Short answer: A mixed methods research design combines qualitative and quantitative data in one study and plans how the two will be integrated. The three core designs are convergent (collect both at once, then merge), explanatory sequential (survey first, interviews to explain the results) and exploratory sequential (interviews first, then build and test an instrument). Choose by which strand answers which question, and when.
Last updated: November 9, 2026
Choosing a mixed methods research design is less about picking a label and more about deciding how two kinds of evidence will talk to each other. A survey can tell you how many people hold a view. Interviews can tell you why they hold it and what it means to them. A mixed methods study is only worth the extra work if those strands are deliberately connected, and many published studies never connect them.
This guide is written for PhD students, research labs and qualitative or market research teams planning a mixed methods study in 2026. It covers the three core designs, the advanced frameworks, the integration techniques reviewers look for, reporting standards, and what changes when AI-moderated interviews and conversational surveys make the qualitative strand cheaper to run at scale. It also includes an original Mixed Methods Integration Planner you can copy into a proposal.
Key takeaways
- Creswell and Plano Clark's widely used typology reduces mixed methods to three core designs: convergent, explanatory sequential and exploratory sequential. More complex studies build on these.
- Integration is the defining feature. Fetters, Curry and Creswell (2013) describe integration at the design, methods (connecting, building, merging, embedding) and interpretation levels.
- Joint displays, tables that put qualitative and quantitative findings side by side, are the most practical tool for showing integration to reviewers.
- Reporting checklists exist: APA's JARS-Mixed (2018), GRAMMS (2008) for health services research and the MMAT (2018) for appraisal.
- LLMs are now being tested inside mixed methods workflows. An AERA 2026 paper found LLM-simulated survey answers matched average human patterns but with much less variability.
What is a mixed methods research design?
A mixed methods research design is a plan for collecting, analyzing and integrating both qualitative and quantitative data within a single study or program of studies to answer research questions that neither approach answers well alone. The key word is integrating. Running a survey and a few interviews and reporting them in separate chapters is a multimethod study, not necessarily a mixed methods one.
Three decisions define any mixed methods design:
- Timing. Are the strands collected at the same time (concurrent) or one after the other (sequential)?
- Priority. Does one strand carry more weight? Morse's notation writes the dominant strand in capitals, for example QUAN → qual.
- Point of integration. Where do the strands meet: in sampling, in instrument design, in analysis or in interpretation?
The integration problem is old. In a frequently cited 2006 analysis of published mixed methods articles, Alan Bryman found that many studies did not genuinely integrate their findings (Bryman, 2006, Qualitative Research).
The three core mixed methods research designs
1. Convergent design
You collect qualitative and quantitative data in the same phase, analyze each separately, then merge the results to compare them. Use it when you want to check whether different kinds of evidence agree, for example survey satisfaction scores and interview accounts from the same program.
Watch out for: different sample sizes and different questions in each strand, which make merging hard. Plan parallel topics in both instruments from the start.
2. Explanatory sequential design
You start with a quantitative phase, then use qualitative data to explain the results. The survey identifies a surprising pattern, outliers or subgroups, and follow-up interviews ask why. This is the most common design in management, education and health services research.
Watch out for: the time gap between phases, and choosing interviewees without a rule. Decide in advance which quantitative results will trigger which interview sample.
3. Exploratory sequential design
You start with qualitative data to explore a poorly understood topic, then build something from it, usually a survey instrument, a typology or an intervention, and test it quantitatively. Use it when no validated measure exists or when you suspect existing measures miss how participants frame the issue.
Watch out for: the building step. Document how each qualitative theme became an item or a construct, or reviewers cannot assess the instrument's content validity.
Advanced frameworks
Fetters, Curry and Creswell list four frameworks that layer on top of the core designs: multistage (several phases over a long project), intervention (qualitative data embedded before, during or after a trial), case study (both strands inside one or more bounded cases) and participatory (communities involved in shaping the questions and use of findings) (Fetters, Curry and Creswell, 2013, Health Services Research).
