Many students begin a thesis in a familiar way: they choose a broad topic they are interested in, read dozens of papers, and then try to invent a question that sounds academic. That approach is not necessarily wrong, but it often leads to the same problem. The more you read, the wider the topic becomes, and the harder it is to explain what you are actually trying to find out.
A strong study usually does not begin with a sophisticated-sounding question. It begins with a clear problem: what do we already know, what do we still not know, and why does that gap matter?
The research question is the point that connects that problem to the rest of the study. It shapes what literature you need to read, what data you need, what design is appropriate, how the analysis should be conducted, and how far the conclusions can reasonably go.
If the question is poorly framed from the start, using SPSS, R, interviews, machine learning, or a more complicated statistical model will not fix the underlying problem.
A Research Question Is Not the Same as a Topic
Suppose you are interested in artificial intelligence in education.
“AI in higher education” is a topic.
“How does AI affect student learning?” sounds more like a question, but it is still too broad to be a useful research question.
What does “AI” mean here? ChatGPT? Automated feedback? Adaptive learning systems? What does “learning” mean? Grades? Writing quality? Critical thinking? Retention? Which students, in which country, in which programme, and over what period?
A more focused version might be:
How is the use of generative AI for draft feedback associated with changes in academic writing performance among first-year university students over one semester?
This is not necessarily the final version, but it already tells the reader much more. The population is clearer, the behaviour of interest is defined, the outcome is more specific, and the time frame is visible.
Simply writing the question at this level of precision can eliminate many irrelevant directions.
A Good Research Question Should Clarify Three Things
Before using PICO, FINER, or any other framework, ask yourself three simple questions.
First, what exactly do I want to know?
Second, what kind of evidence would allow me to answer it?
Third, if I can answer it, what would actually change in knowledge, practice, policy, or theory?
If the first question is still vague, your topic is not yet sufficiently focused.
If the second is difficult to answer, the project may not be feasible.
If the third has no convincing answer, the problem may not yet be important enough to justify a full study.
This is where many thesis projects run into trouble. A topic may be new in a geographic sense without being new in an academic sense.
For example:
No study has examined this topic in Province X.
That sentence alone does not create a meaningful research gap.
You need to explain why Province X matters. Is the economic structure different? Is the education system different? Are demographic conditions different? Is there a policy difference? Do you expect the underlying mechanism to work differently in that setting?
If the location changes but the theoretical logic does not, the contribution may be limited.
Start With the Research Problem Before Writing the Question
A practical way to proceed is to delay the question for a moment and first write a short paragraph of around three sentences.
Sentence one: what is already reasonably well established?
Sentence two: what remains uncertain, inconsistent, or contested?
Sentence three: why does that uncertainty matter?
For example:
Generative AI is increasingly used by university students during academic writing. However, evidence on whether it improves or weakens argumentative quality remains mixed, partly because many studies measure overall AI use without distinguishing how students use it. Separating AI use for content generation from AI use for feedback may help explain these conflicting findings.
From that paragraph, a research question becomes much easier to formulate:
How is the use of generative AI for draft feedback associated with the quality of argumentation in university students’ academic writing?
Now the question is not being invented randomly. It follows directly from the research problem.
Decide What Kind of Question You Are Asking
Not all research questions do the same job.
Some are descriptive.
Some compare groups.
Some examine associations.
Some estimate causal effects.
Some try to explain mechanisms.
Some explore meaning, experience, or interpretation.
This distinction matters because the language of the question should match the design of the study.
Consider this question:
How does TikTok affect political trust among university students?
The word “affect” suggests a causal claim.
But if the study is only a cross-sectional survey measuring TikTok use and political trust at one point in time, the data can mainly show association. They cannot easily establish that TikTok caused a change in trust.
A more defensible wording would be:
How is exposure to political content on TikTok associated with trust in public institutions among university students?
This small change in wording makes the question much more consistent with what the design can actually support.
A useful general rule is simple: do not let the research question promise more than the study can deliver.
Is PICO Useful?
Yes, but not for every project.
PICO usually stands for Population, Intervention, Comparator, and Outcome. It is particularly useful in clinical research, intervention studies, and systematic reviews of interventions.
