Open Coding Descriptions

Barney G. Glaser, PhD, Hon PhD

Open coding is a big source of descriptions that must be managed and controlled when doing GT research. The goal of generating a GT is to generate an emergent set of concepts and their properties that fit and work with relevancy to be integrated into a theory. To achieve this goal, the researcher begins his research with open coding, that is coding all his data in every possible way. The consequence of this open coding is a multitude of descriptions for possible concepts that often do not fit in the emerging theory. Thus in this case the researcher ends up with many irrelevant descriptions for concepts that do not apply. To dwell on descriptions for inapplicable concepts ruins the GT theory as it starts. It is hard to stop. Confusion easily sets in. Switching the study to a QDA is a simple rescue. Rigorous focusing on emerging concepts is vital before being lost in open coding descriptions. It is important, no matter how interesting the description may become. Once a core is possible, selective coding can start which will help control against being lost in multiple descriptions.

Trying to find an indicator for a preconceived, conjectured concept can lead to excessive descriptions. This occurs because there are no indicators usually for a conjectured concept. The descriptions become the study by default. They honor, if possible, a nonexistent concept with no relevance and fit. But usually they just end up a QDA with no concept. In short it is best to stick to open coding for a core concept and then saturating the concept with a few indicators of its properties. This will control and stop conceptual descriptions. Open coding allows the researcher to see the direction in which to take his research so he can become selective and focused conceptually on a particular social problem. When he does focus his research, the relevancy and fit of his indicators will limit them to brief illustrations of his concepts. Excessive conceptual descriptions will be minimized or stop. The data can, once a core category is discovered, then be handled theoretically with minor need for it to be handled descriptively.

The opposite occurs if the core category has no grab and is hard to understand. The reader may request many indicators of it for illustration and understanding purposes. It may take many descriptions to indicate meaning of the core category. The possibility of generating a GT theory is lost. It has not been generated clearly. If many descriptions do not work, a QDA methodology description takes over. Thus it is always best to label a core concept with self-illustrating grab if possible.

Getting out of the data is vital for generating a GT. And staying out of the data (staying abstract of time, place and people) is just as important. It is easier to conceptualize if the researcher does not know the field of the data. He can be more objective and focused. Knowing the field can flood the researcher with descriptive data and lots of conjecture. It is easier to code someone else’s data because of the defacto distance from the data and descriptions are less in mind.

Open coding is guided by several rules and questions which by their proper use limit descriptions to the emergent problem. The first rule is to constantly ask of the data “what is this a study of?” This question severely limits descriptions by having to have them related to the core problem and possible emerging core category. As the core category generation firms up, descriptions get limited. The second question follows closely behind: What category does the problem incident indicate or what property of the core category does the incident indicate. As a GT becomes more and more generated it becomes easier to choose only descriptions that earn their way into the emerging theory. No forcing the data with concepts that do not apply severely limits descriptions to concepts that do apply.

A third question is asking of the emerging analysis what theoretical codes may apply to integrate the emerging theory. This question leaves behind conceptual descriptions and deals only with integrating into a theory the concepts that have emerged. In sum, as the GT analysis proceeds there is less change of excessive descriptions, even if they were excessed in the beginning.

Using the constant comparative method for open coding can lead to excess descriptions. Many descriptions emerge when coding qualitative data line by line. They are attached to nothing until a pattern emerges, so many can occur. But as soon as a pattern emerges excessing descriptions should stop as only a few are needed to illustrate the emerging concept and its properties. Until then the researcher must support the initial confusion and temptation to use existing rhetorical concepts prematurely to be backed up by many descriptions that will not be useable for the resulting GT. In short, using conjectural academic concepts fosters a lot of description, whereas emergent concepts only require a few illustration descriptions.

Up to this point I have discussed excessive descriptions as coming naturally when trying to do GT. A few researchers know the problem yet pursue description while saying they are doing GT research. They boost their Descriptions as GT, and they are just routine QDA. They wallow in story talk rather than discovering patterns and conceptual explanations in their data. This move to descriptions only lets the researcher off the creativity challenge. Instead of using the procedures of GT to reveal latent patterns in his data, he just does data talk claiming accuracy. This copping out on GT in favor of talk story is frequently endorsed by supervisors, committees, academic departments and university evaluation systems. The academic bureaucratic dominant researching is some form of QDA for hypothesis testing with accurate descriptive data. GT is not a testing methodology. It looks for latent patterns abstract of time, place and people. Choosing to search for accurate description stultifies the abstract creativity required by GT.

Coding qualitative data with a preconceived list of codes will result in many descriptions that will not apply to an emergent GT. They will simply describe what is not relevant or fit to an emerging GT. A code list comes from a pure QDA method of which there are several. At least excessive descriptions based on an emergent core concept will fit with relevance.

Some researchers read through their data quickly to get an overall feeling for it. It is natural to get many descriptions from this reading. They come without illustrating concepts yet to be discovered. I do not recommend this overall all reading for doing GT. It wastes time and could easily derail open coding focus on a probable core using the constant comparative method. Descriptions not indicating concepts lack fit and relevance to any grounded patterns. In a quest for an overview of impressions much latent patterns can be missed because of a glossing over of action details. Descriptions can excite and be enjoyable, no matter the pattern or not they indicated. They are grounded but not GT.

The overview approach by itself tends to yield thin theory if at all, with dubious relevance to life and action that leaves the feeling that much has been left out. It fosters conjecture and speculation. The only hope is to attach the descriptions to an academic speculation that is OKd by colleagues. Rich GT theory is denied when not coding properly for a GT. Pet theoretical themes and concepts are forced on the data. Descriptions capture the analysis in QDA fashion yielding many accurate data. Line by line constant comparison of data to yield grounded concepts is a lost procedure.

Facebooktwitterredditpinterestlinkedinmail