03 · What You Need to Know
The Research Question Should Drive the Method
A basic principle of research design is that the method should fit the research question. Different questions seek different kinds of answers, so it should not be surprising when they require different evidence.
If you ask how common a behavior is, you need evidence capable of supporting a defensible estimate. If you ask how participants experience that behavior, you need evidence capable of illuminating experiences. If you ask whether an intervention causes an outcome, the methodological requirements change again.
Trying to force every question through one method simply because the study began with that method reverses the logic of research design.
The same topic can generate questions requiring very different evidence
Consider generative AI in academic writing:
RQ1: “What proportion of undergraduate students use generative AI when preparing assessed written assignments?”
RQ2: “How do undergraduate students decide whether particular uses of generative AI in assessed writing are acceptable?”
The first question seeks a numerical estimate. A survey using an appropriate sampling strategy may be suitable.
The second seeks an understanding of reasoning, interpretations, and decision-making. Interviews, focus groups, observations, documents, or another qualitative approach may be more appropriate depending on the intended inquiry.
Using interviews alone for RQ1 would make it difficult to produce a population prevalence estimate. Using a few closed survey items alone for RQ2 might flatten the contextual reasoning that the question actually seeks to understand.
This is why the distinction between qualitative and quantitative research questions matters. They can concern the same phenomenon while requiring fundamentally different forms of evidence.
Different methods do not automatically mean mixed methods
This is one of the most important distinctions.
Mixed-methods research involves more than the presence of both numbers and words. Methodological literature generally emphasizes intentional integration of quantitative and qualitative approaches within a study or program of inquiry. Integration distinguishes mixed-methods research from studies in which qualitative and quantitative components simply occur alongside one another.
The NIH Office of Behavioral and Social Sciences Research defines mixed-methods research as research in which investigators collect and analyze both quantitative and qualitative data, integrate the two forms of data and their results, and organize the procedures into a rigorous design appropriate to the study's questions.
Multiple methods
More than one method is used to answer one or more research questions.
Mixed methods
Quantitative and qualitative approaches are intentionally combined and integrated so that their relationship contributes to answering the overall research problem.
The distinction is important because a survey in Chapter 4 and interviews in Chapter 5 do not automatically become mixed methods simply by occupying the same thesis.
Integration is what makes the combination analytically meaningful
Suppose your survey shows that students report using generative AI much more frequently for brainstorming than for generating complete drafts.
Your interviews then investigate how students distinguish acceptable assistance from unacceptable substitution of authorship.
If you integrate those findings, you might examine whether students' reasoning helps explain the quantitative pattern. Perhaps brainstorming is commonly viewed as support for one's own thinking, whereas full-text generation is perceived as crossing an authorship boundary.
The qualitative evidence now helps interpret the quantitative result. Conversely, the survey may show how widespread patterns identified qualitatively appear to be.
That interaction between strands is the methodological value of mixed methods.
Without integration, you may simply have two parallel studies: one reporting frequencies and another reporting themes.
Integration can happen at several stages
Mixed-methods integration does not need to occur only in the discussion section.
Fetters, Curry, and Creswell describe integration as potentially occurring through the design, methods, and interpretation or reporting of mixed-methods research. They identify approaches such as connecting, building, merging, and embedding.
For example:
- one phase may determine who is sampled in the next phase;
- results from a quantitative phase may help develop qualitative interview questions;
- qualitative findings may help develop a survey instrument;
- quantitative and qualitative results may be merged during analysis or interpretation; or
- one form of evidence may be embedded within a larger design.
The important issue is that the relationship between methods is planned because it helps answer the research problem.
A sequential design may fit questions that depend on one another
Sometimes the answer to one research question needs to inform how the next question is investigated.
Suppose you ask:
RQ1: “What patterns of generative AI use are reported by undergraduate students?”
RQ2: “How do students explain the most common and least common patterns identified in RQ1?”
The second question depends partly on the results of the first. A sequential explanatory mixed-methods design may be appropriate: collect and analyze quantitative data first, then use qualitative inquiry to help explain the quantitative results.
In another study, qualitative work might come first. Interviews could identify forms of AI use that are poorly represented in existing instruments, and those findings could then inform development of a quantitative survey. This resembles an exploratory sequential logic.
The order should follow the relationship among the questions rather than a generic preference for “quantitative first” or “qualitative first.”
A convergent design may fit questions that can be investigated concurrently
Other questions do not depend on one another temporally.
