “why” may imply a causal explanation that one cross-sectional questionnaire cannot provide.

Primary descriptive question: Among women diagnosed with gestational diabetes who deliver at the study hospital, what proportion complete a documented glucose test within 12 weeks postpartum?

Secondary associational question: Which pre-specified patient and service factors are associated with documented completion within 12 weeks?

Optional qualitative question: How do women who complete and do not complete testing describe the barriers, facilitators and transfer of care after discharge?

Future intervention question: Does scheduling testing before discharge with an automated reminder increase completion compared with the current referral pathway?

The sequence recognises that estimating the magnitude and understanding the pathway may be necessary before selecting an intervention.

Question wording and causal claims

Words such as “effect”, “impact”, “influence”, “determinant” and “leads to” can imply causation. Use them only when the intended design and assumptions support a causal question. For observational questions, “association”, “relationship” or “difference” may be more accurate, although even neutral language does not remove confounding.

Similarly, “predictor” should be reserved for a prediction aim or clearly defined statistical context. A variable associated with an outcome is not automatically useful for prediction, and a predictive variable is not necessarily causal.

Questions can be decomposed into estimandsMMed extension

For a comparative quantitative question, deeper specification includes the population, treatment or exposure strategies, outcome, time point, summary measure and handling of intercurrent events such as treatment discontinuation or death. This estimand thinking prevents apparently precise questions from concealing different possible targets of estimation.

Appraising a research question

APPRAISAL CHECKLIST

Check the proposed question

• Does it arise directly from the documented problem and gap?
• Is the intended inference evident from the wording?
• Are the population and context sufficiently defined?
• Are the main variables, comparison or phenomenon identifiable?
• Is timing specified when it affects interpretation?
• Does the question avoid assuming the result or mechanism?
• Can an ethical and feasible design answer it?
• Is one primary question clearly prioritised?
• Can the question be translated into aligned objectives and measurements?

Suggested protocol wording

Common mistakes

Knowledge check

What makes a question primary rather than secondary?

Answer and reasoning

The primary question is the central inquiry the study is designed and resourced to answer. It usually determines the main outcome, information requirement and principal conclusion. Importance is established prospectively, not by which result later has the smallest p-value.

Why is ‘What is the effect of obesity on depression?’ incomplete?

Answer and reasoning

The population, definitions, temporal relationship, comparison, outcome measurement and intended causal or associational interpretation are unclear. Different designs could answer very different versions of this question.

Does every research question require a hypothesis?

Answer and reasoning

No. A formal hypothesis is often useful for focused comparative or associational quantitative questions. Purely descriptive questions, exploratory qualitative inquiry and some early methodological work may be appropriately guided by questions and objectives without a formal hypothesis.

Why should a predictive question not be phrased as a causal question?

Answer and reasoning

Prediction asks how accurately available information forecasts an outcome. A predictor may be useful without causing the outcome. Causal questions require different assumptions and usually different design and analysis strategies.

Key take-home messages

SUMMARY

Key take-home messages

• The research question converts a defined problem into a precise request for knowledge.
• Its wording should reveal whether the intended inference is descriptive, comparative, associational, causal, predictive or exploratory.
• Prioritise one primary question and keep secondary questions coherent and subordinate.
• Define the population, phenomenon or variables, comparison and timing to the degree needed for design and measurement.
• Avoid wording that assumes a result, cause or solution.
• Questions, objectives, hypotheses, design and analysis should express one aligned study logic.

References

1. Haynes RB. Forming research questions. J Clin Epidemiol. 2006;59(9):881–886. https://doi.org/10.1016/j.jclinepi.2006.06.006
2. Farrugia P, Petrisor BA, Farrokhyar F, Bhandari M. Research questions, hypotheses and objectives. Can J Surg. 2010;53(4):278–281.
3. Browner WS, Newman TB, Cummings SR, Grady DG, Huang AJ, Kanaya AM, et al. Designing clinical research. 5th ed. Philadelphia: Wolters Kluwer; 2022.
4. World Health Organization Regional Office for the Eastern Mediterranean. A practical guide for health researchers. Cairo: World Health Organization; 2004. https://iris.who.int/handle/10665/119703