Mixed Methods Research

When mixing is actually justified

Mixed methods costs roughly twice the work of a single-strand design. It is worth it when the question genuinely has two parts that need different evidence, and it is not worth it when the second strand is there for reassurance.

Good reasons to mix:

• You can measure that something happens but not why, and the why is part of the question.
• You need to develop an instrument for a context where none exists, and must first learn what the relevant dimensions are.
• You need to know whether a pattern found in numbers is recognised by the people it describes.
• A finding is contested, and evidence of two kinds would be more convincing than more of one kind.

Poor reasons, all of which examiners see regularly: wanting the thesis to look thorough; being unsure which paradigm you belong to; adding a small number of interviews to a survey because a reviewer once suggested it. If you could remove one strand and the conclusion would stand unchanged, that strand is decoration and you have paid for it twice.

There is also a philosophical expectation in most disciplines. Mixing strands means combining evidence built on different assumptions about what counts as knowledge, and you are usually expected to say how you reconcile that. Pragmatism is the most commonly cited position — the question decides the tools — but naming it is not enough on its own; it has to be visible in how you actually handle disagreement between strands.

The four designs worth knowing

Almost every doctoral mixed-methods study is a version of one of these. The differences are order, weighting, and where the strands meet.

Explanatory sequential: quantitative, then qualitative

You collect and analyse numbers first, then use qualitative work to explain what they showed — particularly the results that were surprising, weak or contradictory.

Use it when you expect to find patterns you will not be able to interpret from the data alone.

The design decision that matters: who you select for the second strand, chosen on the basis of the first strand’s results. Interviewing a fresh random sample wastes the design’s main advantage. Selecting the cases that were extreme, or that did not fit the model, is where the explanatory power comes from.

Exploratory sequential: qualitative, then quantitative

You explore first, then build something measurable from what you learned — commonly an instrument, or a set of variables worth testing at scale.

Use it when the construct is poorly understood in your setting, or existing instruments were developed somewhere they may not transfer from.

The design decision that matters: how qualitative findings become items or variables. That translation is a genuine methodological step and needs describing in detail, not summarising in a sentence. Building an instrument this way also means validating it, which is a project in itself.

Convergent: both at once, compared afterwards

Both strands are collected in the same period, analysed separately, and then brought together to see where they agree, extend each other, or conflict.

Use it when you want two independent views of the same question and have the capacity to run both properly.

The design decision that matters: deciding in advance what you will do when they disagree. Disagreement is a legitimate and often valuable finding — it usually means the two strands are capturing different aspects of the phenomenon — but only if you planned to treat it that way rather than quietly favouring whichever strand supported your expectation.

Embedded: one strand supporting the other

A smaller strand sits inside a larger design, answering a secondary question — interviews inside a trial about why participants dropped out, for instance.

Use it when the primary design is sound but leaves a specific gap the other method can fill.

The design decision that matters: being honest that the strands are not equal, and saying what the smaller one is for.

Integration: the part that makes it mixed methods

If you take one thing from this page, take this. Integration is what distinguishes a mixed-methods thesis from two studies bound together, and it is what examiners look for first.

Integration happens at three possible points, and a strong study is explicit about which it is using:

• At design — one strand shapes the other: results decide who is interviewed, or qualitative findings decide what is measured.
• At data — material from one strand is transformed to sit alongside the other, such as coding qualitative categories so they can be counted, or grouping cases by their quantitative profile before reading their accounts.
• At interpretation — the strands are brought together in the discussion to build a single argument.

The most practical tool here is a joint display: a table with the quantitative finding in one column, the related qualitative finding beside it, and what the combination means in a third. It forces the comparison to be made explicitly rather than gestured at, and it gives an examiner something concrete to look at. Building one is also a good test — if you cannot fill in the third column, the strands have not actually been integrated.

A useful check on the whole thesis: could you state the conclusion in one sentence that requires both strands? If the sentence works with either strand deleted, integration has not happened.

Practicalities people underestimate

• Two sets of everything. Two literatures on method, two sampling justifications, two analysis chapters, two sets of quality criteria — validity and reliability for one strand, trustworthiness for the other. Applying quantitative criteria to qualitative work, or the reverse, is a small error that signals a larger confusion.
• Sequential designs cannot be parallelised. The second strand depends on the first being finished, and a delay in strand one moves everything. Ethics approval may also be needed twice, since you often cannot specify the second strand until the first is analysed.
• Word count. Two full methodology and results sections inside one thesis puts real pressure on length. Something has to be reported more economically, and deciding what in advance is better than discovering it in the final month.
• Sample sizes serve different logics. The quantitative strand needs enough cases for its analysis; the qualitative strand needs enough depth for its question. There is no requirement that they match, and trying to make them match usually damages one or both.