Scientific models for qualitative research: a textual thematic analysis coding system – Part 1
Intended for healthcare professionals
Evidence and practice    

Scientific models for qualitative research: a textual thematic analysis coding system – Part 1

Frederik Alkier Gildberg Professor, Forensic Mental Health Research Unit, Middelfart, Faculty of Health Science, Department of Regional Health Research, University of Southern Denmark, Denmark
Rhonda Wilson Professor, School of Nursing and Midwifery, The University of Newcastle, Callaghan, NSW, Australia

Why you should read this article
  • To add a high degree of trustworthiness and rigour to your use of thematic analysis

  • To discover how to move your thematic analysis beyond thematic maps and colourful illustrations, to building and learning from models

  • To learn how to rigorously account for scientific models developed from qualitative data

Background Models are central to the acquisition and organisation of scientific knowledge. However, there are few explanations of how to develop models in qualitative research, particularly in terms of thematic analysis.

Aim To describe a new technique for scientific qualitative modelling: the Empirical Testing Thematic Analysis (ETTA). Part 2 describes the ETTA model.

Discussion ETTA generates a semantic structure expressed through theme-code, content and functionality. It highlights the importance of authenticity markings and taxonomical and functional semantic analysis. Its primary advantage is the sequential need to account for taxonomic analysis, functionality factors, preconditioning items, cascade directories and modulation factors; this results in the production of a sound, systematic, scientific development of a model.

Conclusion ETTA is useful for nurse researchers undertaking qualitative research who want to construct models derived from their investigations.

Implications for practice This article provides a step-by-step approach for researchers undertaking research that culminates in the construction of a model derived from qualitative investigations.

Nurse Researcher. doi: 10.7748/nr.2023.e1860

Peer review

This article has been subject to external double-blind peer review and checked for plagiarism using automated software

Correspondence

fgildberg@health.sdu.dk

Conflict of interest

None declared

Gildberg FA, Wilson R (2023) Scientific models generated through a textual thematic analysis coding system – Part 1 Nurse Researcher. doi: 10.7748/nr.2023.e1860

Acknowledgement

The authors would like to thank RD Nissen, JH Kerring, ALW Pedersen and colleagues for contributing with reflections, critical reading and debates of this paper, all the way from draft to finished manuscript. The sharp pen of Roman Frigg has not gone unnoticed in this process. This work is dedicated to the memory of D Bailer-Jones.

Published online: 31 May 2023

Introduction

Theory and logic are verified through empirical inquiry - the rigorous verification by observation, experience or experiment: a thorough examination of phenomena and the conditions in which they occur underlie the scientific process of investigation and knowledge formation. Braun and Clark (2006) built on a long tradition of rich descriptions of thematic analytic techniques, introducing a systematic, stringent alternative version of thematic analysis that has become increasingly popular with qualitative researchers (Braun and Clark 2019). But how can a textual, thematic analysis coding system account for the development of scientific models, and why would that be important to qualitative inquiry in nursing, social, health or medical sciences research? Detailed presentations of how to develop a scientific model that adequately accounts for functionality while using thematic analysis seem to be extremely rare. The aim of this article is to present an approach to scientific qualitative modeling using our Empirical Testing Thematic Analysis (ETTA) technique.

Key points

  • Using ETTA places importance on unearthing and testing potential ‘taxonomic relations’ between themes and sub-themes

  • Using ETTA means identifying and describing ‘functionality factors’ – associations between themes that can be identified as a function or the functional relationship between them that logically describes how either or both are perceived to operate

  • The use of ETTA locates and identifies ‘preconditioning items’ – the conditions that must be in place for the functionality factor to take place

  • Using ETTA to build models means identifying and tracing ‘cascade directories’ – inventories of sequential and spreading effects of functionality factors in a model

  • The use of ETTA means placing importance on identifying ‘modulating factors’ – themes or theme functions that manipulate or modulate other functionality factors or their effects

Empirically testing thematic analysis (ETTA)

ETTA consists of eight steps (Figure 1) that are consistent with Blumer’s (1986) basic requirements for an empirical science. We will emphasise the seventh and eighth steps, as Gildberg (2015) presented the first six steps in greater detail.

Figure 1.

The eight steps of ETTA

nr.2023.e1860_0001.jpg

Step 1. First reading

The aim of the first step is to answer the question ‘what is this text about?’, to ensure that the macro text structures – for example, emerging macro themes – and the proportions of overall theme connections and functions are not overlooked when subsequently developing the coding in the third step.

