Thematic analysis: Choosing a suitable approach


Thematic analysis (TA) is a widely used method for making sense of qualitative data such as interview or focus group transcripts or written responses to open-ended survey questions. Although TA is often thought of as a singular approach, it is better thought of as a family of methods – with some characteristics in common and areas of disagreement. The diversity within the TA family is often poorly understood, which can lead to confusions that compromise the quality of research. The diversity also raises the question of how researchers chose a suitable approach for a particular research project? Virginia Braun and Victoria Clarke guide you through your options.

What is Thematic Analysis?

What thematic analysis (TA) is – is pretty straightforward, right? You use coding to organise data into themes, select some data excerpts to illustrate the themes, write up your report and done! 
In practice, things are a bit more complex. For starters, although TA is often written about as if it is one method, with a standardised set of procedures, it isn’t – there are numerous different approaches to TA, with TA methods codified from the 1990s onwards. But even before then, researchers were reporting the “themes” that emerged from their qualitative data, or were using analytic techniques from grounded theory (such as line-by-line coding and constant comparison) to identify and report themes in qualitative data. Given a messy history, it’s no wonder that there is considerable scope for confusion around what TA is, and potential to do it badly.
The term “thematic analysis” is better thought of as an umbrella term for methods that involve processes of coding and theme development or identification (this distinction in terminology matters, as we’ll go on to discuss). TA methods typically acknowledge:

  • two different types of coding – coding for semantic/manifest/surface meaning and coding for latent/implicit/hidden meaning, 
  • and two different orientations to analysis - inductive or deductive.

TA methods are also distinguished by their flexibility – they can be used to address a wide range of research questions and analyse almost any type of data. TA methods differ from approaches like grounded theory and discourse analysis, which offer a wholesale approach for doing a research project, not just data analytic techniques.

There are  four important areas of difference in the TA family that we will highlight here:

1. the research values that ground and give validity to TA; 
2. understandings of researcher subjectivity; 
3. what constitutes effective coding; and 
4. the conceptualisation of themes. 

Having a sound understanding of these differences helps researchers to choose a suitable approach for a particular research project and avoid common problems in TA research.

Research values
Academic discussions about the philosophies and theories underlying research – paradigms and things like methodology, ontology and epistemology – can seem lofty, abstract, impossible to understand, and far removed from the practicalities of actually doing research. We think these theories or research values are actually very important, as they help us to decide how exactly we go about doing practical things like, for example, data coding, and determining what role we as researchers should have in the research process. But we prefer to think of theory as something practical, rather than something esoteric and entirely separate from the doing of research. 
This means that even if we imagine ourselves as just getting on with the practical business of doing TA, and not getting bogged down in impenetrable literature about research values, theory is still part of our research practice. 
Even if we’re not conscious of doing so we make assumptions about things like our role in the research process (neutral observer? impassioned story teller?), and what our data represent or give us access to (participants’ inner psychology? socially dominant sense-making?). These will be theoretically grounded assumptions but it can be hard to recognise them as such, especially when our assumptions align with dominant understandings. Our position is that there is a lot less potential for confusion and problems if we have a handle on the research values that inform our use of TA.
Researcher subjectivity
We like psychotherapy researcher Linda Finlay’s distinction between TA researchers as scientists or artists. If researchers approach TA as a scientist, certain values shape their research, how they think about their role in the research and the procedures they use. What Finlay terms ‘scientifically descriptive’ approaches to TA share many of the concerns of quantitative research – with things like the reliability and replicability of the findings, striving for objectivity and keeping researcher bias in check. 
What Finlay terms ‘artfully interpretative’ approaches to TA reject the norms and values of science as a model for good practice in research. Instead, researchers approaching TA as an artist embrace the subjectivity of research, and understand themselves as telling artfully interpretative stories about data rather than seeking to scientifically describe the truths within data. 
So, a first step for choosing a TA approach is reflecting on your research values and how you (perhaps implicitly) conceptualise yourself as a researcher – as a scientist or an artist, or something else? How we conceptualise our role in the research is intertwined with our research values. Do we imagine our subjectivity, our positionings, worldviews, experiences and values, as a threat to the reliability and objectivity of the research, something we should seek to contain and control as much as possible? Or as something to be embraced, something valuable, essential even, to the doing of the research? 
The researcher and the coding process
As we’ve already noted, for scientifically descriptive TA, researcher subjectivity is a potential source of bias and distortion. This understanding of researcher subjectivity mandates the use of particular coding procedures – such as the development and agreement of a fixed and delineated codebook or coding frame which directs coding, and having at least two researchers independently code the data using this codebook. A (statistically-tested) high level of agreement between coders is treated as evidence of reliable coding. The scientifically descriptive suspicion of researcher subjectivity is also evident in practices such as theme agreement or consensus – where two or more researchers agree the final set of themes. 
In artfully interpretative TA, researcher subjectivity is understood very differently, as a resource rather than a threat, and as something inevitably and inescapably part of the research. It can’t be separated out, or contained and controlled. From this perspective, practices like fixed codebooks, multiple independent coders, and theme consensus simply don’t make sense. 
When coding and theme development procedures embrace researcher subjectivity (such as in the reflexive TA approach we have developed), coding is an open, organic and relatively unstructured process, and there’s no requirement for two or more researchers to code the data (though they can). 
The researcher creates codes and coding labels – pithy phrases – that capture what’s of interest, and tags the data with these. These codes are often revised as coding develops: tweaked, collapsed with other codes, and split into two or more codes... This is because codes evoke the researcher’s “take” on the data, so often evolve as the researcher’s analytic take evolves. 
If two or more researchers are involved in coding this is understood as a collaborative process - researchers act as critical friends for each other, encouraging reflection on assumptions shaping and potentially limiting data interpretation - rather than one oriented to agreement or consensus.


