Mixed methods research
Introduction to Social Research Methodology
Why mix methods?
Most real organisational problems present two kinds of evidence at once. A company like Meridian knows that its staff turnover has doubled — the HR dashboard reports it precisely and reliably — while its exit interviews hint at why, in the form of a handful of unstructured conversations with people already halfway out of the door. Numbers without stories tell you the size of a problem but not its mechanism; stories without numbers tell you a mechanism but not whether it is typical. Mixed methods research exists to combine the two kinds of evidence by design, rather than by accident, afterwards, in the discussion section.
Definition
Creswell and Creswell define mixed methods research by three components, all of which are required: the researcher collects both qualitative and quantitative data; integrates the two forms of data; and draws interpretations from their combined strengths. The middle component carries the definition. Simply using two methods in one project is not mixed methods research — the strands must actually meet. The underlying premise is that integration yields more than the sum of the parts, because each form of data offsets the known weaknesses of the other. Philosophically, mixed methods sits most comfortably with pragmatism: the research question comes first, and methods are treated as tools rather than loyalties.
The logic of complementary strengths follows directly from the comparison introduced in session 1:
| Quantitative | Qualitative | |
|---|---|---|
| Strength | Breadth, measurement, generalisation | Depth, meaning, discovery |
| Blind spot | Why the pattern exists | How common the experience is |
A survey can establish that turnover is concentrated among staff with under two years’ tenure; only interviews can tell you what those two years feel like. Neither alone answers the board’s question.
Four classic reasons to mix
Greene, Caracelli and Graham (1989) catalogued the purposes that mixing can serve. Four recur constantly, and a good justification names which of them the study is pursuing — “a richer picture” is not a purpose.
| Purpose | What it means |
|---|---|
| Triangulation | Testing whether different methods converge on the same answer, strengthening confidence in the result |
| Complementarity | Using one method to elaborate and illustrate the results of the other |
| Development | Using the results of one method to build the other — interviews that shape a questionnaire, or a survey that selects interviewees |
| Expansion | Widening the study’s range, letting different methods answer different components of one question |
One warning is worth internalising early: when triangulated findings diverge, that is a finding, not a failure — but only if the design has a plan for confronting the two strands with each other.
The three core designs
Notation
Mixed methods has a compact notation (due to Morse) that encodes three design decisions at once. Capitalisation indicates priority: QUAN means the quantitative strand leads, while quan means it plays a supporting role. A plus sign (+) means the strands run concurrently; an arrow (→) means one strand finishes before the next begins and feeds into it. The three core designs in Creswell and Creswell each get one line: QUAN + QUAL (convergent), QUAN → qual (explanatory sequential), and QUAL → quan (exploratory sequential). Whenever you meet a mixed design, decode three things: the timing, the priority, and where the strands meet.
Convergent (parallel) design
In a convergent design, both strands are collected and analysed at the same time, separately, and then merged for comparison. The strands are typically given equal priority, and neither depends on the other’s results. The purpose is usually triangulation (do the two databases agree?) and completeness (the statistical pattern plus the lived experience of the same phenomenon). Its great practical attraction is that everything happens in one data-collection period, making it the most time-efficient of the three designs. Use it when you need both kinds of answer to the same question, and you need them now.
| Phase | Quantitative strand | Qualitative strand |
|---|---|---|
| 1. Collect | CAWI survey, large sample | Interviews or focus groups, small purposive sample |
| 2. Analyse | Statistics, analysed separately | Thematic analysis, analysed separately |
| 3. Merge | Results placed side by side — where do they converge, where do they diverge? | |
| 4. Interpret | One integrated conclusion, drawing on both databases |
Organisational example. Meridian’s customer problem: a CAWI satisfaction survey across all 32 stores runs in the same month as focus groups with shoppers in four stores. The merged results show scores falling everywhere, but the focus groups reveal that queueing at understaffed tills is what customers actually talk about.
Pitfalls. The strands must stand on equal footing — a 1,000-respondent survey merged with three interviews makes the merge decorative. The strands must measure comparable constructs: if the survey measures satisfaction and the focus groups discuss loyalty, they cannot be meaningfully compared. The design needs a plan for divergence, since the strands will sometimes disagree; without one, the study collapses into cherry-picking. Running both strands simultaneously demands qualitative and quantitative expertise at the same time, which usually means a team rather than one researcher. And the classic failure is the pseudo-merge: reporting the two sets of results in adjacent chapters without ever confronting them.
