Choose a convergent, explanatory or exploratory design, then plan how the two strands will connect—from sampling to the final interpretation.
A mixed methods research design plans how qualitative and quantitative evidence will be collected, analyzed and integrated to answer a research question. The design establishes timing, the role of each strand, the relationship between samples and the point at which findings are brought together.
A numerical result can describe a pattern; an interview can explore an experience behind it. A mixed methods study explains how those contributions combine. Two separate reports, one with numbers and one with quotations, do not by themselves show integration.
This is different from a mixed factorial design in experimental statistics, which combines within-subject and between-subject factors. It is also different from mixed-mode collection, such as offering the same survey online and by telephone. Those choices may be part of a study, but they do not alone make it mixed methods.
| Design | Sequence | Reason to choose it |
|---|---|---|
| Convergent | Qualitative and quantitative work in a similar period; compare interpretations | You need complementary accounts of the same issue. |
| Explanatory sequential | Quantitative findings first; qualitative follow-up | You need to understand a pattern or unexpected result. |
| Exploratory sequential | Qualitative inquiry first; quantitative development or testing | Relevant concepts or response options are not sufficiently understood. |
| Embedded or multiphase | A strand nested within a wider design, or linked phases over time | The overall study requires a supporting strand or several connected decisions. |
There is no universally best sequence. The NIH mixed methods research resources provide a foundation for planning quality across both strands. State what integration contributes to your question before choosing a design name.
The core designs and integration approaches are described in Fetters, Curry and Creswell (2013). The examples below illustrate choices; they are not findings from that paper.
In a convergent design, the strands examine a related issue in a similar timeframe. Teams commonly analyze each strand using suitable methods, then compare what the findings mean together. Similar timing does not require identical participants, identical sample sizes or equal weight.
Illustrative example: an organization wants to understand onboarding across locations. A survey describes reported ease and completion status, while interviews explore specific moments of confusion. The combined interpretation asks whether the interview accounts help explain the patterns across locations, and where the accounts differ.
Plan the comparison before collection. If the survey covers the previous month but interviews concern experiences from two years ago, disagreement may reflect the period rather than a contradiction. Decide which concepts are comparable and which evidence adds a different perspective.
The main advantage is that strands can progress together. The challenge is coordination: differences in definitions, access or analysis timing may make integration difficult. Budget time for a joint interpretation meeting, not only separate analysis tasks.
Explanatory sequential design begins with quantitative collection and analysis. Those findings guide qualitative follow-up. The follow-up should address a specific question raised by the first phase rather than merely add a few favorable quotes.
Illustrative example: a training evaluation finds similar test-score improvement across groups but different rates of applying the skill at work. Interviews then explore opportunity, manager support and the situations in which the skill could be used. The second phase helps interpret the difference; it does not automatically establish a causal explanation.
Select follow-up participants deliberately. You might include people with different outcome patterns, unusual cases or those whose experiences challenge the first interpretation. Explain this sampling logic and keep a record of who declined. People willing to be interviewed may not represent everyone in the first phase.
The design requires a real transition between phases. Leave time to analyze the first results, refine the interview guide and obtain permission to contact suitable participants. If follow-up is planned before any results are available, explain how the guide or sample will still respond to the findings.
Exploratory sequential design begins with qualitative inquiry. Findings inform a later quantitative phase—for example, developing survey items, defining categories or investigating how widely a newly identified experience occurs.
Illustrative example: interviews with partner organizations reveal that “late reporting” includes several different problems: unavailable financial records, unclear indicator definitions and difficulty obtaining internal approval. The team develops separate survey questions rather than using one broad question about lateness.
The transition needs more than copying interview phrases into a form. Review whether items reflect the concepts, whether response options cover plausible experiences and whether respondents interpret the wording consistently. Pilot the resulting questions. If creating a scale, appropriate measurement development and validation may be needed.
The quantitative phase can assess patterns in its own sample. It does not prove that the initial interviews captured every relevant experience. Keep room for missing categories and describe which findings informed the final questions.
An embedded approach places a supporting strand within a wider study. For example, interviews may examine how participants experienced an intervention within a larger quantitative evaluation. Explain the strand’s purpose, timing and contribution rather than calling any extra interview an embedded design.
A multiphase study links several phases around a longer program of inquiry. An organization might first explore a problem, develop measures, evaluate a response and investigate implementation across sites. Each transition needs an explicit connection. More phases do not automatically make the evidence stronger.
Complex designs require realistic staffing and decision ownership. If one team owns the numbers and another owns interviews, identify who will resolve interpretive disagreements and document the final reasoning. Keep the simpler core design when it answers the question adequately.
| Your uncertainty | Likely starting point | Check before committing |
|---|---|---|
| A numerical pattern needs explanation | Explanatory sequential | Can you reach suitable people after the first analysis? |
| You do not yet know the relevant concepts | Exploratory sequential | Is there time to develop and test the later measures? |
| You need parallel perspectives on one issue | Convergent | Are the concepts and periods sufficiently aligned? |
| A wider evaluation needs implementation context | Embedded | Will the supporting evidence influence interpretation? |
Specify the decision deadline, the access available and the team’s expertise in both strands. If a decision is needed before the second phase could finish, a sequential design may not fit. If only one source can answer the question, mixed methods may add burden without a useful contribution.
