Mixed Methods Research Design: Types and How to Choose

Published June 20, 2026 | Last updated October 31, 2026

Quick Answer: Choose your mixed methods research design by answering three questions: when the two strands are collected (at the same time, in sequence, or across linked projects), whether one strand has priority, and where the strands meet.¹ Use a convergent design to merge two views of one question, an explanatory sequential design when quantitative results need explaining, an exploratory sequential design when you must explore before you can measure, an embedded design when a secondary dataset sits inside a larger design such as an intervention study, and a multiphase design for linked projects over time.¹^,² The worked case is one open-access paper (CC BY 4.0, BMC Public Health 2026): a survey of 1,500 adults followed by 11 interviews and 8 focus groups in Sierra Leone, redrawn and rechecked in R.³

Mixed Methods Research Design: The Five Designs at a Glance

A mixed methods study intentionally integrates or combines qualitative and quantitative methods to draw on the strengths of each.¹ The NIH report gives reasons to do so: one approach by itself is inadequate, you need to contextualise information, compare or triangulate results, examine processes along with outcomes, or have one database build on another.¹ The figure sets out five designs. The NIH report gives convergent, sequential (explanatory or exploratory), embedded and multiphase designs as examples and says the list is not meant to be exhaustive; we show sequential as two designs, which gives our count of five.¹ Fetters and colleagues name three basic designs and four advanced frameworks.²

Table of five mixed methods designs. Convergent: concurrent, strands meet by merging. Explanatory sequential: quantitative then qualitative, connecting. Exploratory sequential: qualitative then quantitative, building. Embedded: a secondary dataset inside a larger design, embedding. Multiphase: linked projects over time with convergent and sequential elements. — The five designs by timing, when to choose them and where the strands meet. Original Editverse schematic; definitions from the NIH OBSSR report and the four linking approaches from Fetters et al.; the match of each design to an approach is our reading.

Three Questions to Settle Before You Pick a Design

The NIH report lists five ways in which designs differ: analytic logic, timing, priority, point of interface, and single study versus multiphase program.¹ We picked the three below as a starting point; that selection is ours.

Swipe sideways to see every column.

Question	What the NIH report says	What to decide
Timing	Data may be timed so that they are collected concurrently or roughly at the same time, or collected in a sequence with one phase of collection followed by another.	Can both strands run now, or does the second depend on the first?
Priority	In some studies the quantitative and qualitative research is equally emphasised; in others priority goes to one of them.	Which strand carries the main answer, and which supports it?
Point of interface	The point where mixing occurs differs depending on the design; it can be at data collection, analysis or interpretation.	Where do the strands first touch, and what exactly is passed across?

Sampling by design

The NIH report says a detailed discussion of sampling is beyond its scope, but it names challenges specific to concurrent and to sequential designs, shown below.¹ In Fetters’ terms, a link made through sampling is called connecting.²

Swipe sideways to see every column.

Design family	Sampling and analysis challenges named by NIH
Concurrent (convergent)	Having adequate sample sizes for analyses, using comparable samples, and employing a consistent unit of analysis across the databases.
Sequential (explanatory or exploratory)	Deciding what results from the first phase to use in the follow-up phase, choosing samples and estimating reasonable sample sizes for both phases, and interpreting results from both phases.

Our suggestion: write the rule that links phase 2 to phase 1 before you collect any data, for example “interview the 20 respondents whose scores were furthest from the model”, and report it. Sample-size tools help with the quantitative phase, and the free tools below cover that.

The Five Designs in Detail

Convergent design

The intent is to merge concurrent quantitative and qualitative data to address the study aims, and the analysis consists of merging the data and comparing the two sets of results.¹ In the integration terms of Fetters and colleagues, merging means the two databases are brought together for analysis.² The NIH report adds that concurrent collection is attractive to an investigator who must maximise the data collected in the field for the time spent.¹

Explanatory sequential design

Quantitative data come first, and qualitative data then help to explain in more depth the mechanisms underlying the quantitative results.¹ The NIH example is a quality-of-life scale whose results are explained by qualitative follow-up data to understand the responses on the scale.¹ The case below is of this type.

