Some very important things don’t make their way into the data. It’s easier to justify health care decisions in terms of measurable outcomes: increased average longevity or increased numbers of lives saved in emergency room visits, for example. But there are so many important factors that are far harder to measure: happiness, community, tradition, beauty, comfort, and all the oddities that go into “quality of life.”
It’s tempting to use the term intangible when what we really mean is that such things are hard to quantify in our modern institutional environment with the kinds of measuring tools that are used by modern bureaucratic systems. The gap between reality and what’s easy to measure shows up everywhere. Consider cost-benefit analysis, which is supposed to be an objective—and therefore unimpeachable—procedure for making decisions by tallying up expected financial costs and expected financial benefits. But the process is deeply constrained by the kinds of cost information that are easy to gather. It’s relatively straightforward to provide data to support claims about how a certain new overpass might help traffic move efficiently, get people to work faster, and attract more businesses to a downtown. It’s harder to produce data in support of claims about how the overpass might reduce the beauty of a city, or how the noise might affect citizens’ well-being, or how a wall that divides neighborhoods could erode community. From a policy perspective, anything hard to measure can start to fade from sight.
An optimist might hope to get around these problems with better data and metrics. What I want to show here is that these limitations on data are no accident. The basic methodology of data—as collected by real-world institutions obeying real-world forces of economy and scale—systematically leaves out certain kinds of information. Big datasets are not neutral and they are not all-encompassing. There are profound limitations on what large datasets can capture.
I’m not just talking about contingencies of social biases. Obviously, datasets are bad when the collection procedures are biased by oversampling by race, gender, or wealth. But even if analysts can correct for those sorts of biases, there are other, intrinsic biases built into the methodology of data. Data collection techniques must be repeatable across vast scales. They require standardized categories. Repeatability and standardization make data-based methods powerful, but that power has a price. It limits the kinds of information we can collect. (...)
These limitations are particularly worrisome when we’re thinking about success—about targets, goals, and outcomes. When actions must be justified in the language of data, then the limitations inherent in data collection become limitations on human values. And I’m not worried just about perverse incentives and situations in which bad actors game the metrics. I’m worried that an overemphasis on data may mislead even the most well-intentioned of policymakers, who don’t realize that the demand to be “objective”—in this very specific and institutional sense—leads them to systematically ignore a crucial chunk of the world.
Decontextualization
Not all kinds of knowledge, and not all kinds of understanding, can count as information and as data. Historian of quantification Theodore Porter describes “information” as a kind of “communication with people who are unknown to one another, and who thus have no personal basis for shared understanding.” In other words, “information” has been prepared to be understood by distant strangers. The clearest example of this kind of information is quantitative data. Data has been designed to be collected at scale and aggregated. Data must be something that can be collected by and exchanged between different people in all kinds of contexts, with all kinds of backgrounds. Data is portable, which is exactly what makes it powerful. But that portability has a hidden price: to transform our understanding and observations into data, we must perform an act of decontextualization.
An easy example is grading. I’m a philosophy professor. I issue two evaluations for every student essay: one is a long, detailed qualitative evaluation (paragraphs of written comments) and the other is a letter grade (a quantitative evaluation). The quantitative evaluation can travel easily between institutions. Different people can input into the same system, so it can easily generate aggregates and averages—the student’s grade point average, for instance. But think about everything that’s stripped out of the evaluation to enable this portable, aggregable kernel.
Qualitative evaluations can be flexible and responsive and draw on shared history. I can tailor my written assessment to the student’s goals. If a paper is trying to be original, I can comment on its originality. If a paper is trying to precisely explain a bit of Aristotle, I can assess it for its argumentative rigor. If one student wants be a journalist, I can focus on their writing quality. If a nursing student cares about the real-world applications of ethical theories, I can respond in kind. Most importantly, I can rely on our shared context. I can say things that might be unclear to an outside observer because the student and I have been in a classroom together, because we’ve talked for hours and hours about philosophy and critical thinking and writing, because I have a sense for what a particular student wants and needs. I can provide more subtle, complex, multidimensional responses. But, unlike a letter grade, such written evaluations travel poorly to distant administrators, deans, and hiring departments.
Quantification, as used in real-world institutions, works by removing contextually sensitive information. The process of quantification is designed to produce highly portable information, like a letter grade. Letter grades can be understood by everybody; they travel easily. A letter grade is a simple ranking on a one-dimensional spectrum. Once an institution has created this stable, context-invariant kernel, it can easily aggregate this kind of information—for students, for student cohorts, for whole universities. A pile of qualitative information, in the form of thousands of written comments, for example, does not aggregate. It is unwieldy, bordering on unusable, to the administrator, the law school admissions officer, or future employer—unless it has been transformed and decontextualized.
So here is the first principle of data: collecting data involves a trade-off. We gain portability and aggregability at the price of context-sensitivity and nuance. What’s missing from data? Data is designed to be usable and comprehensible by very different people from very different contexts and backgrounds. So data collection procedures tend to filter out highly context-based understanding. Much here depends on who’s permitted to input the data and who the data is intended for.
by C. Thi Nguyen, Issues in Science and Technology | Read more:
