Miscellanea 
321 
less chance of getting their record. Hence the problem is by no means without difficulty, but it 
is of such great interest that we venture to give here our chief results. 
Our health classes were : Very Robust, Robust, Normally Healthy, Rather Delicate, Delicate 
and Very Delicate. These have been verbally defined, and our present definitions are as 
follows : 
SCALE. 
V.R. Very robust. — He has never had to see a doctor, nor been off work through illness. 
R. Robust. — He has only seen a doctor about minor ailments, and has only been off 
work for colds, etc. 
N.H. Normally healthy. — He has not had more than one serious illness, involving, say, 
a fortnight's absence from work during the last ten years. 
R.D. Rather delicate. — He has had more than one serious illness, but not more than one 
involving more than four weeks' absence from work during the last ten years. 
D. Delicate. — He is off work through illness at least four weeks in all every year. 
V.D. Very delicate. — He is in a chronic state of ill-health. 
But these definitions were not used throughout the whole of the records included in the present 
investigation, and we found considerable reluctance to the use of the " Very Delicate " category. 
Accordingly in the present reduction all the delicate categories have been clubbed together. 
As in previous investigations a Gaussian frequency scale was used for Health*, the interval 
on the scale covered by " Normally Healthy " being taken to represent 100 units of health, 
which may be called sanitaces. These hundred sanitaces were supposed to represent the range 
of normal health of each type of individuals — fathers, mothers, sons and daughters, and the rest 
of the distribution calculated in terms of them. At the mean of the parental category was 
then plotted up the mean number of sanitaces of all the offspring of either sex of parents of 
the given category. Thus the round black dots of our four first diagrams were obtained. It 
was found that the three points marking the mean health of offspring of " Very Robust," 
" Robust " and " Normally Healthy " parents were in all four cases closely on a line which passed 
through the mean of parental and filial health, but that the mean health of children of delicate 
parents in all four cases gave a point lying markedly above this line. This line in all four cases 
shows a marked slope indicating that as the health of the parents is worse so the health of their 
offspring is worse. 
To illustrate the probable source, or at least part source, of the anomaly in the case of the 
delicate parents, the percentage of sons dying before 21 years of age and so escaping record 
of health was obtained for each health group of the fathers and this percentage was plotted in 
Diagram V. to a Gaussian scale of Father's health. Again we note a fairly uniform increase of 
the percentage of deaths with decreasing goodness of health of the father, for the three better 
categories, but a very marked excess of deaths — quite off the line — in the category of delicate 
fathers, where some 13 °/ o of sons die as against the 7 °/„ in the case of " Very Robust " fathers. 
(2) Still preserving our Gaussian scales, we may consider that it is the sons of delicate 
fathers who have been especially selected. In this case the health of the sons of the three 
left-hand categories of fathers will give very nearly the true regression line for unselected 
material. We have made no attempt to algebraically tit the best lines to these three points, 
but placed them graphically on the points and through the general means t. 
* Cf. Huxley Lecture, Biometrika, Vol. in. p. 146. 
t This was done because the three left-hand points always fall nearly on the same line with the 
general means, but of course the means of both parents and offspring would be somewhat lowered had 
we the full complement of delicate sons of delicate parents: — to increase the number of such sons 
would ipso facto increase the number of such parents and thus both means of health iu offspring 
and parents would be lowered in the same direction probably approximately that of the true regression 
line. 
Biometrika ix 41 
