8 ETHEL M. ELDEKTON 



for the Family Measurements of six years ago* : they were loaned to College students, 

 personal friends of members of the Biometric Laboratory and others. It was, however, 

 soon obvious that we had miscalculated the ease with which pairs of cousins could be 

 found and measured. The work went forward extremely slowly, most investigators 

 sent in only two or three pairs ; and when the question of repeating or verifying 

 a measurement arose, the delay or even the impossibility of supplementing the data 

 was much more common than in the case of the Family Records. In fact cousins are 

 not like brothers and sisters, or pai'ents and offspring, in daily touch with each other; 

 and at the end of three or four years, we are far from having reached a sufficient 

 supply of cousin pairs. Hardly indeed have 300 pairs been yet measured. Accord- 

 ingly this side of the enquiry is incomplete and will only be used as a control series. 

 We need scarcely say that we shall be very glad indeed to loan spanner and scales to 

 any reader of this memoir who will undertake to measure pairs of adult cousins. 



4. We now turn to the analytical methods by which the material was reduced. 

 We have already pointed out that contingency was used throughout the whole of the 

 first series, but that in the case of Temperament the 3 x 3-fold tables were not finely 

 enough grouped to make contingency thus obtained really comparable with that found 

 from higher-fold tables. Accordingly the temperament tables were only worked out 

 by 3 x 3-fold contingency in the case of the three groups : male cousin pairs, female 

 cousin pairs, and male and female cousin pairs. The fourfold table methodt was used 

 on the same material in thirty cases, namely the three classes of temperament in the 

 ten classes of cousins. 



The mean value of the degree of resemblance between cousins as found from 

 40 contingency tables was "271 + '009 with a standard deviation of "083 + "006. The 

 mean value of the degree of resemblance in temperament as found from 30 tables by 

 fourfold process was - 258 + - 014, but the variability in this case was 35 p. c. greater, 

 the standard deviation being •115 + 'OrO. 



The temperament tables worked by contingency with three types of cousins gave 

 the result -238 ±"010, with the reduced variability indicated by -045 + -007. These 

 results are collected in Table I. They suffice to show that mean square contingency 

 methods give more uniform results than the fourfold tables i. But the mean found 

 from contingency for Health, Ability, Temper and Success does not sensibly differ 

 from that found for the three divisions of temperament by fourfold tables leading to 

 the coefficient of correlation. If we combine the results of both methods so as to 



* Directions and form of schedule for this case are reproduced in Biometrika, Vol. n. pp. 359-60. 



t The fourfold tables were worked for two groupings for the alternatives Excitable or Calm ; the 

 Betwixts being thrown first into one group and then into the other, the average value of the correlation 

 coefficient is that given in Table III. In the case of the alternatives Reserved and Exjjressive the Betwixts 

 were thrown into the Reserved, and in that of the alternatives Sympathetic and Callous into the latter 

 group. This was done after some consideration and enquiry as to the popular weight of terming an 

 individual ' reserved ' or ' callous.' 



\ The corresponding nine fourfold tables were somewhat erratic and gave a mean of only - 19. 



