Research & PolicyFree
Mixed Methods Integration and Joint Display
Send the survey results and the interview findings, get a joint display with a stated verdict on every research question.
River builds the joint display: one row per research question, holding the quantitative result, the qualitative finding, a relation verdict and the inference the pair supports that neither supports alone. Then it does the move most integrations skip. Where your interviewees were drawn from your survey sample, it matches them back to their own survey records. Then it checks each person's rating against what they actually said, which is what makes a divergence explainable rather than merely noted. Both artifacts arrive together.
Search this and you get design taxonomies: convergent, explanatory sequential, exploratory sequential, with diagrams. Useful for planning and silent on what to write. The NIH's own best-practices guidance for the health sciences is more direct. It tells investigators to include a discussion of how potential divergent or inconsistent findings will be managed and interpreted, and it names data displays showing both strands as a way integration actually happens. Divergence is a required section, not an embarrassment.
Written for the author with two finished results sections and no idea how to join them, the evaluator whose funder asked for mixed methods, and the doctoral student defending a design choice. It runs after both strands are analysed: themes with negative cases and participant spread on one side, and whatever produced your survey estimates on the other. Before submission, the reporting checklist located in your actual draft catches what the integration section still omits.
The two strands agreed in aggregate and not per person
A programme evaluation with a survey of 412 participants and 22 follow-up interviews. On the key item, 305 of 412 rated the programme helpful, which is 74.0 per cent. The interviews were recruited from the survey frame, so 19 of the 22 match back to a survey record and three came from outside it. Of those 19, fourteen had rated the programme helpful, 73.7 per cent, so the interviewed subsample is representative on the item that matters.
Now cross each person's rating against what they described. Nine rated it helpful and described it that way. Five rated it helpful and spent the interview describing problems. One rated it unhelpful and described it positively. Four were negative in both. Thirteen of 19 are concordant and six, 31.6 per cent, are not. The aggregate agreement was 0.3 percentage points off and hid a third of the cases disagreeing with themselves.
Five of the six discordant cases run the same direction: a positive rating with a negative account. That asymmetry, 83.3 per cent one way, is a finding about the instrument rather than about the participants, and the interviews say why. The joint display then carries seven rows, one per research question, with two verdicts of confirmation, three of expansion and two of discordance. The two discordant rows are the ones a reader learns from. A results section that reported 74 per cent and moved on would have missed all of it.
How it works
Send both strands
The quantitative results and the qualitative themes, plus the research questions they were meant to answer.
Match the samples
Interviewees traced back to their own survey records wherever the two samples overlap.
Build the display
One row per research question, with a relation verdict and the joint inference on each.
Explain the divergence
Where the strands disagree, the cases that caused it and the direction they run.
What you get
- One row per research question, with the quantitative result and the qualitative finding side by side
- A relation verdict on every row, including the rows where the two strands disagree
- The inference each pair supports that neither the survey nor the interviews support alone
- Interviewees matched back to their own survey records, where the samples actually overlap
- Case-level concordance, so aggregate agreement cannot hide the individuals who contradict themselves
- Whether the discordance runs one direction, which is a finding about your instrument
Common questions
What is a joint display and why does my paper need one?
It is a table that puts the quantitative result and the qualitative finding for the same question in the same row, with a verdict on how they relate. It is the artifact that shows integration happened. Two separate results sections and a paragraph saying the findings converged is the default, and a reviewer reads it as two studies stapled together.
What if my two samples do not overlap at all?
Then the case-level join is not available and the run says so rather than faking it. You still get the joint display, the relation verdicts and the joint inferences, all at the aggregate level. What you lose is the ability to explain a divergence by pointing at the people who caused it, which is worth knowing before you design the next study.
Should I report a divergence or work around it?
Report it, and NIH guidance for mixed-methods applications expects a plan for exactly that. A divergence between a rating and an account is usually the most informative thing in the study, because it tells you what the instrument measures rather than what it claims to. Working around it produces a paper that agrees with itself and teaches nobody anything.
Does it count how often themes appear so I can compare?
Only where you ask for it, and it flags the cost. Converting themes to counts makes the two strands comparable and throws away the reason you collected qualitative data. The run will do it as a labelled secondary view, never as the primary one, and it never reports a theme frequency as though it estimated a population.
My design was sequential, not concurrent. Does that change anything?
Yes, and the display changes shape. In a sequential design one strand informed the other, so the row records what the first phase contributed to the second rather than whether they agree. The run asks which came first and builds accordingly, because a convergence verdict on a strand you designed from the other one is circular.
Can I use it on qualitative data I have not coded yet?
Better to code first. The display needs themes with their evidence and their participant spread, and building it from raw transcripts produces impressions rather than findings. A codebook with definitions and a change log is the input, and the coding work is a separate job worth doing properly.
What do I get back?
A Sheet holding the joint display, one row per research question with both results, the relation verdict and the joint inference, plus the case-level match table where the samples overlap. A Doc writing the integration by research question, with the divergences explained and their direction stated.
Mixed Methods Integration and Joint Display
Fill in the form and your workspace opens with the work already underway.