How to write a good research question

A research question is the question your study is designed to answer. Everything else in the project hangs from it, like the canvas of a tent from its centre pole. The papers you read, the method you choose, the data you collect and the chapters you write are all there to answer that one question.

Research questionnoun

A clear, focused question that a study sets out to answer, using evidence that can be collected and analysed.

Deepak's first attempt was "What is the impact of UPI on India?" Professor Mehta replied with three questions of his own: impact on what, on whom, and measured how? A question that broad cannot tell you what to read, whom to ask or when you have finished. It needs national data and a team of researchers. A dissertation needs a question one student can answer in a few months.

A good question decides three things for you: what you read, what data you collect and when you are done. Most dissertations have one main question and two to four sub-questions that break it into smaller steps.

Few people write a good research question in one go. It is reached in steps, and each step makes the question smaller and sharper. This is how Deepak got there.

1. Name the topic in a few words."UPI and small shopkeepers" is a topic, not yet a question. If you do not have a topic yet, start with how to choose a research topic.
2. Write down what puzzles you.In ten minutes Deepak listed twelve questions: why his uncle refused UPI, whether shopkeepers trust it, whether it changes how they keep accounts, whether customers push them into it. Do not judge them yet.
3. Read what is already known.Check which of your questions have been studied and which have not. Deepak found many studies of people paying with UPI and fewer of the shopkeepers receiving the money. How to spot this is covered in how to find a research gap.
4. Narrow it with who, where and when.Not "shopkeepers" but "owners of small grocery shops"; not "India" but "two markets in East Delhi"; not "ever" but "in the past year".
5. Write it as one open question.Start with what, how or why, and include the key terms: "What factors influence whether owners of small grocery shops in East Delhi accept UPI payments?"
6. Test it and show it to your guide.Run it through the checks in the next section before you plan anything else.

Most people rush the fifth step. Write three or four versions of the same question and compare them side by side. The clearest one is usually the plainest.

For a head start, Fiberarticle plans three to five research questions from a topic and searches arXiv, OpenAlex, Semantic Scholar and Crossref in parallel for papers on it. Treat those questions as drafts to test and rewrite, not as your final question.

A common checklist for research questions is FINER, from Designing Clinical Research, a textbook by Stephen Hulley and colleagues. It was written for medicine but works in any field. A good question is feasible, interesting, novel, ethical and relevant.

A question that fails one test can usually be fixed. If it is not feasible, narrow the place or the group. If it is not novel, change the setting or the method. If it raises ethical problems, change what you ask or how you ask it, and check your university's ethics rules before you collect any data.

Research questions come in a few main types, and the type decides the method. One topic can produce every type. Here they are with Deepak's examples.

Each type points to a method. Descriptive and comparative questions usually need a survey or records. Relationship questions need data on both things from the same shops, analysed with correlation or regression. Causal questions need an experiment or a careful before and after design. Exploratory questions need interviews or observation.

Deepak's main question mixes two types, which is common in business research: a survey to describe and compare the shops, then interviews with a few owners to understand their reasons.

Some fields use frameworks: fixed slots, one for each part of a question. Fill every slot and the question almost writes itself. The two best known are PICO and SPIDER.

PICOframework

Population, Intervention, Comparison, Outcome. Developed in the 1990s for clinical questions in evidence-based medicine, and now used widely in health research.

A health example: in adults with type 2 diabetes (population), does a daily 30-minute walk (intervention), compared with no exercise plan (comparison), lower blood sugar after three months (outcome)? Deepak borrowed the same slots for his causal question: among small shop owners in East Delhi, does a one-hour training session on UPI, compared with no session, change the share of payments received through UPI over the next month?

SPIDERframework

Sample, Phenomenon of Interest, Design, Evaluation, Research type. Proposed in 2012 for qualitative and mixed-methods research, where there is often nothing to compare.

Deepak used SPIDER to plan his interviews:

• Sample: shop owners who tried UPI and then went back to cash.
• Phenomenon of interest: their reasons for going back.
• Design: in-depth interviews.
• Evaluation: their experiences, worries and views.
• Research type: qualitative.

Other frameworks, such as PEO (population, exposure, outcome) and SPICE, work the same way. The filled slots also give you ready-made search terms, which makes your literature search far more precise.

Deepak's notebook from that month holds seven crossed-out drafts. Most first drafts fail in the same few ways.

Two more mistakes are harder to spot. The first is asking several questions at once. "How and why do shop owners and customers use UPI and cash?" is really four questions: split it and keep the one that matters most. The second is a question your method cannot answer. A survey of a hundred shops cannot tell you in depth why each owner decided as they did, and interviews with ten owners cannot tell you what share of all shops accept UPI.

Professor Mehta had a simple test: "If I gave you the data tomorrow, would you know exactly what to do with it?" If the answer is no, the question needs another draft.

A research question asks. A hypothesis predicts: it is a testable statement of the answer you expect, written before you collect the data.

Quantitative studies often state a hypothesis in two forms. The null hypothesis says there is no difference or relationship: UPI acceptance is the same for owners under and over 40. The alternative hypothesis says there is one. A statistical test then shows whether the data give enough reason to reject the null hypothesis.

Exploratory and qualitative studies usually have questions but no hypotheses, because they aim to understand something before predicting it. Deepak's interviews had no hypothesis at all.

In a thesis or paper, the question usually appears at the end of the introduction. The methods section is built to answer it, and the conclusion answers it directly. The structure of a research paper shows how the parts fit together.

Deepak's final main question read: "What factors influence whether owners of small grocery shops in two East Delhi markets accept UPI payments?" Professor Mehta asked how he would measure each factor, listened to the answer and approved the proposal that afternoon. The first shop Deepak visited for his pilot survey was his uncle's.