Thematic analysis
An introduction to analysing qualitative data
What is thematic analysis?
Thematic Analysis (TA) is a popular method of analysing non-numerical qualitative datasets. These datasets could consist of primary research data - such as interviews, focus groups or survey responses – or they could consist of secondary research data – such as research papers or newspaper articles. The aim of TA is to identify themes across a dataset.
To conduct TA a researcher must ‘code’ the dataset. A code captures a single idea associated with a segment of the data. A code could be a small as single word or as long as a sentence or an entire paragraph. ‘Coding’ is the process of identifying different codes in a dataset. Codes are the building blocks of themes. A ‘theme’ captures a common, recurring pattern across a dataset, typically with a central organising concept.
What type of thematic analysis should I use?
There are different types of TA which have different approaches to the coding process. Some forms of TA are more inductive. In these methods, coding and theme development are directed by the content of the data rather than an overarching research question, theory, or conceptual framework. In inductive TA, the emphasis is on engagement with the data and acknowledging the subjectivity of the coder.
Other forms of TA are deductive. In these methods, coding and theme development are directed by existing concepts or ideas, either drawn from theory, personal experience, or research questions. In deductive TA, the emphasis is on coding reliability.
There are methods which merge inductive and deductive approaches to TA. The method proposed by Jennifer Fereday and Eimar Muir-Cochrane (2006) is a hybrid inductive-deductive TA which is suitable to a wide range of different datasets. Using their method, a researcher can identify codes inductively through engagement with the data, and deductively through engagement with the key concepts and theories related to the topic.
Steps to conduct a hybrid inductive-deductive TA
Step 1: Develop a code manual.
A code manual initially consists of theory-driven codes based on the research question and theoretical concepts which underpin the research. These are deductive theory-driven codes.
Step 2: Test the reliability of the codes.
The reliability of the theory-driven codes is tested by coding a segment of the dataset.
Step 3: Summarize data
If the dataset is too large to code completely, summarize the data and then code.
Step 4: Code the dataset and identify any inductive codes.
Code the entire dataset or the summarized dataset. Identify examples of theory-driven codes in the code manual and identify any potential data-driven codes in the dataset.
Step 5: Update code manual with deductive and inductive codes. Connect codes and code the dataset a second time.
Add the inductively identified codes into the code manual. Inspect for overlaps between codes, and potentially synthesise several codes together. Once this is complete, code the dataset a second time with the complete code manual as a guide.
Step 6: Identify and corroborate themes.
After the second coding, identify and corroborate any themes. Themes will most likely be amalgamations of different codes with a central organising concept.
Example: Freshwater policy implementation research
The project was guided by the research question: “What are the barriers to local government implementation of freshwater policy in New Zealand”.
A code manual was developed with theory-driven codes collected from New Zealand specific sources (such as Ministry for the Environment documents), as well as policy implementation research internationally (see table below).
| Code | Definition |
|---|---|
| Identifying regional freshwater objectives | The Land and Water Forum identified an implementation lag and argued that one way of resolving this would be to focus on the translation of public values into reginoal freshwater objectives |
| Alignment with national policy | MfE identified in a review of the NPSFM that councils were struggling to align plans to the NPSFM |
| Identifying community and iwi values in freshwater | The Land and Water Forum argued that translation of public values into objectives was lagging |
| Ensuring staff implement policy as managers and senior staff envision | Policy implementation research argues that when long chains-of-command are present, the discretionary practices of street-level bureaucrats will distort the implementation of policy |
The reliability of these codes was tested on a sample of the dataset. On theory driven code, “alignment with national policy”, was identified in the sample coding, and an example code from the data was placed next to the code in the code manual (see Table below).
| Code | Definition | Example Code |
|---|---|---|
| Alignment with national policy | MfE identified in a review of the NPSFM that councils were struggling to align plans to the NPSFM | "The challenge we faced is, in particular, the National Policy Statement for Freshwater Management [that] has changed on a regular basis. So, our current PNNRP [is] based on the 2014 NPS, but of course now we have the 2017 version and a 2019 version as well. So, there has been some shifting of the goal posts, which has proved challenging for us, especially when you are always playing catch up." |
In this example, the dataset was small enough that no summarizing was needed.
In the first round of coding, we identified a range of different inductive data-driven codes. One of these was ‘acknowledging uncertainty’. We defined this code in the code manual and provided a quote example next to the definition to illustrate its presence in the dataset (see Table below).
| Code | Definition | Example Code |
|---|---|---|
| Acknowledging uncertainty | Uncertainty makes it difficult to implement policy because making changes on their land want certainty it will have a positive effect, planners need certainty for legal reasons, and politicians want certainty so they are praised for action | "One of the challenges is being to work in that constrained and ambiguous environment. But a lot of our staff and communities want certainty, and at the moment we can't quite deliver it because we don't know what the solutions look like. That makes some people uncomfortable, but it can make politicians uncomfortable as well because they can't say 'this is what we're doing, we're going to deliver on this, this and this'. The nebulous and ambiguous is difficult to communicate." |
After the code manual was updated with data-driven codes, we coded the dataset a second time. We found, when conducting the final process of corroborating and legitimising themes, that only one of the data-driven codes formed a theme, “alignment with national policy”. The other themes were amalgamations of codes discovered inductively in the coding process itself. The 23 codes in the code manual were synthesised into four themes and seven sub-themes (see Table below with themes bolded).
| Theory-driven codes | Data-driven codes |
|---|---|
| Identifying regional freshwater objectives | Acknowledging uncertainty |
| Alignment with national policy | Public conversation on freshwater |
| Identifying community and iwi values in freshwater | Conservative approaches to water management |
| Ensuring staff implement policy as managers and senior staff envisage | Underinvestment |
| Ensuring public compliance with policies and plans | Local government and community capacity |
| Lack of technical guidance | Difficulties with collaboration |
| Conflicting guidance from senior staff | Planning requirments |
| Encouraging behaviour change | |
| Local context different from national context | |
| Path dependence | |
| Prohibitive costs | |
| Council size | |
| Submitters commandeering planning processes | |
| Too many variables to manage | |
| Translating science to the public | |
| Unitary council issues |
References
References
Fereday, J. and Muir-Cochrane, E., 2006. Demonstrating rigor using thematic analysis: A hybrid approach of inductive and deductive coding and theme development. International journal of qualitative methods, 5(1), pp.80-92.
