Two dashboards say engagement is down four points. The number is real, and it tells you nothing about what to do. To act, someone has to find out why, and that is a different kind of evidence entirely. Most bad decisions at work aren't caused by missing data; they're caused by reaching for the wrong type of data and trusting the answer anyway.

The quick version

  • Quantitative = counting. It answers what, how many, how often, how much, and lets you measure, compare and generalise. It cannot tell you why.
  • Qualitative = meaning. It answers why, how, what does it feel like through words, observation and interviews. It explains; it can't tell you how widespread something is.
  • Mixed methods = both, on purpose, so each covers the other's blind spot, numbers find the pattern, words explain it (or the reverse).
  • The choice isn't ideological. Start from your question. "How many?" is a counting question. "Why did they leave?" is a meaning question. Most real decisions need both.

The idea in depth

The cleanest way to hold the three apart comes from John W. Creswell and J. David Creswell, whose Research Design: Qualitative, Quantitative, and Mixed Methods Approaches (Sage, now in its sixth edition) is the standard reference taught in graduate programmes worldwide. They frame the three not as techniques but as whole approaches, each with its own logic about what counts as good evidence.

Quantitative research turns the world into numbers so it can be measured, compared and tested for statistical significance. Surveys, experiments, A/B tests, analytics, financial data, anything you can count, average or correlate. Its great strength is generalisability: a well-sampled survey of 800 people can stand in for thousands. Its great weakness is that a number is mute. Knowing that churn rose 12% tells you the size of the fire, not where it started.

Qualitative research works in the opposite direction. It collects words, behaviour and context, interviews, open-ended questions, observation, diary studies, and looks for meaning and pattern rather than frequency. It is hard to beat on the why, and it surfaces things you didn't know to ask about. Its limit is the mirror image of quantitative's: ten rich interviews explain a behaviour beautifully but cannot tell you whether it's true of ten people or ten thousand. Here's the shortcut: match the method to the verb in your question. "How many / how often / how much" is a counting question, go quantitative. "Why / how / what's it like" is a meaning question, go qualitative. Get the verb right and the method usually picks itself.

flowchart TD
    Q("What are you trying to learn?") --> A{"What's the verb?"}
    A -->|"How many / how much / how often"| QN(["Quantitative, measure & compare"])
    A -->|"Why / how / what's it like"| QL(["Qualitative, explain & explore"])
    QN --> M("Need the other half?")
    QL --> M
    M -->|"Yes"| MX(["Mixed methods, pattern + reason"])
Start from the question, not the method. Leaders Loop

Mixed methods: the third approach

For decades these two camps fought what methodologists actually call "the paradigm wars", quantitative researchers dismissing interviews as anecdote, qualitative researchers dismissing surveys as shallow. Abbas Tashakkori and Charles Teddlie named the resolution: mixed methods as the third methodological movement, sitting alongside the quantitative and qualitative traditions. Their case, set out in works including Mixed Methodology: Combining Qualitative and Quantitative Approaches (Sage, 1998), rests on a deliberately practical philosophy, pragmatism. The pragmatist position is blunt: stop arguing about which paradigm is "true," and use whatever combination of methods actually answers the question in front of you.

But "use both" is not a method. The seminal work on why you'd combine them is Jennifer Greene, Valerie Caracelli and Wendy Graham's 1989 study in Educational Evaluation and Policy Analysis (Toward a Conceptual Framework for Mixed-Method Evaluation Designs). Reviewing 57 real mixed-method evaluations, they found people mixed methods for five distinct reasons, and that naming the reason is what separates a designed study from a pile of data:

  • Triangulation, use both to check the same thing from different angles; if they converge, you trust the finding more.
  • Complementarity, use each to illuminate different facets of one phenomenon (the number and the felt experience).
  • Development, use the results of one method to build the other (interviews to design a better survey).
  • Initiation, deliberately hunt for contradiction, because where the two methods disagree is where the interesting questions hide.
  • Expansion, use different methods for different parts of a programme, to widen scope.

