Qualitative and Quantitative Integration

What do I do when the survey and the interviews say different things?

Investigate it. Work through eight candidate explanations in order and record the assessment for each rather than stopping at the first plausible one: different populations, different periods, different constructs, stated versus observed, prompted versus volunteered, social desirability differing by mode, sample composition, and analytical error in one stream. Then take one of four defensible positions. Averaging, splitting the difference, or preferring the larger base are not among them.

Should I trust the survey over the interviews when they conflict?

Not by default. Authority follows the question, not the sample size. The quantitative stream leads on how many, how often and how it varies. The qualitative stream leads on why, how and under what conditions. A large survey cannot overrule a mechanism finding, and a set of interviews cannot establish an incidence. Assign leadership per research question before you look at the results, because assigning it afterwards is choosing the answer you prefer.

What is a joint display?

The working artefact of integration: a matrix with one row per research question and columns for the quantitative result with its base, the qualitative finding with its prevalence and denominator, the relationship between them, and the integrated statement. Build it for every finding before writing anything, including the ones where the streams disagree and the ones where one stream is silent. If the integration exists only in the narrative, it has been asserted rather than done.

Can I report percentages from qualitative research?

No. Qualitative samples are recruited for range and depth, not incidence, so a proportion from them estimates nothing, and it will be quoted as though it did. Report counts with denominators, keep the base visible, and let the quantitative stream carry any claim about how widespread something is.

What is the difference between convergence and complementarity?

Convergence means both streams point the same way about the same thing. Complementarity means they address different facets and neither confirms nor contradicts the other: the survey gives the size, the interviews give the mechanism. Complementarity is the most common relationship in good mixed-method work and is routinely reported as convergence, which takes a confidence uplift the evidence does not support.

Does agreement between qual and quant mean I can be more confident?

Only if the streams were independent. In a sequential design where the survey was built from the qualitative findings, agreement is the design working, not corroboration. In an embedded design the same people answered both at the same sitting, so it is one piece of evidence, not two. Test independence before claiming any uplift.

What are the mixed-method designs and what does each let me conclude?

Sequential exploratory (qual then quant) licenses claims about the incidence of what the qual discovered, and nothing about what the qual missed. Sequential explanatory (quant then qual) licenses mechanism claims about the pattern, but cannot disconfirm it, since participants were often selected for exemplifying it. Concurrent triangulation licenses the strongest convergence claim, because neither stream shaped the other. Embedded designs license enrichment of the dominant stream, not independent corroboration of it.

How do I stop qualitative work becoming decoration in a quantitative report?

Give it a question to lead on and mean it. The test is simple: if no qualitative finding is permitted to change a quantitative conclusion, the study was not integrated. Structure the report by research question rather than by method, so no section belongs to one stream, and make sure at least some qualitative findings stand on their own rather than appearing under a chart.

Why do people say one thing in a survey and something else in an interview?

Usually because the two are measuring different things: prompted response versus volunteered salience, a general attitude versus a specific recent event, or a self-completion form versus a conversation with a person in the room, where social desirability operates much more strongly. Stated and observed evidence are supposed to differ, and treating that gap as an error to be reconciled deletes one of the most reliable findings in applied research.

What do I do if the divergence cannot be explained?

Report it as unexplained, with both figures, both sources, and reduced confidence, and say what would resolve it. This costs credibility with nobody who matters and protects the client from acting on half the evidence. A confident average would be worse, because it reports a number neither stream produced.