Skip to content

Start here · Beginner

Turn an interesting topic into a research question

Turn an interest into a feasible question with an explicit outcome and contribution.

11 min read · Oct 26, 2026

Key points

  • Define a question that evidence can answer.
  • Match the claim to the design.
  • Identify a feasible contribution.
Contents

Turn an interest into a feasible question with an explicit outcome and contribution.

A publishable paper answers a question with evidence that readers can examine. An interesting topic is only the beginning. “I want to study robotics” identifies a field, but it does not tell a reader what will be compared, measured or learned. Your first task is to turn curiosity into a question whose answer would have a clear meaning.

This tutorial explains how to build that question, choose the kind of claim you can support, define the important terms and decide what contribution is feasible. The examples are hypothetical planning exercises. They are not descriptions of completed studies.

Start with a decision that matters

Ask what someone should understand differently after reading your paper. A useful contribution might clarify why two previous studies disagree, test whether a finding survives a change of setting, compare a proposed method with a credible baseline, or document a phenomenon that has been poorly measured. A project does not need to discover an entirely new field to be worth reporting.

For example, “test a new water filter” is an activity. “Determine whether filter B reduces turbidity more than filter A under the same inlet conditions” is an answerable question. “Assess whether the improvement is large enough to justify greater pressure loss” adds an engineering decision. Each version needs different measurements. If pressure loss matters to the conclusion, it cannot appear for the first time in the Discussion as an unmeasured advantage.

Write an initial sentence beginning, “The reader needs to know whether, how much, how, or why…” Then ask which evidence could answer it. If your only answer is “I will find information on the internet”, you probably still have a topic rather than a research question.

Identify the kind of question

The same subject can support several legitimate kinds of investigation. Choose deliberately because the question determines the design and the language of the paper.

Aim Question it answers Hypothetical example Main boundary
Description What exists, and how often or how much? What proportion of sampled campus buildings have step free entrances? The sample and measurement determine what is represented.
Association How do measured variables vary together? How is study time associated with examination performance? An association alone does not identify the effect of changing study time.
Causal estimation What would change under an intervention? What is the effect of assigning a study schedule compared with usual practice? The design and assumptions must justify the causal comparison.
Prediction How accurately can an outcome be predicted for new cases? Can sensor readings predict equipment failure in another operating period? Evaluation must reflect genuinely new cases and relevant conditions.
Interpretation How do people understand or experience something? How do students describe belonging during their first laboratory placement? The sampling, context and interpretive approach shape the account.
Evidence synthesis What does a defined body of research support? Why do studies disagree about the educational benefits of mentoring? Search, eligibility and appraisal determine which evidence contributes.

These aims can coexist, but one should organise the paper. A model that predicts grades accurately does not thereby reveal the effect of increasing study time. Interviews that explain participants’ experiences do not estimate the prevalence of those experiences in all students. An observational analysis may address a causal question, but it must make and defend the additional assumptions required; causal wording is not created by adding covariates to a regression.

Specify five boundaries

Write down the population or system, condition or exposure, comparison if needed, outcome or phenomenon, and setting or time. These are prompts rather than a compulsory formula. A qualitative study may need a context and process instead of a numerical comparator. A descriptive study may have no intervention at all.

Consider the progression “Are robots accurate?”, “Does controller B improve navigation?”, and “Across independently generated indoor layouts, what is the mean change in positioning error with controller B compared with controller A under a specified sensor noise condition?” The final question identifies a comparison, outcome and target setting. It remains incomplete until “positioning error”, “independent layout” and the noise condition are defined, but the missing decisions are now visible.

A research question connects a defined system, a comparison, an outcome and the claim the design supports.

Keep a distinction between the target population, the cases about which you want to learn, and the study sample, the cases you can actually observe. A convenience sample of students from one course may be useful, but it does not automatically represent all young people. Likewise, a collection of easy simulation environments does not represent all real deployment conditions.

Turn concepts into measurements

An operational definition states how a concept will be observed or calculated. “Academic confidence” is a concept. A specified questionnaire score, with named items, response options, scoring rules and measurement time, is an operational measure. “Plant growth” could mean height, dry mass, leaf area or change in one of these. Different choices can lead to different conclusions.

Reliability concerns consistency of measurement under specified conditions. Validity concerns the evidence supporting a particular interpretation and use of the measurement. A scale can produce repeatable scores while measuring the wrong construct. Using a published questionnaire does not establish that every translation, shortened version or new population is equally suitable. Flake and Fried explain why measurement choices and modifications need explicit justification. S104

For a hypothetical study of research mentoring, a confidence questionnaire and a scored research proposal answer different questions. Students might feel more confident without writing stronger proposals, or improve their work while becoming more aware of their limitations. Decide which outcome addresses your objective, then consider whether the other is a useful secondary outcome. Do not silently replace the planned outcome with whichever one looks more favourable.

Name the quantity you want to estimate

An estimand is the precisely defined quantity that the analysis aims to learn about. An estimator is the calculation used to estimate it, and an estimate is the value obtained from the observed data.