To do this, the data are transcribed and read through to establish an overview of the overall structure and any emerging themes (Zoglowek 1999, Morse et al 2002). Notes are systematically labelled with a subject heading (‘code’) and notated with an authenticity marking to indicate the degree of abstraction from the data (Spradley 1980, Whittemore et al 2001, Gildberg 2015)(Box 1). A codebook is developed by iterating back and forth between code, inclusion criteria and data, similar to Braun and Clark’s version of thematic analysis (Braun and Clark 2006, Braun et al 2019). The type of marking is noted and kept throughout the analysis, so that it can be used as an element in the functional semantic analysis:

  • 1. This indicates content or condensation and code that are close to the words, concepts and terms the participants used rather than as abstract interpretations – for example, little interpretation or elaboration is made regarding the original data, rather than literature or an existing theory being referenced.

  • 2. This is assigned to condensations and codes that include empirical content as well as analytical abstractions and/or propositions of relationships.

  • 3. This is given if words, concepts and terms that the participants used are condensed and coded as abstract.

A codebook is developed by iterating back and forth between code, inclusion criteria and data, similar to Braun and Clark’s version of thematic analysis (Braun and Clark 2006, Braun et al 2019). This ensures all the codes and notes generated during analysis are labelled with an authenticity marking score. The type of marking is noted and kept throughout the analysis, so that it can be used as an element in the functional semantic analysis.

Step 2. The analytical question

After the ‘first reading’ an analytical question is constructed from the project aim, exclusion/inclusion criteria, operationalised concepts and terms, theory and positioning, semantics, and other attributes belonging to the abstract problem being investigated (Gildberg 2015). This is done to ensure data are stringently decontextualised from the empirical material by providing an answer to the study’s aim (Morse et al 2002).

Step 3. Coding

The decontextualisation of data relevant to the study aim is carried out by reading through the data and coding any data that represents an answer to the analytical question. A mark is used – [AB248188-195] in Figure 2 – to identify specific data or type of data used in the original material. The mark follows the data through analysis, making it possible to return to the original source at any point.

Figure 2.

Example of coding and decontextualisation

nr.2023.e1860_0002.jpg

Step 4. Condensation and authenticity markings

The original text is then interpreted into an essential condensed text that answers the analytical question; an authenticity marking is also assigned to it. Codes or subject headings are developed using a codebook for initial processing. The data are systematically searched with the intention of gathering ‘necessary data through careful and disciplined examination’ of the textual data (Blumer 1986).

Step 5. Categorisation

After coding and condensing, the established codes are sorted alphabetically into categories.

Step 6. Thematisation of categories

Overlapping with Step 5 is the immediate thematisation of the categories, in which categories with identical codes are merged into a coherent theme under the same subject heading (Gildberg 2015). This is done by moving between data, theme and first reading, looking at how the theme interrelates with or diverges from other themes using clear, explicit exclusion and inclusion criteria and how the merged categories fit or do not fit the themes, while refining the logic of relationships to other codes (Buetow 2010). The themes are then tested against the original text to avoid skewed representations and to ensure unexplained variance is accounted for (Whittemore et al 2001).

This form of inspection should result in a nuanced descriptive answering of the analytical question using themes (Blumer 1986). Researchers are encouraged to keep a reflective and self-critical record of their work and decisions throughout analysis.

Step 7. Taxonomical relations

In this step, themes represented by their subject headings are analysed to unearth and test potential taxonomic relations between themes and sub-themes (Blumer 1986, Whittemore et al 2001). To do this, the semantic relationship ‘A [is a part of] B’ and the structural question ‘what are all parts of B?’ (Spradley 1980) are used, where B is the main theme and A is the unknown sub-theme (Gildberg 2015).

This is done systematically by placing the subject headings into a matrix, where a column corresponds to a potential theme and a row to a potential sub-theme of that theme (Figure 3) (Ritchey 2018). Illogical taxonomic entries such as ‘A is a part of A’ and ‘B is a part of B’ are shaded to indicate they have been removed from the analysis. Every affirmative answer to ‘Is row X [a part of] column Y?’ is marked ‘X’.

Figure 3.

