Best Practice

Using a mix of deductive and inductive coding techniques to analyse data with ATLAS.ti

This article was written by Natasha Katunga, a Certified ATLAS.ti Professional Trainer from the University of the Western Cape, South Africa. She shares about how she used ATLAS.ti to carry out her deductive and inductive coding. We hope you enjoy reading about her experience!
Natasha Katunga
Certified ATLAS.ti Professional Trainer Junior
  1. Introduction
  2. Data coding technique
  3. Filtering codes in ATLAS.ti
  4. Using networks to visualise the relationships across entities
  5. Lessons learned
  6. References

Introduction

Rich qualitative data can be a treasure-trove of valuable information for the curious mind with the right analysis tools. There are various data analysis techniques available to qualitative researchers including thematic analysis, grounded theory, and phenomenological analysis. The techniques can be tailored to suit specific research needs and objectives. My doctoral study focused on the use of social media by digital inclusion intermediaries to communicate for development in under-resourced communities. For my research design, I adopted an interpretive qualitative research approach that was guided by thematic data analysis techniques. This analysis technique entails coding data and identifying themes. The themes are patterns found in the information that describe, organise and interpret aspects of the phenomenon. They can be generated inductively from the raw information, or deductively from theory and prior research (Boyatzis, 1998).

My data analysis, therefore, required intensive deep dives into a considerable amount of raw narrative data collected through in-depth interviews and focus group discussions, and the use of mixed deductive and inductive coding techniques. As a multi-purpose qualitative data analysis software program, ATLAS.ti was a useful tool that organised and centralised the process. Moreover, I was able to utilise the code grouping feature to filter my codes into themes and sub-themes, all of which I visualised using networks.

Data coding technique

The coding aspect of data analysis can be intimidating, particularly if one is dealing with large sums of data. The process itself entails attaching a conceptual and/or descriptive label to a word or segment of data that is particularly interesting to the researcher. The label can fulfil different purposes and can be applied at different levels of abstraction – this is at the discretion of the researcher. The labelled data and labels are analysed further by, for instance, sorting them into themes that facilitate insight, comparison and theory development (Kaplan & Maxwell, 2006). A good code is a label that captures the qualitative richness of the phenomenon, it should be useable in the data analysis and interpretation, as well as the presentation of the findings (Boyatzis, 1998). In ATLAS.ti this label is the code and the segment of data that has been labelled is the quotation (illustrated in Figure 1).

Figure 1. Display of quotation and codes

To suit the data analysis needs of my study I adopted a hybrid coding technique which made use of inductive and deductive coding approaches. In essence, this is the integration of data-driven codes and theory-driven codes in the data analysis process. The deductive coding process was made up of several key steps. Since the objectives of my study were multidisciplinary it was necessary to develop a multi-theory framework that drew theories from the domains of development, communication, and technology to guide the study. Each theory used was broken down to its core concepts – with each concept being interpreted and described. This description informed the development of relevant research questions for the investigation. For each research question, I identified labels (codes) that could be applied to data (participants’ responses) that would be collected based on the specific research question. Table 1 illustrates this step as it was applied to one of the theories used, Community Development Theory which had 5 core concepts.

Table 1. Identifying theory-driven codes

I then compiled a list of 37 codes from the multi-theory framework and saved it in an Excel document (illustrated in Figure 2). All the codes were in one column of a spreadsheet, they were not grouped or ranked. To differentiate the inductive and deductive codes, the deductive codes each shared a similar prefix.

