ABSTRACT
The vexing practical problems associated with navigating an uncertain world have been with us for millennia, going back to the very beginnings of our written historical record (as an example, see the discussion in Rubin [1971], “Quantitative Commentary in Thucydides,” and on the latter’s monumental work, The History of the Peloponnesian War, dating from the fifth century B.C.).1 By the 17th century, work began to appear that was statistical (or more accurately, stochastic): the 1650s correspondence between Fermat and Pascal on probability (Devlin, 2008); Huygens’ 1656/1657 tract on probability and reasoning in games of chance; and John Graunt’s Natural and political observations mentioned in a following index, and made upon the bills of mortality (1662), which contained both data and wise suggestions about how to interpret variability, and generally established the value of careful observation of imperfectly collected human data. These isolated incidents were but the start of what developed quickly. In 1710, John Arbuthnot’s “Argument for Divine Providence” used Graunt’s data to illustrate how to do formal hypothesis testing. In the same century, Adrian Marie Legendre (1752-1833) used least squares to do astronomical calculations in the face of the uncertainty inherent in such measurements (Legendre, 1805); also, in that golden age of science, Pierre Laplace (1749-1827), Abraham De Moivre (1667-1754), and Jacob Bernoulli (1634-1705), among many others, were developing and using the tools that to-
day help us measure, understand, and control uncertainty. Despite this long preamble, statistics, the science of uncertainty, is primarily a 20th century creation, dating its modern progenitors from Francis Galton (1822-1911), Karl Pearson (1857-1936), Francis Ysidro Edgeworth (1845-1926), George Udny Yule (1871-1951), and then to R. A. Fisher (1890-1962). (The definitive history of statistical thought before 1900 is available in Stigler [1986].)
The emerging field of statistics is marked by the publication of a handful of critically important books. Arguably, the two most important are R. A. Fisher’s (1925) Statistical Methods for Research Workers, and a half century later, John Tukey’s (1977) Exploratory Data Analysis. The former laid out the analytic tools for generations of scientists; the latter provided the formal justification for a new way of thinking about statistical methods. Taken as a pair, they provide the basis for modern statistics. The focus of this book is principally on the view of statistics as expressed by Fisher. This is not to denigrate exploratory methods, which we believe are at least as important, and usually a lot more fun, but only because of length restrictions. We leave a parallel discussion of exploratory procedures to other accounts, although we touch on them here and there when we cannot resist the temptation.
