Qualitative Methods of Analysis: 7 Approaches and How to Choose

Qualitative analysis is the work of turning non-numerical data, such as interview transcripts, open-ended survey answers, support chats and session recordings, into patterns you can explain and act on. The method you pick depends on your question. Seven methods cover most work: thematic, content, narrative, discourse, grounded theory, framework analysis and interpretative phenomenological analysis. For most product and UX questions, thematic analysis is the right starting point, and the rest are specialist tools you add when the question demands them.

This guide explains what qualitative data analysis is, compares the seven methods, walks through the steps they all share, and shows how to keep your conclusions defensible when no p-value is available to do it for you.

What qualitative data analysis actually is

Quantitative analysis asks how many and how often. Qualitative analysis asks what is happening, why, and what it means to the people involved. The data is words, images, behavior and context, and the analysis is iterative and interpretive: it focuses on meaning, not numerical frequency.

Two consequences follow.

First, you are not extracting truth from a dataset. One recent reflection on research practice argues that analysis is closer to co-constructing meaning with participants, and that themes are inferences the researcher develops, going beyond the raw data. That is why they need validating.

Second, the process rarely runs in a straight line. Collection and analysis often overlap, and early findings are commonly used to refine later interview guides instead of waiting for a finished dataset. If your fifth interview shows the first four missed a question, you fix the guide. You do not wait.

The field is also not limited to academia. The CDC’s field epidemiology manual, for instance, treats qualitative work as a rigorous way to answer specific questions in public health investigations, and notes that good analysis needs enough time and immersion in the data. Product teams face the same trade-off: depth costs time.

The seven main qualitative methods of analysis

The table summarizes what each method is best for. Details follow.

The fit descriptions above follow the ResearchProspect overview of the seven methodologies. The effort column is our own judgment, not a published rating.

Thematic analysis

You read, code, and group codes into themes. It is flexible, works across most disciplines, and does not require you to adopt a theoretical stance first. If you have forty interview transcripts and a question like “why do trial users stall during setup?”, this is where to begin.

A widely used structure is the Braun and Clarke six-phase model, introduced in their 2006 paper. Its phases run from familiarizing yourself with the data through coding and theme building to writing up.

Content analysis

Content analysis sorts material into categories and often counts how frequently each appears. It suits large, structured sets of text or media where prevalence and replicability matter. Think of coding 2,000 open-ended NPS comments into categories like “pricing,” “onboarding” and “missing integration,” then comparing how often each appears month over month. It sits at the border of qualitative and quantitative work.

Narrative analysis

Narrative analysis treats a story as the unit of analysis: its sequence, turning points and the way the person frames events. It works when the order of events matters, such as how a customer describes the path from first frustration to switching tools.

Discourse analysis

Discourse analysis examines how language constructs meaning, including who gets to speak, how problems are framed and which words signal authority. In product work it is rarer, but useful for reading how a community talks about your category in forums or how support agents and customers talk past each other.

Grounded theory

Grounded theory builds a theory from the data instead of testing one you brought. You collect, code, compare and collect again until new data stops changing the picture. Choose it when the area is poorly understood and no existing framework fits.

Framework analysis

Framework analysis uses a matrix: rows are cases, columns are themes. Because the structure is explicit, several people can work on the same dataset and reach comparable results. That makes it a strong choice for applied teams with a defined brief and a deadline.

Interpretative phenomenological analysis

IPA goes deep on a small number of people, often fewer than ten, to understand how they make sense of a specific experience. It is slower and demands careful interpretation. Use it when you need to understand one significant experience, such as how clinicians or administrators feel when adopting a new system, rather than map a whole user base.

How to choose a method

Start with the question, not the method.

• “What themes come up across our users?” Thematic analysis.

• “How often do users mention X, and is it growing?” Content analysis.

• “What happened, in the user’s telling, from start to finish?” Narrative analysis.

• “We have no framework for this behavior.” Grounded theory.

• “Five researchers need to synthesize the same interviews by Friday.” Framework analysis.

• “We need to understand a few people’s experience in depth.” IPA.

A lighter ladder also exists. One university course handout lays out options from simpler to more demanding: identify critical incidents, simple categorization, then categorization into themes and patterns. If you are short on time, a critical-incident pass (find the moments where things went badly wrong or unusually right) gives you real insight in an afternoon.

The steps almost every method shares

The methods differ in philosophy, but the working sequence is similar. The standard core, as summarized by ResearchProspect, looks like this:

1. Organize the data. Transcribe, label files, and put everything in one place.

2. Immerse yourself. Read or watch repeatedly before you code anything.

3. Code. Attach short labels to segments of data and keep a codebook that defines each label.

4. Build categories and themes. Group related codes into patterns, and name each pattern in a sentence, not one word.

5. Interpret and report. Draw conclusions, test them, and write up what you found.

Iteration runs through all five. Initial coding often shows that a code is too broad, or that you need another interview to understand a case.

A worked example

Say you have twelve onboarding interviews for a project management tool. Early codes might be “unclear first step,” “invited teammate but nothing happened,” and “imported spreadsheet, columns wrong.” Grouped, the first two become a theme: users do not know what to do after the empty state. The third belongs to a different theme about data import. Notice the theme is an interpretation. The transcripts never say “empty state problem.” You inferred it, which is why you then check it against more data and against the participants themselves.

