Statistical Investigations 383
assignment of rank in a class, or by making other deductions that I have not space to refer to here, such as the numerical values by which the nearness of different degrees of kinship may be expressed, or the closeness of correlation between different parts of the body. There is no intrinsic difficulty in grasping the conceptions of which I speak, but they are foreign to present usage and look strange at first sight. They are consequently very difficult to express briefly and intelligibly to those to whom they are wholly new."
It will be seen that as early as 1889 Galton was fully assured that the ideas of a correlational calculus opened a new world of thought, not only to the trained scientist, but to every man of education who could master his natural inertia and endeavour to grasp the new conceptions.
Chapter II of the pamphlet is entitled " Human Variety." This was Galton's final address to the Anthropological Institute at the end of the four years during which he had held the office of president'. The paper opens with a paragraph on finger-prints which shows that Galton was working at the subject, and already fully recognises} its importance for identification. He especially refers to Sir William Herschel's use of the method in India, and suggests its use in North Borneo for identifying the coolies, and in other cases where there may be fraud from impersonation of pensioners and annuitants. Galton then turns to correlation,
"a very wide subject indeed. It exists wherever the variations of two objects are in part due to common causes; but on this occasion I must only speak of such correlations as have an anthropological interest."
He tells us that the particular problem he first had in view -was to ascertain the practical limitations of the ingenious method of anthropometric identification due to M. Alphonse Bertillon, which was then in habitual use in the criminal administration of France. Correlation between the various measurements would obviously be a serious defect of the Bertillon system, and Galton suspected strongly the existence of this source of error'. An element of history is now revealed
"The first results of the inquiry, which is not yet completed, have been to myself a grateful surprise., Not only did it turn out that the measure of correlation between any two variables is exceedingly simple and definite, but it became evident almost from the first that I had unconsciously explored the very same ground before. No sooner did I begin to tabulate the data than I saw that they ran in just the same form as those that referred to family likeness in stature,
2 1 Delivered Jan. 22,1889. See Journal of the Anthropological Institute, Vol. xviir, pp. 401-19. The first complete analysis of these correlations was given by the late Dr W. R. Macdonell and he indicated how an index could be constructed of artificial functions of the Bertillon measurements in which this difficulty of correlation would be satisfactorily surmounted, Biometrika, Vol. i, 1902, pp. 177-227, "On Criminal Anthropometry and the Identification of Criminals." He shows that the correlations of the Bertillon measurements are high, far too high for indexing purposes. Galton first expressed his doubts on this point after Bertillon's discourse on his system before the Anthropological Institute. "There maybe room for reasonable doubt among anthropologists whether the precision with which the living body can be measured is quite as great, and whether its dimensions are quite as permanent, as they are considered to be by M. Bertillon; and again there may be some hesitation in believing that a very large collection of measures would admit of being so surely catalogued on the Bertillon system as to be ransacked with a promptitude at all corresponding to that with which a word may be found in a huge dictionary" (Journal of the Anthropological Institute, Vol. xx, p. 198). This was one of the early clashes in the contest between the Bertillon and Galton systems, which was to end ultimately with the victory of the latter.