The revised name we’ve recently given our approach – reflexive TA – emphasises both: 
A. the researcher’s (inescapably) active role in the analytic process, and 
B. the need for the researcher to reflect on and interrogate their assumptions, and decisions, and the ways these might have delimited their analysis, closing off interpretative possibilities.

What is a theme?

Differences are also evident in the TA family in how themes are conceptualised. When we first wrote about TA we drew on US educator researcher Richard Boyatzis’s definition of themes as patterns of meaning across a dataset. We assumed everyone interpreted this as we did – but we were wrong. Very wrong! This prompted us to dig deeper into different understandings of themes within TA, and we discuss three here:

  • themes as diamonds in the sand, 
  • themes as topic summaries, and
  • themes as patterns of shared meaning. 

By themes as diamonds we refer to an often implicit understanding of themes as real things in data, that pre-exist the analysis. This understanding is evident in descriptions of themes as things that were identified, found or discovered, and in the framing of themes as “emerging” from data like Venus arriving at the shore having “emerged” from the sea in Botticelli’s famous painting. Scientifically descriptive TA researchers tend to write about themes in this way, as if they are entities separate from the researcher, which the research uncovers.

Themes as topic summaries describes an understanding of themes as containers – like a bucket – for data relating to a particular topic or issue. In topic summary themes, quite meaning-disparate data can be reported, including very divergent views.  

A classic topic summary theme would be “barriers to the implementation of policy X”. The ‘theme’ would report all the data relevant to the barriers, some of which might be divergent. Different barriers might be organised into sub-themes (barrier A, barrier B etc.). Such ‘themes’ tend not to tell a story about the data, but summarise and report. 

Because they centre a topic in the data, topic summary themes can usually be ‘identified’ or established before data coding, and they often map very closely onto research and/or data generation questions. In scientifically descriptive TA, topic summary themes are common, and coding is often understood as a process for allocating data to these predetermined themes, hence the phrasing “coding for themes”.

Finally, themes can be understood as patterns of shared meaning organised around a central idea or concept – this is how we understand themes in reflexive TA, and how we naively assumed everyone else understood themes when we first wrote about TA! 

In this understanding, themes can also draw together seemingly disparate data, but there’s an idea and a story that unites all the observations. To imagine a shared meaning theme, consider something like a sunflower, with a central core, with multiple “petals” attached. These shared meaning themes capture multiple facets and variations of meaning related to the central idea, like a sunflower has multiple petals. 

Let’s consider an example from our in-progress research exploring young adults’ meaning-making around physical disability and dating, using the story completion method. We gave participants the start of a story based on a hypothetical scenario in which two people - one a wheelchair user - have been set up on a date by a mutual friend and invited them to complete the story. 

When analysing the stories written by participants, we created a theme called “the good disabled person”. This theme tells a story of the way dates which had an imagined happy ending often relied on a portrayal of the disabled dating partner as having triumphed over the adversity of their disability, demonstrating a positive and optimistic outlook, having a fun personality, and an endless capacity for performing emotional labour around their disability. We used this theme name because demands to be the “good X” are shared across minoritised identities in our cultural context (the good gay, the good Muslim, etc). 

Because shared meaning themes tell interpretative stories, rather than identify and summarise data relating to a topic, they cannot be developed until the researcher has engaged in considerable analytic work. This means it’s impossible to “code for themes” in reflexive and artfully interpretative TA –  themes are only developed from and through coding. So, a core methodological difference is that whereas scientifically descriptive TA tends to move from themes to coding, artfully interpretative TA moves from coding and codes to themes.

Deciding on a TA approach that’s right for you

Different conceptualisations of themes (and the wider TA process) connect to the goals and purpose of the research being done. In research with a clearly delineated analytic purpose – such as identifying the barriers and facilitators to the implementation of “policy X” – an approach to TA that conceptualises themes as topic summaries might be most appropriate. By contrast, reflexive and artfully interpretive TA is best suited to research with a more open and exploratory purpose, without a pre-determined thematic structure.

To help readers identify and choose between different TA approaches, we have developed a tripartite distinction in TA approaches. 

  • Those we term ‘coding reliability TA’ align with Finlay’s scientifically descriptive approach - examples include the approaches developed by US education researcher Richard Boyatzis and US public health researchers Greg Guest and colleagues.
  • We also designate ‘codebook TA’ approaches that occupy somewhat of a middle-ground between these. 

These ’codebook TA’ approaches are sometimes referred to as TA, but often by other names such as framework analysis or template analysis. These “middle-ground” approaches typically use a structured coding approach (the use of a codebook or coding frame), and early theme development with reporting of topic summary type themes – in common with scientifically descriptive/coding reliability TA. 

But they combine this with the embracing of researcher subjectivity evident in artfully interpretive/reflexive TA, and wider qualitative research values. Codebooks also tend to be used to map or chart the developing analysis, rather than for the purpose of measuring and facilitating intercoder agreement. 

To read more about different approaches to TA, you can explore our website (which also includes resources for quality standards for, and evaluation of, TA), or our book Thematic Analysis: A Practical Guide.

About the authors:

Virginia Braun (she/her, they/them) is a Professor in the School of Psychology/Te Kura Mātai Hinengaro, Waipapa Taumata Rau/The University of Auckland, New Zealand. Victoria Clarke (she/her, they/them) is an Associate Professor in Qualitative and Critical Psychology in the School of Social Sciences at the University of the West of England, Bristol, UK. They are the authors of numerous outputs on thematic analysis including the award-winning and best-selling book Thematic Analysis: A Practical Guide with an extensive open-access companion website (2022, Sage). Emails: [email protected] [email protected]