Figure 2. Code list in Excel

The next step was importing the code list (Excel document) into ATLAS.ti as the initial set of codes.
Importing a code list/ codebook into ATLAS.ti

  1. With ATLAS.ti Windows open, click on the Import & Export tab
  2. Click once on Codebook in the ribbon
  3. From the two options that appear select Import from Excel (that is if your document with the codes is saved as an Excel document)
  4. From the pop-up window which appears, search for your document
  5. Click import

After importing the code list, I uploaded my interview and focus group transcripts into ATLAS.ti and the reading process began. The theory aspect of the research was only part of the investigation, the other being the practical aspect of social media use by the intermediaries. It was therefore very necessary to read through the transcripts intensively to understand the story from the respondents’ point of view and interpret their responses. Guided by (1) my research questions, (2) the literature review and (3) the multi-theory framework, I analysed the data. As I read through the transcripts relevant and interesting words and segments of data were coded with existing codes (deductive coding) and new codes (inductive coding) that originated from concepts that were identified from the data. Unlike theory-driven codes, the data-driven codes came from the stories told by the text itself. I made use of comments during the coding process. Quotation comments helped to remind me why those words and statements were important and relevant to my research. I used code comments to describe the code and my thoughts, I even attached images to remind me of the context (illustrated in Figure 3).

Figure 3. Code comments

I also utilised memos to keep track of my research ideas, thoughts, and reflections. Even in cases where I felt I needed more information or clarity from a respondent, I made use of memos as my to-do list.

Filtering codes in ATLAS.ti

Having strategically analysed and coded the transcripts to a point of saturation I had 184 codes and 727 quotations. The coding process helped me to gain a better understanding of the data. However, a code on its own is simply a conceptual or descriptive label (Friese, 2017), it does not tell us the full story. I further developed my coding system by analysing the codes to dig deeper into the data and ask more specific questions using ATLAS.ti's tools. I filtered my codes into groups and themes using both interpretive efforts and deductive theory-driven approaches to identify the themes. I had three levels of themes. Level one was the overarching theme - identified based on the research sub-questions. Level two was the sub-themes - identified based on the theoretical underpinnings of each overarching theme. Level three was the code group themes – based on the surface meaning of the grouped codes. For example, the codes: cost of mobile data a challenge and no budget for paid media were grouped under the code group theme: financial challenges (illustrated in Figure 4). The process of assigning codes to a particular code group theme was facilitated in ATLAS.ti using the code and code group manager.

Figure 4. Grouping codes

Using networks to visualise the relationships across entities

ATLAS.ti networks not only visualise the relationships between the entities, but they enable other forms of exploration that advance the analysis process on another level. The networks support creativity and help in the detailing of an idea or by developing a researcher’s line of reasoning (Friese, 2020). Guided by the principles of ‘mind-mapping’ I used networks to visually map the codes to code group themes, the group themes to sub-themes, and the sub-themes to the overarching themes (illustrated in Figure 5). This was a form of conceptual analysis. A systematic colour pattern was applied to different theme levels. The networks highlighted the links, patterns and relationships across the codes, group themes and sub-themes. This enabled me to tell the complicated story of the data and demonstrate the validity of the analysis.

Figure 5. Using a network to visualise the relationships across entities

Lessons learned

As an ATLAS.ti Trainer and avid user of the software I always seem to find new ways of using the different features of the tool. The more you use the software, the more you understand the functionalities, and this enables one to dive deeper and explore their data at different conceptual levels. Segmenting the collected data, dealing with it part by part also makes the analysis process less daunting and more manageable. In addition, using a particular technique to guide the identification of codes and the coding approach helps – some techniques are illustrated in Friese (2017). The use of networks can also be intimidating but if you identify a systematic approach which works for you, this feature can be a powerful concept analysis tool that can also tie everything together.

References

Boyatzis, R. E. (1998). Transforming Qualitative Information: Thematic Analysis and Code Development. SAGE Publications Inc.

Friese, S. (2017, March 16). How to Make the Best of Codes in ATLAS.ti. The ATLAS.Ti Research Blog. https://atlasti.com/2017/03/16/make-best-codes-atlas-ti/

Friese, S. (2020, December 11). Visualizing Relationships in Networks. The ATLAS.Ti Research Blog. https://atlasti.com/2020/12/11/visualizingrelationshipswithnetworks/

Kaplan, B., & Maxwell, J. A. (2006). Qualitative research methods for evaluating computer information systems. In J. G. Anderson & C. E. Aydin (Eds.), Evaluating the organizational impact of healthcare information systems (2nd ed., pp. 30–55). Springer.