Deliberately combines quantitative and qualitative strands in one study — so that numbers gain explanation and narratives gain scale.
Research designMixed / integrative
What it is
Mixed methods research combines quantitative and qualitative strands within a single study or programme — not as decoration, but with an explicit integration logic: one strand builds the other, explains the other, or corroborates the other. The design answers questions neither strand can answer alone: not just whether flexible work predicts retention, but what “flexibility” means to employees and why the effect appears in some units and not others.
Three core designs cover most needs. Explanatory sequential (QUAN → qual): collect and analyse quantitative data first, then use qualitative follow-up to explain the results — especially surprising, extreme or contradictory ones. Exploratory sequential (QUAL → quan): explore qualitatively first, then build an instrument or model and test it quantitatively at scale. Convergent (QUAN + QUAL in parallel): collect both concurrently, analyse separately, then merge and compare — asking where the strands converge, diverge or complement each other. Notation conventions (capitals for the dominant strand, arrows for sequence) help you communicate the structure precisely.
The cost is real: mixed designs need competence in both paradigms, more time, and a genuine integration step — a joint display, a meta-inference, a discussion that weaves strands together. A survey and a few afterthought interviews reported in separate chapters is not mixed methods; it is two studies stapled together.
✓ When to use
Neither numbers nor narratives alone can answer the question — you need both prevalence and mechanism
Instrument development — qualitative grounding first, then quantitative validation (exploratory sequential)
Quantitative results need explaining — why did the effect appear, vanish or reverse in subgroups? (explanatory sequential)
Triangulation matters — findings will face skeptical audiences and convergent evidence strengthens the claim
Evaluations of interventions where outcomes (quant) and implementation experience (qual) both matter
✗ When NOT to use
One strand would fully answer the question — added methods mean added cost without added insight
You lack time, resources or skills for two rigorous strands — one strong study beats two weak ones
No integration is actually planned — if strands will never speak to each other, run separate studies honestly
The word-count or timeline of the thesis/report cannot accommodate reporting both strands properly
The epistemological framing is muddled — decide how you will reconcile conflicting findings before they occur
Typical research questions
RQ — What proportion of employees underuse parental-leave policies (QUAN), and what workplace signals explain the hesitation (qual)?
RQ — What does “seamless omnichannel experience” mean to shoppers (QUAL), and do the derived dimensions predict loyalty at scale (quan)?
RQ — Did the sales-force automation increase productivity (QUAN), and how did reps adapt their workflow around it (qual)?
RQ — Why do exit-survey scores and exit-interview narratives disagree about reasons for leaving (convergent comparison)?
Key characteristics
Purpose
Answer questions requiring both breadth and depth: measure the what, explain the how and why, or build then test
Typical data
Both — survey/archival numbers and interview/observation/document text, linked by design
Researcher control
Varies by strand: experimental or survey control in the quantitative strand; interpretive flexibility in the qualitative strand
Temporal aspect
Sequential designs unfold in phases (months); convergent designs run strands in parallel
Typical sample
A larger quantitative sample plus a smaller, purposefully linked qualitative sample — often nested (interviewees drawn from survey respondents)
Variables & measurement
Each strand keeps its own logic: operationalized variables, validated scales and power planning on the quantitative side; sensitizing concepts, saturation and reflexivity on the qualitative side. The distinctive design work is the linkage — deciding in advance how constructs in one strand map to the other.
In exploratory sequential designs, qualitative themes become constructs and items: participants' language is distilled into a pool, refined by expert review and cognitive interviews, then validated (EFA → CFA → reliability). In explanatory sequential designs, quantitative results define the qualitative sampling frame (e.g. interview high, low and off-diagonal scorers) and the interview guide's probes.