Explanatory sequential design
In an explanatory sequential design, the quantitative strand comes first and leads; a smaller qualitative phase follows to explain its results. The second phase is deliberately targeted at the first phase’s puzzles — unexpected results, extreme groups, significant differences that beg for a mechanism. The typical logic: the survey establishes what and for whom; the interviews establish why. This is the natural choice when good numbers already exist — a dashboard, an annual survey — that raise questions they cannot themselves answer.
| Phase | What happens | What it produces |
|---|---|---|
| 1. QUAN | CAWI survey, statistical analysis | The pattern — and the puzzles |
| 2. Connect | Use the results to choose whom to interview and what to ask | Purposive sample and interview guide |
| 3. qual | Interviews, thematic analysis | Explanations of the pattern |
| 4. Interpret | Qualitative findings read against the quantitative results | An explained result |
Organisational example. A hotel chain’s annual engagement survey finds night-shift staff scoring far below every other group. Follow-up interviews with night workers — selected because of their scores — trace the gap to unpredictable rota changes, something no survey item had asked about.
Pitfalls. The phases cannot overlap by definition, so the study takes roughly twice as long, and boards and clients must be warned at the start. Phase 2 cannot be fully specified in advance, because whom you interview depends on results you do not yet have — which complicates planning and ethical review. The classic failure is a weak connection: interviews that could have been conducted without the survey, ignoring its puzzles and merely gathering general impressions. There is also a statistical risk — an “unexpected” subgroup difference may be sampling error, and explaining a result qualitatively presupposes that the result is real. Finally, re-contacting respondents for phase 2 requires identifying them in phase 1, an anonymity trade-off that must be disclosed and consented to.
Exploratory sequential design
In an exploratory sequential design, the qualitative strand comes first and leads; a quantitative phase follows to measure, generalise, or test what it discovered. The classic use is instrument development: interviews establish what the phenomenon is and what words people use for it; the survey then measures it at scale, in the participants’ own vocabulary. It is the right design when the phenomenon is new, sensitive, or poorly understood — when you do not yet know enough to write good closed questions. The middle step is a genuine development phase — turning themes into items, piloting, refining — not a formality. The design builds session 1’s inductive–deductive cycle into a single study: inductive logic feeding deductive logic.
| Phase | What happens | What it produces |
|---|---|---|
| 1. QUAL | Interviews or focus groups, thematic analysis | The constructs — and the vocabulary |
| 2. Build | Turn themes into survey items; pilot and refine the instrument | A grounded questionnaire |
| 3. quan | CAWI or CATI survey of a large sample | How widespread each theme is; tests of relationships |
| 4. Interpret | Do the discovered constructs hold, and generalise, at scale? | A tested, generalisable account |
Organisational example. A bank migrating its SME clients to a new platform wants to measure “digital trust” — but nobody knows what that means to clients. Twenty interviews surface four distinct concerns, which become the scales of a CATI survey fielded to the full client base.
Pitfalls. This is the longest of the three designs, with three genuine stages (explore, build, measure) — plan for it, and say so in any proposal. Skipping the middle stage — bolting survey items together straight from interview quotes, without piloting — produces an instrument with unknown validity and reliability. If the qualitative sample misses a key group, the survey will scale up that blindness: it measures precisely, but measures the wrong things. Deciding which qualitative findings become survey items is a consequential judgement that should be made explicitly, not silently. And the quantitative phase must test what the qualitative phase found, rather than wandering off to new questions.
Embedded (nested) designs
In an embedded design, one strand is nested inside a design led by the other, playing a clearly supporting role — for example QUAN(qual). The supporting strand answers a secondary question that the primary design cannot reach on its own. Two organisational forms are common: a qualitative strand inside an experiment or pilot (Meridian pilots a new bonus scheme in six stores, with embedded interviews asking how staff experienced it and why it worked or failed), and a quantitative strand inside a case study (a deep single-site study that embeds a short staff survey to check how widely the observed views are held). Priority is unmistakably unequal, and integration happens at the level of the overall design — the nested strand serves the host. The three core designs cover most needs; the embedded design is noted here for completeness.
Justifying the choice
This course requires not just the principles of combining qualitative and quantitative methods but the justification of the choice — and so will every board, client, reviewer, and examiner you ever face. A research design is an argument, not a preference: “we like interviews” and “mixed methods is more thorough” justify nothing. Mixing is expensive, so it must earn its cost against the alternative of one method done well. Any serious justification answers five questions.
| Question | What it asks | A good answer sounds like |
|---|---|---|
| Fit | Does the question need both strands? | “The question has a how much and a why component; here is each” |
| Sequence | Why this order — or why no order at all? | “Phase 2 could not be designed until phase 1 reported, because…” |
| Priority | Which strand carries the main answer? | “The QUAN strand leads: the board’s decision turns on the number” |
| Integration | Where do the strands actually meet? | “Phase 1 results select the phase-2 sample; a joint display merges the findings” |
| Feasibility | Can we actually do it? | “Two phases fit the six-month budget because the survey runs as CAWI” |
Fit to question
The starting point is always the question, never the method: does the question genuinely have both a quantitative and a qualitative component? Decompose it and check. How much has turnover risen, and where is it concentrated? is quantitative. Why are those employees leaving? is qualitative. Is the problem widespread or local, and what would make people stay? requires both. If honest decomposition yields only one kind of component, the question does not need mixing, however fashionable mixing may be. A fit argument names the specific components and shows that neither strand alone can answer the whole question.
