Joanna Drucker’s “Humanities Approaches to Graphical Display,” a classic among critical digital humanists, is ostensibly a paper on visualization, with nice, original figures. The author discusses the implicits of “objective” visualizations; those often endorse simple conceptions of time (e.g. linear or even with no past) and categories (exclusive from one another, with firm boundaries, etc.). Doing this, she in fact addresses the implicits of “objective” categorization generally – and she explicity makes the point that data are capta (constructed), that humanists know this and that using computers/quantification should not make them forget it. Her paper might free your imagination for the categorization of your own data. If you want to begin with an even shorter piece, she makes some of the same points with incisive clarity in this interview (with Miriam Posner, by Miriam Kienle).
Marten Düring’s “From Hermeneutics to Data to Networks” is based on the same idea of not applying supposedly objective categories, but devising a coding scheme based on specific data and research questions. It very clearly guides you through the non-straightforward process (which occupied a large part of his PhD research) of categorizing help provided to Jews in Nazi Germany. The section on categorization is relevant for all research in historical sources and probably all research in the humanities and social sciences – even if you do not care about networks.
In our Chapter 3, we often use a great paper [sadly enough, I found no legal access to an open version] as our main example: Cameron Lynne Macdonald and Karen V. Hansen’s exploration of visiting patterns in nineteenth-century New England, based on the diaries of working-class men and women. The paper is full of careful discussions of choices made both during the input phase and the categorization phase – discussions that are substantively as well as methodologically interesting. How do you define a “visit” when it is you main unit of analysis, but men and women tended to view it differently? Who can we consider as part of the “working class” when people often had two or three very different jobs? The authors even explained that (and why) they decided not to categorize the visitors as “kin vs. non-kin.” Another must-read (if you can’t get access, write to us).
We had our share of lively collective discussions on categorization, not only in our own research, but in the context of teaching quantitative history by doing it. Digital humanists Miriam Posner and Marika Cifor have reported on a similar experience (you can read here a shorter but openly accessible version of their discussion). In order to build, with students, a database on early African-American silent “race films,” they had to decide on a definition of the genre. What could be viewed as a technical issue (what to include and not to include) led them to a thorough substantive discussion of existing, more or less implicit definitions in the (completely qualitative) literature. They settled on a definition of the “race film” as “a film with African American cast members, produced by an independent production company, and discussed or advertised as a race film in the African American press.” [quotation from a different, also not openly accessible paper in The Moving Image] They also kept a separate file for “discarded data” so that other scholars could make different decisions. In her interview by Miriam Kienle (with Joanna Drucker), Posner makes more general comments based on this example that very much resonate with our book. “part of the point of the humanities is that one person’s ontology won’t be the same as another person’s;” there must be a solution to this problem that does not involve “ever more detailed and complex data models.” And she explains the gains from the “iterative process” of making choices in order to create a digital product. “When one has to think systematically about how to (…) categorize one’s sources, one realizes how much it’s possible to elide the details of these decisions in narratives. Sometimes this tension between what goes unsaid in text-based scholarship and what needs to be made explicit in a data-based project becomes the real question at the heart of your work.”
Posner and Cifor more briefly mention another issue with categorization: they decided not to include a “race” variable in their “People” table, because “individual people could and did change races during their lifetimes – a fluidity that exceeded our own ability to capture the data properly.” The fact that people fall, during their lives (or even at one given moment), in two or more different categories does not, however, in itself preclude categorization in a database, especially during the input phase. It would be perfectly OK to have several (many) columns for “race” giving details as to how a person was characterized in different periods, by different observers (or by themselves), according to different sources, and then to decide on how to simplify (or not) this information. Perhaps the data available in the “race film” project did not allow it. But if someone tells you that it’s impossible for reasons intrinsic to database, computers, or quantification, they are lying – or lack imagination. Data does tolerate nuance, but it requires nuance to be made explicit.