03 · What You Need to Know
Research Question Development and Literature Reviewing Are Usually Iterative
The tidy textbook sequence is appealing: choose a topic, formulate the research question, review the literature, select the method, collect the data.
Actual research development is often less linear.
You need some initial question or problem to know what literature to search. But what you find in that literature can change your understanding of the problem, which changes the question, which changes what literature you need to search next.
Methodological guidance consequently describes research-question development as an iterative process. Literature searching, consultation with experts, feasibility assessment, and refinement of concepts can all lead researchers to revise an initial question before settling on the question that will guide the study.
You need an initial question before you need a final question
Beginning a literature review without any direction can produce an enormous collection of papers with no clear criterion for relevance.
At the other extreme, treating your first research question as untouchable can prevent the literature from doing one of its most important jobs: showing you whether the question is sensible.
A useful distinction is:
Working research question
A provisional question that gives direction to early searching and thinking while remaining open to refinement.
Finalized research question
The question that reflects the developed rationale, literature, concepts, feasible evidence, and methodological decisions of the study.
The first helps you search. The second helps you conduct the study.
The literature may show that your question has already been answered
Suppose your initial question is:
“Is academic self-efficacy associated with academic achievement among university students?”
You begin searching and discover a substantial literature, including multiple meta-analyses, examining essentially that relationship.
This does not necessarily mean that no further study is justified. Perhaps an important population, context, measurement issue, mechanism, or temporal dimension remains uncertain.
But simply reproducing the original question because it was the question you started with would ignore what the literature has revealed.
You might instead ask:
“Does the association between academic self-efficacy and achievement differ during students' transition into university?”
or investigate a theoretically important context in which existing evidence is genuinely limited.
The revised question should arise from a defensible gap rather than from increasingly elaborate attempts to make an already well-studied relationship look new.
A literature review can reveal that your supposed gap is not actually a gap
Researchers sometimes begin with a statement such as “There are no studies on X.” That is a dangerous claim to make before searching carefully.
You may discover that studies exist but use different terminology. The phenomenon may appear in another discipline. A concept may have been studied under an older theoretical label. Relevant evidence may exist in another population or methodological tradition.
The literature review therefore does more than supply citations for the introduction. It tests your assumptions about what is already known.
If the presumed gap disappears, the research question should be reconsidered rather than defended out of loyalty to the original proposal.
The literature may reveal a better gap than the one you started with
Sometimes the original question is not wrong, but the literature reveals a more consequential uncertainty.
Suppose you plan to ask:
“Do university students have positive attitudes toward generative AI?”
After reviewing the literature, you find many attitude surveys. What appears less understood is how students translate those attitudes into decisions when instructors communicate conflicting rules about AI use.
Your question might shift toward:
“How do university students navigate conflicting expectations concerning acceptable generative AI use in assessed work?”
The revised question asks something substantially more informative because the literature helped locate where uncertainty actually remains.
The literature can reveal that your concepts are too vague
Researchers often begin with broad concepts such as “technology use,” “student performance,” “engagement,” “AI literacy,” “well-being,” or “academic success.”
Reading may reveal that the field distinguishes several forms of the concept.
For example, “generative AI use” might include brainstorming, explanation, summarization, editing, feedback, coding, drafting, or complete task generation. Treating these behaviors as one undifferentiated exposure could obscure meaningful differences.
The research question may therefore become more specific:
Initial question:
“How is generative AI use related to student learning?”
Revised question:
“Is students' use of generative AI for explanatory feedback associated with subsequent performance on independently completed learning tasks?”
The literature has not merely provided background. It has changed what the researcher means by the phenomenon.
The literature may show that your terminology is conceptually wrong
Sometimes refinement requires more than adding specificity.
You may discover that two terms you have been using interchangeably represent different constructs. “Engagement” may not mean “participation.” “Achievement” may not be equivalent to “learning.” “Adoption” may differ from sustained use. “AI literacy” may encompass more than technical skill.
If the conceptual literature shows that the wording of your research question collapses distinctions important to the field, revise the question.
Keeping inaccurate terminology simply because it appeared in your original proposal would make the study less coherent, not more consistent.
The literature can reveal that your question is too broad
Suppose you begin with:
“How does artificial intelligence affect higher education?”
A literature search quickly produces work on assessment, tutoring, academic integrity, administration, accessibility, learning analytics, writing, coding, feedback, policy, faculty work, student experience, and institutional strategy.
The literature has demonstrated empirically what was already visible conceptually: the question contains too many phenomena for one study.
You may need to narrow by population, phenomenon, outcome, setting, process, or another meaningful boundary.
This connects directly with recognizing when a research question is too broad. Reviewing existing scholarship can show which distinctions are necessary rather than narrowing arbitrarily.
The literature can also reveal that your question is too narrow
The reverse is possible.
Perhaps your original question concerns one particular AI platform because that is the tool your institution currently uses. Reading reveals that the relevant theoretical phenomenon is automated feedback rather than the brand of software providing it.