For example:
Among adults with stage 1 hypertension, does app-supported home blood pressure monitoring, compared with usual care, reduce mean systolic blood pressure after six months?
Here:
Population is adults with stage 1 hypertension.
Intervention is app-supported home blood pressure monitoring.
Comparator is usual care.
Outcome is systolic blood pressure.
When a time frame is added, the framework is often called PICOT.
PICO is helpful when the study genuinely has that structure. But if you are studying postgraduate students’ experiences of using AI, or how ASEAN states respond to US China strategic competition, forcing the question into PICO may not improve the study.
For scoping reviews, JBI commonly uses PCC, meaning Population, Concept, and Context.
For qualitative or mixed-method evidence synthesis, SPIDER can sometimes be more useful.
The point is not to memorise acronyms. The point is to choose a structure that matches the nature of the question.
FINER Is a Useful Reality Check
A research question can sound impressive and still be impossible to carry out.
FINER is a simple framework that asks whether a question is Feasible, Interesting, Novel, Ethical, and Relevant.
Feasible means: can this actually be done?
Do you have enough time? Can you access the participants? Do you have the data? Do you have the necessary analytical skills?
A five-year longitudinal question may be excellent scientifically but completely unrealistic for a one-year master’s thesis.
Interesting means the answer should matter to someone.
Not everyone needs to care, but at least a scholarly community, professional group, or decision-making audience should find the result useful.
Novel does not mean nobody in the world has ever studied the topic.
A study can still be novel if it uses better data, improves measurement, tests a mechanism, compares competing theories, performs a justified replication, or resolves conflicting evidence.
Ethical refers to whether the study can be conducted responsibly.
A question may be scientifically interesting but inappropriate if it creates unreasonable risks for participants or requires access to sensitive data without adequate protections.
Relevant asks what the result would contribute.
Would it improve theory, measurement, policy, practice, or decision-making? Would it change the way the problem is understood?
If a research question fails several parts of FINER, the right response is usually to revise the question rather than force the study forward.
Translate the Question Into the Data You Would Need
One of the best practical tests is to imagine that the perfect dataset arrives tomorrow.
What would it need to contain?
Suppose your question is:
How is trade dependence on China associated with foreign policy choices among ASEAN states?
You would probably need some measure of trade dependence, perhaps the share of exports and imports involving China.
But how will “foreign policy choices” be measured?
UN voting behaviour?
Official statements?
Defence agreements?
Responses to maritime disputes?
If you cannot define what evidence would represent the main concepts in your question, the question is probably still too vague.
This is why operationalisation should be considered early.
You do not need to decide the full statistical model immediately, but you should have a realistic idea of what evidence could answer the question.
The same principle applies to qualitative research.
If your question is:
How do PhD students experience pressure to publish internationally?
You should already be thinking about who needs to be interviewed, what contexts matter, what differences between participants may be important, and what kinds of data would allow you to distinguish between competing interpretations.
A Thesis Question and an Article Question Are Not Always the Same
A thesis usually has more room.
You may have one primary question and several related sub-questions, as long as they all contribute to one central argument.
A journal article usually needs tighter focus.
A strong article tends to have one central message. If one manuscript tries to answer four major questions, use three theories, and analyse five groups of variables, the argument can become difficult to follow.
For example, a thesis might investigate:
How does generative AI influence academic writing practices among university students?
That thesis could later generate several different papers.
One paper might examine AI use and argumentative quality.
Another might focus on students’ attitudes toward AI.
Another might analyse differences across disciplines.
A thesis does not always need to be compressed into one single paper.
Three Examples of Improving a Research Question
In medicine, a weak question would be:
Is telemedicine effective?
There is not enough information to design a study.
A stronger version would be:
Among adults with type 2 diabetes receiving outpatient care, does telemedicine with remote glucose monitoring, compared with usual care, reduce HbA1c after six months?
Now the population, intervention, comparator, outcome, and time frame are clear.
In education:
Does AI help students learn better?
is too broad.
A more researchable version might be:
Among first-year university students, how is the frequency of using generative AI for draft feedback associated with changes in academic writing scores over one semester?
In political science:
How does US China competition affect ASEAN?
is also too broad.
