Chapter 15
Thematic Analysis
Helene Joffe
In Qualitative Research Methods in Mental Health and Psychotherapy: A Guide for Students and Practitioners. Edited by David Harper and Andrew Thompson,
Chichester: Wiley-Blackwell, 2012, pp. 209-223.
Description of the method
Thematic Analysis (TA) is a method for identifying and analysing patterns of meaning in a dataset (Braun & Clarke, 2006). It illustrates which themes are important in the description of the phenomenon under study (Daly et al., 1997). The end result of a thematic analysis should highlight the most salient constellations of meanings present in the dataset. Such constellations include affective, cognitive and symbolic dimensions. If one were looking at how those who do not take up the services of mental health professionals view them, for example, a thematic analysis of interviews with a carefully chosen sample of such people would reveal how they represent the various mental health professionals. This, in turn, would reveal what keeps them away from the services offered by those such as psychotherapists and psychologists. Thus a thematic analysis can tap the manifest and latent drivers concerning an issue such as uptake of mental health professional services.
Since a TA refers to themes, the notion of a theme must be examined more closely. A theme refers to a specific pattern of meaning found in the data. It can contain manifest content - that is something directly observable such as mentions of stigma across a series of interview transcripts. Alternatively, it can contain more latent content, such as references in the transcripts, which refer to stigma implicitly, via mentions of maintaining social distance from a particular group, such as certain mental health professionals. Specific criteria need to be stipulated concerning what can and cannot be coded within such themes; otherwise this form of content is highly subjective. Themes are thus patterns of explicit and implicit content. Thematic analyses tend to draw on both types of theme. Often one can identify a set of manifest themes, which point to a more latent level of meaning. The deduction of latent meanings underpinning sets of manifest themes requires interpretation (Joffe & Yardley, 2004).
A further important distinction in terms of the demarcation of a theme is whether it is drawn from a theoretical idea that the researcher brings to the research (termed deductive) or from the raw data itself (termed inductive). While theoretically derived themes allow researchers to replicate, extend and refute existing studies (Boyatzis, 1998), there is little point in conducting qualitative work if one does not want to draw on the naturalistically occurring themes evident in the data itself. So one utilises the two together – one goes to the data with certain preconceived categories derived from theories, yet one also remains open to new concepts that emerge. It is important to approach each dataset with knowledge of previous findings in the area under study to avoid ‘reinventing the wheel’. However, in addition, one wants to take seriously findings that do not match with previous frames and have the potential to revolutionise knowledge of the topic under investigation. Thus a dual deductive/inductive and latent/manifest set of themes are used together in high-quality qualitative work.
Thematic analysis has recently been recognised as a method in its own right. Previously it was widely used in psychology and beyond, often without acknowledgement or demarcation (Boyatzis, 1998; Braun & Clarke, 2006). It has also been used in this way in the evaluation of mental health services. Some argue that the ability to thematise meaning is a necessary, generic skill that generalises across qualitative work (Holloway & Todres, 2003). Like other qualitative methods, TA facilitates the gleaning of knowledge of the meaning made of the phenomenon under study by the groups studied and provides the necessary groundwork for establishing valid models of human thinking, feeling and behaviour. However, TA is among the most systematic and transparent forms of such work, partly because it holds the prevalence of themes to be so important, without sacrificing depth of analysis. Thus TA not only forms the implicit basis of much other qualitative work, it strives to provide the more systematic, transparent form of it.
Historical origins and influences
Thematic analysis is rooted in the much older tradition of content analysis (CA). TA shares many of the principles and procedures of CA, a historically quantitative tradition that dates back to the early 20th century within the social sciences, but further back in the humanities (Smith, 2000). CA involves establishing categories and then counting the number of instances in which they are used in a text or image. It determines the frequency of the occurrence of particular categories. Many content analyses rely purely on counting attributes in data (e.g. particular words or images). CA is appealing because it offers a model for systematic analysis of both elicited and naturally occurring data. It has been widely used for the analysis of mass media material. However, the results it generates have been judged as `trite' (Silverman, 1993) when they rely exclusively on the frequency outcomes it generates. It is also accused of removing codes from their context, thereby stripping data of its meaning.
The concept of ‘thematic analysis’ was developed, in part, to go beyond observable material to more implicit, tacit themes and thematic structures (Merton, 1975). For the founder of thematic analysis, Gerald Horton, such material can be termed ‘themata’ and these tacit preferences or commitments to certain kinds of concepts are shared in groups, without conscious recognition of them.
Ideally, contemporary TA is able to offer the systematic element characteristic of CA, but also permits the researcher to combine analysis of the frequency of codes with analysis of their more tacit meanings, thus adding the advantages of the subtlety and complexity of phenomenological pursuits.
Key epistemological assumptions
Thematic analysis is not tied to a particular theoretical outlook and so can be applied when using a range of theories and epistemological approaches. It is well suited to use with social phenomenology (see Fereday & Muir-Cochrane, 2006) as well as with social representations theory (SRT) (see Farr & Moscovici, 1984; Joffe, Washer & Solberg, in press). It is well matched to theories with weak constructionist (Lupton, 1999) tenets like SRT. Weak constructionism assumes that how people engage with a particular issue is socially constructed though the issues themselves have a material basis. This is broadly in keeping with the critical realist position, though with a less dichotomous view concerning the need to be either realist or social constructionist. In addition, many of the tenets of phenomenology are compatible with weak constructionism (see Willig, in press). A key feature of SRT is that it focuses on the content of people’s thoughts/feelings regarding the issue under study without reference to the ‘reality’ of the issue. For example, regarding lay conceptualisations of a health service professional, the concern is not with the accuracy of the representation but with what meanings people attach to this profession and the consequences of such meanings for themselves, for others and for the society.