If nothing in the research problem depends on that particular platform, retaining the brand-specific restriction may unnecessarily limit the study's relevance.
The literature can therefore help identify restrictions that make a research question too narrow without adding conceptual value.
The literature may change the population you need to study
Suppose you initially plan to study all university students. Prior research reveals that the phenomenon appears particularly consequential during the transition into first-year academic writing, when students are learning institutional expectations about authorship and source use.
You might narrow the population to first-year students because the literature provides a substantive reason for doing so.
Alternatively, the literature may reveal that nearly all existing studies involve first-year students and little is known about advanced students. Your contribution might therefore lie elsewhere.
Population refinement should follow the research problem and gap rather than the assumption that narrower populations automatically produce better questions.
The literature may change the outcome you should examine
You begin with “academic performance” and intend to use final course grades.
The literature shows that previous studies consistently use course grades but that these measures combine many factors and provide limited information about independent performance on the skill of interest.
You may decide that an independently completed task provides a better outcome for your question.
The wording of the research question may then change because the intended outcome has changed conceptually, not merely procedurally.
This is part of the broader issue of whether the data you plan to collect can actually answer the research question.
The literature may change the relationship you think is important
Perhaps you begin by asking whether X is associated with Y. Reading reveals that the relationship is already well established, while the mechanism through which it occurs remains uncertain.
The next useful question might therefore concern mediation, process, experience, or contextual variation rather than another estimate of the same association.
Alternatively, the literature may reveal that a proposed mechanism lacks evidence and that even the basic association remains uncertain.
Research-question refinement should follow the state of knowledge rather than an assumed ladder in which every study must automatically progress from description to correlation to causation.
The literature may change the methodology that makes sense
Suppose you plan a survey because previous work has established several measurable predictors of student AI use.
Your literature review reveals dozens of surveys measuring attitudes and intentions but little understanding of how students actually negotiate ambiguous rules in real academic tasks.
You may decide that another survey would add relatively little and formulate a qualitative question instead.
Conversely, extensive qualitative work may have identified recurring constructs that are now ready for quantitative estimation or testing.
The literature can therefore shift not only the wording of the question but the kind of question worth asking.
That does not violate the principle that methods should follow questions. Question and method development are often iterative: the emerging state of knowledge helps determine what question is useful, and the question determines what evidence and method are appropriate.
The literature may reveal that your causal language is too strong
Suppose your initial question asks:
“What is the impact of social media use on academic achievement?”
As you review previous studies, you discover that most evidence is cross-sectional and observational, with substantial uncertainty about temporal ordering and confounding.
If your own proposed study is similar, you may refine the question to:
“Is social media use associated with academic achievement among undergraduate students?”
This is not merely editing. The literature has helped clarify what level of inference the available design can support.
Alternatively, the review may reveal methodological developments that allow a more explicitly causal observational design. In that case, the question may become more rather than less causally specific.
The important issue, as discussed in the guide on causal language in observational research questions, is alignment between the inferential target and the design.
The literature may reveal that the question cannot be answered with the data you expected to use
Perhaps your proposed outcome requires longitudinal evidence, but the accessible dataset is cross-sectional. Perhaps the construct you want to investigate was measured using one weak proxy. Perhaps the population represented in the dataset differs substantially from the population in your question.
At that point, you can change the data source, redesign the study, or refine the question.
What you should not do is preserve the original question while quietly allowing the available data to answer a different one.
A scoping search and a full literature review serve different stages of refinement
You do not necessarily need to complete an exhaustive literature review before writing any research question.
A preliminary or scoping search can help determine whether the broad topic is viable, identify terminology, locate major debates, and reveal whether an obvious question has already been answered.
You can then formulate a stronger working question and conduct more focused searching.
This creates a cycle:
Initial problem → preliminary question → preliminary search → refined question → focused review → further refinement.
The cycle ends not because the literature has somehow become complete, but because the question is sufficiently developed to support the next stage of the study.
A systematic review has different constraints once its protocol is established
There is an important distinction between reviewing literature to develop a primary study and conducting a systematic review as the study itself.
During early development of a systematic review, scoping searches can help refine the review question and eligibility criteria. Once the review protocol has been finalized or registered, however, substantial changes to the question, outcomes, eligibility criteria, or methods should be documented transparently rather than made invisibly.
PRISMA 2020 asks systematic-review authors to provide registration and protocol information and to describe and explain amendments to information provided at registration or in the protocol.
So the principle remains: questions can change, but the transparency requirements become greater once the study has formally committed to a protocol.
Changing the question before data collection is different from changing it after seeing results
Timing matters.
Suppose the literature review leads you to revise the question before participants are recruited. That is ordinary study development.
Now suppose you collect data, inspect the results, discover that the original relationship is weak, and rewrite the research question around a different statistically significant association.
That is methodologically different.