Analyzing behavioral data, not just transcripts

Product teams hold qualitative data that is not text: session recordings, heatmaps, screenshots, click paths. Text-first coding methods such as open and axial coding struggle with this kind of mixed UX material. Design researchers instead use visual and participatory approaches, where triangulation happens through color-coding and mapping rather than text labels alone.

In practice that means:

• Watch recordings in batches of 10 to 15 around one question, such as users who abandoned the checkout form.

• Note each friction moment as a short code (“rage clicks on disabled button,” “scrolls past pricing, returns”).

• Put the codes on a board and cluster them visually, the way you would sticky notes from a workshop.

• Check the pattern against numbers: how many sessions show it, and does the funnel agree?

That last step is where qualitative and quantitative work meet. If you want more on pairing the two, our piece on the sweet spot between quantitative and qualitative analysis goes deeper, and our guide to qualitative UX research covers how to collect the data in the first place.

It also helps to treat observed behavior as one lens among several. Connecting what people do with how they describe it is a common thread in work on understanding user behavior, and it is what keeps a recording from being read as a verdict.

Keeping qualitative analysis rigorous

No p-value rescues a weak qualitative study. Rigor comes from verification strategies that you build into the work. The standard set, described in a book on challenges and solutions in qualitative research, is:

• Triangulation: corroborate findings with more than one source or method, for example interviews plus session recordings.

• Member checking: return your interpretations to participants and see whether they resonate.

• Audit trails: document each analytical decision so someone else can follow it.

• Reflexivity: record your own assumptions and how they may shape your reading.

• Peer debriefing: have someone outside the study challenge your conclusions.

Two cautions keep these from becoming rituals.

Member checking does not mean asking people “do you agree?” The better version shares categories or interpretations and asks whether they match the person’s experience. Participants sometimes offer a view that changes your analysis, and that is the point.

Reflexivity is also not a check-box. A short reflexive journal kept during the project, noting what surprised you and what you expected, does more than a paragraph of declared bias at the end.

Finally, following procedure does not by itself guarantee rigor. Peer reviewers point out that rigor lives in consistent analytical logic, conclusions that are justified, and alternative explanations that were actually considered. If you can say “here is the competing explanation and here is why I rejected it,” you are on solid ground.

If your findings feed a formal report, the SRQR reporting standards give a checklist of what to document, from the approach to the analytic procedures.

Our perspective: analysis is only useful if it changes a decision

We build a product analytics platform, so we see qualitative analysis from the product side. Three things stand out.

The most common failure is not a wrong method. It is analysis that never reaches a decision. A beautifully coded set of interviews that nobody on the roadmap team reads is a cost, not an insight. We would rather see a team run a rough thematic pass on eight sessions and ship a fix than polish a codebook for a month.

The second is that qualitative and quantitative evidence answer different halves of the same question. Numbers show you where users drop off. Watching the sessions, and talking to people, shows you why. Treating one as the “real” data and the other as color tends to produce either confident wrong answers or interesting stories with no scale.

The third is that behavioral data deserves the same discipline as interviews. A recording is not self-explanatory. Code what you see, cluster it, check it against counts, and ask a second person whether they read it the same way.

Putting it to work in LiveSession

If your qualitative data includes what users do inside your product, LiveSession gives you session replay, heatmaps and click maps, and a product analytics dashboard so you can code behavior and then check how widespread it is. There is a Free plan to start with, and you can compare the Basic, Pro and Enterprise options on the pricing page.

Common mistakes to avoid

• Coding before reading. Without immersion, your codes reflect your first impression, not the data.

• Themes that are just topics. “Onboarding” is a topic. “Users cannot tell what to do after signing up” is a theme.

• Ignoring sampling. If your participants are only your most engaged users, your themes will be too. Our article on selection bias in user research shows how it creeps in.

• Counting as proof. Frequencies of codes help, but a rare, severe issue can matter more than a common, mild one.

• Skipping the audit trail. Six months later, nobody remembers why two codes were merged.

For related method choices on the collection side, see our overview of usability testing methods and the wider user experience analysis process.

FAQ

What is qualitative data analysis?

It is the process of organizing, coding and interpreting non-numerical data such as interviews, open-ended answers, observations and recordings. The goal is to understand meaning and context, not to measure frequency. The output is usually a set of themes or a theory, supported by quotes and examples.

What is the difference between qualitative and quantitative analysis?

Quantitative analysis measures how much or how often, using numbers and statistics. Qualitative analysis explains why and how, using words, behavior and context. Many teams use both: numbers to find the problem, qualitative work to understand it.

Which qualitative method is best for beginners?

Thematic analysis. It is flexible, does not demand a specific theoretical framework, and follows the same core steps as other methods: organize, read, code, build themes and report. Once you are comfortable, you can move to framework analysis for team projects or grounded theory for new territory.

How many interviews do I need?

There is no universal number, because it depends on method and question. IPA often works with a handful of participants, while content analysis can handle thousands of comments. In iterative work, many researchers keep collecting until new data stops changing the themes.

Sources

• Qualitative Data Analysis: Methods & Steps, ResearchProspect

• Qualitative data analysis: reflections, procedures, and some points for consideration, Frontiers

• Collecting and Analyzing Qualitative Data, CDC Field Epi Manual

• Braun and Clarke’s Six-phase Framework, Vappingo

• Qualitative analysis techniques, course handout

• A Designerly-Way of Conducting Qualitative Research in Design Studies

• Challenges and Solutions in Qualitative Research

• Validity strategies in qualitative research, KCI

• Standards for Reporting Qualitative Research (SRQR)