Sequence logic
Something must justify the arrow: why this order — or why no order at all? Each design carries its own sequence logic. In an explanatory design, the qualitative phase cannot be designed until the quantitative results exist, because you do not yet know whom to interview or about what. In an exploratory design, the quantitative instrument cannot be written until the qualitative phase supplies constructs and vocabulary. In a convergent design, neither strand depends on the other’s results, and running them together saves time. The test is simple: could the second phase have been specified, in full, before the first phase reported? If yes, the sequence has no logic — only a schedule. “We did the interviews first because they were easier to arrange” is logistics, not sequence logic.
Priority
In most mixed designs one strand carries the main answer and the other supports it — the capitalisation in the notation is a real decision, not typography. Priority should follow from the question (is its core how many or why?) and from the audience (a board deciding on budget usually needs the generalisable number, with the qualitative strand explaining and humanising it). Priority drives resources: the leading strand gets the larger share of time, money, and sample. Declaring priority in advance also protects the study later, because it settles which strand rules when the two point in different directions and no further data can be gathered.
The integration point
Integration is the moment the strands actually meet. Creswell and Creswell distinguish three forms:
| Form | Mechanism |
|---|---|
| Merging | Results are brought together and compared, often in a joint display — one table with the statistics and the corresponding themes side by side |
| Connecting | One phase’s results build the other’s design — its sample (explanatory) or its instrument (exploratory) |
| Embedding | One strand is built into the other at the level of the design itself |
The integration point should be named in advance, in the proposal: at which step, using what procedure, producing what output. The brutal test: if you cannot point to the place where one strand changed what the other did or meant, you do not have a mixed design — you have two studies.
Practical constraints
Feasibility is a legitimate part of the justification, not an embarrassing afterthought. Time: sequential designs roughly double the calendar, and the exploratory design adds an instrument-building stage on top. Cost: two strands mean two rounds of collection and analysis; CAWI is cheap, while CATI and face-to-face interviewing are not. Skills: the team must be competent in both traditions, and a lone researcher strong in only one is a real constraint. Access: can you re-contact survey respondents, enter the stores, and get release time for interviews? Constraints may legitimately push a study from the ideal design to a simpler one, and the justification should say so openly. What they never justify is faking the integration while keeping the mixed-methods label.
When not to mix
The stapler problem
The most common failure in mixed methods is two disconnected studies stapled together, with integration consisting of one hopeful paragraph in the conclusion. The symptoms are recognisable: the strands answer different questions rather than components of one question; they are run by separate people who never confront each other’s findings; the report presents them in separate chapters, with an “integration” section that merely summarises each in turn; and divergent findings are quietly ignored rather than investigated. The diagnostic question is: would either half change in any way if the other were deleted? If not, the staple is all that connects them. A stapled study costs as much as a mixed one and delivers the insight of neither done well.
The costs of mixing
Mixing is not free, and an honest justification prices it in. It brings complexity — two designs, two samples, two analyses, plus the integration work that neither strand needs alone. It is resource-intensive, demanding more time and money than a single-method study of the same question. It requires expertise in both traditions, which usually means a team rather than an individual. And interpretation is genuinely harder: synthesising two kinds of evidence is more demanding than reading one, especially when they disagree. The uncomfortable implication is that a weak mixed study is worse than a strong single-method study — two half-done strands do not add up to one whole answer.
When a single method is the right answer
There are recognisable situations in which not mixing is the better choice:
- The question is single-strand — has the new rota reduced overtime? needs payroll data, not focus groups.
- Resources stretch to one strand done well — a rigorous survey beats a rushed survey plus three token interviews.
- The timeline cannot fit the sequence — if the board decides in six weeks, an explanatory sequential design is a promise that cannot be kept.
- Routine monitoring — a quarterly KPI dashboard needs consistency and comparability, not a fresh qualitative strand each quarter.
- Fashion — “mixed methods looks rigorous” is a badge, not a reason, and examiners and reviewers see through it immediately.
Choosing not to mix, for stated reasons, is itself a defensible methodological justification.
Conclusion
Mixed methods research is defined by integration: both kinds of data, deliberately combined, interpreted together — not two methods in one binder. The three core designs are three answers to one pair of questions — when does each strand run, and what does each do for the other: the convergent design (QUAN + QUAL), the explanatory sequential design (QUAN → qual), and the exploratory sequential design (QUAL → quan), with embedded designs nesting one strand inside the other. The choice among them is an argument with five parts — fit to question, sequence logic, priority, integration point, and feasibility — and the failure mode to fear is the stapler: two disconnected studies whose only integration is the binding. When mixing cannot be done properly, the honest alternative is one method done well.