How to Write a Research Question That Actually Works

A strong research question is specific, answerable with available methods and data, and focused on a feasible scope. Start from a broad topic, narrow it with context, define variables and population, choose an appropriate question type (descriptive, comparative, causal, etc.), then stress-test for relevance, ethics, and practicality.

Table of contents

• What a Good Research Question Does

• From Topic to Testable Question: A Simple Process

• Types of Research Questions (and How to Choose)

• Evaluating and Refining Your Question (the R.A.I.S.E.D. test)

• Examples by Discipline (and a Quick Comparison Table)

---

What a Good Research Question Does

A good research question is the north star of your project. It tells you what to collect, which methods to use, and when to stop. If your question is vague (“social media is bad for teens”), your data will be all over the place. If it’s overly ambitious (“solve climate change”), you’ll drown before you begin. The sweet spot is clear, bounded, and method-ready.

Here are the essential qualities:

• Focused: Narrow enough to handle within your word limit, time, and resources.

• Researchable: Answerable with credible data and methods available to you (surveys, experiments, text analysis, archives, datasets).

• Specific: Names the population, context, variables, and time frame.

• Ethical and feasible: Respects participants and access constraints; feasible with your schedule and budget.

• Original in angle: You don’t need to discover a new planet; you need a question that adds a fresh angle (new sample, context, timeframe, comparison, or method).

A helpful rule of thumb: if you can imagine a results section that reports a pattern, effect, difference, or explanation tied to the question, you’re likely on the right track.

---

From Topic to Testable Question: A Simple Process

Step 1 — Start broad, then place boundaries.
Write a one-sentence topic (“remote work productivity”). Add context (industry, country, time), population (software engineers in mid-sized firms), and outcome (feature delivery velocity). Your sentence might become: “How remote work affects feature delivery velocity among software engineers in mid-sized U.S. SaaS firms in 2024–2025.”

Step 2 — Clarify the key concepts and proxies.
Abstract words must be operationalized. Productivity could mean story points closed per sprint, bugs per release, or lead time to change. Choose measures you can access. Define each measure in one line.

Step 3 — Choose the question type by goal and data.
If you aim to describe (“what is happening?”), ask about prevalence or patterns. If you want to compare groups or test an effect, frame a comparative or causal question. If you need reasons (“why?”), plan for explanatory or qualitative designs. Align the type with the data you can realistically collect.

Step 4 — Scope the design.
Sketch the method (survey? experiment? interview set? corpus analysis?), data source, sample size/criteria, and timeframe. If you can’t fill this in quickly, the question is still too fuzzy.

Step 5 — Draft, then iterate.
Write two or three versions of your question: one broader, one tighter, one alternative angle (e.g., switch outcome or population). Pick the one you can defend in a paragraph explaining data, method, and contribution.

Step 6 — Pre-commit to exclusion.
List tempting but out-of-scope paths and stick to the boundaries (e.g., “No cross-country comparisons,” “Exclude 2020 due to lockdown confounds”). This protects the question from ballooning later.

---

Types of Research Questions (and How to Choose)

Choosing the right type is half the battle. Below are common types with mini-templates and example phrasings.

1) Descriptive (“what/ how much/ how often?”)
Goal: quantify or map a phenomenon.
Mini-template: What is the prevalence/ pattern of [Outcome] in [Population/Context] during [Timeframe]?
Example: “What proportion of first-year biology students use generative AI weekly for homework in 2025?”

2) Comparative (“is there a difference between…?”)
Goal: compare groups, programs, or time periods.
Mini-template: How does [Outcome] differ between [Group A] and [Group B] in [Context]?
Example: “How do completion rates differ between online-only and hybrid sections of Intro to Psychology at community colleges?”

3) Correlational/Associational (“is X related to Y?”)
Goal: identify relationships without claiming causation.
Mini-template: To what extent is [Predictor] associated with [Outcome] among [Population]?
Example: “To what extent is sleep consistency associated with GPA among first-year engineering majors?”

4) Explanatory/Causal (“does X cause Y?”)
Goal: estimate a causal effect (needs design to handle confounds).
Mini-template: Does [Intervention/Exposure] lead to changes in [Outcome] in [Population]? Under what conditions?
Example: “Does a four-day workweek pilot change defect rates in enterprise software teams?”

5) Evaluative (“does this program/policy work?”)
Goal: assess effectiveness, often with predefined criteria.
Mini-template: How effective is [Program/Policy] at improving [Outcome] for [Population]?
Example: “How effective is peer-mentoring at reducing dropout rates among first-generation college students?”

6) Exploratory/Qualitative (“how/why?”)
Goal: develop explanations, themes, or theory.
Mini-template: How do [Participants] make sense of [Phenomenon] in [Context]?
Example: “How do novice teachers make sense of classroom management advice during their first semester?”

