Photographic Researches and Portraiture 305
Bertillon divides each of his four fundamental characters into three groups -large, medium, small-and Galton points out that the difference between the men at the extremes of the medium group is, for stature, say 2.3 inches while the possible error of determining stature may be ± 0.5 inch ; that is to say, that there is a total doubtful range of 2 inches, while the medium range itself is only 2.3 inches. He further pointd out that nearly all Bertillon's characters, we may anticipate, will be highly correlated together and accordingly his 81 (= 34) groups will contain very unequal numbers.
"No attempt has yet been made to estimate the degree of their interdependence. I am therefore having the above measurements (with slight necessary variations) recorded at my anthropometric laboratory for the purpose of doing so." (p. 175.)
I do not think these measurements were ever taken in adequate numbers or that Galton ever determined actually their correlations. This was, I believe, first done by the late Dr Macdonell, on actual criminal data, and he pointed out how, by the use of proper "independent variates," the trouble of correlation in the characters could be eliminated'.
The first difficulty, however, of the border-line cases, which involve such a large proportion of the population and therefore the multiplication of cards, in several groups, Galton got over by what he termed a "mechanical selector." I have not found any `selector' described before 1888, but many since, all involving Galton's principle, some patented, without any recognition of Galton's priority. The idea is indeed a very simple one; each individual has a card 8 to 9 inches long. If there are 4 or 6 indexing characters each is allotted something less than a quarter to a sixth of the card. This portion of the card represents the range of the corresponding variate and a notch is cut into the card at the value of the variate within this range. The breadth of this notch represents twice the possible error of measurement, once in excess and once in defect, for that variate. The cards are placed vertically and loosely in a box divided into batches by partitions so that there is not sufficient friction to interfere with their independent motion. The bottom of the box, except sufficient at the ends for the cards to rest on, is replaced by a "keyboard" as Galton termed it; this keyboard is of the breadth of the variate portion of the cards, and can be elevated by a lever. Adjustable wires can be arranged across a gap in the keyboard of the size of the series of cards, and these wires are adjusted to give the measurements of an individual to be selected, just as the notches are cut in the cards. When the keyboard is elevated its wires pass into the notches of those cards which are within possible errors of the individual set on the keyboard-all the other cards but these are raised and thus discriminated from those which require examination. It is clear that the cards do not require classification by size of organs, but may be placed by age or alphabetically. Galton considered that this mechanical
' Biometrika, Vol. i, pp. 177-227. The Bertillon system of indexing by physical measurements has now been replaced by direct indexing of finger-prints.
2 Actually the notch would not be cut at the exact value of the variate except when near the boundary of the sub-range; in other positions it would suffice to cut it at the middle of the sub-range. For Bertillon's index it would suffice practically to have pin points. marked for each
variate on the card, where notches should be cut.