Qualitative coding for computer scientists.
Qualitative coding is a form of feature extraction in which text (or images, video, etc.) is tagged with features of interest. Sometimes the codebook is defined ahead of time, other times it emerges through multiple rounds of coding. For more on how and why to use qualitative coding, see Emerson, Fretz, and Shaw's Writing Ethnographic Fieldnotes or Shaffer's Quantitative Ethnography.
Most of the tools available for qualitative coding and subsequent analysis were designed for non-programmers. They are GUI-based, proprietary, and don't expose the data in well-structured ways. Concepts from computer science, such as trees, sorting, and filtering, could also be applied to qualitative coding analysis if the interface supported it. Furthermore, a command-line based tool can be combined with other utilities into flexible pipelines.
Qualitative Coding, or
qc, was designed to address these issues. I have used
qc as a primary coding tool in a SIGCSE
how K-12 schools define and design computer science courses. The impetus for
packaging and releasing a stable version was my own dissertation work.
- Due to its nature as a command-line program,
qcis only well-suited to coding textual data.
qcuses line numbers as a fundamental unit. Therefore, it requires text files in your corpus to be hard-wrapped at 80 characters. The
inittask can handle this for you.
- Currently, the only interface for actually doing the coding is a split-screen in vim, with the corpus text on one side and comma-separated codes adjacent. This works well for me, but might not work well for you. I have other ideas in the pipeline, but they won't be around soon.
pip install qualitative-coding
- All the source files you want to code should be in a directory (possibly nested).
- Choose a working directory. Run
qc init. This will create
corpus_dirwith the directory holding your source files. This may be relative to
settings.yamlor absolute. Similarly, specify directories for
memos_dir, and the YAML file where you want to store your codebook. Unless you're particular, the default settings are fine.
qc init --prepare-corpus --prepare-codes --coder yourname. This will hard-wrap all the text in your corpus at 80 characters and create blank coding files.
qc is designed to give you a powerful terminal-based interface. The general
workflow is to use
code to apply qualitative codes to your text files. As you
go, you will start to have ideas about the meanings and organization of your
memo to capture these.
Once you finish a round of coding, it's time to reorganize your codes. Use
codebook to refresh the codebook based on new coding. Use
stats to see the
distribution of your codes. If you want to move codes into a tree, make these
changes directly in the codebook's YAML. If you realize you have redundant
After you finish coding, you may want to use your codes for analysis. Tools are provided for viewing statistics, cross-tabulation, and examples of codes, with many options for selecting and filtering at various units of analysis. Results can be exported to CSV for downstream analysis.
--coder argument supports keeping track of multiple coders on a project,
and there are options to filter on coder where relevant. More analytical tools,
such as inter-rater reliability, are coming.
Create a new directory somewhere. We will create a virtual environment, intstall
qc, and download some sample text from Wikipedia.
$ python3 -m venv env $ source env/bin/activate $ pip install qualitative-coding $ qc init $ qc init $ curl -o corpus/what_is_coding.txt "https://en.wikipedia.org/w/index.php?title=Coding_%28social_sciences%29&action=raw" $ qc init --prepare-corpus --prepare-codes --coder chris
qc init three times? The first time creates a prepopulated
settings.py file. You could then change any settings. The second time
settings.py and creates the specified files and directories.
And the third run, with the flags, processes the corpus file and creates a
corresponding coding file.
Now we're ready to start coding. This next command will open a split-window vim session. Add comma-separated codes to the blank file on the right. I usually page-up (control+u) and page-down (control+d) each file to keep their line numbers synchronized. Once you've added some codes, we can analyze and refine them.
$ qc code chris -f $ qc codebook $ qc list - a_priori - analysis - coding_process - computers - errors - grounded_coding - themes
Now that we have coded our corpus (consisting of a single document), we should
think about whether these codes have any structure. All data in
qc is stored
in flat files, so you can easily modify it by hand. Re-organize some of your
codebook.yaml. When you finish, run
codebook again. It will go
through your corpus and add any missing codes.
