On data and capta

On the latest episode of First Draft, we revisited the distinction between capta and data. Most of you should know what data is, but if you're unfamiliar with capta, it's a term created by Johanna Drucker. Capta is used to distinguish the uncertain and incomplete raw material used to study history or literature (or the like) from the presumably complete and unambiguous raw material used to study the habits of Netflix viewers (or the like). Data visualization is a term invented by industry, for doing useful things like identifying trends, and when it's used without reflection in humanities scholarship, for doing useless things like understanding the shape of the Roman World, then, to butcher the expression, the medium can massage the message, and make readers think that the scholarship is far more exact than it really is. There is also a laudable impulse amongst digital humanities scholars to push back against being useful, and in bring a critical gaze on anything that pretends to be purely practical.

It's been half a year since I moved on from Stanford and away from doing digital humanities scholarship and into more traditional business intelligence and data visualization tasks that leverage data in the former sense. In that time, I had the chance to attend a workshop by Stephen Few, who teaches participants how to effectively create data visualization, and was surprised to hear him mention Graphesis, a recent text by Johanna Drucker, but then not surprised to hear him lambast it as incomprehensible and useless.

I find the concept of capta useful, and likewise the emphasis on the incompleteness and uncertainty of data also useful, but without practical implementation, it's little more than haranguing. To be clear, I also disagree with the orthodox view of data visualization typified by Stephen Few, which focuses solely on numerical precision as the purpose of data visualization, and almost wholly on the summary embedded in a presentation or the quick decision-making of a dashboard as the only modes of data visualization. But I'm more sympathetic to that side because it is more rooted in practical constraints: you cannot expect readers to have strong data visualization literacy right now, and trying to implement exotic data visualization methods is very risky.

In contrast, the idea that incompleteness and uncertainty looms so large in the production of data visualization--somehow moreso than it does in the production of a textual representation of a process or phenomenon--to the point that it cripples the practical application of data visualization, signals a lack of imagination. In my time working in digital humanities scholarship, I saw much ink spilled about uncertainty and incompleteness (as I mention in the podcast, a few clouds and blur filters and opacity scales) and very little real progress in the creation of visual elements or the application of visual channels.

This is readily apparent in the best practices of the respective camps. There are now innumerable books telling you just how to create effective charts and graphs. Most of them speak ill of pie charts and push back against Tufte as too radical and artistic.^* As I noted, I don't agree with such a conservative take, but I know that it's improved my line charts and horizontal bar charts. Where are the equivalent best practices in representing uncertain and incomplete data?

This isn't an idle question. In industry we still need to tackle this problem--and not only in visually representing significance and approximate values like that produced by HyperLogLog. It's especially the case when we have a complete dataset but, due to the limitations of the data delivery or visual complexity of the result, we can only represent on-screen some portion of that dataset.

The work in digital humanities has done a good job criticizing the implication of exactitude in data analysis and visualization, and has produced a few terms, like "capta", that express some of these concepts. But it's done little to establish best practices in how to deal with them. Which is a shame, because they're so very useful.

Links

The First Draft - Season 2, Episode 9 "If We Have Opinions, Let's Go With Mine"

^*I fondly remember a demonstration of ggplot2 where the presenter took the time to show the audience that Edward Tufte was now creating sculpture and other art installations, as a way of indicating that he could not be trusted for good charting principles.

Notes

This originally, erroneously, referred to Drucker's term as "captcha" when it is, of course, "capta".