coding and counting

The Learning Sciences places great emphasis on the qualitative coding of verbal data as a means to capture and make claims about learning. In so-called “verbal analyses” (Chi, 1997) the goal is to uncover what an individual is “doing,” as opposed to comparing against an a priori model. In most cases, this implies the development of a coding scheme that segments and categorizes data by mapping participant talk to a code that meets some chosen criteria. From there, researchers locate and interpret patterns in and across codes. A potential next step is the quantification of qualitative analysis – with the quantitative analyses serving as a necessary step in confirmation.

While common, this approach is not taken for granted, and has been elaborated, challenged, and reconceptualized. For example, Hammer and Berland (2014) note, among other things, that the development of codes should be considered a finding in and of itself, and not atheoretical tools for analyses.  In some sense, their critique of qualitative coding mirrors that of Wiggins and Potter (2003): it forces people to “play semantic games” with categories. As Hammer and Berland point out, asked to code an oval as a square or a circle, coders would all agree that “circle” was the right code but also all agree that an oval is not a circle. Central to their point is that coded data (even after IR has been calculated) is not error-free data or neutral data. Nonetheless, in the Learning Sciences coding verbal and textual data is a relied upon method--steeped in lingering cognitive psychological epistemologies and practices.

Potter (2000) pushes even harder on cognitivist assumptions of coding and counting. He is critical of a cognitivist “competence before performance” approach to knowledge and activity. That is, when cognitivist researchers are not explicitly privileging models or researcher-generated theories and explanations, inevitably what happens is that “some record of practice does enter the research process (via a tape or transcript), it is counted and coded using some form of content analysis (or, at best, grounded theory) designed to recover the background factors or themes” (p. 33-34). With this in mind, a discursive psychologist would claim that even when learning sciences leaves the psychology lab, it still conducts work that is biased toward a researcher generated “reality” (even when we problematize taken for granted methods, e.g., Hammer and Berland).

This is not to say that social constructionist research simply does not code. However, “coding” in the learning sciences might as well be a different word than “coding” in discursive psychology. Potter and Wetherell (1987) state that the goal of coding is “to squeeze an unwieldy body of discourse into manageable chunks.” This is much different than learning sciences, which tends toward considering the results of coding as claims rather than segments of data. For example, Wiggins and Potter (2003) likely “coded” for instances of evaluative talk. This pared down 40 hours of talk to a data corpus that could be further transcribed and analyzed. Potter and Wetherell are clear that this stage of pre-analysis should be “inclusive.” Keeping with Hammer and Berland, this would mean to include everything that vaguely resembles a “circle” or “square” or “circular-square.”

Potter and Wetherell would likely agree with Hammer and Berland that coding is more than a reduction of qualitative data to simplify the job of the researcher. Yet, coding has an entirely different endgame. If I use discursive psychology for, say my dissertation, if, how and why I code must be methodologically justified. Even at my current stage of research I am finding coding to be problematic. For example, I am working on my early inquiry project. The first round of analysis is coding pre-service teacher talk and writing with somewhat of an a priori coding scheme: material, relational, and ideational identity resources. (Briefly, I am considering how pre-service teachers differently construct the roles of student and teacher in play and in formal academic writing.) This coding scheme is being repurposed from a coding scheme that was used in previous learning sciences work. So, the scheme is already “LS” and therefore an appropriate tool. However, it is assumes that pre-service teachers use identity resources, in particular those I am coding for. Now, material, relational, and ideational resources are very broad (and vague) categories. I expect that I still can orient toward participants and ground claims in interactions. Nonetheless, I am applying my analytical lens to the data before analysis…tisk, tisk!

Where does this leave me? Can I justify using an a priori coding scheme if it is used to segment the data à la Potter and Wetherell? In other words, can I use material, relational, and ideational as broad categories, which can be elaborated, retooled, or removed all together, situated in how resources are (or are not) taken up by pre-service teachers? Does this begin to strike a balance between LS and DP? Can “coding” exist in relative harmony across fields or is it a deal breaker? Is there any room for the a priori in DP? Could not one argue that subjective and objective constructions is an a priori analytical category?

Non-DP references:

Chi, M. T. (1997). Quantifying qualitative analyses of verbal data: A practical guide. The journal of the learning sciences, 6(3), 271-315.

Hammer, D., & Berland, L. K. (2014). Confusing claims for data: A critique of common practices for presenting qualitative research on learning. Journal of the Learning Sciences, 23(1), 37-46.