Exploratory sequential design

You begin with a qualitative exploration followed by a quantitative follow-up.¹ In Fetters’ terms the first database informs the data collection approach of the second, which they call building.² A typical use is shaping a questionnaire; that example is ours.

Embedded design

A dataset of secondary priority is embedded within a larger, primary design.¹ In intervention studies, qualitative data may be used before the intervention to inform recruitment strategies, or within the intervention procedures to understand how participants experience the treatment.¹ The report adds that embedded designs may be a variation of a convergent or sequential design.¹

Multiphase design

A multiphase design emerges from multiple projects conducted over time and linked by a common purpose. Such designs commonly involve convergent and sequential elements, with links in place so that one phase builds on another.¹

How to Choose: A Decision Table

The first column is your situation, the second is the design we would consider, and the third is what to write down before collecting data. This table is our suggestion, built on the NIH and Fetters definitions above.¹^,²

Swipe sideways to see every column.

Your situation	Design to consider	Write down before you collect
One approach by itself answers the question	A single-method study	Why mixing would add nothing; the NIH reasons for mixing apply when one approach is inadequate.
You can run both strands now and want to compare them	Convergent	The unit of analysis, how samples compare, and how results will be merged.
A survey or trial result needs explaining	Explanatory sequential	Which results trigger follow-up and how phase 2 participants are chosen.
You do not yet know what to measure	Exploratory sequential	How phase 1 findings will shape the phase 2 instrument or sample.
An intervention or program study where you also want participant experience	Embedded	Which dataset has secondary priority and when it is collected.
A program of linked studies over several years	Multiphase	How each phase builds on the one before.

Planning Integration at the Design Level

Fetters, Curry and Creswell describe integration at three levels: study design, methods, and interpretation and reporting.² At the design level it happens through the three basic designs and four advanced frameworks: multistage, intervention, case study and participatory.² At the methods level they list four approaches: connecting, building, merging and embedding.² The BMJ article by O’Cathain, Murphy and Nicholl names three further techniques for integrating data: a triangulation protocol, following a thread, and a mixed methods matrix.⁴

This page is about choosing and planning. For the reporting level, including reporting standards and a joint display redrawn from another open-access study, see How to Write a Mixed Methods Research Paper: Report Integration.

A Real Explanatory Sequential Case, Redrawn in R

Moseray, Gogra and Ansumana studied the socio-economic and psychological burden of onchocerciasis (river blindness) in Bo District, Sierra Leone, with an explanatory sequential design written QUAN → qual.³ The paper is CC BY 4.0, which allowed us to redraw its numbers. Everything below is from the printed article; derived numbers were computed by Editverse in R.

How the study was built

Swipe sideways to see every column.

Element	As reported in the article
Phase 1: survey	1,500 adults aged 18 or older in 26 villages across five chiefdoms. Multistage cluster sampling: chiefdoms chosen purposively, villages stratified, households sampled systematically, one adult per household chosen with a Kish grid. The minimum sample was 640 and the target was raised to 1,500 to allow chiefdom-level comparisons.
Phase 1: measures	Closed binary items summed into a socio-economic index (range 0 to 2) and a psychological index (range 0 to 3), translated into Mende, piloted and collected on tablets.
Phase 2: qualitative	11 key informant interviews (drug distributors, health officers, chiefs, a traditional healer, the district focal person) and 8 focus groups of 5 community members (n = 40), analysed thematically to explain the survey findings.
Link between phases	The paper’s Figure 1 caption says the survey findings were used to develop the interview and focus group guide. We did not find a statement that interview participants were chosen from survey respondents or by survey result.

Phase 1: how common was each burden?

Of the 1,500 adults sampled, 71.7% reported economic hardship and 95.8% scored at least one psychological impact.³ We recomputed each percentage and its 95% interval from the printed counts.

Dot and whisker chart of the share of 1,500 adults sampled in endemic villages reporting each burden, from 59.1% for self-esteem shifts and 71.7% for economic hardship to 89.9% and 95.8% for any impact on the socio-economic and psychological indices, with 95% Wilson intervals — Phase 1 prevalence, redrawn from Table 2 of Moseray et al. (CC BY 4.0). Percentages and 95% Wilson intervals computed in R from the printed counts; they match the printed intervals.

