An introduction to qualitative research: what it means to analyse qualitative data, the main analytical approaches, key concepts, and how to establish rigour in qualitative work.
Qualitative data analysis is the process of making sense of non-numerical data, interview transcripts, focus group discussions, open-ended survey responses, field notes, documents. It is not about counting or measuring. It is about understanding meaning: what people say, why they say it, and what that tells us about the phenomenon under study.
This guide introduces qualitative data analysis as a field: what distinguishes it from quantitative research, the main analytical approaches and when to use each, the core concepts you need to understand before beginning, and what rigour looks like in qualitative work. The final section describes how AnalyZ Solutions supports this kind of analysis.
The distinction between qualitative and quantitative research is not simply about whether you use numbers. It runs deeper than that: to the kinds of questions each approach is designed to answer.
Quantitative research asks: how many, how much, how often, how strongly? It works by measuring variables across a large sample and using statistical analysis to identify patterns, test hypotheses, and make generalisable claims. Its strength is precision and breadth.
Qualitative research asks: what, why, and how? It works by studying a smaller number of cases in depth: through interviews, observation, or document analysis: and using interpretive methods to understand meaning, process, and context. Its strength is depth and explanatory power.
A survey can tell you that 60% of beneficiaries report low satisfaction with a programme. It cannot tell you why they are dissatisfied, what satisfaction means to them, or what they would need for the programme to work better. Qualitative research can answer those questions.
Qualitative data is any data that captures meaning, experience, or perspective in non-numerical form. The most common types in social research and evaluation are:
Qualitative analysis is not a single method, it is a family of approaches united by a commitment to interpretive, meaning-centred inquiry. The approach you choose should be guided by your research question, your theoretical framework, and what you want the analysis to produce.
Most applied social research and programme evaluation uses thematic analysis or framework analysis, because they are practical, transparent, and produce findings that are accessible to non-academic audiences. The remaining sections of this guide focus primarily on thematic analysis.
Before beginning a thematic analysis, it helps to be clear about some foundational concepts. These are not jargon, they describe real distinctions that matter for how you design and conduct the analysis.
Inductive coding means the codes emerge from the data itself. You read the transcripts without a fixed framework and allow the language, content, and patterns in the data to guide what you notice and label. This approach is suited to exploratory research where you want to let participants' own categories drive the analysis.
Deductive coding means you bring a framework in advance: a theoretical model, a programme logic, a set of constructs from the literature, or the dimensions of your evaluation framework: and apply it to the data, looking for evidence of each construct. This approach is suited to evaluation and applied research where the questions are set in advance.
Most applied qualitative research uses a mixed approach: a starting set of deductive codes derived from the research questions, supplemented by inductive codes that emerge during the analysis. The initial framework provides structure; the inductive codes capture what the framework did not anticipate.
A code is a short label applied to a specific passage of text. It captures what is happening in that excerpt: the topic being discussed, the experience being described, the attitude being expressed. Codes are specific and close to the data.
A theme is a broader pattern that emerges across multiple codes and multiple participants. Themes answer the research question at a higher level of abstraction. A theme is something you argue from the data, it requires interpretation, not just labelling.
The relationship between codes and themes is not fixed. Some studies use a flat structure (codes are the themes). Others use a hierarchical structure (codes group into sub-themes, which group into themes). What matters is that the structure is transparent and consistent throughout the analysis.
Semantic analysis stays close to the surface of what participants said. It codes the explicit content of a passage: the participant described a barrier, mentioned a benefit, expressed dissatisfaction. This level of analysis is accessible and defensible.
Latent analysis goes beneath the surface to interpret the underlying meaning, assumption, or ideology. It asks not just what was said, but what assumptions are revealed by how it was said. This level requires more interpretive engagement with theoretical frameworks and is more appropriate for academic research than for applied evaluation.
Applied researchers typically work at the semantic level, occasionally moving to the latent level when a pattern in the data warrants deeper interpretation.
Whatever approach you use, qualitative analysis follows a broadly similar sequence. The steps are iterative, you move back and forth between them rather than following them strictly in order.