$ qc list - analysis - coding_process - a_priori - grounded_coding - computers - errors - themes
I decided to group a priori coding and grounded coding together under coding process. Let's see some statistics on the codes:
$ qc stats Code Count ------------------ ------- analysis 2 coding_process 7 . a_priori 2 . grounded_coding 2 computers 2 errors 1 themes 2
stats has lots of useful filtering and formatting options. For example,
qc stats --pattern wiki --depth 1 --min 10 --format latex would only consider files
having "wiki" in the filename. Within these files, it would show only
top-level categories of codes having at least ten instances, and would output a
table suitable for inclusion in a LaTeX document. Use
--help on any command to
see available options.
Next, we might want to see examples of what we have coded.
$ qc find analysis Showing results for codes: analysis what_is_coding.txt (2) ================================================================================ [0:3] In the [[social science|social sciences]], '''coding''' is an analytical process | analysis in which data, in both [[quantitative research|quantitative]] form (such as | [[questionnaire]]s results) or [[qualitative research|qualitative]] form (such | [52:57] process of selecting core thematic categories present in several documents to | discover common patterns and relations.<ref>Grbich, Carol. (2013). "Qualitative | Data Analysis" (2nd ed.). The Flinders University of South Australia: SAGE | analysis Publications Ltd.</ref> | |
Again, there are lots of options for filtering and viewing your coding. At some
point, you will probably want to revise your codes. You can easily rename a
code, or collapse codes together, with the
remane command. This updates your
codebook as well as in all your code files.
$ qc rename grounded_coding grounded
At this point, you are starting to realize some of the deeper themes running
through your corpus. Capturing these in an "integrative memo" is an important
part of qualitative coding.
memo will open a preformatted document for you in vim.
$ qc memo chris --message "Thoughts on coding process"
Congratulations! You have finished the first round of coding. Before you move
on, this would be an excellent time to check your files into version control.
I hope you find
qc to be powerful and efficient; it's worked for me!
--help for a full list of available options for each command.
Initializes a new coding project, as described above.
$ qc init
Checks that all required files and directories are in place.
$ qc check
Opens a split-screen vim window with a corpus file and the corresponding code file. The name of the coder is a required positional argument. After optionally filtering using common options (below), select a document with no existing codes (for this coder) using --first (-1) or --random (-r)
$ qc code chris -1
Scans through all the code files and adds new codes to the codebook.
$ qc codebook
Lists all the codes currently in the codebook.
$ qc list --expanded
Goes through all the code files and replaces one or more codes with another. Removes the old codes from the codebook.
$ qc rename humorous funy funnny funny
Displays all occurences of the provided code(s).
$ qc find math science art
Displays frequency of usage for each code. Note that counts include all usages of children.
List code names to show only certain codes. In addition to the common options below,
code results can be filtered with
$qc stats --recursive-codes --depth 2
Displays a cross-tabulation of code co-occurrence within the unit of analysis, as counts or as probabilities (--probs, -0). Optionally use a compact (--compact, -z) output format to display more columns. In the future, this may also include odds ratios.
$qc crosstab planning implementation evaluation --recursive-codes --depth 1 --probs
Filtering the corpus
pattern(-p): Only include corpus files and their codes which match (glob-style)
-invert (-i): Only include corpus files that do not match
filepath(-f): Only include corpus files listed in
filepath(one per line).
coder(-c): Only include codes entered by
coder(if you use different names for different rounds of coding, you can also use this to filter by round of coding).
Filtering code selection
- code [codes]: Many commands have an optional positional argument in which you may list codes to consider. If none are given, the root node in the tree of codes is assumed.
- --recursive-codes (-r): Include children of selected codes.
depth(-d): Limit the recursive depth of codes to select.
unit(-n): Unit of analysis for reporting. Currently "document" and "line" are supported by most commands.
- --recursive-counts (-a): When counting codes, also count instances of codes' children. In contrast to --recursive-codes, which controls which codes will be reported, this option controls how the counting is done.
Output and formatting
format(-m): Formatting style for output table. Supported values include "html", "latex", "github", and many more.
- --expanded (-e): Show names of codes in expanded form (e.g. "coding_process:grounded")
outfile(-o): Save tabular results to a csv file instead of displaying.