Did any group differ?

The socio-economic index did not differ significantly across chiefdoms, with F(4,1495) = 1.91 and p = 0.106.³ Table 3 also reports tests by sex, age group, exposure and education, and none of the ten tests in the figure reached p = 0.05. A pattern like this leaves a clear question for a qualitative follow-up: why is the burden so even? That reading is ours; the authors state that the interviews and focus groups were analysed to explain the quantitative findings.³

Lollipop chart of ten recomputed p-values for tests in the case study, from 0.106 to 0.795 for index comparisons and 0.162 to 0.498 for single indicators, all above the 0.05 line — Ten p-values recomputed in R from the test statistics printed in Table 3 of Moseray et al. (CC BY 4.0). This checks the printed values; it is not a new analysis of the raw data.

Phase 2: what the qualitative strand added

Three themes were reported: a productivity and poverty cycle, stigma experienced as “social death”, and an unrecognised gendered caregiving burden.³ Stigma was the most prevalent psychological item in the survey (69.6%). The authors say these accounts explain the prominence of stigma and why the burden is felt community-wide.³

Computed by Editverse in R

All eight printed 95% intervals match Wilson intervals recomputed from the counts. A simple normal-approximation interval differs by 0.1 in several rows (for economic hardship it gives 69.5 to 74.0 against the printed 69.4 to 74.0).
All ten printed p-values were reproduced from the printed test statistics to within 0.004. The largest gap is 0.003, for the Spearman correlation (printed 0.311, recomputed 0.314), which is expected when the statistic is rounded.
Cramér’s V recomputed as the square root of chi-square divided by 1,500 gives 0.047, 0.050 and 0.059, matching the printed values for these 2-by-k tables.
Near-ceiling check: 63 of 1,500 respondents (4.2%) scored 0 on the psychological index and 151 (10.1%) on the socio-economic index. The authors note that near-ceiling prevalence limits the variance available for subgroup contrasts.
The group of respondents who were diagnosed or knew someone with the disease (46.2% of the sample) is about 693 people.

What this case shows: how a quantitative phase with null group comparisons can hand a clear question to a qualitative phase, and how a design is described by sample, measures and sequence. What it does not show: a causal effect, or that the qualitative participants were selected from the survey. The survey sampled all adults in the endemic villages and, as the authors note, did not separate respondents with personal experience of the disease from those reporting through relatives, neighbours or community knowledge, so the percentages describe adults sampled, not people with onchocerciasis. The study is cross-sectional with self-reported, closed single-response items, and we did not have the raw data.³

Need the Analysis Run or Checked? Consult Editverse

Editverse subject-matter experts work to COPE, ICMJE and GPP 2022 guidance, with institutional and country-specific ethics followed.

Editverse

Methodology Consultation

Not sure which design fits your question?

You send: your research question, setting and any data you already have.

You get: three 45-minute sessions with a PhD methodologist covering research design review, sampling strategy and a written methodology framework document.

When: You need a design you can defend to a committee or reviewer.

Methodology Consultation Package · $400

Book my consultation →

Editverse

Statistical Analysis

Need the quantitative strand run and reported?

You send: your dataset and your research questions.

You get: the analysis run in SPSS, R or Stata, with annotated code, complete output files, an interpretation guide and publication-ready tables.

When: You need numbers you can place beside your themes.

SPSS/R/STATA Statistical Analysis · $400

Analyse my data →

Editverse

Manuscript Writing

Both strands finished and the paper still unwritten?

You send: your data from both strands, your design notes and your target journal.

You get: a manuscript written and edited by PhD experts, with statistical review, journal selection and submission support.

When: Your project is finished and the paper is not.

Manuscript Writing Services · $1,500

Write my paper →

Free Research Tools from Editverse

Free Sample Size and Power Analysis Calculator: size the quantitative strand before you plan the follow-up.
Which Statistical Test Should I Use?: the test, its assumptions and what to report for the survey strand.
Free Research Tools for Medical Researchers: the full set of browser-based tools.