You might simultaneously investigate:
RQ1: “How frequently do students use generative AI for academic writing?”
RQ2: “How do students experience negotiating acceptable AI use?”
The quantitative and qualitative evidence can be collected during roughly the same phase and later compared or integrated.
A convergent design can be useful when different forms of evidence illuminate complementary aspects of the same phenomenon and neither needs to precede the other.
Again, collecting both types of data at the same time is not enough. The design should explain how the findings will be brought together.
One method can also answer multiple questions
The fact that different questions can require different methods does not mean every research question needs its own method.
A single survey might answer:
RQ1: “What proportion of students use generative AI for academic writing?”
RQ2: “Is frequency of generative AI use associated with writing self-efficacy?”
RQ3: “Does reported generative AI use differ across year levels?”
The analyses differ, but the same general quantitative data-generation strategy may provide the necessary evidence for all three.
Similarly, one set of qualitative interviews might address several related subquestions about participants' experiences.
Methods should multiply only when the questions require them, not because methodological variety looks impressive.
Different analyses are not necessarily different methods
Researchers sometimes say they are using “multiple methods” because one question uses descriptive statistics and another uses regression.
Those are different analytical techniques, but they may still belong to the same quantitative methodological approach and dataset.
Likewise, thematic analysis of interviews and document analysis may represent multiple qualitative methods without making the project mixed methods in the quantitative-plus-qualitative sense.
Terminology varies somewhat across fields, but the conceptual distinction remains useful: using several analytical procedures is not automatically the same thing as combining different methodological traditions.
Two qualitative questions can require different qualitative methods
Consider:
RQ1: “How do students experience being accused of inappropriate generative AI use?”
RQ2: “How are acceptable and unacceptable AI uses represented in institutional policy documents?”
The first may require interviews. The second may require document analysis.
Both are qualitative, but the evidence sources differ because the questions concern different aspects of the same broader problem.
A multimethod qualitative study could combine them if understanding the relationship between institutional discourse and student experience is central to the research purpose.
There is no methodological requirement that all questions in a qualitative study must be answered through the same interview dataset.
Two quantitative questions can require different quantitative methods
The same applies within quantitative research.
A study might ask one question using survey data and another using administrative records. One question might require cross-sectional prevalence estimation while another requires longitudinal modeling.
The project remains quantitative, although it uses multiple data sources or quantitative methods.
What matters is whether the combined design has been planned to answer the questions coherently.
Different methods create additional feasibility requirements
Adding a method is not merely adding another section to the methods chapter.
Each method can introduce:
- new sampling requirements;
- additional recruitment;
- new instruments or protocols;
- different data-management procedures;
- additional ethical considerations;
- specialized analytical expertise;
- more time for data collection and analysis; and
- the additional task of integrating findings.
A mixed-methods study may therefore answer a complex research problem more comprehensively while also requiring substantially more resources than a single-method study.
This returns to the issue of how many research questions become too many. The burden of a question depends partly on what evidence and methods it creates.
Methodological competence matters
A study using several methods needs credible expertise in each of them.
Adding interviews to a quantitative study does not make qualitative analysis straightforward. Likewise, adding a survey to a qualitative project introduces measurement, sampling, statistical, and potentially psychometric issues.
Mixed-methods methodology also requires competence in integration. A research team may conduct strong qualitative and quantitative components separately yet still produce a weak mixed-methods study if the strands are never meaningfully connected.
Methodological breadth should therefore be matched by methodological capacity.
Different methods should not be used merely for “validation”
Researchers sometimes justify mixed methods by saying, “We will conduct interviews to validate the survey results.”
That can oversimplify the relationship between forms of evidence.
Qualitative findings do not automatically function as a truth test for quantitative findings, nor do numerical results automatically validate participants' accounts. The two approaches may examine different dimensions of a phenomenon and can legitimately produce findings that are complementary, divergent, or apparently contradictory.
Mixed-methods integration can involve convergence, but it can also reveal dissonance that requires explanation.
Fetters and colleagues emphasize that integration can generate insights through confirmation, expansion, or discordance between quantitative and qualitative results.
Contradictory findings are not automatically a methodological failure
Suppose a survey finds that most students report understanding the university's AI policy, while interviews reveal substantial uncertainty about how the policy applies to particular writing practices.
