Book

Description

Analytics is taught as inference from a dataset. Much of an analyst's work is the opposite problem: the quantity a decision requires is absent, and no further query will produce it. This book develops three methods for constructing missing values, proxy-based temporal disaggregation, projection by analogy to a comparable unit, and synthetic generation of a rare class, alongside a procedure for designing measurements that do not exist until derived. Its claim is that every constructed quantity is part observation and part assumption, and that the division can be established before implementation begins.