Mixed Methods Research: When One Approach Is Not Enough

A few months ago I was reviewing a paper that had two clearly separate studies stitched together. The first was a survey with structural equation modeling. The second was a set of semi-structured interviews. The authors called the whole thing a mixed methods study. But the two parts never talked to each other. The interviews did not explain the survey results. The survey did not test anything that emerged from the interviews. They were just sitting next to each other in the same document. That is not mixed methods. That is two studies with a shared introduction.

The defining feature of mixed methods research is integration, not combination. Tashakkori and Teddlie (1998, 2010) are clear on this in their foundational work on mixed methods as a research paradigm: mere coexistence of qualitative and quantitative components does not make a study mixed methods. Integration is what makes it mixed methods. The two strands have to inform each other at some meaningful point in the research process, whether that is at the design stage, the data collection stage, the analysis stage, or when drawing conclusions.

Why does this matter for IS research? Because IS phenomena are complicated in ways that resist a single methodological lens. Take a typical adoption study. You survey 400 users and find that perceived usefulness and social influence predict behavioral intention. That is a clean quantitative result. But it leaves obvious questions unanswered. Why does social influence matter more for some user groups than others? What are users actually thinking when they assess usefulness? Why do some people with high intentions never actually use the system? A follow-up qualitative component, designed to explain those residuals, can answer questions the numbers cannot. That is a legitimate integration point.

Venkatesh, Brown, and Sullivan (2016) extended earlier mixed methods guidelines specifically for IS research, identifying fourteen design properties that distinguish rigorous mixed methods designs from studies that are just two things happening in parallel. One of the key distinctions they draw is between concurrent designs, where qualitative and quantitative data are collected at roughly the same time and synthesized afterward, and sequential designs, where one strand informs the other in a deliberate order. In a sequential explanatory design, you run the quantitative analysis first and then use qualitative data to explain surprising or nuanced findings. In a sequential exploratory design, you build theory from qualitative work first and then test what emerged with a quantitative study. Both are real and useful. The choice depends on the state of knowledge in the area and what kind of contribution you are trying to make.

The pragmatic argument for mixed methods is simple enough. If your theory says "use of X leads to better outcomes" and you want to test that, a quantitative design works. But if you also want to understand why some users get better outcomes than others, or why the relationship holds in some organizations but not in others, qualitative data is more useful than adding more control variables. The two questions need different epistemological tools. Forcing one into the other produces a worse answer than designing for both.

There is a cost to this, and I do not want to pretend otherwise. Mixed methods studies are genuinely harder. They require competence in two methodological traditions. They take longer. The integration point, the moment where the two strands actually influence each other, is hard to design and even harder to execute convincingly. I have seen plenty of mixed methods papers where the integration section reads like an afterthought. The authors list the qualitative themes, point out that they are consistent with the quantitative findings, and call it convergent evidence. That is weak integration. Real integration means the qualitative findings either explain variance that the quantitative model leaves on the table, or they generate hypotheses that the quantitative component tests, or they challenge the quantitative results and force a reinterpretation.

Industry analysts face the same integration problem. Gartner, for instance, regularly publishes research that combines large-scale surveys of CIOs and enterprise buyers with qualitative case studies and expert interviews. You can see examples of this mixed approach across their reports and briefings at the Gartner newsroom. The goal is triangulation: using different methods to converge on a finding so that no single method's weaknesses undermine the conclusion. When Gartner reports a finding about enterprise AI adoption and backs it with both survey percentages and case study examples, they are doing something structurally similar to what mixed methods researchers aim for. The difference is that academic mixed methods research documents the integration protocol explicitly and subjects it to peer review. Industry reports usually do not show their methodology at all.

The paradigm question comes up in almost every methods discussion. For a long time, quantitative research was associated with positivism and qualitative research with interpretivism, and the two camps had real philosophical disagreements about what counts as knowledge. Mixed methods research does not resolve those disagreements. What it does is adopt a pragmatic position: use the methods that best answer the question. I wrote more about how paradigm choices shape method choices in my post on why your paradigm is not neutral. The key move in a mixed methods defense is explaining the integration logic. Not just "I used both methods" but "I used both methods in this sequence because my research question requires this kind of evidence at this point in the analysis."

The paper I was reviewing at the start did not have that logic. Both studies were real. Both were executed competently. But neither needed the other to exist. That is the test I now apply when I read a mixed methods paper: could either strand stand alone as a complete study? If yes, you probably have two studies, not one mixed methods design.