What makes a good research question in a PhD thesis?

The property people underweight is the second one. A question whose answer is already obvious is not a research question, it is a demonstration. If everyone in your field would confidently predict the result, and they would be right, you have committed several years to confirming what was already known.

The test is to write down the answer you expect, then write the opposite, and ask whether the opposite is defensible. If a competent colleague could argue for it, the question is live. If nobody could, either find a sharper version or accept that the contribution is methodological rather than substantive, and say so.

The fourth property, scale, is the one supervisors correct most often. Doctoral questions fail in both directions. Too large, and the thesis becomes a survey with no findings of its own. Too small, and there is a good paper but not a thesis, which is a real risk in fields where a single experiment is a normal unit of work.

The reliable calibration is the sub-question structure. A doctoral question is one that decomposes naturally into three or four smaller questions, each of which would make a solid chapter. If yours does not decompose, it is probably a paper. If it decomposes into nine, it is probably a research programme and you should take a piece of it.

If you want a longer checklist than four properties, the standard one is FINERMAPS: feasible, interesting, novel, ethical, relevant, manageable, appropriate, potentially valuable and publishable, and systematic (Ratan et al., 2019, pp. 15-16). Two of those items are usually collapsed and should not be. Feasibility asks whether the study is possible under the conditions you have, and manageability asks whether you personally can organize and finish it (Ratan et al., 2019, p. 17). A question can pass the first and fail the second, which is the common shape of a stalled third year.

The same source gives a concrete test for scale in the other direction. "What is the protein-energy malnutrition in schoolchildren in Delhi?" is answerable with a single statistic, while "How does the education level of the parents impact protein-energy malnutrition in schoolchildren of Delhi?" names a relationship you can argue about (Ratan et al., 2019, p. 16). The narrowing move is the same in any field: start from prior studies, state what is still unknown, then specify the population, the outcome, and the timeframe until the method you propose could actually produce an answer (Ratan et al., 2019, p. 18). Their worked clinical example fixes patients aged 2 to 9 years, 34 applications of mitomycin C at 0.4 mg/ml for five minutes over six months, and symptomatic and radiological relief as the outcome (Ratan et al., 2019, p. 19). Copy the level of detail, not the content.

How many research questions should a thesis have?

One main question and two to four sub-questions. More than that and the thesis loses a spine, because a reader cannot hold five independent lines of inquiry and see them as one contribution. If you have six, two of them are usually the same question at different levels of detail.

The main question is the one your abstract answers in a sentence. If you cannot write that sentence, you have a collection of studies rather than a thesis, which is survivable but makes the introduction and conclusion much harder to write.

Test the set for coherence by asking whether answering all the sub-questions would answer the main one. If a sub-question could be dropped without weakening the answer, it belongs in a paper rather than in this thesis.

Sub-questions do not all have to take the same form. Questions can concern existence, description or classification, composition, relationships, comparisons, or causality, so a descriptive sub-question is legitimate as long as it feeds the main argument (Ratan et al., 2019, p. 15). What is not legitimate is a question whose answer is pure retrieval, such as how much time children spend exercising, where a question about the relationship between physical activity and childhood obesity does real work instead (Ratan et al., 2019, p. 16). The two sources part company on timing. Ratan treats question formulation as a front-loaded design step, while Henderson and Chambers, writing about Registered Reports, require all research questions to be stated in the Stage 1 manuscript and any question that arises from the data to be labelled as having emerged later (Henderson and Chambers, 2022, p. 8). If your field is moving toward preregistration, that labelling is the difference between an honest exploratory finding and a question you look like you had all along.

Should sub-questions map onto chapters?

Usually yes, and it makes the thesis much easier to read. One sub-question per empirical chapter gives each chapter an obvious job and an obvious conclusion. Where a sub-question needs two chapters, say so early rather than leaving the reader to work out the structure.

The mapping also helps you write. A chapter with a question at the top has a finish line: it ends when the question is answered. Chapters defined by topic rather than question are the ones that sprawl, because nothing tells you when they are done.

State the mapping explicitly in the introduction, in one short paragraph or a small table. Examiners read that passage and refer back to it, and a thesis whose structure is legible on page ten is read more generously throughout.