Frequently Asked Questions

What are the main types of mixed methods research design?

The NIH best-practices report gives convergent, sequential (explanatory or exploratory), embedded and multiphase designs as examples and says the list is not meant to be exhaustive. It adds that an embedded design may be a variation of a convergent or sequential design. Fetters and colleagues name three basic designs, convergent, explanatory sequential and exploratory sequential, and four advanced frameworks: multistage, intervention, case study and participatory. The five designs on this page are our count, with sequential shown as two.

How do I choose between an explanatory and an exploratory sequential design?

Start from what you already know. Choose explanatory sequential when quantitative results come first and qualitative data are needed to explain in more depth the mechanisms behind them. Choose exploratory sequential when you begin with a qualitative exploration and follow it with a quantitative phase.

When should I use a convergent design?

Use it when you want to merge concurrent quantitative and qualitative data to address the study aims. The NIH report lists three challenges for concurrent designs: adequate sample sizes for analyses, comparable samples, and a consistent unit of analysis across the databases.

Which strand should have priority in a mixed methods study?

Either choice is allowed. The NIH report says some studies emphasise the quantitative and qualitative strands equally and others give priority to one of them. In an embedded design a dataset of secondary priority sits inside a larger primary design.

How should I sample in a sequential design?

The NIH report names the sampling challenges: deciding what results from the first phase to use in the follow-up, and choosing samples and estimating reasonable sample sizes for both phases. Fetters and colleagues call a link made through sampling connecting. Our suggestion is to write the rule that links the phase 2 sample to phase 1 before you collect data.

What does integration at the design level mean?

At the design level, Fetters and colleagues say integration happens through the three basic designs and four advanced frameworks. At the methods level it happens through four approaches: connecting, building, merging and embedding. Reporting is the third level and is covered in our separate guide on writing a mixed methods paper.

What did the Sierra Leone explanatory sequential case show?

In a survey of 1,500 adults sampled in endemic villages of Bo District, 71.7% reported economic hardship and 95.8% reported at least one psychological impact. The socio-economic index did not differ significantly across chiefdoms, with F(4,1495) = 1.91 and p = 0.106. Interviews and focus groups then described a productivity and poverty cycle, stigma experienced as social death and an unrecognised gendered caregiving load.

What is a multiphase mixed methods design?

The NIH report says a multiphase design emerges from multiple projects conducted over time and linked together by a common purpose. Such designs commonly involve convergent and sequential elements, with links in place so that one phase builds on another. Its example is developing, testing, implementing and evaluating a health prevention program for adolescents.

References

Creswell JW, Klassen AC, Plano Clark VL, Smith KC, for the Office of Behavioral and Social Sciences Research. Best practices for mixed methods research in the health sciences. National Institutes of Health; August 2011. Full report PDF; section The nature and design of mixed methods research, printed pages 4 to 9. obssr.od.nih.gov/sites/g/files/mnhszr296/files/Best_Practices_for_Mixed_Methods_Research.pdf
Fetters MD, Curry LA, Creswell JW. Achieving integration in mixed methods designs: principles and practices. Health Serv Res. 2013;48(6 Pt 2):2134-2156. doi:10.1111/1475-6773.12117. Abstract read on PubMed; the full text was not available to us. pubmed.ncbi.nlm.nih.gov/24279835/
Moseray AB, Gogra AB, Ansumana R. The pervasive socio-economic and psychological burden of onchocerciasis in endemic communities of Bo District, Sierra Leone: an explanatory sequential mixed-methods study. BMC Public Health. 2026;26(1). doi:10.1186/s12889-026-28720-3. Open access, CC BY 4.0; figures redrawn and derived numbers computed by Editverse in R. pubmed.ncbi.nlm.nih.gov/42907833/
O’Cathain A, Murphy E, Nicholl J. Three techniques for integrating data in mixed methods studies. BMJ. 2010;341:c4587. doi:10.1136/bmj.c4587. We read the article summary page, not the full text. pubmed.ncbi.nlm.nih.gov/20851841/