flexible coding for field notes

As I tell my grad students, almost no one does “pure” grounded theory. It’s technically impossible (we can’t forget everything we’ve read before going into the field). And, as Nicole Deterding and Mary Waters explain in their recent Sociological Research & Methods article, it’s a poor fit for twenty-first century qualitative research (often large-scale projects with fixed protocols, as required by IRBs and grant funding agencies).

Given the limitations of grounded theory, Deterding and Waters offer a different approach, which they call “flexible coding.” I won’t go through their method in detail, but the basic approach has a couple of steps:

1. Organization
2. Indexing
3. Memoing
4. Data Reduction

The whole point of this approach is to allow for a more systematic analysis of qualitative data. And Deterding and Waters lay out what these three steps can look like with semi-structured interview data. It’s a terrific, hands-on guide to qualitative methods. But it left me wondering – what does flexible coding look like with ethnographic data, instead?

Reflecting on that question, I realized that I actually did a version of “flexible coding” with my research for Negotiating Opportunities. So I thought I’d share here in case it’s useful for other students and scholars embarking on ethnographic projects (or trying to climb their way up the mountain of ethnographic data they’ve already produced).

For some background, Negotiating Opportunities looks at how middle- and upper-middle-class white students persuade their teachers to give them unfair advantages in school. The project involved ethnographic observations across multiple grades, multiple classrooms, multiple subjects, and multiple students. That’s a lot of units of analysis and a lot of potential comparisons to explore.

Early in the data collection/analysis process, I tried a more grounded-theory approach. I dumped all my field notes (just a handful at that point) in Atlas.ti, opened the first document, and started free-form coding. I quickly ended up with almost a hundred codes. There were overlapping codes. Codes that didn’t have a clear meaning. Codes that made no sense at all.

So I knew I needed a different approach. Consistent with Deterding and Waters’s model, I started by organizing the data. Rather than write each 2-3 hour observation as a single field note, I created separate field note documents for each 40ish-minute subject period I observed. For example, if I got there first thing in the morning and stayed until lunch, I’d end up with separate field note files for “morning work,” “science,” “gym,” and “math.”

To keep all those files organized, I also developed a labeling system. Each field note was labeled as follows: [GRADE]_[TEACHER]_[SUBJECT]_[DATE]. So, for example, one file might be called “5_Fischer_Math_2010.03.19.”

Organizing the data that way led me to what Deterding and Waters call “indexed coding.”

The first step, for me, involved applying document-level codes to each field note. (I used Atlas.ti, but this can be done with most qualitative data analysis (QDA) software). Each file got a document-level code for grade level, teacher/classroom, subject, and marking period (fall, winter, spring).

I also created within-document index codes for each student and for common in-class activities (e.g., test-taking, group work, direct instruction). And I went through and coded the interactions described in each field note to indicate which students were involved and what the whole class was doing, activity-wise. 

Side Note on Coding: I write my field notes as short stories. So I apply codes to the whole story. With QDA software, the output makes more sense that way, as it includes the context needed to make sense of a particular interaction.

That indexed coding step made it much easier to use the QDA software to find and analyze particular types of interactions. For example, I could easily pull all the stories from Mr. Fischer’s math class. Or all the stories involving Jared during a science test.

Another key step, for Deterding and Waters and for me, was to write memos. Lots and lots of memos. I tried doing this a bunch of different ways. But I ended up settling on standard header and footer memos that I included in each field note document. My header memos include a bullet-point list of key themes in the field note, with brief descriptions of key interactions that illustrate those themes. My footer memos include a more free-form discussion reflecting on what happened in the field, how it relates to other things I’ve seen and read, and what next steps I need to take.

Those memos, and especially the header memos, became the basis for the  focused coding schemes I developed for each article and chapter I wrote with the data. The themes typically became broad codes (e.g., help-seeking), and often pointed to more specific sub-codes, as well (e.g., calling out for help, approaching the teacher for help, raising a hand to ask for help, not asking for help, etc.).

Now, Deterding and Waters suggest that indexed coding is helpful, at least in part, because it allows researchers to apply focused/thematic codes to only the relevant portions of the text. And that works for interviews, because the index codes correspond to the themes and specific questions in the interview guide.

Unfortunately, ethnographic data, even when carefully indexed, doesn’t work that way. Help-seeking doesn’t only happen in math. Or in Mr. Fischer’s class. Every interaction (or almost every interaction) is potentially relevant. So all the focused/thematic codes still have to get applied to the whole, giant data set. And it takes *forever.*

But it’s worth it. Because that kind of systematic, focused-coding approach, especially when combined with carefully indexing, makes it much easier to see and analyze patterns in the data.

One particularly useful way to see and analyze those patterns is with data matrices. I created an Excel file for each paper or chapter. Within that file, I had tabs for each relevant code (e.g. calling out for help). Then, within that tab, each student got a row. In each student’s row, I included basic background data, with social class, gender, race/ethnicity, and math level in separate columns. Then, in the columns beyond that, I included a brief description of/link to each relevant field note excerpt, each in a separate cell. With all those pieces of information linked together, I could use the “sort” function in Excel to rearrange the spreadsheet and see how patterns changed when I looked at them by social class or race or gender. That approach made it easier to identify outliers and also check alternative explanations for the patterns I was seeing in the field.

Of course, there isn’t a magic algorithm that turns coded data into an argument or an article. But “flexible coding” can make it much easier to identify and analyze patterns and find clear examples to illustrate them. At the very least, it gives you the tools you need to climb a mountain of data and not give up or get lost along the way.