hope of measuring parental resemblance in moral character. I confined my attention entirely to fraternal resemblance. My argument was of this kind. Regarding one species only, then if fraternal resemblance for the moral and mental characters be less than, equal to, or greater than fraternal resemblance for the physical characters, we may surely argue that parental inheritance for the former set of characters is less than, equal, to, or greater than that for the latter set of characters.

In the next place it seemed impossible to obtain moderately impartial estimates of the moral and mental characters of adults. Who but relatives and close friends know them well enough to form such an estimate, and which of us will put upon paper, for the use of strangers, a true account of the temper, probity and popularity of our nearest? Even if relatives and friends could be trusted to be impartial, the discovery of the preparation of schedules by the subjects of observation might have ruptured the peace of households and broken down life-long friendships. Thousands of schedules could not be filled up in this manner. The inquiry, therefore, resolved itself into an investigation of the moral and mental characters of children. Here we could replace the partial parent or relative by the fairly impartial school teacher. A man or woman who deals yearly with forty to a hundred new children, rapidly forms moderately accurate classifications, and it was to this source of information that I determined to appeal.

...

To illustrate the method I will examine a little at length the degree of resemblance of brothers in a physical character. I choose cephalic index and this for two reasons:

(a) Because from the first few years of life onwards the cephalic index scarcely changes with growth.

I have not yet investigated my own school data from this standpoint, but I have every confidence in the care taken by the late Dr. W. Pfitzner in his elaborate system of measurements, and the above is the conclusion he reaches.

(b) Several great authorities have recently stated that they do not "believe" in the cephalic index, i.e., consider it of small value for anthropometric purposes.

In the Appendix,' we have the cephalic index given for 1982 pairs of brothers. This table is, I hope, perfectly intelligible. Taking the boys, for example, with cephalic indices between 74 and 75, these boys had 78 brothers who were distributed according to the arrangement in the column headed 74 to 75. Brothers are not alike in cephalic index, but distributed with a considerable range of variation. We now take in the usual way the arithmetic mean of this array of brothers, and find it to be 77.45. The average brother of a boy with cephalic index 74.5 has an index of 77.45. This is the phenomenon of regression towards the general 2 Omitted here.

population mean (78.9) as discovered by Francis Galton. Now turning to Diagram 1 we plot to 74.5, the mean brother 77.45, and doing this for all arrays we get the series of points there exhibited. You will see at once that they lie almost exactly on a straight line. This is the wellknown regression line. If that line has a slope of 1 in 1, the brother of 74.5 would have a mean brother of 74.5 cephalic index. If it had no slope at all the brother of 74.5 would have a brother like the mean of the general population. In the one case we have absolute resemblance, in the

67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 OBSERVED INDEX, FIRST BROTHER

SLOPE of REGRESSION LINE=-49

Diagram 1. Resemblance of Brothers in Cephalic Index.

other case no resemblance at all. The actual degree of resemblance, our brothers being equally variable, is measured by the steepness of this regression line. In our case that steepness is .49, almost 5 or 1 in 2. That is the measure of fraternal resemblance in brothers for cephalic index-the correlation between brothers as we term it.

Now we have learnt two great features of inheritance in man. First, that the points in Diagram 1, within the limits of observation, are on a line, and secondly, that the slope of this line is about .5. Are these results true for characters other than the cephalic index? Undoubtedly for all the physical characters yet worked out in man. . . . We cannot hesitate about the regression line being essentially linear. Has it for brethren usually a slope of about .51

In Table 1 are given my observations on some 1000 families for adult brothers and sisters. You will see that the steepness of the regression line is essentially about .5.

In table 23 are given my observations on the head measurements of school children. We note at once precisely the same convenient number .5. I think we, therefore, may safely conclude that for the measurable physical characters in man, we have quite a definite regression line, and that it ascends 1 in 2. . . .