Write the mixed-methods question explicitly, alongside the strand-specific questions
Prepare a mapping table: construct → quantitative measure → qualitative data source
Decide priority (which strand dominates) and the point of integration before fieldwork
Plan the joint display (themes × statistics matrix) you intend to publish — it forces integration thinking early
Sampling approaches that fit
Quantitative strand: probability or criterion-based sampling sized for the planned analysis (power, precision or SEM benchmarks)
Qualitative strand: purposive sampling to saturation — often nested within the quantitative sample for direct linkage
Explanatory sequential: sample interviewees from the survey by results (extreme scorers, discrepant cases, typical cases)
Exploratory sequential: the qualitative informants and the later survey population should belong to the same universe
Report both sampling logics separately and the linkage between them explicitly
Data collection methods that fit
Surveys with validated (or newly developed) scales for the quantitative strand
In-depth interviews or focus groups for the qualitative strand, guided by the integration purpose
Archival and system data to anchor outcomes independently of self-report
Identifiers (with consent) that allow individual-level linking of survey and interview data
Sequence logistics: build in time between phases for analysis that shapes the next phase's instruments
Appropriate statistical & analytical methods
Each strand is analysed with its own rigor first. The quantitative strand draws on the full statistical toolkit — these Atlas methods are the frequent workhorses:
Each link below opens the matching method page (opens the Statistical Methods Atlas) with assumptions, worked examples and reporting guidance.
The qualitative strand uses thematic analysis, grounded-theory coding or content analysis as its tradition dictates (see the Qualitative Research page). The distinctive mixed-methods step is integration: joint displays that array themes against statistics, following a thread across strands, or data transformation (quantitizing themes into counts, qualitizing profiles into narratives). End with meta-inferences — conclusions that neither strand supports alone — and address divergence between strands head-on rather than burying it.
Quality criteria
Quantitative strand judged by its own criteria — validity, reliability, statistical conclusion validity
Qualitative strand judged by trustworthiness — credibility, transferability, dependability, confirmability
Legitimation (mixed-specific): does the design fit the mixed question? Are samples appropriately linked? Are meta-inferences genuinely grounded in both strands?
Integration quality — visible integration products (joint displays, threaded narratives), and explicit handling of convergence and divergence
Transparency — the design named and diagrammed (e.g. QUAL → QUAN), with timing, priority and integration points stated
A worked mini-example
A researcher studies why a company's engagement scores stagnate despite generous benefits. Phase 1 (QUAN): analysis of the annual survey (n = 2,450) shows engagement unrelated to benefits satisfaction but strongly associated with perceived voice (β = .41); the association is weakest in the operations division. Phase 2 (qual): 20 interviews sampled purposively from operations — high and low engagement scorers — explore how voice is experienced.
Interviews reveal that formal voice channels exist but supervisors' shift-scheduling discretion makes speaking up feel risky — a mechanism invisible to the survey. A joint display aligns each quantitative pattern with its qualitative explanation. The meta-inference: voice predicts engagement only where daily dependence on supervisors is low — a moderation hypothesis the company tests the following year with a revised survey. This is an explanatory sequential design: QUAN → qual, quantitative priority, integration at interpretation.
Common pitfalls
Two studies stapled together — no research question, sampling link or discussion that actually integrates the strands
Token qualitative work — three convenience interviews summarized in a paragraph to “add depth”
Ignoring divergence — when strands disagree, the discrepancy is data, not an embarrassment to hide
Underestimating workload and skills — both strands done at half rigor
Sequencing errors — building the survey before the qualitative phase it was supposed to be built from
Reporting that buries the design: readers cannot tell what was collected when, from whom, or how strands connect
Using mixed methods as a hedge because the researcher could not choose a question
What to report
The named design (convergent, explanatory sequential, exploratory sequential) with a phase diagram and notation (e.g. QUAN → qual)
The mixed-methods research question in addition to strand-specific questions
Each strand's methods reported to its own community's standard (sampling, instruments, analysis)
The linkage: how qualitative participants relate to the quantitative sample, and what carried across phases
An integration product — joint display or equivalent — and explicit meta-inferences
How convergent and divergent findings were handled
Limitations of each strand and of the integration itself