A history of regression: actors, networks, machines, and numbers

In this paper the history of correlation and regression analyses, both in the discipline of statistics generally and in human geography particularly, is examined. It is argued that correlation and regression analysis emerged from a particular social and cultural context, and that this context entered into the very nature of those techniques. The paper is divided into three sections. First, to counter the idea that mathematics and statistics are somehow outside the social, the arguments put forward by David Bloor and Bruno Latour suggesting that mathematical propositions are socially constructed are briefly reviewed. Second, using the ideas of both Bloor and Latour I turn to the development of statistics as an intellectual discipline during the 19th century, and specifically to the invention of correlation and regression at the end of that period. It is argued that the development of statistics as a discipline and its associated techniques are both stamped by, but also leave their stamp on, the wider society in which they are set. Last, the importation of correlation and regression analyses into human geography which occurred in the 1950s is examined. Following my general social constructionist argument, it is suggested that because of the difference in context the correlation and regression analyses devised in the late 19th century were often inappropriate for mid-20th century spatial science.