A more focused question could be:
How are trade dependence on China and security dependence on the United States associated with foreign policy choices among ASEAN states in maritime disputes between 2016 and 2026?
The last question still requires careful operationalisation, but at least the researcher now knows what kinds of relationships need to be examined.
Common Mistakes When Writing a Research Question
One common mistake is making the question too broad.
If you need an entire chapter just to explain what the question means, the scope is probably too large.
Another mistake is including too many variables and too many goals.
A question containing several “ands”, multiple outcomes, and several populations often signals that the project is trying to do too much.
A third problem is using causal language with a non-causal design.
This is especially common in survey research.
Another mistake is treating “not previously studied in location X” as an automatic contribution.
A new location matters only when there is a credible academic reason to expect differences in theory, mechanism, measurement, context, or application.
Finally, some researchers choose the question based on whatever data happen to be available.
If a dataset contains twenty variables, that does not mean the research question should be designed to use all twenty.
Ideally, the research question should guide the data selection, not the other way around.
A Simple Checklist Before Finalising the Question
Before submitting a proposal or starting data collection, ask yourself:
- Can I explain the research problem in two or three sentences?
- Do I know what specific evidence gap I am addressing?
- Is the population or unit of analysis clear?
- Do I know whether I am describing, comparing, examining association, estimating causality, or interpreting experience?
- Can the main concepts be measured or observed credibly?
- Can I actually access the data or participants I need?
- Can the proposed design genuinely answer the question?
- Is the scope realistic for the available time and resources?
- Would the study still be informative if the result contradicts my expectation?
- If I answer the question, what exactly will we understand better?
If several answers are unclear, that does not mean the project has failed. It usually means the research question still needs revision.
How Long Should a Research Question Be?
There is no fixed length.
A strong research question can be one sentence.
The important thing is that it gives enough information to guide the study without trying to compress the entire methodology section into one line.
A useful drafting structure is:
Among or within [population/case/context], how is [exposure/intervention/phenomenon] associated with or experienced in relation to [outcome/meaning], compared with [comparator if needed], over [time if needed]?
Then remove anything unnecessary.
A research question does not need to sound complicated to be academically strong.
In fact, the strongest questions are often easy to understand.
The Most Important Thing Is Alignment
A research question does not stand alone.
It needs to align with the research gap.
The research gap needs to align with the design.
The design needs to produce the right kind of data.
The data need to support an appropriate analysis.
And the analysis should only lead to conclusions that the design can actually justify.
A simple way to remember this is:
Research Gap → Research Question → Research Design → Data → Analysis → Conclusion
If one link does not fit the next, it is better to revise the question early.
Rewriting a research question before data collection may take a few hours or a few days. Discovering six months later that the data cannot answer the original question is far more costly.
References
Farrugia, P., Petrisor, B. A., Farrokhyar, F., & Bhandari, M. (2010). Research questions, hypotheses and objectives. Canadian Journal of Surgery, 53(4), 278–281.
Fandino, W. (2019). Formulating a good research question: Pearls and pitfalls. Indian Journal of Anaesthesia, 63(8), 611–616. https://doi.org/10.4103/ija.IJA_198_19
Covvey, J. R., McClendon, C., & Gionfriddo, M. R. (2024). Back to the basics: Guidance for formulating good research questions. Research in Social and Administrative Pharmacy, 20(1), 66–69. https://doi.org/10.1016/j.sapharm.2023.09.009
Rios, L. P., Ye, C., & Thabane, L. (2010). Association between framing of the research question using the PICOT format and reporting quality of randomized controlled trials. BMC Medical Research Methodology, 10, 11. https://doi.org/10.1186/1471-2288-10-11
Cooke, A., Smith, D., & Booth, A. (2012). Beyond PICO: The SPIDER tool for qualitative evidence synthesis. Qualitative Health Research, 22(10), 1435–1443. https://doi.org/10.1177/1049732312452938
Cochrane. Cochrane Handbook for Systematic Reviews of Interventions. Sections on developing review questions and PICO-based eligibility criteria.
JBI. JBI Manual for Evidence Synthesis. Sections on developing review questions and eligibility criteria using Population, Concept, and Context.