Those findings are not necessarily inconsistent in a useless way. Students may believe they understand the general rule while remaining uncertain in specific situations. The apparent contradiction may expose an important distinction between perceived general clarity and practical interpretability.
One advantage of multiple forms of evidence is precisely that they can complicate an initially simple conclusion.
Integration should therefore ask why findings converge or diverge rather than forcing them into artificial agreement.
The methods should correspond clearly to the questions
A useful planning tool is a research-question-to-method matrix.
| Research question |
Evidence needed |
Possible method |
Analysis |
| How frequently do students use generative AI? |
Numerical reports from an appropriate sample |
Survey |
Descriptive estimation |
| Is AI-use frequency associated with writing self-efficacy? |
Measures of AI use and self-efficacy |
Survey or other quantitative measurement |
Appropriate associational analysis |
| How do students decide whether AI use is acceptable? |
Detailed accounts of reasoning and experience |
Semi-structured interviews |
Qualitative analysis appropriate to the methodological approach |
| How do institutional rules define acceptable AI use? |
Policy and guidance documents |
Document analysis |
Qualitative or content-analytic approach appropriate to the question |
The table makes an important problem visible: if every row requires a new population, dataset, method, and analytical tradition, the study may be expanding rapidly even when the questions remain topically related.
The questions still need to belong to the same study
Suppose a project asks:
RQ1: “How frequently do students use generative AI?”
RQ2: “How do faculty members experience burnout?”
You could use a survey for the first and interviews for the second. Both methods might be perfectly appropriate.
But methodological appropriateness does not create conceptual coherence.
This is why multiple research questions must still address a coherent research problem. Different methods can serve connected questions; they cannot rescue questions that never belonged together.
Sometimes the correct solution is two studies
Imagine one question requires a national survey of students while another requires year-long ethnographic observation of faculty practice. Both concern educational technology, but each could independently justify a substantial study.
Combining them may add little unless their findings need to interact to answer a larger question.
Separating the studies can allow each to use an appropriate sampling strategy, methodological rationale, analytical framework, and publication format without forcing artificial integration.
“Different methods” is therefore not a problem to eliminate. It is a signal to ask whether the larger design has a reason to contain both.
The research questions should be formulated before the methods are assigned
A common mistake is beginning with available tools:
“I want to use a survey and interviews. What research questions can I make?”
That sequence risks producing questions whose main justification is methodological convenience.
A stronger sequence is:
Research problem → research question → evidence needed → method.
The process can certainly be iterative. Feasibility may force you to refine the question, and methodological possibilities may reveal ways of studying a problem you had not considered. But the final design should be explainable in terms of why each method is needed to answer the question rather than why each question was invented to justify a method.
Mixed methods needs a reason for mixing
Before calling a study mixed methods, complete this sentence:
“We need both quantitative and qualitative evidence because...”
Possible answers might include:
“...we need to estimate how widespread the pattern is and understand how participants explain it.”
“...qualitative findings are needed to develop a measure that will subsequently be tested quantitatively.”
“...quantitative results identify a pattern whose underlying processes require qualitative investigation.”
“...the research problem requires both outcome estimates and an understanding of implementation.”
If the only answer is “because using two methods makes the study stronger,” the rationale is incomplete.
Mixed methods is useful when integration produces knowledge that either strand alone would leave incomplete.
Integration should appear in the research questions when it is central to the design
Some mixed-methods studies formulate separate quantitative and qualitative questions plus an explicit mixed-methods question.
For example:
Quantitative question: “What patterns of generative AI use are reported by undergraduate students?”
Qualitative question: “How do students explain their decisions about when to use generative AI?”
Mixed-methods question: “How do students' explanations help interpret the quantitative patterns of generative AI use?”
This makes integration visible as a research task rather than something that will somehow occur after both analyses are finished.
Not every mixed-methods study needs exactly this wording or three-question structure. The useful principle is that integration should be planned, not accidental.
The final interpretation should answer the whole study, not just each method separately
A common weakness in multimethod projects is that the quantitative findings are discussed, then the qualitative findings are discussed, and the paper ends.
If the study was justified as mixed methods, the reader should eventually learn what becomes visible when those findings are considered together.
Fetters and colleagues describe joint displays as one strategy for bringing quantitative and qualitative findings together and generating integrated interpretations.
The final contribution should therefore return to the overall research problem. Otherwise, different methods may have answered their individual questions successfully without producing the integrated knowledge the mixed-methods design promised.