TA serves as a useful tool to illuminate the process of social construction. In particular, combining thematic analyses of a range of data can trace how a particular representation develops. Mass media material (both text and image) can be thematically analysed in parallel to the TA of interviews with lay people and professional groups to examine the circulation and transformation of representations in the process of communication. Unlike cognitive approaches, which do not generally take into account the symbolic meanings that people attach to issues (Lupton, 1999), SRT in combination with TA, can provide an inroad into these symbolic meanings.
Symbolic meaning is best accessed via subtle methods. The material accessed via surveys taps consciously available cognitions that do not necessarily play the major role in driving behaviour. In other words, when explicit questions are asked one taps reason-based explanations, attitudes and beliefs, which tend to be easily accessible but may hide not only the symbolic, but also the emotional and experiential material that drives cognition and behaviour.
What kind of research questions is thematic analysis most suited to addressing?
Thematic analysis is best suited to elucidating the specific nature of a given group’s conceptualisation of the phenomenon under study. In my own work this has ranged from publics’ conceptualisations of emerging infectious diseases (EID) such as AIDS (Joffe, 1999), the Ebola virus (Joffe & Haarhoff, 2002), and MRSA (Washer et al., 2008; Joffe et al., 2010), to mass media conceptualisations of these entities (e.g. Washer & Joffe, 2006). It has been used in the mental health arena in a similar way, for example Morant’s (2006) exploration of the social representations of mental illness from the perspective of French and British mental health practitioners. I use TA to discern possible identity issues associated with the representations of each disease and their impact on lay people’s sense of personal and societal concern. More specifically, a key thread running through the EID findings is that there is a tendency to distance self and in-group from vulnerability to such diseases via a set of symbolic associations to marginalised, non-dominant groups and foreigners. The nuances of such associations are well tapped by TA, a method that can capture latent meaning while remaining systematic.
What kinds of data are most appropriate and from whom should they be collected?
Verbal interview (or focus group) data or textual newspaper data tend to be at the root of thematic research. However, open-ended responses to questionnaire items, diaries, video material, images and essays can also be thematically analysed. Interview data are usually collected via semi-structured interviews: an interview with 5-7 topics that the respondent is prompted to talk about (see Wilkinson et al., 2004). This imposes topic areas on people’s thinking, where it may be preferable to gain a more naturalistic inroad into people’s meaning systems concerning the phenomenon under study.
Instead of using topics introduced by the researchers as the basis for the interview, I have developed a more naturalistic method to elicit material. It produces data that follow the pathways of the respondent’s thoughts and feelings rather than imposing questions and topic areas. To obtain this data the meeting with each respondent begins with a task that elicits first thoughts: Respondents are presented with a grid containing four empty boxes. They are prompted to write or draw in each box any word, image or feeling that comes to mind concerning the research issue. Prior to this they are only given a very general sense of the field of study, for example, being invited to an interview on `a public health issue’ in the example that follows.
Instruction given to elicit free associations:
The following is an example of the instructions given for this grid method. The grid presented to respondents in a study of public engagement with MRSA in Britain was preceded by the following instruction ‘We are interested in what you associate with MRSA. Please list the different images and words you associate with MRSA using these boxes. Include everything you associate with one image and/or word into one box.’ (Joffe et al., 2010; see also Solberg, Rosetto, & Joffe, 2010).Once first associations have been written or drawn respondents are asked to talk about the content of each box in the order that the boxes have been completed. The aim is to elicit subjectively relevant material with a minimum of interference, to tap ‘stored’, naturalistic ways of thinking about a given topic and to then pursue the chains of association or pathways of thought that the respondents go down. Each interview is then transcribed and entered into a qualitative software package such ATLAS.ti, NUD*IST or NVivo.
In terms of who such data should be collected from, the decision concerning how many participants are required has vexed researchers who use thematic analysis. There is no notion of ‘power’ for the choice of the sample. A power analysis, for those working quantitatively, can be used to calculate the minimum sample size the researcher requires to accept the outcome of a statistical test with a particular level of confidence. The choice of sample size for a TA rests upon certain guiding principles: since the researcher is generally looking at group-based variation and/or similarity across groups, sufficient numbers of participants in each group are needed to make valid comparisons that are likely to reveal group-based threads in the data rather than idiosyncratic tangents of meaning. Furthermore, the sample size generally needs be divisible - for equal cell sizes to be used – so a primary number is not desirable. Since the idea is to look at patterning, sufficient numbers are required to discern patterns within the dataset as a whole and across sub-groups thereof. According to such criteria numbers such as 32, 48, 60 and 80 are appropriate and when work is cross-cultural one multiplies these sample sizes by the number of cultures one is studying. These are large sample sizes in comparison to most qualitative approaches. However, with the aid of computer packages large datasets can be handled. Such packages also allow for systematic examination across the data at co-occurring themes, the sequence of themes and other more complex relations between themes, in a way that would be very difficult manually.