Positivism and Data Visualization

NOTE: While I've been interested in data visualization and digital humanities more generally for a long time, various commitments have kept me from really coming to grips with what’s going on in the field in a deep way. This blog post is, as usual, thinking out loud and I’m sure overlooks many contributions in the existing literature covering the same ground.

One of the things I’ve kept at the back of my mind is whether “the data turn” might serve to draw humanities scholarship in the direction of positivism and away from traditional overarching orientations like hermeneutics. This is a question that lies at the heart of debates around close reading vs. distant reading, and I think it’s generally covered by arguing that both methods need to complement each other rather than being mutually exclusive.

However, I was struck by this following passage at the beginning of Lev Manovich’s Cultural Analytics. Manovich is discussing the One Million Manga Pages project, and in particular a data visualization that maps out two particular graphical properties of the pages, the standard deviation of each page’s greyscale values, and the entropy of all the pixels’ greyscale values. Manovich notes that the resulting visualization shows that among the one million pages, “we find every possible stylistic variation”.

This suggests to me that our basic concept of “style” may not be appropriate when we consider large cultural datasets. The concept assumes that we can partition a set of cultural artifacts into a small number of discrete categories. In the case of our One Million Manga Pages dataset, we find practically infinite graphical variations. If we try to divide this space into discrete stylistic categories, any such attempt will be arbitrary.

I think Manovich is making a mistake here. He is assuming that “our basic concept of ‘style’” is based on properties of the object under consideration - in other words, he is applying a positivist approach to his understanding of the dataset. In many ways, this makes sense; when we analyze a dataset, what we have in front of us are the positive properties of “the objects themselves”. Any conclusions we want to draw from the dataset must be based on observations of those properties.

But this approach by definition excludes any understanding of, say, “style” that relies on phenomena that lie outside our observations of the objects under consideration. But this does not make such an understanding “arbitrary”, simply social or historical - in other words, hermeneutic.

One of the characteristics of postmodern culture is an attempt to avoid the kind of “arbitrary” abstraction that Manovich here implicitly rejects. Postmodernism thought seeks to comprehend the infinite variations of empirical life without flattening, erasing, or aggregating. Fredric Jameson succinctly describes postmodernism’s challenge to abstraction and schematization in Marxism and Form. In his discussion of an earlier philosophical attempt to not oversimplify and reduce the richness of empirical life, existentialism, Jameson remarks that

it may be maintained that in a sense all understanding, all abstract thought is reductive: indeed, the very process of abstraction itself is in its very essence a reduction through which we substitute for the four-dimensional density of reality itself simplified models, schematic abstract ideas, and thereby do violence to reality and to experience. On the other hand, it is difficult to see how we could understand or deal with reality in any other way than by such reduction.

This gets at a major problem for the philosophy of science: empiricism requires that we take account of every observation without drawing unsupported conclusions; but the very fact of taking multiple observation requires that we seek to describe them in some common fashion - this is all a scientific theory or law really is. It is hard to understand how we could even do science without combining multiple unique observations under a single description. In many was, the discipline of statistic was developed as a way to side-step the problem: a statistical (positivist) description of multiple observed phenomena was considered less “arbitrary” than other kinds of description. The rift between positivism and hermeneutics follows this line of argument.

It must be remembered that postmodernism/neoliberalism developed alongside the increase in computational capacity and the construction of large data sets. It makes sense, then, that the postmodern insistence on the irreducible richness of empirical reality should find support in computational techniques of analysis. The data visualization Manovich is discussing is no less abstract and reductive, but because it is positivist it is acceptable; other kinds of abstraction are dismissed as arbitrary.

But human pattern-recognition and sense-making is never purely positivist, as the Jameson quote underlines. Our understanding of “style” cannot be circumscribed or limited by positive characteristics present in the objects under observation. But this doesn’t make the concept of style “arbitrary”, it simply places the criteria for abstraction - for human pattern-recognition and sense-making - outside the objects being observed. In other words, such abstraction and categorization are social and historical. This makes them less tractable to positivist analysis, which is why we have hermeneutics in the first place.

Indeed, historians of style insist on the social and historical determinants of style. Charles Rosen, in The Classical Style, makes the point that style cannot be understood in a positivist way: “the concept of a style does not correspond to a historical fact but answers a need: it creates a mode of understanding” (i.e. it is hermeneutic).

In many ways, positivism in cultural studies is part and parcel of the anti-interpretive tendency Jameson notes in The Political Unconscious. Jameson writes that this tendency sees interpretation/hermeneutics “as a reduction and a rewriting of the whole rich and random multiple realities of concrete everyday experience into the contained, strategically limited terms” of a given hermeneutic approach. Jameson’s work is a defence of interpretation - specifically Marxist interpretation - in the face of postmodernist tendencies in literary studies, but his defence applies also to the positivist approach described here.

Both postmodernism and computer-aided positivism, then, support each other in their resistance to hermeneutics, which they see as various attempts to impose a totalizing (indeed, a totalitarian) finality on cultural artifacts. Postmodernism seeks to insulate culture through play, instability, and radical contingency; positivism by outsourcing human comprehension to a “non-arbitrary” function of “neutral” or “objective” computational analysis. Not only does this woefully misunderstand hermeneutics itself (as Jameson argues), but it obscures the very ways that both postmodernism and positivism continue to abstract and reduce empirical reality for the sake of comprehension. Hermeneutics argues that such abstraction should not be rejected, but embraced with a full understanding of its limitations.

All of this has broader repercussions, of course. When we want to describe structures of oppression we violate the insistence on positivism that is characteristic of bourgeois society, as well as the postmodern decentralization (and neutralization) of power itself. Our descriptions of structures of power and how they operate are “arbitrary” (because not visible properties of objects under observation) and therefore dismissible. They require comprehension, and therefore abstract understanding; they can’t simply be read off empirical properties of things. So while I don’t think positivism is necessarily the only approach one can take with respect to cultural analytics, but it is a risk, and we need to remain vigilant when we see it uncritically promoted.