So far we have seen surprising uniformity in the inheritance of the measurable physical characters. How are we to extend our results to physical characters not capable of accurate measurement, and to psychical characters? Clearly the whole problem turns on this: Can we find the steepness or slope of this regression line without all the paraphernalia of the correlation table and the means of arrays? The answer is: Yes; providing we assume a certain distribution of frequency for the GaussLaplacian normal curve of deviations from the mean. Grant this distribution, and by very simple classifications indeed we can determine the steepness of the regression line. Now the problem before us is the following one: Is this assumption legitimate? It is certainly not true for organs and characters in all types of life. But it really does describe in a remarkable manner the distribution of most characters in mankind. We have shown that within the limits of random sampling, it is very true for a great variety of characters in the human skull. Dr. Macdonell has demonstrated it also for measurements on criminals, and you can be fairly convinced of its suitability by looking at one or two diagrams. . . . I should be the last to assert that no human characters can be found that do not diverge sensibly from this Gaussian distribution. But I believe they are few, and that for practical purposes we may with nearly absolute safety assume it as a first approximation to the actual state of affairs. This being once granted we can obtain the slope of our regression line by an exceedingly simple process. We can make a mere classification of

Omitted here.

the following kind, say, into boys with breadths of head below 145 mm., and boys with breadth of head above 145 mm. .

Now from such a division the mathematician can deduce the slope of the regression line on the assumption of normal distribution. Here, to give us confidence, are the results for head breadth and height in boys, which were worked out both ways:

...

For practical purposes these results are identical. I now come to the fundamental idea of my comparison of the psychical and physical resemblance of brothers. Suppose we assume that moral and mental qualities in man, like the physical, follow a normal law of distribution, and that the regression is linear. What results shall we obtain by thus assuming perfect continuity between the physical and the psychical? No doubt the drums will begin to beat the tattoo, we shall hear talk of the hopeless materialism of some men of science. But to use Huxley's appropriate words: "One does not battle with drummers." I cannot free myself from the conception that underlying every psychical state there is a physical state, and from that conception follows at once the conclusion that there must be a close association between the succession or the recurrence of certain psychical states, which is what we judge mental and moral characteristics by, and an underlying physical confirmation be it of brain or liver. Hence I put to myself the problem as follows: Assume the fundamental laws of distribution which we know to hold for the physical characters in man, and see whither they lead us when applied to the psychical characteristics. They must: (a) Give us totally discordant results. If so, we shall conclude that these laws have no applications to the mental and moral attributes. Or, (b) Give us accordant results. If so, we may go a stage further, and ask how these results compare with those for the inheritance of the physical characters: are they more or less or equally subject to the influence of environment? Here are the questions before us. Let us examine how they are to be answered. As an illustration I take Ability in Girls. I measured intelligence by the following seven classes: (i) Quick Intelligent; (ii) Intelligent; (iii) Slow Intelligent; (iv) Slow; (v) Slow Dull; (vi) Very Dull; and a quite distinct category: (vii) Inaccurate-Erratic. . .

...

My next stage was to ask two or three different teachers in several schools to apply the classification to 30 to 50 pupils known to each of them. The classifications were made quite independently, often by teachers of quite different subjects, and a comparison of the results showed that 80 to 85 per cent of the children were put into the same

classes by the different teachers, while about 10 per cent more only differed by one class. This gave one very great confidence not only in the value of this scale, but of other psychical classifications when used by observant teachers. The next stage was to obtain exactly, as in the case of Health, a general scale of intelligence.

MEAN ABILITY of GIRLS

VERY

-30

DULL SLOW INTELLIGENT INTELLIGENT QUICK INTELLIGENT

FIRST SISTER'S ABILITY

Diagram 12. Resemblance of Sisters in Ability.

Diagram 11' gives the normal distribution of intelligence in a population of 2014 girls. It is a curious, if a common result of experience, to find that the modal ability is on the borderland between the Intelligent and Slow Intelligent. We have here for the first time a quantitative scale of intelligence, and we can at once apply it to the problem of the degree of resemblance between sisters as regards ability. Just as in the case of Health, all the girls of a given class are taken, say the Slow Intelligents, 'Omitted here.

Group Mean

30