How to Build a Research Question for a Thesis or Research Paper
Many students begin a thesis in a familiar way: they choose a broad topic they are interested in, read dozens of papers, and then try to invent a question that sounds academic. That approach is not necessarily wrong, but it often leads to the same problem. The more you read, the wider the topic becomes, and the harder it is to explain what you are actually trying to find out.
A strong study usually does not begin with a sophisticated-sounding question. It begins with a clear problem: what do we already know, what do we still not know, and why does that gap matter?
The research question is the point that connects that problem to the rest of the study. It shapes what literature you need to read, what data you need, what design is appropriate, how the analysis should be conducted, and how far the conclusions can reasonably go.
If the question is poorly framed from the start, using SPSS, R, interviews, machine learning, or a more complicated statistical model will not fix the underlying problem.
A Research Question Is Not the Same as a Topic
Suppose you are interested in artificial intelligence in education.
“AI in higher education” is a topic.
“How does AI affect student learning?” sounds more like a question, but it is still too broad to be a useful research question.
What does “AI” mean here? ChatGPT? Automated feedback? Adaptive learning systems? What does “learning” mean? Grades? Writing quality? Critical thinking? Retention? Which students, in which country, in which programme, and over what period?
A more focused version might be:
How is the use of generative AI for draft feedback associated with changes in academic writing performance among first-year university students over one semester?
This is not necessarily the final version, but it already tells the reader much more. The population is clearer, the behaviour of interest is defined, the outcome is more specific, and the time frame is visible.
Simply writing the question at this level of precision can eliminate many irrelevant directions.
A Good Research Question Should Clarify Three Things
Before using PICO, FINER, or any other framework, ask yourself three simple questions.
First, what exactly do I want to know?
Second, what kind of evidence would allow me to answer it?
Third, if I can answer it, what would actually change in knowledge, practice, policy, or theory?
If the first question is still vague, your topic is not yet sufficiently focused.
If the second is difficult to answer, the project may not be feasible.
If the third has no convincing answer, the problem may not yet be important enough to justify a full study.
This is where many thesis projects run into trouble. A topic may be new in a geographic sense without being new in an academic sense.
For example:
No study has examined this topic in Province X.
That sentence alone does not create a meaningful research gap.
You need to explain why Province X matters. Is the economic structure different? Is the education system different? Are demographic conditions different? Is there a policy difference? Do you expect the underlying mechanism to work differently in that setting?
If the location changes but the theoretical logic does not, the contribution may be limited.
Start With the Research Problem Before Writing the Question
A practical way to proceed is to delay the question for a moment and first write a short paragraph of around three sentences.
Sentence one: what is already reasonably well established?
Sentence two: what remains uncertain, inconsistent, or contested?
Sentence three: why does that uncertainty matter?
For example:
Generative AI is increasingly used by university students during academic writing. However, evidence on whether it improves or weakens argumentative quality remains mixed, partly because many studies measure overall AI use without distinguishing how students use it. Separating AI use for content generation from AI use for feedback may help explain these conflicting findings.
From that paragraph, a research question becomes much easier to formulate:
How is the use of generative AI for draft feedback associated with the quality of argumentation in university students’ academic writing?
Now the question is not being invented randomly. It follows directly from the research problem.
Decide What Kind of Question You Are Asking
Not all research questions do the same job.
Some are descriptive.
Some compare groups.
Some examine associations.
Some estimate causal effects.
Some try to explain mechanisms.
Some explore meaning, experience, or interpretation.
This distinction matters because the language of the question should match the design of the study.
Consider this question:
How does TikTok affect political trust among university students?
The word “affect” suggests a causal claim.
But if the study is only a cross-sectional survey measuring TikTok use and political trust at one point in time, the data can mainly show association. They cannot easily establish that TikTok caused a change in trust.
A more defensible wording would be:
How is exposure to political content on TikTok associated with trust in public institutions among university students?
This small change in wording makes the question much more consistent with what the design can actually support.
A useful general rule is simple: do not let the research question promise more than the study can deliver.
Is PICO Useful?
Yes, but not for every project.
PICO usually stands for Population, Intervention, Comparator, and Outcome. It is particularly useful in clinical research, intervention studies, and systematic reviews of interventions.