The problem is not that new questions can never emerge from data. Exploratory findings are scientifically useful. The problem is presenting a post hoc question as though it had been the prespecified question all along.
The next guide considers directly whether a research question can change after data collection has started. The evidentiary and transparency implications become considerably more consequential at that stage.
Qualitative research may remain more iterative for longer
Some qualitative methodologies deliberately permit research questions to evolve as researchers become more familiar with the phenomenon, setting, participants, and emerging data.
Methodological guidance describes qualitative research as iterative and notes that questions may be fine-tuned as understanding develops. The researcher may move between literature, fieldwork, analysis, and conceptual development rather than treating question formulation as permanently completed before the first interview.
This flexibility is methodological, not casual.
The evolving question should remain coherent with the study's purpose, qualitative approach, ethical approvals, sampling strategy, and data being generated. A major shift to an entirely different phenomenon or population may require more than ordinary refinement.
Some qualitative traditions handle prior literature differently
The relationship between literature reviewing and question development also varies among qualitative traditions.
Researchers may differ in how extensively they engage with particular theoretical or empirical literatures before or during data collection, partly because of concerns about prematurely imposing existing categories on emerging analysis.
That does not mean qualitative researchers conduct studies in ignorance of prior scholarship. Rather, the timing, role, and intensity of literature engagement may depend on the methodology and epistemological commitments of the study.
The practical implication is that there is no universal moment at which every qualitative research question must become permanently fixed.
Do not change the question merely because one paper is interesting
Iteration can become drift.
You read one fascinating article about AI anxiety and suddenly your study about AI-supported writing feedback becomes a study about anxiety. The next week, another article redirects you toward academic integrity. Soon the research question follows the most recent PDF you opened.
A revision should have a substantive reason.
Ask:
- Does the new evidence undermine an assumption in the existing question?
- Does it reveal that the question is already answered?
- Does it clarify a concept or theoretically important boundary?
- Does it reveal a stronger knowledge gap?
- Does it change what evidence or inference is feasible?
If not, the article may simply belong in your broader understanding of the field rather than in the research question itself.
Do not chase novelty by repeatedly shrinking the question
Another form of drift occurs when researchers discover that prior studies exist and respond by adding increasingly arbitrary restrictions:
“This relationship has been studied among university students, so I will study only third-year students.”
Then:
“It has also been studied among third-year students, so I will study third-year information-technology students.”
Then:
“Someone studied them too, so I will restrict the study to one university and one semester.”
Eventually the question is technically novel because almost no one has studied precisely those 37 people on a Tuesday.
Novelty is not simply the absence of an identical prior study. A useful research gap should have substantive significance. The literature should help you identify what remains uncertain and why resolving that uncertainty matters.
The literature can justify broadening the question
Refinement does not always mean narrowing.
You may discover that a phenomenon previously studied within one discipline appears conceptually similar across several disciplines. If the theoretical question concerns a broader process, expanding the population or setting may be justified.
You may also find that prior research is excessively fragmented into narrow contexts and that a comparative or multisite study would provide a more useful contribution.
The appropriate change depends on what the literature reveals, not on the assumption that research questions become better every time another restriction is added.
Keep a record of substantial question revisions
A simple research log can be surprisingly useful.
| Version |
Research question |
Reason for revision |
| Initial |
How does generative AI affect student writing? |
Initial broad area of interest |
| Revision 1 |
Is generative AI use associated with academic writing performance? |
Clarified that the proposed observational design does not itself identify an effect |
| Revision 2 |
Is use of generative AI during drafting associated with independently assessed writing performance among first-year students? |
Literature distinguished forms of AI use and identified first-year academic writing as a theoretically relevant context |
| Final |
Is frequency of generative AI-assisted drafting associated with subsequent independently completed writing performance among first-year undergraduate students? |
Clarified exposure, temporal ordering, outcome, and population after literature and feasibility review |
This record helps you explain how the study developed and prevents forgotten assumptions from quietly returning later.
The final literature review should support the final question, not the abandoned one
Once the research question changes, the literature review may need to change with it.
If your original review focused broadly on attitudes toward AI but your final question concerns how students negotiate conflicting institutional expectations, the final literature review should establish what is known about policy interpretation, student decision-making, academic norms, and the relevant context.
Do not preserve several pages of literature merely because you already wrote them. Academic prose is not a loyalty program.
The review should build the rationale for the study you are actually conducting.
Know when refinement has become a different study
Some revisions are modest:
- clarifying a construct;
- narrowing the population for a substantive reason;
- changing an outcome measure to better represent the concept;
- replacing causal wording with associational wording appropriate to the design.
Other changes are more fundamental:
- changing from student learning to faculty workload;
- changing the population entirely;
- moving from prevalence to causal effectiveness;
- replacing the central phenomenon;
- adopting a methodology designed to answer a fundamentally different question.
At some point, you are no longer refining the original study. You are designing another one.
The final guide in this sequence examines when changing the research question means you are actually doing a different study.
