What is mixed-methods data analysis?
Mixed-methods data analysis brings quantitative and qualitative findings together to answer a question that neither could answer as well alone. You analyze each strand using a suitable method, then examine how the findings agree, explain one another, add context or conflict. The final interpretation must account for both.
A satisfaction score can show how respondents rated an experience. Interviews may reveal why people with similar ratings describe different problems. Putting a chart beside a quotation is a start, but the analytical work is explaining what their relationship changes about your conclusion.
This guide shows how to plan that integration, build a joint display, investigate disagreement and report a defensible finding. It applies to member research, customer feedback, employee listening, training and program evaluation. For choosing the overall study, begin with mixed-methods research designs.
Decide where the two strands will meet
Write the combined question before selecting software. For example: “How useful are regional services, and what explains differences between newer and established members?” The rating and the accounts of service use have distinct jobs. Decide which populations, periods and experiences they cover, and how the two findings will inform one interpretation.
Fetters, Curry and Creswell describe integration at the design, methods and interpretation levels. Their framework includes connecting samples, building one strand from another, merging findings and embedding one within a wider design. This is broader than matching individual records. See their research on integration principles and practices.
Scroll horizontally to see all columns →
| Design | How analysis connects | Practical planning question |
|---|---|---|
| Convergent | Analyze both strands and compare their findings | Do they cover sufficiently comparable experiences and periods? |
| Explanatory sequential | Use quantitative findings to guide a later qualitative investigation | Which result needs explanation, and whose experience could clarify it? |
| Exploratory sequential | Use qualitative learning to help develop measures or a later quantitative study | Which concepts need clearer questions, and how will the new instrument be tested? |
| Embedded | Use one strand within a larger study to address a complementary question | What can the smaller strand explain that the main measures leave open? |
Choose the design that fits the question and resources. A short comment box does not automatically make a survey a rigorous mixed-methods study. Explain how the text will be analyzed and integrated, rather than collecting it because the form allows it.
Choose the right level of connection
Some analyses connect the rating and comment from the same response. Others compare survey results for a region with interviews from a purposively selected group in that region. Both can be useful, but they support different claims.
Record-level connection requires an appropriate key and a legitimate reason to link. It need not identify someone by name. An anonymous response can contain both a rating and a comment. Linking several waves to the same person requires additional planning and must fit the privacy promise made during collection.
Group-level integration does not require every interviewee to have completed the survey. It does require a clear account of who contributed to each strand. Never describe an interview explanation as the reason every survey respondent chose a score when that connection was not established.
In federated networks, agree a small shared core: the measure, scale, relevant group, period and denominator. Local teams can ask additional questions in their own forms. A data dictionary makes the shared fields comparable without requiring one identical survey across every chapter, school or site.
Analyze each strand on its own terms
Quantitative analysis
Check eligibility, missing answers, duplicates and the unit of analysis. Describe distributions, valid denominators and relevant groups before fitting a more complex model. A mean can hide very different experiences, so inspect the spread where it matters.
If comparing periods, check the question wording, population and collection method. Matched-person change and change in successive respondent groups answer different questions. Statistical tests must fit the sampling, measurement and dependence in the data; significance alone does not establish practical importance or cause.
Qualitative analysis
Choose a method that fits the question and material. You might use a structured codebook to compare recurring operational concerns, or develop themes through close reading when the important concepts are not yet known. Explain how codes were developed, revised and reviewed.
Keep enough source context to interpret a passage accurately. “The training was excellent, but my manager will not let me use it” contains praise and an implementation barrier. Reducing it to positive sentiment would lose the part that matters to a decision.
Count coded responses only when the counting rules and denominator are meaningful. A theme appearing in six interviews is evidence that it occurred in those interviews. A purposive interview sample generally does not establish its prevalence in the wider population.
Build a joint display that changes the interpretation
A joint display places related findings in a table or visual and adds the interpretation produced by reading them together. It is an analysis aid, not simply a way to decorate a report. Research by Guetterman, Fetters and Creswell illustrates how joint displays support intentional integration.
The following example is fictional. A network receives 120 valid service-usefulness ratings: 80 established members and 40 newer members. Ratings of 4 or 5 on a five-point scale count as favorable. Separately, the team interviews 12 members selected to explore varied experiences. The interview sample is not a representative estimate of the network.
Scroll horizontally to see all columns →
| Question | Quantitative finding | Qualitative finding | Integrated interpretation and next check |
|---|---|---|---|
| Who finds the service useful? | 60 of 80 established members are favorable: 75%. For newer members, 20 of 40 are favorable: 50%. | Some newer interviewees describe difficulty finding the right contact; established interviewees describe personal referral routes. | The 25-percentage-point gap merits investigating access. Interviews suggest a possible explanation, not proof of its contribution to the entire gap. |
| Is helpful advice enough? | Several respondents rate advice highly. | Some interview accounts distinguish helpful advice from delays in obtaining it. | Usefulness and speed may be separate dimensions. Check whether the survey measures both before treating the accounts as contradictory. |
| What should change? | The newer-member group has a less favorable distribution. | Interviewees suggest a clearer first contact and examples of when to use the service. | Test clearer onboarding with an accountable owner. Review access and usefulness separately, with respondent coverage noted. |
The overall favorable share is 80 of 120, or 66.7%. It is not the simple average of 75% and 50%, because the groups differ in size. Report the group results as well as the aggregate. The interviews help the team decide what to investigate; they do not turn the survey into a causal evaluation.
Treat disagreement as a finding
When ratings and accounts appear inconsistent, first check whether they actually concern the same thing. A rating may cover the whole service, while a comment describes one recent incident. Different periods, groups, question wording or missing responses can also explain a mismatch.
