The people referred to are taken from a collection of histories of long life made by the Genealogical Record Office of Washington.[189] One hundred individuals were picked out at random, each of whom had died at the age of 90 or more, and with the record of each individual were placed those of all his brothers and sisters. Any family was rejected in which there was a record of wholly accidental death (e.g., families of which a member had been killed in the Civil War). The 100 families, or more correctly fraternities or sibships, were classified by the number of children per fraternity, as follows: Number of Total number Number of children per of children fraternities fraternity in group 1 2 2 11 3 33 8 4 32 17 5 85 13 6 78 14 7 98 9 8 72 11 9 99 10 10 100 3 11 33 2 12 24 1 13 13 --- --- 100 669 The average at death of these 669 persons was 64.7 years. The child mortality (first 4 years of life) was 7.5% of the total mortality, 69 families showing no deaths of that kind. The group is as a whole, therefore, long-lived. The problem was to measure the resemblance between brothers and sisters in respect of longevity,--to find whether knowledge of the age at which one died would justify a prediction as to the age at death of the others,--or technically, it was to measure the fraternal correlation of longevity. A zero coefficient here would show that there is no association; that from the age at which one dies, nothing whatever can be predicted as to the age at which the others will die. Since it is known that heredity is a large factor in longevity, such a finding would mean that all deaths were due to some accident which made the inheritance of no account. In an ordinary population it has been found that the age at death of brothers and sisters furnishes a coefficient of correlation of the order of .3, which shows that heredity does determine the age at which one shall die to considerable extent, but not absolutely.[190]