How to do a Thematic analysis

Braun and Clarke’s thematic analysis has become a staple of qualitative HCI research. Here’s how to get started with their reflexive Thematic Analysis method and why themes don’t emerge.

In 2015 at the very beginning of my PhD, my advisor gave me a simple yet essential advice for academic writing: look at papers similar to what you want to achieve. New to qualitative methods, I analyzed a sample of qualitative papers published at CHI that year. Qualitative analysis either referenced Grounded Theory by Charmaz, thematic analysis by Braun and Clarke (B&C), or simply stated using open coding. This is overall confirmed by this review of qualitative analysis approaches in CHI and CSCW. During CHI reviewing this year, Samantha and I noticed many references to thematic analysis used language and concepts Braun and Clarke have often disavowed. They have in face expressed frustrations regarding how their paper is interpreted and used. We thought it would be helpful to summarize their recent writings on the methods, in hope it would be helpful to researchers new to thematic analysis as we were.

These past few years, their 2006 paper took off, reaching 71 739 citations according to Google Scholar at the time we write this article. Some time last year, they even gave their approach a new name: reflexive thematic analysis. They’ve also vigorously opposed that ‘themes emerge from the data’. As junior researchers, we found applying thematic analysis both easy (a way to annotate data) and difficult (there are theoretical and methodological ramifications we don’t have a clear grasp on). We’ve also kept up with discussions on thematic analysis in psychology and social sciences. Here’s a summary of Braun and Clarke’s concerns regarding uses of their approach to TA and how it applies to HCI research.

Before we continue, let’s just note this article will not get into the matter of Grounded Theory and how to do it. We’ve included more resources at the end of this article.

What is Braun and Clarke’s Reflexive Thematic Analysis and when to use it?

Reflexive thematic analysis is an approach to analysing qualitative data to answer broad or narrow research questions about people’s experiences, views and perceptions, and representations of a given phenomena. It’s theoretically flexible: it can be guided by concepts from a variety of fields, as well as being used in a variety of research approaches (inductive, deductive, semantic…). It does not pretend to be neutral: all analysis is influenced by the researcher or researchers. The result of the analysis is a theme explaining people’s experiences, perceptions, views or representations of a given topic. Braun and Clarke propose resources for both researchers and reviewers on their website. They have further discussed when (not) to use Reflexive Thematic Analysis, and how it differs from other approaches: discourse analysis, qualitative content analysis, interpretative phenomenological analysis and grounded theory.

Reflexive analysis begins with a choice of approach

Reflexive thematic analysis starts with a theme or research question and an approach to investigate it. Reflexive TA can be used for research questions or themes necessitating to “describe the ‘lived experiences’ of particular social groups” or “examine the ‘factors’ that influence, underpin, or contextualize particular processes or phenomena” (source). As such, “it is often understood as belonging to the phenomenological or experiential qualitative research tradition, common in counselling and psychotherapy research” and is not particularly adapted for summarising qualitative usability data (see below for a summary of alternative approaches).

In terms of theoretical approach to analysis, B&C’s approach is agnostic: they list possible approaches as inductive/deductive, semantic/latent, and critical realist/constructionist or a mix of those.

• Deductive/Inductive: codes would be informed by the hypothesis of the researchers and the theoretical framework. Let’s say you’re studying women’s negative perceptions of technology for mobility. This would be a deductive approach, informed by previous research about women having negative experiences of public spaces. That said, an inductive research about representations about mobility technology could reveal a pattern of gendered experiences, with women having more negative experiences than men.
• Semantic/latent: The analysis just look at what people say (semantic), or try reporting on the assumptions underpinning the data, i.e. the “ideas, assumptions, and conceptualizations — and ideologies — that are theorized as shaping or informing the semantic content of the data” (latent — source). With our example it might mean looking at how the #metoo movement shapes these representations in implicit ways.
• Critical realist/constructionist: Finally, researchers might want to make a claim about the objectivity of people’s experiences as they report it (critical realist) — or they might frame it as a study into how people perceive a situation (constructionist). The former could be done to develop guidelines for designers and policy makers, while the latter could be useful for writing participatively a position paper challenging current design or methodological approaches to this topic.

Reflexive thematic analysis is not theory-agnostic. Both the research question/theme and the approach are influenced by the researchers’ position in the academic field, including for inductive approaches, and this should be reported. For instance, maybe the inquiry about mobility technology is grounded in critical disability studies and urban theory; maybe it’s rooted in pragmatic urban sociology.

Then the data

The data suitable for this type of inquiry and research approach is qualitative. It could be diaries, interviews, surveys, data from participatory design, visual methods such as drawings and storyboards or secondary sources such as “online forums, blogs, websites, magazines, newspaper articles, and police reports”. It can also be used across datasets as pertinent for analysis across different sets of data (source). However, there’s no such thing as data saturation nor is there a strict guide about the quantity of data collected.

Unfortunately, there are no magic formulas for determining sample size in TA research!

(Source)

Braun and Clarke argue this concept is more adequate for Grounded Theory approaches or other versions of thematic analysis. When reviewing a qualitative paper using reflexive thematic analysis, the goal is to verify the study is adequately designed to answer the research question, not make sure it doesn’t miss anything. A paper isn’t the final answer to a question. Braun and Clarke do provide rules of thumb regarding the number of interviews or research data in their guide for beginners — minimum four or five dense interviews for a paper.