The discipline, then, is to write down your reason before you collect anything. If you can't say which of those five you're doing, you're not running mixed methods, you're collecting two datasets and hoping. Greene and colleagues found exactly this failure in the field: studies that claimed "triangulation" but used a design that couldn't actually triangulate. Jobs-to-be-Done research, for instance, is essentially a development design, qualitative interviews surface the "job," which then shapes what you measure at scale.

flowchart LR
    subgraph EXPLAIN["Explanatory sequence"]
      Q1(["Survey: what & how many"]) --> Q2(["Interviews: why behind the number"])
    end
    subgraph EXPLORE["Exploratory sequence"]
      E1(["Interviews: discover the themes"]) --> E2(["Survey: how widespread are they"])
    end
The two everyday mixed-methods shapes: explain a number, or scale up a theme. Leaders Loop

An honest limitation. Mixed methods is not a free upgrade. It costs more time, two skill sets, and the harder job of reconciling findings when the numbers and the stories disagree, which they will. Tashakkori, Teddlie and others have acknowledged that combining paradigms remains philosophically contested, not a solved problem. For a small, low-stakes, reversible call, one good method beats a half-done version of both. Reach for mixing when the decision is big enough, or contested enough, to justify the second lens.

A worked example

A regional sales director sees the dashboard: this quarter's customer-satisfaction score (CSAT) dropped from 82 to 74. That is a clean quantitative signal, measurable, comparable, and clearly bad. It is also useless on its own. "Satisfaction is down eight points" supports no action except panic.

So she runs an explanatory sequential design, quantitative first, then qualitative to explain it. The numbers do the locating: she segments the CSAT data and finds the drop is concentrated almost entirely in accounts onboarded in the last 90 days; long-tenured accounts barely moved. Now she has a where, but still no why. She and a colleague run eight 30-minute interviews with recently-onboarded customers (figures here are illustrative). A pattern surfaces fast: the product is fine, but a recent change to the setup flow meant new customers were going live without ever being shown a feature they'd specifically bought it for. They felt misled, not dissatisfied, a distinction no survey scale would ever have captured.

That is the whole argument in one story. The number told her where and how big. The interviews told her why, and pointed at a specific, fixable cause. Either method alone would have left her stuck: the dashboard with a problem she couldn't act on, the interviews with a story she couldn't size. Together, they're a decision. Before rolling out the onboarding fix, she'd then go back to quantitative, an A/B test on the setup flow, to confirm the fix actually moves the number. Numbers, words, numbers again.

Numbers tell you where to dig. Words tell you what you'll find. Don't confuse one for the other.

Frequently asked questions

Is qualitative research just "soft" or unscientific?

No. Good qualitative work is rigorous in its own terms, systematic sampling, structured coding, checks against bias and researcher influence. It answers a different question than quantitative work, not a lesser one. The genuine difference is generalisability: qualitative findings explain deeply but don't tell you how common something is. Treat that as a scope limit, not a quality flaw.

How many interviews or survey responses do I actually need?

They scale completely differently. For finding usability problems, Nielsen Norman Group's research suggests around five qualitative sessions surface roughly 85% of issues, while a quantitative study usually needs 30-plus participants to produce trustworthy numbers (see Budiu, NN/g, 2017). Rule of thumb: qualitative needs enough to stop hearing new things (saturation); quantitative needs enough for the statistics to hold.

Which should come first?

It depends on your reason for mixing. If you have a number and need to explain it, go quantitative-then-qualitative (explanatory). If you're exploring an unknown and need to know how widespread your findings are, go qualitative-then-quantitative (exploratory). Decide the sequence from the question, not from habit.

Can't I just ask AI or pull a big dataset and skip the talking-to-people part?

Big datasets are quantitative, brilliant at the pattern, silent on the cause. Scale doesn't change the category. A billion rows still can't tell you why someone behaved as they did; only listening to people does that. More data makes the "what" sharper, never the "why."

What's the single most common mistake?

Answering a meaning question with a counting method, or the reverse, then trusting the answer. A survey can't tell you why people churn; eight interviews can't tell you what your churn rate is. Match the method to the verb in your question first, every time.

Related in the Toolkit

Where to go next