Also decide which strand carries which inference. A purposive interview sample may explain a range of experiences without estimating prevalence. A large survey may estimate a pattern without explaining the mechanism. The combined conclusion must respect both limits.
Integration can connect samples, use one strand to build the other, bring results together, or embed evidence within a wider design. Fetters, Curry and Creswell’s methods paper on integration describes these approaches across design, methods and interpretation. The practical planning question is what will be connected, when and for what inference.
Write an integration question, such as: “How do accounts of onboarding difficulty explain the locations with lower completion?” Then define the unit of comparison. It could be a person, organization, site, event or time period. Do not force person-level linkage when the research question is about sites or when respondents were promised anonymity.
Where person-level linkage is appropriate, establish it with suitable permission and a reliable identifier. Separate the key used for analysis from identifying details where possible. Record unmatched cases and ambiguous links. Similar names do not establish that two records describe the same person.
Preserve the question version, timing, source and relevant context for each strand. A technically successful database join can still connect measures that mean different things. Data management supports integration; it does not replace a reasoned interpretation.
A joint display places relevant findings beside one another and records the combined interpretation. It can be organized by theme, group, location or time. It should do more than decorate a report with a quotation next to a chart.
| Quantitative finding | Qualitative evidence | Combined interpretation and next check |
|---|---|---|
| Lower completion at one location | Accounts describe unclear handoff responsibilities | Investigate handoff records before attributing the difference to staff performance. |
| High satisfaction but low reported application | Participants describe few opportunities to use the skill | Separate training experience from opportunity to apply learning. |
| Little change in an average score | Some accounts describe improvement; others describe deterioration | Examine variation and whether an average obscures different trajectories. |
Keep the source and basis of each entry available. State the sample, period and uncertainty rather than presenting the table as a self-contained proof. A useful display also records unanswered questions and evidence that challenges the emerging explanation.
Disagreement is a finding to examine, not a formatting problem to remove. Begin by checking whether the strands refer to the same concept, population and period. Review missing data, response patterns, wording and how qualitative material was selected or coded.
Next, consider substantive explanations. A favorable score may coexist with a detailed criticism because the person values the overall service but encountered a specific difficulty. Different groups may experience the same process differently. Neither account needs to be discarded to create a single positive message.
If the evidence cannot resolve the difference, report the uncertainty and specify what additional information would help. Do not count quotations as votes or assume that the larger dataset must be right. The strength of an inference depends on the question and quality of the evidence, not only sample size.
A mixed methods workflow may use a survey tool, recording or transcription service, qualitative analysis software, statistical software and a shared record system. The question is whether the arrangement preserves the information needed to combine interpretations reliably.
AI may help propose codes, organize material or identify candidate relationships. It cannot invent a missing phase, supply consent or establish that an explanation is causal. Review generated interpretations against the underlying evidence and retain uncertainty.
Sopact is relevant when recurring responses and documents need to remain connected across a lifecycle. Evaluate that fit using a real example rather than a generic summary demonstration. For learning and program teams, see Training and Programs.
Use the mixed methods research tools guide to continue evaluating the software requirements.
Describe why both strands were needed, their sequence and emphasis, how each sample was selected, what was collected and how each strand was analyzed. Explain the integration point and show what the combined interpretation contributed beyond either strand alone.
Report limitations in each strand and in their connection. If follow-up participants differ from the initial sample, say so. If definitions changed, explain the effect. Avoid presenting a diagram of the intended design as evidence that the study actually followed it.
For the practical next step, work through connecting quantitative and qualitative survey data. If your main task is designing a questionnaire with both response types, start with mixed method surveys instead.
Convergent, explanatory sequential and exploratory sequential are common core designs. Embedded and multiphase arrangements support more complex purposes.
Explanatory starts with quantitative findings and follows with qualitative inquiry to interpret them. Exploratory starts with qualitative inquiry and uses it to develop a later quantitative phase.
Not exactly. Convergent describes an arrangement of strands. Triangulation describes examining evidence through multiple perspectives or methods; its purpose and implementation should be explained.
No. Samples may overlap, be nested or differ. Explain how the sampling relationship supports the combined question and the unit at which findings are integrated.
No. Each sample must fit its purpose. Equal numbers do not establish equal quality or make the resulting evidence comparable.
No. In experimental statistics, mixed design commonly refers to within-subject and between-subject factors. Mixed methods research refers to integrating qualitative and quantitative inquiry.
Yes, but specify what the interviews will explain, how people will be selected and how the findings will be integrated. Report the design as conducted, including decisions made after the first results.