For example:
Among adults with stage 1 hypertension, does app-supported home blood pressure monitoring, compared with usual care, reduce mean systolic blood pressure after six months?
Here:
Population is adults with stage 1 hypertension.
Intervention is app-supported home blood pressure monitoring.
Comparator is usual care.
Outcome is systolic blood pressure.
When a time frame is added, the framework is often called PICOT.
PICO is helpful when the study genuinely has that structure. But if you are studying postgraduate students’ experiences of using AI, or how ASEAN states respond to US China strategic competition, forcing the question into PICO may not improve the study.
For scoping reviews, JBI commonly uses PCC, meaning Population, Concept, and Context.
For qualitative or mixed-method evidence synthesis, SPIDER can sometimes be more useful.
The point is not to memorise acronyms. The point is to choose a structure that matches the nature of the question.
FINER Is a Useful Reality Check
A research question can sound impressive and still be impossible to carry out.
FINER is a simple framework that asks whether a question is Feasible, Interesting, Novel, Ethical, and Relevant.
Feasible means: can this actually be done?
Do you have enough time? Can you access the participants? Do you have the data? Do you have the necessary analytical skills?
A five-year longitudinal question may be excellent scientifically but completely unrealistic for a one-year master’s thesis.
Interesting means the answer should matter to someone.
Not everyone needs to care, but at least a scholarly community, professional group, or decision-making audience should find the result useful.
Novel does not mean nobody in the world has ever studied the topic.
A study can still be novel if it uses better data, improves measurement, tests a mechanism, compares competing theories, performs a justified replication, or resolves conflicting evidence.
Ethical refers to whether the study can be conducted responsibly.
A question may be scientifically interesting but inappropriate if it creates unreasonable risks for participants or requires access to sensitive data without adequate protections.
Relevant asks what the result would contribute.
Would it improve theory, measurement, policy, practice, or decision-making? Would it change the way the problem is understood?
If a research question fails several parts of FINER, the right response is usually to revise the question rather than force the study forward.
Translate the Question Into the Data You Would Need
One of the best practical tests is to imagine that the perfect dataset arrives tomorrow.
What would it need to contain?
Suppose your question is:
How is trade dependence on China associated with foreign policy choices among ASEAN states?
You would probably need some measure of trade dependence, perhaps the share of exports and imports involving China.
But how will “foreign policy choices” be measured?
UN voting behaviour?
Official statements?
Defence agreements?
Responses to maritime disputes?
If you cannot define what evidence would represent the main concepts in your question, the question is probably still too vague.
This is why operationalisation should be considered early.
You do not need to decide the full statistical model immediately, but you should have a realistic idea of what evidence could answer the question.
The same principle applies to qualitative research.
If your question is:
How do PhD students experience pressure to publish internationally?
You should already be thinking about who needs to be interviewed, what contexts matter, what differences between participants may be important, and what kinds of data would allow you to distinguish between competing interpretations.
A Thesis Question and an Article Question Are Not Always the Same
A thesis usually has more room.
You may have one primary question and several related sub-questions, as long as they all contribute to one central argument.
A journal article usually needs tighter focus.
A strong article tends to have one central message. If one manuscript tries to answer four major questions, use three theories, and analyse five groups of variables, the argument can become difficult to follow.
For example, a thesis might investigate:
How does generative AI influence academic writing practices among university students?
That thesis could later generate several different papers.
One paper might examine AI use and argumentative quality.
Another might focus on students’ attitudes toward AI.
Another might analyse differences across disciplines.
A thesis does not always need to be compressed into one single paper.
Three Examples of Improving a Research Question
In medicine, a weak question would be:
Is telemedicine effective?
There is not enough information to design a study.
A stronger version would be:
Among adults with type 2 diabetes receiving outpatient care, does telemedicine with remote glucose monitoring, compared with usual care, reduce HbA1c after six months?
Now the population, intervention, comparator, outcome, and time frame are clear.
In education:
Does AI help students learn better?
is too broad.
A more researchable version might be:
Among first-year university students, how is the frequency of using generative AI for draft feedback associated with changes in academic writing scores over one semester?
In political science:
How does US China competition affect ASEAN?
is also too broad.
