How do you connect quantitative and qualitative survey data?
Link the relevant scores and comments to the same person, group or period, then examine them together using a clear analysis question. A joint display—a table that brings the numerical result, qualitative evidence and interpretation together—helps show where findings agree, add detail or conflict. Comments can suggest explanations, but do not automatically prove why a score changed.
Use this reference when scores and comments must inform the same decision. Build a linked analysis view, compare the evidence and write a conclusion with its limits. For the broader study-design choices, use the mixed-method design guide.
In a fictional tutoring program, some learners improve on an assessment and others do not. Families describe caring tutors, limited session availability and transport difficulties. An overall theme list is useful for understanding the service. To investigate differences in progress, the team also needs to know which feedback belongs to which learner, period and respondent.
Choose the question before choosing the split
Ask something specific: “Do learners with different assessment changes describe different access barriers?” That question supports an outcome-group comparison. “How do families experience tutoring?” may be better answered by an overall qualitative analysis. Neither replaces the other.
Mixed-method integration does not require every project to split outcomes before reading comments. You may use qualitative findings to develop a survey, use interviews to investigate a numerical pattern, or analyze both strands and then bring them together. Fetters, Curry and Creswell’s integration framework describes these different routes.
For a planned group comparison, write the grouping rule before interpreting differences. If exploratory reading suggests a new comparison, keep it and label it exploratory. Do not present a threshold selected after seeing the results as if it were specified beforehand.
Connect the right records, not just matching names
Keep a permitted stable identifier, the measure and question versions, response dates and the period being described. A parent’s comment about a child is not the child’s own account. Preserve the respondent’s role and relationship rather than putting both voices into an unlabeled comment field.
One person may provide several comments, attend several programs or have several assessments. Decide the unit of analysis before joining files. A simple spreadsheet can use one row per learner-period with separate source links; a relational structure may be needed when there are multiple respondents and events.
- Person or case: whose outcome is being analyzed?
- Respondent: who supplied the comment, and in what role?
- Observation: which assessment, question and time period does it concern?
- Relationship: how is that respondent or source connected to the person?
- Match status: confirmed, unresolved or not designed for individual linkage?
Do not force uncertain matches to eliminate an unmatched pile. Report the unmatched count and investigate its pattern. Anonymous feedback can be integrated at an appropriate group or theme level without being re-identified; individual linkage is not a requirement for every mixed-method study.
A common spreadsheet mistake is to repeat a learner’s score once for every comment. The most vocal learners then receive more weight in the average. Check the expected row counts after joining and calculate participant outcomes at the participant level.
Define outcome groups without hiding the original scores
Use a meaningful, justified criterion where one exists. In this fictional exercise, the team defines “met the improvement criterion” as an increase of at least five points on its example assessment. This is an invented teaching rule, not a recommendation for any real instrument.
Retain the actual change scores as well as the group label. Grouping can simplify a display while hiding differences near a cutoff. Where there is no established criterion, explain the exploratory rule and assess whether reasonable alternatives change the interpretation.
Keep missing outcomes separate from “did not improve.” A wave-status register identifies who lacks a usable pair; use the linked missing-waves reference if you need to build one. A person with no endline score cannot be classified as unchanged merely because the software requires a group.
Practice: build a joint display
In the fictional cohort, 120 learners have usable paired assessment scores. Seventy meet the improvement criterion and 50 do not. Sixty of the first group and 40 of the second have a linked, substantive family comment. The following counts use one family comment per learner for the exercise.
| Theme | Met criterion: 60 comments | Did not meet: 40 comments | Interpretation to investigate |
|---|---|---|---|
| Tutor described as caring | 36/60 = 60% | 24/40 = 60% | Shared service experience; may matter even without a group difference |
| Requested more sessions | 18/60 = 30% | 16/40 = 40% | Explore need, expectations and session access |
| Bus timing conflicts with session end | 3/60 = 5% | 12/40 = 30% | Check schedules, attendance and site differences before claiming a cause |
Transport timing appears more often in the comments from learners who did not meet the criterion: 30% versus 5%, a 25 percentage-point difference among these commenters. It is a lead for investigation. It does not prove that transport caused the assessment difference.
The comment coverage is also different: 60/70 = 85.7% of learners meeting the criterion, and 40/50 = 80% of those not meeting it. Report those bases. Theme percentages describe the available comments, not necessarily every learner or family in either outcome group.
The theme categories can overlap, so their percentages need not add to 100%. Count learners, responses or passages consistently. Do not mix a count of mentions in one group with a count of people in the other.
Guetterman, Fetters and Creswell’s study of joint displays shows how bringing findings together can support interpretation. The display should contain an analytical connection, not merely a numerical table placed beside an unrelated quotation.
Read shared themes and contradictions as evidence
A theme present in both groups is not worthless. Caring tutors may support participation, matter to families in ways the assessment misses, or be too similar across groups to explain this particular contrast. The data do not justify dismissing it as “explaining nothing.”
Likewise, a negative comment from someone whose score improved is not an error to remove. The score and comment may concern different aspects of the experience. Check the question, timing and context before deciding that the measure failed or the comment contradicts it.