In the controller example, the estimand might be the mean difference in positioning error across a specified distribution of layouts, calculated as controller A error minus controller B error. Positive values would indicate lower error with B. The estimator could be the average of the observed paired layout differences. It is not the largest improvement in the most attractive example.

Ask what happens when a controller fails to finish. Removing failed layouts changes the question to performance among successful cases and may favour an unreliable controller. A more useful plan might report completion probability and error conditional on completion separately. The choice should follow the scientific decision and be described before the results are known.

For a review, an analogous discipline is to define the phenomenon and comparison precisely. “Effects of mentoring” is broad. “Evidence concerning research skill development during structured undergraduate mentoring” makes it easier to decide whether a paper on job satisfaction belongs in the synthesis.

Work through three complete examples

Natural science. A student begins with “Does light affect plants?” The supervisor asks whether the objective concerns wavelength, duration or intensity. The student chooses duration because the available apparatus can control it while maintaining comparable intensity. The revised question is: “Under the specified growth conditions, what is the difference in mean dry mass after 21 days between seedlings assigned to two daily light durations?” The protocol must identify the species, independent unit, growth conditions and drying procedure. If all seedlings in one condition occupy a single chamber, the design must address the confounding of chamber and condition before the wording promises a treatment effect.

Social science. A student asks, “Does social media cause stress?” Available data are a voluntary survey taken once. The feasible question becomes: “Among respondents from the sampled courses, how is reported daily social media use associated with scores on a specified stress measure?” The Methods must explain recruitment, measurement and planned adjustment. The Discussion must recognise selection, imperfect recall and uncertain temporal order. If the real interest is how students experience online pressure, interviews and an interpretive question may be more suitable than forcing that interest into a correlation.

Engineering. A student claims, “Our new controller is better.” Better could mean accuracy, speed, energy use or reliability. The revised primary question compares mean positioning error at a fixed operating condition; secondary questions examine completion rate and computation time. A credible baseline receives a stated tuning budget. The student identifies which scenarios were used during development and reserves other scenarios for evaluation. The paper can then explain a specific performance tradeoff rather than declaring universal superiority.

The lesson across all three is that a precise question exposes design problems early. Precision in wording is valuable because it forces precision in decisions.

Build an evidence gap, not an empty space

Make a brief evidence map with three fields: what existing studies establish, what remains uncertain, and what your project can resolve. Attach a source to each claim about previous research. Distinguish a knowledge gap, such as uncertainty about a mechanism, from a methodological gap, such as inconsistent measurement, and a context gap, such as unclear transfer to another setting.

A context gap needs a reason. Testing an established relationship at a different school is informative when something about the setting could change the relationship, or when replication itself is the objective. Merely changing the location does not prove a major theoretical advance. A careful replication can still be a useful contribution when reported honestly.

For review articles, the gap may concern organisation of evidence. Perhaps studies use the same label for different interventions, or measure outcomes at incompatible times. A synthesis that explains those differences can teach readers more than a catalogue of “positive” and “negative” papers.

Separate the objective, hypothesis and analysis plan

An objective states what you intend to learn. A hypothesis states a testable expectation. An analysis plan states how the data will address it. These are related but not interchangeable.

For the controller project, the objective could be to estimate the mean paired reduction in error. The hypothesis could predict lower error with B. The analysis plan would define subtraction direction, treatment of failures, uncertainty estimation and any secondary comparisons. A directional expectation does not automatically justify a test that ignores changes in the opposite direction.

Label analyses developed after inspecting outcomes as exploratory. Exploration is useful for discovery, but readers need to distinguish it from a prediction specified in advance. Preregistration records planned decisions before the relevant analysis and supports this distinction; it does not validate an unsuitable question or design. S07

Finish with a feasibility and claim check

List the equipment, access, permissions, supervision, time and skills required. Identify the smallest useful version of the project. If the central question requires data you cannot obtain, revise the question before collecting an inadequate substitute. Consult a supervisor or methods adviser while changes are still inexpensive.

Write two possible conclusions: one for a clear result and one for an uncertain result. If both would be scientifically uninformative, reconsider the project. A useful study should be capable of teaching something even when the preferred hypothesis is unsupported.

Practice. Revise “Does our app improve learning?” for a project that can observe voluntary users and nonusers but cannot allocate access. Identify one measurement problem and one claim boundary.

Model answer. “Among participating students, how is use of the app associated with change in a specified assessment score over the study period?” The assessment must reflect the learning objective and be scored consistently. Voluntary use may be related to motivation or prior attainment, so the comparison alone does not establish the effect of assigning the app. A stronger causal question would need an appropriately designed study and defensible assumptions.

Use Template 1 to record your question, definitions, contribution and boundaries. A peer should be able to identify what evidence would answer it without asking you to explain what “better” means.

Sources

S07: Center for Open Science. Preregistration. Official guidance, accessed 11 November 2026.

S104: Flake, J. K., and Fried, E. I. (2020). Measurement Schmeasurement: Questionable Measurement Practices and How to Avoid Them. Advances in Methods and Practices in Psychological Science, 3(4), 456 to 465. DOI: 10.1177/2515245920952393.