If Big Data reasoning relies on the predictive power that comes from repetitive processing of unstructured data, this data can be generated, directly or indirectly, from whole populations. It no longer relies on the carefully constructed samples of the general population on which classical statistical knowledge relied. This new model of knowledge requires collecting data continuously, so that N might as well be all.

Nick Couldry & Ulises A. Mejias, The Costs of Connection, loc. 398