Comparison: which design fits your question?
| Design | Sequence | Typical question | Integration point | Main risk |
|---|---|---|---|---|
| Convergent | QUAN + QUAL at the same time | Do the numbers and the accounts agree? | Merging in analysis and joint display | Strands ask different things and cannot be compared |
| Explanatory sequential | QUAN → qual | Why did this result occur? | Connecting: results select interviewees and topics | Interviews chosen ad hoc, not from the results |
| Exploratory sequential | QUAL → quan | What should we measure, and does it generalize? | Building: themes become items or constructs | Weak audit trail from theme to item |
| Embedded / intervention | qual inside a QUAN trial (or the reverse) | How and why did the intervention work? | Embedding at several points | Qualitative strand treated as an afterthought |
How to integrate qualitative and quantitative data
Reviewers increasingly ask one question of mixed methods manuscripts: where is the integration? Fetters and colleagues describe four ways to integrate at the methods level:
- Connecting: one strand selects the sample for the other, for example interviewing survey respondents with extreme scores.
- Building: one strand informs the data collection of the other, for example interview themes become survey items.
- Merging: both datasets are brought together for analysis and comparison.
- Embedding: data collection and analysis are linked at multiple points, common in intervention studies.
At the interpretation level, integration happens through narrative (weaving themes and statistics together, or presenting them in separate but linked sections), data transformation (for example turning coded themes into counts) and joint displays. When you compare strands, label the "fit" honestly: confirmation, expansion or discordance. Discordance is a finding, not a failure, and it often produces the most interesting discussion section.
A minimal joint display
A joint display can be as simple as one row per construct with four columns: the quantitative result, the qualitative theme with an anchoring quote, the fit (confirm, expand, discord) and the meta-inference you draw. If you cannot fill the last column, the strands are not yet integrated.
The Mixed Methods Integration Planner
This is a Qualitati-original planning checklist. Complete it before data collection and attach it to your proposal or ethics application.
- Rationale. Write one sentence on why neither strand alone answers the question (complementarity, explanation, development, triangulation or expansion).
- Questions by strand. List the quantitative question, the qualitative question and the mixed methods question that requires both.
- Design and notation. Name the core design and write the notation, for example QUAN → qual.
- Sampling link. State whether samples are identical, nested, parallel or multilevel, and the rule that selects participants for the second strand.
- Integration point. Choose connecting, building, merging or embedding, and say when it happens.
- Joint display draft. Sketch the rows and columns before you have data.
- Fit rule. Decide how you will handle discordance: a further analysis, additional interviews or reporting it as a finding.
- Quality criteria. Name the appraisal or reporting standard you will follow (JARS-Mixed, GRAMMS or MMAT).
- AI use. Record which steps use AI tools, which model, and how a human checked the output.
Reporting and appraisal standards
Three tools cover most needs:
- JARS-Mixed. APA's first reporting standards for mixed methods, published in American Psychologist in 2018 by Levitt and colleagues (APA Style, JARS-Mixed).
- GRAMMS. Six reporting items from O'Cathain, Murphy and Nicholl (2008) in the Journal of Health Services Research & Policy, including justification, design, integration and what was gained from mixing.
- MMAT, version 2018. An appraisal tool for systematic reviewers covering qualitative, quantitative and mixed methods studies, with five core criteria per study category (Hong et al., 2018, Education for Information).
What AI changes in mixed methods research in 2026
The biggest practical change is cost. AI-moderated interviews and conversational surveys let a team collect open-ended, probed answers from hundreds of participants, so the qualitative strand no longer has to be limited to 12 to 20 interviews. That makes nested samples and convergent designs with matching samples easier to run.
Researchers are also testing LLMs inside the integration step itself:
- At AERA 2026, Zhang and colleagues used interviews and BREQ survey data from 19 after-school program staff to see whether GPT-4.1, Gemini 2.0 Flash and Claude 3.7 Sonnet could simulate the same people's survey answers. The models reproduced average patterns but showed significantly less variability. Prompts with personal interview data reached Pearson correlations of 0.5 to 0.73 with human answers (AERA 2026 paper summary).
- An April 2026 preprint by Gonzalez and colleagues used LLM-assisted sentiment analysis on 151 student reflections, tested differences across seven identity variables, then used the statistical results to guide a qualitative analysis of why they appeared, an explanatory logic inside a single dataset (arXiv 2605.27403).
Our reading: AI is most defensible in the connecting and building steps, where it speeds up data collection and first-pass coding that a researcher then checks. It is least defensible as a replacement for one strand. Simulated survey answers with compressed variance are not a quantitative strand. See our guide to interview-informed LLM survey responses for more detail.
Where Qualitati fits
Qualitati is a European, budget-friendly qualitative research platform for universities and research companies, with GDPR and data privacy as central priorities. In a mixed methods study it supports the qualitative strand and the link between strands:
- Conversational surveys combine closed items with AI-driven follow-up questions and branching, so one instrument can collect both scaled answers and probed explanations from the same participants.
- AI-moderated interviews in text or voice run the follow-up phase of an explanatory sequential design at scale, in 10 languages.
- ThemeLens maps codes to research questions and synthesizes themes with participant-anchored quotes across up to 100 transcripts at once, which helps fill the qualitative column of a joint display.
- QDA Workspace supports inductive and deductive coding and codebook generation, useful when themes must become survey items in an exploratory sequential design.
Qualitati does not replace your statistics software. Run the quantitative analysis in R, Stata, SPSS or Python, and keep the integration decisions in your own memos.
Limitations and methodological concerns
- Paradigm tension. Mixing methods does not resolve philosophical differences. State your stance, often pragmatism, and do not claim interpretive depth a large-N open-ended dataset may lack.
- Scale is not saturation. Hundreds of short AI-collected answers are not the same as a few long interviews. Decide which the question needs. Our guide to data saturation covers this.
- AI coding needs checking. Have a human review a sample of AI codes and report how, following the LLM reporting guide.
- Workload. Mixed methods studies take longer than single-method ones. A PhD timeline should budget for two analyses plus integration.
Human-review note: methodological recommendations here are general guidance. Confirm design and reporting choices with your supervisor, ethics board or target journal.
Who this is for, and when not to use it
Use a mixed methods design when your question has a "how much" and a "why" part that must be answered together. Do not use it when one approach answers the question well, when you lack the time or skills for two rigorous strands, or when the qualitative strand would only supply decorative quotes for a survey.
FAQ
What are the three main mixed methods research designs?
The three core designs are convergent (qualitative and quantitative data collected at the same time and merged), explanatory sequential (quantitative first, then qualitative to explain results) and exploratory sequential (qualitative first, then a quantitative phase built from it).
What is the difference between mixed methods and multimethod research?
Multimethod research uses more than one method, possibly all qualitative or all quantitative, without necessarily integrating them. Mixed methods research combines qualitative and quantitative data and plans how they will be integrated.
What is a joint display in mixed methods research?
A joint display is a table or figure that puts qualitative and quantitative findings side by side for each construct or case, shows whether they confirm, expand or contradict each other, and states the combined inference.
How many interviews do I need in an explanatory sequential design?
There is no fixed number. Select interviewees with a rule linked to the quantitative results, such as extreme cases or subgroups, and continue until the explanation for each pattern is adequately supported.
Can AI be used in mixed methods research?
Yes, for data collection, transcription, first-pass coding and connecting strands, with human checking. A 2026 AERA paper found LLM-simulated survey answers had much less variability than human answers, so AI should not replace a strand.
Bottom line
A good mixed methods research design is defined by its integration plan, not by having two kinds of data. Pick the core design from your question, decide where the strands connect, draft the joint display before fieldwork and report against JARS-Mixed or GRAMMS. If you want to run the qualitative strand with AI-moderated interviews or conversational surveys, start free with 30 credits or view transparent pricing.
Last updated: November 9, 2026
Study findings are summarized from the public abstracts and pages linked above as of November 9, 2026. Qualitati is not affiliated with the authors.