Using Quotation Names for Coding: An Illustration From Grounded Theory
Author: Nicholas Woolf, PhD
The problem with coding
Most qualitative methodologies involve coding as a central activity, and coding is also a central software feature of ATLAS.ti. So they would seem to be the same activity. But they are not. When using ATLAS.ti, “coding” is always the same mechanical process. It involves creating code names in the CODE MANAGER, and linking the CODES to stretches of text. (To avoid ambiguity I’m putting all the things that exist inside ATLAS.ti in SMALL CAPS).
But when thinking about the data analysis process, “coding” can involve very varied mental activities. In a content analysis it might involve counting the number of times various topics are mentioned. In more interpretive methodologies it might involve creating codes to subtly conceptualize the meaning of a participant’s experience, leading to a quite different kind of coding scheme from a content analysis. It would be better if the mental coding activity of the research process and the mechanical coding activity of operating ATLAS.ti had different names. A good name for what is done mentally in qualitative research would be “conceptualizing data”, and a good name for its mechanical equivalent when using the software would be “coding data”. Usually these activities are the same, with each concept in a research project getting its own CODE in ATLAS.ti. But not always, because a CODE in ATLAS.ti is not always the best way to represent a concept. Here is an example of when it is not.
The Context: Kathy Charmaz’s Approach to Grounded Theory Coding
This illustration is from grounded theory, but it illustrates a principle that applies to many other coding situations.
Grounded theory includes several phases of coding activity, with each phase built on the results of the previous phase. The overall goal is to reduce a large quantity of text to a small number of concepts that give an account of the whole body of data from a particular perspective. One approach is Strauss & Corbin (1998)’s, which involves the three phases of open coding, axial coding, and selective coding. An increasingly popular approach is Charmaz (2006)’s, which involves the somewhat different phases of initial coding, focused coding, and theoretical coding. One reason for Charmaz (2006)’s popularity is that the process and practices are clear and it is easy to understand what you are supposed to do. But the book does not get involved in how to use software to accomplish the method. This article concerns how to accomplish the first phase of initial coding in ATLAS.ti.
In a nutshell, the purpose of initial coding is to name codes that reflect actions or processes in each individual segment of data. A segment can be a sentence, a line, or even a single word. Each data segment has its own unique initial CODE that is specific to that segment. For example, one text segment is: “I had been in bed for days and she called me up, ‘you never tell me, and I have to find out from Linda…”, for which the suggested initial code is Receiving second-hand news.
When the initial coding is done, the second phase involves deciding on the most significant initial codes for categorizing all the data to this smaller number of codes, called the focused codes. The third phase is to relate the focused codes together with theoretical relationships, in order to integrate various sets of focused codes and name each set as a theoretical code. These theoretical codes are few in number and form the core concepts of the account to be given of the whole body of data.
Implementing Initial Coding in ATLAS.ti
The most obvious approach to using ATLAS.ti is to assume that if there is a code in the data analysis then it must be represented by a CODE in the program. If we do that with Charmaz-style initial coding, then each data segment – each QUOTATION – is linked to its own unique CODE. If there are 1,000 QUOTATIONS in a grounded theory project, there would be 1,000 CODES in ATLAS.ti, with each CODE linked to a single QUOTATION. This is a very cumbersome outcome and produces an extremely long list of CODES. These CODES do not lend themselves to their usual purpose, which is to accumulate many QUOTATIONS per CODE in order to reduce the data. But the Charmaz (2006) initial coding phase is not for data reduction; it is to analyze the “theoretical possibilities” in each individual segment of data (p. 47). This initial coding activity is therefore not well-suited to being represented by CODES.
Figure 1: Not the best way to implement the initial coding illustrated on page 44 of Charmaz (2006)
A better approach is to recognize that data reduction is not the purpose of the first phase of the initial coding phase, and instead create FREE QUOTATIONS in ATLAS.ti. These are QUOTATIONS that mark off a segment of text in the usual way, but are not linked to anything yet. Now it is a matter of naming these FREE QUOTATIONS with their unique initial codes.
Everything in ATLAS.ti has a name – everything – and everything in ATLAS.ti can be renamed. When creating a code the researcher provides the name of the CODE in the process of creating it. It is different when creating QUOTATIONS. Here the program automatically provides an initial name for the QUOTATION, which is the first few words of the QUOTATIONS’s text. In most projects we are not too concerned about the names of QUOTATIONS, and rarely if ever look at them. Renaming each QUOTATION from its automatically provided name to the initial code name makes sense because there is a unique initial code for each QUOTATION. The list of QUOTATIONS – the QUOTATION MANAGER – then effectively becomes for us what we could call the “Initial Coding Manager”. Then we can write the analytic memos that Charmaz (2006) recommends writing for each initial code in the comment area of each QUOTATION
Figure 2: A better way to implement the initial coding illustrated on page 44 of Charmaz (2006)
Implementing Focused Coding in ATLAS.ti
In the second phase of analysis, data reduction does comes into play, and the usual purpose of CODES is needed. In this second phase the initial codes are re-read and further studied, and the most significant ones that are considered best-suited for coding large amounts of data become the smaller set of focused codes.
It now makes sense to create a regular CODE for each focused code, and use each one to code all the QUOTATIONS that are embraced by its concept. There are a number of ways this could be done. One efficient way would be to select all the QUOTATIONS in the QUOTATION MANAGER that would be embraced by a focused code, using the control key to select non-contiguous QUOTATIONS. Then drag-and-drop the whole group on to the focused code in the CODE MANAGER. This serves to link the QUOTATIONS to the focused code – that is, to code them. This procedure avoids having to jump to each QUOTATION in each interview in the text area in order to code each one individually from the text area.
This is one example when conceptualizing data is not accomplished most efficiently by creating CODES and linking them to QUOTATIONS. The underlying principle that applies to all methodologies is: what is the best way to represent a concept in ATLAS.ti? Usually this is with a CODE linked to a QUOTATION, but often it is not. It depends on the purpose, or in other words, what will be done next with the concept. There are other ATLAS.ti features that offer alternative ways to represent concepts and that allow different follow-up activities, for example:
- QUOTATION names as concepts, as described in this article.
- MEMOS used to represent the big-picture themes in a project, which can also by linked to QUOTATIONS, in effect turning the MEMO MANAGER into a kind of second CODE MANAGER.
- CODES used as high-level concepts in a hierarchy of concepts, for the purpose of relating them to other CODES in NETWORKS, with no intention of linking them to QUOTATIONS.
Overall, this approach to using ATLAS.ti powerfully reverses the direction of thinking about ATLAS.ti that is common. It starts by considering the purpose of each analytic strategy and then deciding how this purpose can best be accomplished in ATLAS.ti, rather than the opposite approach of learning how the software works and looking for the most obvious ways to use its features in the data analysis.
About the Author
Nicholas Woolf has taught several hundred ATLAS.ti workshops and has consulted to dozens of ATLAS.ti research projects and individual researchers. He is currently at work on a textbook, How to Use ATLAS.ti Powerfully, to be published at the end of this year. His Ph.D. from the University of Iowa is in Instructional Design. More information about his work can be found at www.learnatlas.com, or contact firstname.lastname@example.org to stay informed about the publication date of the textbook.
Charmaz, K. (2006). Constructing grounded theory. Thousand Oaks, CA: Sage.
Strauss, A., & Corbin, J. (1998). Basics of qualitative research: Techniques and procedures for developing grounded theory (2nd ed.). Newbury Park, CA: Sage.
© Woolf Consulting 2014