bq. We cut nature up, organize it into concepts, and ascribe significances as we do, largely because we are parties to an agreement to organize it in this way — an agreement that holds throughout our speech community and is codified in the patterns of our language […] all observers are not led by the same physical evidence to the same picture of the universe, unless their linguistic backgrounds are similar, or can in some way be calibrated.
As expressed by Benjamin Whorf, that’s the essence of what has come to be known as the Sapir-Whorf or “linguistic relativity” hypothesis. It’s what came to my mind recently when I read an ingenious new study by Thomas Hammond, Kyle Jen, and Ko Maeda in the latest issue of the Journal of Theoretical Politics (abstract here).
The basic idea underlying the study is that how an organization structures the information that it possesses is likely to affect what decision makers learn from this information. Lacking an ability to create experimental organizations that structure information in different ways, the researchers turned to the library catalogue — which, after all, is a hierarchical organizer of knowledge — as a testing site. The research question boiled down to whether users in a university library that uses the Library of Congress cataloguing system could be led by these two different information structures to behave differently than those in a Dewey Decimal-based university library. (The Northwestern and Michigan State libraries, where they conducted the study, contain virtually the same number of books overall — approximately 4,000,000 apiece.)
Hammond and his colleagues began by identifying 40 “classic” political science books (e.g., Essence of Decision, Who Governs?, Presidential Power, and The Politics of the Budgetary Process. They then identified the 50 books immediately surrounding each of the 40 “classics” in the two libraries’ catalogues. (This task required considerable cross-checking to ensure that both libraries contained all the surrounding books.)
The two sets of books in the vicinity of the target book turned out to be “almost completely disjoint, that is, the 50 books around the target book in the Library of Congress library were usually quite different from the 50 books around the same target book in the Dewey Decimal library.” Across the 40 “classics,” the mean number of surrounding books in common was just 3, the median was 2, and the mode was 0. Follow-up analyses of various sorts established the robustness of these results.
So what? Well, it may be a bit of a stretch, but consider the following scenario:
bq. …let us interpret the target book and the 50 books surrounding the target book as 51 bits of data that are potentially relevant to the possibility of a surprise attack, and let us assume that these 51 bits of data, if analyzed together, would provide compelling evidence of an impending surprise attack. …But now assume that the intelligence community is reorganized so that its new structure is as different from the old structure as Northwestern’s Dewey Decimal catalogue is different from Michigan State’s Library of Congress catalogue. The critical organizational question is this: would any analysts in the second structure be likely to see these same 51 bits of data and thus be as likely to draw the same inference that an attack is imminent?
The answer, the authors conclude, is clearly no: “Analysts in the second structure who managed to identify one initial bit of critical data … might be expected to see, on average, only three additional bits of data that are also relevant; the other 47 relevant bits of data seen by the analysts in the first structure would be effectively invisible to the analysts in the second structure.”