Subject heading classification matrix

nr.2023.e1860_0003.jpg

The matrix then enables the theme hierarchy to be derived (Figure 4). The main themes are initially located by identifying blank rows in the matrix. Row A is blank in Figure 3, so theme A is a main theme. Sub-themes of the main themes are then identified by finding cells in the corresponding columns that contain an X. For example, rows F and H in Figure 3 contain an ‘X’ in column A, so F and H are sub-themes of A. Sub-themes of F and H can then be identified by searching for ‘X’ in columns F and H, and so on.

Figure 4.

Theme hierarchy

nr.2023.e1860_0004.jpg

In this example, a ‘mono-dimensional’ theme string was identified – that is, the sub-themes all connect to only one main theme. However, themes may also connect across several strings and this approach can also identify such multi-dimensional theme strings and their crossover semantic relationships. Rows identified as blank will indicate main themes, and a search for ‘X’ in the matching column will reveal subthemes that were identified as a part of the main theme during the taxonomic analysis. In this way, by using the semantic analysis ‘A [is a part of] B’ as a way of generating potential hierarchy propositions, all identified themes are searched sequentially and tested against each other, using the matrix logic. Thereby themes are assigned a location into a mono-dimensional or multi-dimensional structural position in the model under development (Figures 3 and 4).

Step 8. Functional semantic analysis

The final step (Step 8) is to carry out a functional semantic analysis on the established model structure to identify and formulate propositions regarding functional relationships between themes and to weave such propositions into a ‘theoretical scheme’ (Blumer 1986, Morse et al 2002). Functional semantic analysis of the established model structure and the matrix created in the previous step are used to account for the following issues:

Functionality factors

Functionality factors are the associations between themes that can be identified as function or the functional relationship between them that logically describes how either or both are perceived to operate. They can relate to manifest variables (authenticity 1) and latent variables (authenticity 3). They are identified by looking for the functional semantic relationship using the matrix created in the taxonomical analysis.

For example, suppose that in Figure 5, theme ‘G’ is ‘perceived threat’ and theme ‘I’ is a ‘perceived fear’. Looking through data (codes, transcripts, first reading) for the semantic relationship A [is a function between] X and Y and using the cross subclassification matrix, a proposition grounded in the data could be: ‘An increase in G’ leads to ‘an increase in I’, thereby formulating the simple qualitative associational proposition that informants perceived that an increase in ‘threat’ brings about an increase in perceived ‘fear’.

Figure 5.

Multi-dimensional theme strings

nr.2023.e1860_0005.jpg

Functionality factors can relate to both manifest variables (authenticity 1) and latent variables (authenticity 3) and are identified by looking for the functional sematic relationship through the cross subclassification matrix. For each identified functionality factor, preconditioning items are identified by searching for B [is a precondition for] A in the data regarding the specific function.

Preconditioning items

‘Preconditioning items’ are conditions that must be in place for the functionality factor A to take place. For each functionality factor identified, preconditioning items can be found by searching for ‘B [is a precondition for] A’ in the data regarding the specific function.

In the example shown in Figure 5, themes E and H might have to be in place for the function between G and I to take place; if we take away E or H, the function should no longer be possible.

Preconditioning items are co-determinant of and co-define observable functionality characteristics. They can also act as markers of underlying functionality variables.

Cascade directories

Preconditioning items precondition the ‘cascade’ from one theme to the next so are also found in ‘cascade directories’. Figure 6 shows an example of a cascade in which theme A sets off theme G, which makes theme D possible for an effect on theme I. ‘Cascade directories’ are inventories of sequential and spreading effects of functionality factors and enable new preconditioning factors to precondition the next function in the cascade and illuminate new functionality factors.

Figure 6.

Locating cascade directories and modulating factors

nr.2023.e1860_0006.jpg

By searching through the themes and their functional relationships for the semantic relationship ‘C [ is the cascade directory of] A’, any cascade in functions will be noted in the data. Considering the possible cascade directories allows for logic that describes how models form sequentially or as maps, thereby tracking the cascade of functionality through data.

Figure 6 illustrates C is logically identified as the cascade directory of A – a cascade from A through G and D to I, each of them revealed by being a precondition of the next function.

Modulating factors

Modulation factors are themes or theme functions that manipulate or modulate other functionality factors or their effects. These are searched for in the data using the semantic relationship ‘D [is a modulation factor of] A’.

Figure 6 shows that the cascade directory is controlled by theme C’s functionality factor as, when present, it blocks the cascade from theme A through G, D from reaching I. This means that modulation factor C is revealed as a preconditioning factor for I not to occur.

Putting the model together

The factors, items and directories described above must now be integrated into a model. This requires a great deal of substance (data), a critical approach and creativity, as the authenticity of themes and their functions is central to the logic involved in the model’s development. The chain of logic arrived at must therefore be critically and carefully considered. In the same way as with categories and themes, functions are labelled to denote authenticity and are based on empirical data, while other functions can be based on extrapolation from the literature or theory.

Once all aspects are described, and propositions formulated regarding structure and content, selection of the optimal representational strategy proceeds using trial and error. The model must be adjusted to achieve the best possible fit, while accounting for possible errors and deviations. Surface and content validation or adjustment can be achieved by adding in new data from the population under investigation, to optimise, evaluate and improve the model further.

This process should respect the nature of the empirical phenomena under investigation (Blumer 1986). A check-recheck iterative process between data and model regarding the empirical phenomena is required to improve the research integrity.

In our experience, a solid model is typically based on themes and functions marked by close empirical authenticity as a basis or foundation for the higher taxonomic levels, with higher levels of authenticity (steps 1 and 7). However, a model based on authenticity markings of 1 would be very descriptive, but a model based purely on authenticity markings of 3 would be purely speculative as it would not be grounded in data.

Conclusion

The Empirical Testing Thematic Analysis is useful for nurse researchers undertaking qualitative research who want to construct models derived from their investigations. It has been developed by nurse researchers and trialled in mental health nursing and multidisciplinary contexts. It attaches particular importance to authenticity markings and taxonomical and functional semantic analysis and generates a structure of semantics representing a target system that is expressed in theme-code, content and function.

This article has provided a step-by-step guide to using the approach. Our next article – part 2 - will discuss the model in depth. Together, the two articles will provide a comprehensive presentation of new theoretical work that will be of interest to nurse researchers and students who wish to apply a rigorous approach to qualitative research design leading to model formulation.

References

  1. Blumer H (1986) Symbolic Interactionism: Perspective and Method. University of California Press, Berkeley CA.
  2. Braun V, Clarke V (2006) Using thematic analysis in psychology. Qualitative Research in Psychology. 3, 2, 77-101. doi: 10.1191/1478088706qp063oa.
  3. Braun V, Clarke V (2019) Reflecting on reflexive thematic analysis. Qualitative Research in Sport, Exercise and Health. 11, 4, 589-597. doi: 10.1080/2159676X.2019.1628806.
  4. Braun V, Clarke V, Hayfield N et al (2019) Thematic analysis. In Liamputtong P (Ed) Handbook of Research Methods in Health Social Sciences. Springer Singapore, Singapore, 843-860.
  5. Buetow S (2010) Thematic analysis and its reconceptualization as ‘saliency analysis’. Journal of Health Services Research & Policy. 15, 2, 123-125. doi: 10.1258/jhsrp.2009.009081.
  6. Gildberg FA, Bradley SK, Tingleff EB et al (2015) Empirically Testing Thematic Analysis (ETTA) – methodological implications in textual analysis coding system. Nordic Nursing Research. 5, 2, 193-207. doi: 10.18261/ISSN1892-2686-2015-02-10
  7. Morse JM, Barrett M, Mayan M et al (2002) Verification strategies for establishing reliability and validity in qualitative research. International Journal of Qualitative Methods. 1, 2, 13-22. doi: 10.1177/16094069020010.
  8. Ritchey T (2018) General morphological analysis as a basic scientific modelling method. Technological Forecasting and Social Change. 126, 81-91. doi: 10.1016/j.techfore.2017.05.027.
  9. Spradley JP (1980) Participant Observation. Holt, Rinehart and Winston, New York NY.
  10. Whittemore R, Chase SK, Mandle CL (2001) Validity in qualitative research. Qualitative Health Research. 11, 4, 522-537. doi: 10.1177/104973201129119299.
  11. Zoglowek H (1999) Tematisk analyse. En Fremgangsmåte for å Analysere Kvalitative Intervju. Nordisk Pedagogik. 19, 3, 156-167.

Share this page

Related articles

The biology of cancer
Cancer research is moving fast. Understanding of the biology...

Exploring the education and information needs of patients on oral anticancer medicines
Aim To explore the education and information needs of...

Research and the knowledge and skills framework
Carrying out research is part of many academic courses. As...

Accessing specialist early intervention services for pre-school children
The aim of this phenomenological study was to investigate...

The use of e-learning in continuing professional development
Aim To record and evaluate the experiences of a group of...