190 Life and Letters of Francis Galton 



these terms tends to obscure finer resemblances. This peculiarity of the 

 loops recurs in further investigations made by Galton with the aid of 

 Howard Collins, and the former writes : 



" I am unable to account for this curious behaviour of the loops, which can hardly be due 

 to statistical accident, in the face of so much concurrent evidence." (p. 185.) 



But I think the explanation lies in the fact that resemblance is lost when 

 a very broad category such as "loops" is taken. 



Galton, however, did see the difficulties of the Arch-Loop-Whorl classifica- 

 tion, though not as far as I can judge of the limitation due to "corresponding 

 finger." He accordingly prepared a set of 53 standard patterns, of which 46 

 are in pairs for "inner" and "outer," i.e. each pair is a mirror reversal. They 

 are for the right hand, and the numbers of each pair of the last 46 must be 

 reversed when we deal with the left hand. He calls this the " (7-set ol 

 Standard Patterns," as Mr Howard Collins performed most of the tabulation 

 under the C-set of patterns. The data consisted of right fore, middle and 

 ring fingers in 150 couplets of siblings*, 900 digits in all. Unluckily the 

 "C-set of Standard Patterns" is in one, the most important, respect almost 

 as defective as the Arch-Loop-Whorl classification. While in the former 

 treatment 129 out of 210 finger-prints fell into the loop category here 291 

 out of 900 finger-prints fall under the pattern No. 42, which is practically 

 the simple loop; it is clear that this standard set of 53 patterns has failed to 

 meet the inherent difficulty of breaking up the bulk of the loops. 



Our author proceeds in the same way to deal only with complete re 

 semblances, i.e. he deals only with the diagonal of his contingency table, 

 disregarding the possibility that a deBciency below the random value may 

 be as important in measuring association as an excess above that value. Com- 

 paring in this way random values, observed values and maximum possible 

 values, and applying his method of the centesimal scale, Galton obtains the 

 following results: 



Resemblance of Siblings, 150 Couplets: forefinger, 9°; middle finger, 

 10° ; ring finger, 12°. We have no probable error given for this method of 

 computing association, but it may be to some extent estimated by the fact 

 that an additional 50 couplets, worked out for middle finger only, gave a 

 value of 21°. For loops on the middle finger only, the 150 couplets gave r25°, 

 and the 50 couplets 8°, indicating little if any association. In nearly all 

 cases the random values were below the observed; in the few cases where 

 they are not so they were only slightly in excess. I think there is enough 

 to show that fraternal resemblance exists, but I personally hold that the 

 classification is rather inadequate, and the statistical method of reduction is 

 unsound. 



Galton next turns (p. 185) to the degree of resemblance in twins. Here 

 he has two series, each of 1 7 sets of twins for the fore, middle and ring fingers 

 of the right hand. In the first series 19 of the 51 finger-print pairs gave the 

 same pattern for the same fingers of both twins, 13 gave partial resemblance 

 and 19 disagreement. Or, as he puts it, of 17 sets of three fingers, two 

 * Pairs of children with the same parents without regard to sex. 