If no theme differs between groups, report that result. It may reflect similar experiences, an insensitive question, limited sample size, coarse coding or other factors. It does not prove participants cannot explain the difference. Consider whether another question, source or method is needed.
Review both supportive and disconfirming examples. Do not select only the most vivid quotes for a preferred explanation. A quotation illustrates an account; it does not establish how common the account is.
Choose a coding process that limits expectation bias
Develop or refine the codebook on a varied set of responses. Use clear inclusion and exclusion rules, preserve source passages and record changes. The response-cleaning lesson explains why valid “none” answers and missingness should remain distinct.
Reading by group can help explore a pattern, but it can also prime the reviewer to see what they expect. Depending on the purpose, code comments without showing outcome labels first, then compare the coded results. Another option is to review a subset independently and discuss differences before completing the analysis.
If you code with AI, inspect the assigned themes and supporting text rather than judging the summary’s fluency. Review ambiguous responses, rare themes and exceptions. Changing the prompt or rubric may change the outputs; record the version and decide whether earlier responses need reprocessing.
For multilingual material, use the language-review workflow. An apparent group difference may partly reflect translation or interpretation quality. Do not attribute it to the participants without checking that possibility.
Investigate the explanation with additional evidence
For the bus example, check whether affected learners attend the same site or session, whether the timetable actually overlaps, and whether attendance records show missed learning time. Keep the period consistent: a later timetable change does not explain an earlier assessment result.
Site, starting ability, program exposure and other factors may relate to both the feedback and the outcome. A simple table does not separate those influences. If the decision requires a causal claim, use a design and analysis capable of supporting it; do not promote an observational association into proof.
A practical next step may still be justified without proving the entire causal pathway. For example, verify and address a documented scheduling conflict, then collect follow-up evidence about access and outcomes. State that the team is testing an explanation, not that the first table has settled it.
A customer context: the King Center
The King Center customer story describes pre- and post-survey evidence and qualitative feedback across seven programs. It provides a relevant setting for connecting different kinds of evidence. The tutoring figures above are fictional and are not results from that customer.
Plan the recurring coding and reporting work
The joint display is useful only while its codes, ratings and record relationships stay current. Plan a codebook revision as part of the workflow so the next comparison does not require another manual coding and joining project.
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
Watch this 2-minute 58-second explanation of repeated coding, revised definitions and connected analysis. Then estimate the work for your own collection cycle.
Test the workflow in your data system
Start with one cohort, one comparable outcome measure and a relevant open question. Produce the linked-record count, unmatched count, usable outcome pairs and comment coverage before asking for themes. That reconciliation prevents a polished report from concealing who was excluded.
For a configured Sopact workflow, ask the team to demonstrate how an incoming comment stays connected to the appropriate person or relationship, how rubric-based analysis can be reviewed, and how a group comparison links back to the original evidence. Then add a later response and check the effect on the report.
Change the display from outcome group to site and confirm the counts still reconcile. Existing codes may be reusable, but a new research question can require new coding or review. Do not assume every useful comparison can be made without revisiting the source.
A pass does not require discovering a one-sided theme. It requires correct relationships, visible missingness, consistent counting rules and an interpretation that the evidence supports—even when the result is inconclusive.
Write the finding, limitation and next action together
For the fictional display: “Bus-timing conflicts were reported in 12 of 40 comments linked to learners who did not meet the improvement criterion, compared with three of 60 among those who did. This association does not establish a cause. We will check site schedules and attendance before deciding whether a timing change should be tested.”
Add the comment coverage and grouping rule alongside the table. Preserve links to supporting passages for authorized reviewers, with identifying details handled appropriately. If you need a reporting structure, use the impact-report writing guide and report examples.
Watch: keeping qualitative evidence connected
Watch the video · 2 minutes 34 seconds. This companion explains why qualitative evidence needs to stay connected to the records it describes. Browse more videos in the video library.
Frequently asked questions
Must numbers and comments be on the same row?
No. They need a valid relationship appropriate to the analysis. One-row views can help, but multiple respondents or periods may need related records. Aggregate integration is possible when individual linkage is not intended.
Should outcome groups always be defined before coding?
For a planned comparison, define the rule before interpreting results. Exploratory analysis is also legitimate when labeled. Coding without outcome labels can help limit expectation bias.
Are shared themes useful?
Yes. They can describe common experiences or suggest factors worth keeping. They may not explain a particular difference, but that does not make them unimportant.
How do I compare groups of different sizes?
Show counts and percentages using a clearly stated base for each group. Report comment coverage and do not confuse mentions, responses and people.
Can comments prove why an outcome changed?
They can support or challenge explanations, but an association alone does not prove causation. Check timing, alternative explanations and additional evidence.
What should I do with unmatched records?
Keep them visible, investigate uncertain relationships and report exclusions. Do not force matches just to produce a complete-looking dataset.
What if no differences appear?
Report the result honestly. Consider the question, coding, sample and other evidence. A sound analysis does not have to discover a difference.
For a three-checkpoint analysis
Carry the linked records, codebook and reporting bases into the pre, mid and post analysis lesson. The next step is to examine the sequence of change while retaining what participants said at each stage.
Reviewed September 12, 2026. Tutoring scenarios and numerical examples are fictional. Customer and research sources are identified separately.