Coding

The coding process inherits from the approach. In an inductive approach, the research theme is progressively refined, whereas in a deductive approach, codes would be informed by the hypothesis of the researchers. It can be done at a semantic or latent level. It might also mix these approaches, hence the coding. B&C suggest deductive and latent approaches often are tied to a constructionist approach (source).

Collaboration in coding in reflexive TA is not focused on inter-reliability — but rather on how different perspectives on the same data helps reflect on codes and develop themes.

As for codebooks:

From our perspective, the use of a structured codebook, determining themes in advance of analysis or following only data familiarization (using themes as analytic inputs) and conceptualizing themes as domain summaries, delimits the depth of engagement and flexibility central to qualitative research practice.

There are forms of thematic analysis working well with the requirement that the codebook be provided (see below). In reflexive thematic analysis, more often only an excerpt of the coding and the description of the coding process provided (source).

Themes don’t emerge!

Themes are conceptualized based on the data, based on the research question — which may evolve and lead to start a new analysis process. Themes express the meanings and representations participants hold, as interpreted by the researcher who is “a storyteller [..] interpreting data through the lens of their own cultural membership and social positionings, their theoretical assumptions and ideological commitments, as well as their scholarly knowledge.” Braun and Clarke suggest it is well indicated for work with a “social justice motivation–be it ‘giving voice’ to a socially marginalized group, or a group rarely allowed to speak or be heard in a particular context, or a more radical agenda of social critique or change.” Braun and Clarke also describe them as abstract (source).

What themes are not, in the words of B&C:

The most common problem we see is topic summaries being treated as themes–a student once memorably called these ‘bucket themes’ because they’re effectively a ‘topic dump’. There’s a topic in the data, and the theme becomes everything participants said about it. That’s not how we conceptualise themes, but we see that type of theme so much, especially in applied research. Topic summary themes cluster around experiences of X, benefits of Y, barriers to Z, and so on. That type of analysis doesn’t tell a thematic story.

(Source)

A common pitfall in (reflexive TA) theme development is identifying a feature of the data, rather than meaning-based patterns.

(Source)

So for instance in HCI, reflexive thematic analysis could be applied to understand how a marginalised population perceive wearable health technologies and how this may reflect larger health inequalities, but not to simply summarize what they said about a prototype of a wearable health technology (e.g., that it was easy to use and how they would use it).

Themes are iteratively developed from the coding. For instance, related to health technology, an initial theme would be community-specific forms of humor to talk about inequalities and health technologies, but not that people used humor to talk about the topic during the interview, which is a pattern in the data. Initial themes might be mapped and linked to tell a story about the data, what Braun and Clarke call the central meaning-based concept, the story about the data. The final story relates to the research question: maybe the overall theme for the paper is the community’s role in individual’s perceptions and choices of wearable health technologies.

Checklist for authors and reviewers

• Twenty questions to evaluate the quality of a thematic analysis.
• A FAQ by Braun and Clarke

When not to use Reflexive Thematic Analysis and what to do instead?

A challenge in keeping up with evolving uses of thematic analysis is that it was designed to be versatile, to adapt to different theoretical and epistemological frameworks, and adapt to many research questions. There’s bound to be disagreements about when and how it should be applied. Braun and Clarke identify three main approaches to thematic analysis: a coding reliability approach, a codebook approach, and their reflexive approach. They also provide an overview of how thematic analysis compare to other qualitative analysis approaches, discourse analysis, qualitative content analysis, interpretative phenomenological analysis and grounded theory.

Other thematic analysis approaches

Coding reliability approach

This is closer to quantitative approaches, in that it attempts to eliminate researchers’ biases and emphasizes replicability as a main indicator of quality. Themes are hypothesis that are developed using, or checked against, the data. This analysis is adapted to ‘Understanding Users’ type of studies with research questions that don’t focus on representations, since this type of questions is a core focus of the Reflexive TA approach. Braun and Clarke point to Boyatzis (1998) as a good example of this approach. See also Grounded Theory approaches below.

Codebook approach

This approach is well suited for describing and summarizing qualitative data, and participants’ views on a topic or technology. These codebooks may be presented as a template (which can also provide an article structure), a framework (which insists on showing each stage of the analysis) or a matrix. It does not measure (inter)reliability.

Multimodal, Interaction, Interpretive and General Inductive analysis

When analyzing how people make sense of a given system while interacting with it, or make sense of information using a given system, based on observations or video recording, you could consider Interaction Analysis or multimodal analysis. Interaction analysis focuses on how participants develop knowledge or act using both social and material features of their environment — multimodal analysis emphasizes the importance of considering all different aspects and modes of communication (visual, embodied, speech…). Interpretive phenomenological analysis is another option — note that reflexive TA can be done within a phenomenological framework, but the type of paper and the write-up differ given IPA approaches rely on few and highly detailed reports of interaction. Finally, Thomas’ General Inductive Approach for Analyzing Qualitative Evaluation Data is also sometimes used.

Grounded theory

Grounded theory, as the name suggests, emphasizing the development of an explanatory model — and the themes are developed based on mapping of relations between the codes, testing alternative explanations or outliers. There’s been some push back recently that it is possible to do grounded theory with no prior knowledge or hypothesis. Useful to HCI could be this paper on GT for literature reviews, and the proposal to use flexible coding for analysing interviews or observations.

Anything else to suggest? Let us know in the comment below!

Written with the help of Samantha Finnigan! Many thanks to Ignacio Avellino and Tom Giraud for their comments and edits.

To cite: Brulé, E., Finnigan, S. 2020. Thematic Analysis in HCI. https://sociodesign.hypotheses.org/555