Read through all your transcripts before coding anything. Take notes on what strikes you as interesting, surprising, or significant. This immersion in the data is not a preliminary step to be rushed, it is where initial analytical ideas begin to form. Researchers who skip directly to coding without reading miss patterns that are only visible across the dataset as a whole.
Work through the data systematically, attaching short labels to relevant passages. At this stage, code generously, it is easier to merge or discard codes later than to recover passages you did not code. Do not yet try to organise codes into themes; focus on being thorough and consistent.
Review all your codes and the passages associated with them. Look for groups of codes that address the same underlying idea. Begin to construct candidate themes that answer your research question. This is the most interpretive stage of the analysis.
Test each candidate theme against the data. Does it hold up across the full dataset? Are the passages assigned to it genuinely similar? Are there passages in other themes that belong here? Revise, merge, split, or discard themes until the thematic structure accurately reflects the data.
Write a clear definition of each theme: what it is, what it captures, and what it does not include. A well-defined theme has a name that conveys its essence (not just its topic) and a definition that would allow another researcher to recognise it in the data. The theme names you use in your report should be vivid and specific enough to be meaningful to a reader who has not read the transcripts.
Qualitative findings are written as analytical narratives: structured around themes, illustrated by quotes, and interpreted in relation to the research question. The quotes are evidence; the interpretation is the finding. A report that presents quotes without interpretation is data, not analysis.
Qualitative research does not use validity, reliability, and generalisability in the same way quantitative research does: but it is no less concerned with rigour. The key criteria are different:
Are the findings an accurate representation of participants' experiences and perspectives? Credibility is established through prolonged engagement with the data, triangulation across sources or methods, member checking (sharing findings with participants), and peer review. In AnalyZ Solutions, analytical memos stored alongside each code create a transparent record of how interpretation developed.
Can the findings be applied to other contexts? Qualitative research does not generalise statistically, it offers thick description: enough detail about the study context and participants that readers can judge whether the findings might apply to their own situation. This requires clear reporting of sampling strategy, participant characteristics, and data collection procedures.
Would another researcher following the same process reach consistent findings? Dependability is established through transparent documentation of the analytical process: a clear audit trail from data to findings. This is why codebook documentation, analytical memos, and inter-rater reliability checks are important, not merely procedural.
Are the findings grounded in the data rather than in the researcher's preconceptions? Confirmability requires reflexivity: an explicit acknowledgement of how the researcher's background, assumptions, and theoretical commitments may have shaped the analysis. It is not a claim to objectivity, qualitative research does not pretend to that: but a transparent account of the interpretive standpoint from which the analysis was conducted.
Using a chatbot to "find the themes" in your data is not qualitative analysis. Several specific problems arise:
AnalyZ Solutions uses AI only to assist with synthesis after the researcher has completed the analysis, and never receives raw transcript text.
Creating a separate code for every slightly different phrasing results in a codebook so fragmented that patterns become invisible. If two passages feel different but you cannot articulate how, they probably belong under the same code. Group related content together and use the analytical memo to note internal variation.
Skipping passages that seem peripheral or difficult to code is a common source of analytical blind spots. Code generously in the first pass, then review for gaps. The passages that resist easy coding are often the most analytically interesting.
Code definitions that change implicitly during analysis, without the researcher noticing, produce inconsistencies that undermine the credibility of the findings. When you change a code's definition, document the change and go back to recode earlier transcripts under the revised definition.
Looking only for evidence that confirms your hypotheses and overlooking contrary evidence is a risk in all research, but it is particularly acute in qualitative research because the data is rich enough to support almost any interpretation if you select carefully enough. Actively look for passages that contradict your emerging themes. Disconfirming cases are analytically productive.
AnalyZ Solutions provides a single browser-based environment for the full qualitative analysis workflow. Everything happens in your browser, no data is sent to a server, and no transcript text is ever shared externally.
The qualitative modules map onto the analytical process described above:
Watch a quick snippet of qualitative analysis in AnalyZ Solutions
Upload your transcripts and begin coding in AnalyZ Solutions.
Get started