A measurement is one of those scientific constructs like time and mass: you think your intuition has it covered until you start to build with the construct and discover that your intuition is not as well informed as you thought it was. Stephen Stigler’s The History of Statistics: The Measurement of Uncertainty before 1900 is the story of the discovery and bridging of gaps between intuitions about and reasoning with quantitative measurements.
The book begins with a question that bedeviled experimental philosophy: how do you combine two measurements of the same thing that are different? Book III of Newton’s Principia Mathematica is for the most part devoted to this problem. Regarding Problem XXI, From three observations given to determine the orbit of a Comet moving in a parabola, for example, Newton writes (in translation),
This being a problem of great difficulty, I try’d many methods of resolving it; and several of those problems, the composition whereof I have giv’n in the first book, tended to this purpose. But afterwards I contrived the following solution, which is something more simple.
Newton was analyzing observations of the comet of 1680 made by people with different (and largely unknown to Newton) levels of metrology expertise, using different mathematical instruments, and in different locations around the world.
Using detailed analysis of the works of Tobias Mayer on the libration of the moon, Euler on the orbits of Saturn and Jupiter, and Roger Boscovich on the shape of the earth, among others, Stigler recounts how the method of least squares slowly emerged as method for combining observations in a way that enhanced rather than dissipated their utility
This methodology frames each chapter in Stigler’s book. First, Stigler describes the then-current state of thinking about some aspect of measurement in a particular scientific context. Then he recounts examples of a conceptual conundrum that people encountered when they tried to extend the science based on their understanding. Next, comes an analysis of the various attempts to resolve the conundrum. And finally, there is a convergence to a consensus understanding of the nature of the barrier as well as an agreed-upon way of resolving it and moving to a new level of understanding.
After the combination of observations and least squares, Stigler takes up probability and the measurement of uncertainty, inverse probability and Bayes theory, the central limit theorem, sampling and the law of large numbers, distributions other than the normal, and, finally, correlation and regression. In each case Stigler uses detailed analyses of examples drawn from the literature of the day to explore the feedback loop between an intuition and its mathematization.
The book’s bibliography spans twenty-five pages and includes twenty-three articles by Stigler himself. (This is one of those rare situations wherein self-citation is to be lauded.) Together with four-pages of suggestions for further reading we have at hand many starting points for further exploration of the history of statistics. There are two appendices, each being a syllabus for a series of lectures by F. Y. Edgeworth at King’s College.
The strength of Stigler’s history is that we witness an aspect of doing mathematics that is, in my view, too infrequently written about; the difficulties in getting from scientific phenomena and their measurement to mathematical notations and their calculus. Chances of events, observation error, and correlated events are all aspects of measurement about which everyone has an intuitive understanding. There are huge difficulties attendant to rendering this understanding mathematically so that as you manipulate the mathematics the intuition comes along. Stigler’s book is a scholarly telling of the early history of statistics, but more than that: it is collection of case studies of how mathematics is really done that is as relevant today as it was when Newton was trying to figure out how to combine comet observations from Boston, Massachusetts, and Cambridge, England.
Scott Guthery is the author of Practical Purposes: Readers in Experimental Philosophy at the Boston Athenaeum (1827–1850), in which he uses the book borrowing registers of the Athenaeum to characterize the scientific and technical reading preferences of Boston’s antebellum mathematical practitioners. His previous book, A Motif of Mathematics, explored the history and application of the mediant and the Farey sequence. Guthery received a PhD in probability and statistics from Michigan State University and worked for Bell Laboratories, Schlumberger, and Microsoft before co-founding two of his own companies. He can be reached by e-mail at scott@docentpress.com.