7) Predictive (“can we predict Y?”)
Goal: build models that forecast outcomes.
Mini-template: Can [Set of Predictors] predict [Outcome] for [Population]? With what accuracy?
Example: “Can early-term attendance and LMS activity predict passing grades in statistics courses?”

How to choose wisely:

• If your access to experimental control is limited, avoid strong causal framings and prefer comparative or associational questions.

• If your timeframe is tight and sample sizes will be small, qualitative or descriptive designs may be smarter than underpowered experiments.

• Whenever ethics or privacy limit data granularity, pivot to questions answerable with aggregate or anonymous data.

---

Evaluating and Refining Your Question (the R.A.I.S.E.D. test)

Use R.A.I.S.E.D. to stress-test any draft question:

Relevant — Does it matter to a real audience (field, organization, community)? In one sentence, state who benefits and how. If you can’t, narrow or reframe.

Answerable — Do you have a plausible method and data path? Write the method in 3 lines: design, data source, sample. If this looks unrealistic, adjust the scope, variables, or population.

Interesting/Impactful — Will the answer change a decision, approach, or understanding? Add a “so what” clause: “If we find X, then Y should…”

Specific — Are population, outcome, and timeframe explicit? Replace generic nouns (“students,” “performance”) with crisp definitions (e.g., “first-year biology majors,” “exam-weighted GPA in Spring 2025”).

Ethical — Does the plan respect consent, privacy, and risk minimization? If not, change the question or method before you get attached.

Doable — Can you execute with the resources and calendar you actually have? If not, shrink the scope (fewer variables, narrower population, shorter window) or switch to a more feasible design.

Before/After example with R.A.I.S.E.D.:

• Before: “Do smartphones harm student learning?”

• After: “Among first-year psychology majors at a public university, does a 48-hour lock-screen notification freeze during midterms change average exam scores compared with usual phone settings in Fall 2025?”

The After version specifies population, intervention, outcome, comparison, and timeframe—and hints at a feasible experimental or quasi-experimental design.

---

Examples by Discipline (and a Quick Comparison Table)

This section shows how vague topics evolve into solid, answerable questions. Each pathway moves from broad interest → narrowed scope → testable wording.

Business/Marketing

• Topic: Social media ads effectiveness.

• Refined: Focus on video length in short-form ads for DTC skincare brands.

• Question: “Among DTC skincare brands on TikTok in Q2–Q3 2025, how does a 6–9 second video ad compare with a 15–18 second ad in click-through rate?”

• Design hint: Observational comparison with ad-level controls; or A/B tests if accessible.

Education

• Topic: Study strategies in intro STEM courses.

• Refined: Emphasize spaced practice vs cramming among first-year calculus students.

• Question: “Does a required spaced-practice schedule in the LMS change pass rates compared with voluntary study habits in first-year calculus?”

• Design hint: Natural experiment if policy varies by section; otherwise matched groups.

Psychology

• Topic: Sleep and mood.

• Refined: Sleep regularity as predictor of PHQ-9 scores among undergraduates.

• Question: “To what extent is week-to-week sleep regularity associated with PHQ-9 depressive symptom scores among undergraduates in Spring 2025?”

• Design hint: Repeated-measures correlation; control for baseline stress.

Computer Science/HCI

• Topic: Code review quality.

• Refined: Checklist-based reviews vs free-form reviews in mid-sized teams.

• Question: “Do checklist-based code reviews reduce post-release defects compared with free-form reviews in mid-sized SaaS teams over two release cycles?”

• Design hint: Clustered comparison by team; defect data from issue tracker.

Environmental Studies

• Topic: Urban heat islands.

• Refined: Tree canopy coverage and surface temperature on school grounds.

• Question: “Is census-tract tree canopy percentage associated with average school-day surface temperature on K-12 campuses in Phoenix, AZ, Summer 2025?”

• Design hint: Public satellite data; tract-level regression.

To consolidate these transformations, here’s a compact table that maps weak → better → best phrasings and the implied question type.

Common polishing moves you can copy:

• Name the population and time window (“first-year majors, Spring 2025”).

• Define the outcome with a measurable proxy (“pass rate,” “CTR,” “PHQ-9 score,” “defects per release”).

• Declare the comparison or relation (“A vs B,” “associated with,” “changes in”).

• State the context (institution, city, industry).

• Keep claims aligned with design (if you can’t randomize, don’t promise causal language).

How to write your final question quickly:

1. Draft a sentence with population + outcome + timeframe.

2. Add the comparison (or relation) and measurement.

3. Run R.A.I.S.E.D. once.

4. Replace any vague words with concrete proxies.

5. Cut any extra clause that doesn’t affect the method.