Return to the source before changing the code or excluding the case. Review whether the rating was imported correctly, whether the passage was truncated and whether the question invited a different interpretation. If the discrepancy remains, describe it openly and consider a follow-up question.
Do not force agreement to make a cleaner story. A useful conclusion may be that the organization delivers valuable support unevenly, or that the survey misses a concern participants consider important. Conflicting findings can reveal a measurement problem or a real difference in experience.
How Sopact reduces coding and reporting work
Integration has an ongoing maintenance cost. A new coding definition should lead to a reviewed update, not another manual pass followed by a fresh spreadsheet join.
A workflow with repeated manual work
- Define from an initial sampleRead material and agree on the codebook.
- Apply it across the datasetCode responses and check the result.
- Revise a definitionReturn to affected material and recode it.
- Reconnect the numbersReconcile coded results with ratings and context, then rebuild the view.
The Sopact workflow
- Your team owns the definitionsDecide what each code means and improve it as you learn.
- Apply coding across the eligible dataAutomate application; people review quality and exceptions.
- Reprocess after a definition changesReapply the revised definition across the configured scope instead of recoding each response by hand.
- Ask across coded text and numbersKeep the response, rating and relevant record context connected; inspect the evidence behind the result.
This compares workflow patterns, not a claim that every research tool requires manual coding or separate files. Some already automate parts of this work; compare the complete cycle.
For this codebook-based workflow, the main saving is repeated application and reconnection—not the removal of human judgment. A changed definition can be reapplied across the configured data while reviewers concentrate on quality, exceptions and interpretation. Coded text stays connected to the relevant ratings and context.
Count the recurring work in ownership cost. Include setup, coding, recoding after revisions, source reconciliation, review and reporting, plus your actual platform and processing expenses. A worked scenario of four cycles of 4,000 responses illustrates 272 fewer annual staff hours; it is an assumption-based example, not a customer benchmark. Existing automation, review needs and implementation effort can substantially change the result.
Adjust the workload assumptions and compare total effort →
A reliable assistant should calculate from the selected records and let a reviewer open the supporting evidence. Check the data scope, definition, denominator and access permissions. Reproducible arithmetic does not make every AI interpretation correct.
Watch: Why Qualitative Analysis Stays Small — And How to Scale It
See why revising a codebook creates repeat work, and how connected coding and quantitative analysis change that workload.
Use software and AI to reduce preparation work
Spreadsheets, statistical packages and qualitative analysis tools can all play a role. A well-designed workflow can integrate their outputs. The practical question is how much recurring effort your team spends preparing data, maintaining definitions, checking joins and tracing conclusions to evidence.
Sopact is relevant when recurring collection, comments and documents need to remain connected with usable context and team-controlled definitions. Evaluate it against an actual analysis: ask reviewers to inspect a rating, its relevant source material, the code applied and the explanation for a combined finding. Confirm which steps are configured, integrated or still require an analyst.
AI can help suggest codes, locate passages and draft summaries. Review those suggestions against varied examples, including disagreement, mixed sentiment and different languages. Preserve original text and track meaningful changes to the codebook. A fixed codebook does not guarantee identical model output, and a source citation does not guarantee that the interpretation is correct.
Test repeated collection as well as the first import. Add a new wave, a changed local question and a document covering a different period. Check whether the analyst can distinguish comparable records from those that need separate interpretation. Include review time and correction effort in the evaluation.
Report the integrated conclusion and its limits
A useful report states the combined question, the purpose of each strand, who contributed and when, the analysis methods and where integration occurred. Include a joint display or a short narrative that explicitly compares the findings. Readers should not have to infer the relationship between two separate result sections.
Separate observation, interpretation and proposed action. In the network example, the rating gap is an observation. Difficulty finding contacts is a possible explanation supported by selected accounts. A revised onboarding process is a proposed response that still needs testing.
Keep a record of the source version, calculation, coding decisions and review. Share quotations only where appropriate and remove identifying details when needed; changing a name may not be enough if the experience is distinctive. Give authorized reviewers access to supporting evidence without publishing unnecessary personal information.
For the broader reporting process, use the impact report guide and report examples. Continue with survey analysis steps or worked mixed-methods research examples.
Watch: combining qualitative and quantitative evidence
This video introduces mixed-methods research. Use the worked joint display above to plan how your own findings will connect.
Frequently asked questions
Is mixed-methods analysis the same as triangulation?
Triangulation is one purpose or approach to integration. Mixed-methods work can also use one strand to explain findings, develop measures or address a complementary question. Describe what combining the strands contributes to your particular study.
Must the qualitative and quantitative samples contain the same people?
No. Some designs connect individual records; others integrate at a group, case or conceptual level. Explain the relationship between the samples and avoid attributing one group's explanations to another without evidence.
What is a joint display?
It is a table or visual that brings related quantitative and qualitative findings together and records the interpretation developed from them. Include uncertainty and the next question, rather than only placing results side by side.
What if the findings disagree?
Check the sources, periods, groups, questions and analysis decisions. If the disagreement remains, report it and investigate what it means. Do not discard evidence merely to make the two strands agree.
Can mixed-methods analysis prove what caused an outcome?
Combining numbers and accounts can strengthen understanding, but causal claims depend on the study design and evidence. A plausible explanation from interviews is not automatically proof of causation.
Can AI complete the analysis without a researcher?
AI can assist with organization, coding and summaries. People still need to judge the method, inspect sources, examine conflicting evidence and take responsibility for the interpretation.
Choose your next step
ASME
Explore how quantitative and qualitative stakeholder evidence supports a membership organization’s work.
Read the customer story →