A more focused question could be:
How are trade dependence on China and security dependence on the United States associated with foreign policy choices among ASEAN states in maritime disputes between 2016 and 2026?
The last question still requires careful operationalisation, but at least the researcher now knows what kinds of relationships need to be examined.
Common Mistakes When Writing a Research Question
One common mistake is making the question too broad.
If you need an entire chapter just to explain what the question means, the scope is probably too large.
Another mistake is including too many variables and too many goals.
A question containing several “ands”, multiple outcomes, and several populations often signals that the project is trying to do too much.
A third problem is using causal language with a non-causal design.
This is especially common in survey research.
Another mistake is treating “not previously studied in location X” as an automatic contribution.
A new location matters only when there is a credible academic reason to expect differences in theory, mechanism, measurement, context, or application.
Finally, some researchers choose the question based on whatever data happen to be available.
If a dataset contains twenty variables, that does not mean the research question should be designed to use all twenty.
Ideally, the research question should guide the data selection, not the other way around.
A Simple Checklist Before Finalising the Question
Before submitting a proposal or starting data collection, ask yourself:
- Can I explain the research problem in two or three sentences?
- Do I know what specific evidence gap I am addressing?
- Is the population or unit of analysis clear?
- Do I know whether I am describing, comparing, examining association, estimating causality, or interpreting experience?
- Can the main concepts be measured or observed credibly?
- Can I actually access the data or participants I need?
- Can the proposed design genuinely answer the question?
- Is the scope realistic for the available time and resources?
- Would the study still be informative if the result contradicts my expectation?
- If I answer the question, what exactly will we understand better?
If several answers are unclear, that does not mean the project has failed. It usually means the research question still needs revision.
How Long Should a Research Question Be?
There is no fixed length.
A strong research question can be one sentence.
The important thing is that it gives enough information to guide the study without trying to compress the entire methodology section into one line.
A useful drafting structure is:
Among or within [population/case/context], how is [exposure/intervention/phenomenon] associated with or experienced in relation to [outcome/meaning], compared with [comparator if needed], over [time if needed]?
Then remove anything unnecessary.
A research question does not need to sound complicated to be academically strong.
In fact, the strongest questions are often easy to understand.
The Most Important Thing Is Alignment
A research question does not stand alone.
It needs to align with the research gap.
The research gap needs to align with the design.
The design needs to produce the right kind of data.
The data need to support an appropriate analysis.
And the analysis should only lead to conclusions that the design can actually justify.
A simple way to remember this is:
Research Gap → Research Question → Research Design → Data → Analysis → Conclusion
If one link does not fit the next, it is better to revise the question early.
Rewriting a research question before data collection may take a few hours or a few days. Discovering six months later that the data cannot answer the original question is far more costly.
References
Farrugia, P., Petrisor, B. A., Farrokhyar, F., & Bhandari, M. (2010). Research questions, hypotheses and objectives. Canadian Journal of Surgery, 53(4), 278–281.
Fandino, W. (2019). Formulating a good research question: Pearls and pitfalls. Indian Journal of Anaesthesia, 63(8), 611–616. https://doi.org/10.4103/ija.IJA_198_19
Covvey, J. R., McClendon, C., & Gionfriddo, M. R. (2024). Back to the basics: Guidance for formulating good research questions. Research in Social and Administrative Pharmacy, 20(1), 66–69. https://doi.org/10.1016/j.sapharm.2023.09.009
Rios, L. P., Ye, C., & Thabane, L. (2010). Association between framing of the research question using the PICOT format and reporting quality of randomized controlled trials. BMC Medical Research Methodology, 10, 11. https://doi.org/10.1186/1471-2288-10-11
Cooke, A., Smith, D., & Booth, A. (2012). Beyond PICO: The SPIDER tool for qualitative evidence synthesis. Qualitative Health Research, 22(10), 1435–1443. https://doi.org/10.1177/1049732312452938
Cochrane. Cochrane Handbook for Systematic Reviews of Interventions. Sections on developing review questions and PICO-based eligibility criteria.
JBI. JBI Manual for Evidence Synthesis. Sections on developing review questions and eligibility criteria using Population, Concept, and Context.
◉ 0 Thảo luận Khoa học