]\gt ; secondary bodies of the nature of comets round the stars themselves .
The place on the temperature curve assigned to these stars on the meteoritic hypothesis is fully confirmed by the more detailed inquiry , and the hypothesis is thereby strengthened .
I am indebted to Mr. Fowler for assistance in the determination of the species and the construction of the map ten years ago , and for additional assistance in discussing the recent work .
I have also to express my thanks to Mr. Shackleton for a detailed examination of the recent photographs .
" " Mathematical Contributions to the Theory of Evolution.\mdash ; On the Law of Beversion By KARL PEARSON , F.R.S. ( A New Year 's Greeting to Francis Galton.\mdash ; January 1 , 1900 .
) Received December 28 , 1899 , \mdash ; Read January 25 , 1900 .
( 1 ) Intdty.\mdash ; In a memoir recently presented to the Royal Society , I have endeavoured to emphasise the importance of distinMathematical Contributions to the Theory of Evolution .
141 guishing between three diverss types of heredity , namely ( i ) , Blended Inheritance , ( ii ) Exclusive Inheritance , and ( iii ) Particulate Inheritance .
In a memoir printed in vol. 62 , pp. 386\mdash ; 412 of the 'Proceedings , ' I have dealt at length with the theory of blended inheritance , generalising for this purpose Mr. Galton 's Law of Ancestral Heredity .
Allowing a certain degree of variation in the constant , or " " coefficient of heredity there discussed , I consider that this theory gives a fairly good first approximation to the facts hitherto observed in this field .
But blended inheritance certainly does not cover the whole field of heredity .
When a character , then this law of ancestral heredity tells us the most probable blend for the offspring of given ancestry .
It shows us the offspring of exceptional parents regressing towards mediocrity , owing to the fact that without stringent selection the great bulk of their ancestry must be mediocre and not exceptional .
Thus the main feature of the law of ancestral heredity is regression .
Such regression is not what most biologists would understand by reversion .
In fact , when the inheritance from a variety of ancestry is blended , the idea of reversion becomes very obscure ; I venture to think meaningless .
Let us suppose stature a blended character , then the array of offspring of a definite short statured ancestry will have a ean regressing ( here progressing ) towards the population mean and a definite variability .
Hence the theory of chance enables us at once to determine the frequency of a very tall man born of such short ancestry .
The frequency may be small , but sooner or later the tall man will appear .
Now let us suppose one distant ancestor in the otherwise short ancestry to have been tall .
Clearly his existence will hardly affect at all the mean of the array of offspring .
He will not materially influence the chance of a very tall man appearing among the offspring ; yet a superficial observer might easily describe the appearance of the very tall man as a case of reversion to the distant tall ancestor .
The absurdity of this attribution is manifest when we remember that persons like him would have had sensibly equal frequency with or without the distant tall ancestor .
In fact , it seems to me that in the case of characters which continuously vary , and which blend their inheritance , it is hopeless to look for any evidence whateyer of reversion .
The term is , then , meaningless .
To find reversion we must investigate oases in which characters do not blend , i.e. , the individual takes exclusively after some one member of the ancestry .
In this case the appreciation of reversion becomes possible and its meaning intelligible .
Cases of this kind are by no means unindividual has 1024 great parents , and these can ardly be else but , fair sample of the population of their generation , if there has not been an excessi ' amount of in-and-in breediug or much selection .
VOL. LXVI .
142 Prof. Karl Pearson .
common .
Thus , .
Galton writes in his ' Natural Inheritance ' " " Parents of different statures usually transmit a blended heritage their children , but parents of different eye-colours usually transmit an lternative heritage if one parent has a light eye- colour and the other a dark eye-colour , some of the childrcn will , as a lule , be light and the rest dark ; they will seldom be medium eye- coloured like the children of medium eye-coloured parents Again , in his paper on " " Basset Hounds , \ldquo ; Mr. Galton classifies these hounds as tricolour and non-tricolour ( N ) , remarking , " " I am assured that transitional cases between and are very rare , and that experts would hardly ever disagree about the class to which any particular hound should be assigned In other words , Mr. Galton appears to assume exclusive inheritance.erhly , in such exclusive inheritance , the offspring takes after one or other parent , or reverfs to more distant ancestry .
It becomes somewhat difficuIt to see how the law of ancestral heredity , which applies to blended inheritance , can be transferred to this different field .
Yet .
Galton in his ' Inheritance ' ( p. 153 ) writes : " " The broad sion to lvhich the present results irresistibly lead , is that the same peculiar hereditary relation that was shown to subsist between a man and each of his ancestors in respect to the quality of stature , also subsists in respect to that of eye-colour.\ldquo ; Further , in the paper on Basset Hounds , he actually endeavours to demonstrate the truth of the law on the exclusive colour of these hounds .
Now I think we must keep these two matters quite apart .
The average stature of an individual is a blend of all his progenitors ' characters ; even in a single individual we find contributions from many ancestors ; this is not the case with an exclusive inheritance , and it does not accordingly seem to me possible that " " the same peculiar hereditary relation that was shown to subsist between a man and each of his ancestors " " for a blended character can also hold for an exclusive character .
It is no longer of the proportions of a character in one individual that we speak , but of the frequency of various types of individuaIs among the total offspring of a given ancestry .
The one statement is a law of blending characters , and the other is a law of distributing the exclusive characters among a group of individuals .
In the first case we deal with regression , in the second with reversion .
What Mr. Galton really asserts is , that the proportions of reversion in an array of offspring are identical with the proportions of blend in the average * ' Roy .
Soc. Proc vol. 61 , p. 403 .
A remark in the ' tural Inheritance ' that ' Stature is due to its being the aggregate of the quasi-independent inheritances of many separate parts , while eye-colour } , to be much less various in its origin ] seem to indicnte that Mr. Galton oonsiders that blended inheritance is ultimately based exclunive of parts\mdash ; a suggestion well worth investigation .
Mathematical Contributions to the Theory of Ev ) 143 individual .
If this be true , then his law , or possibly some generalisation of it , is very comprehensive ; it embraces the two distinct types of heritage , blended and exclusive .
But I think it most desirable to keep the two ideas quite separate , and speak of the one dealing with blended inhoritance as the Law of Ancesfr .
the second , with exclusive inheritance , as the Law of Reversion .
If this be done , we shall , I venture to think , keep not only our minds , but our points for observation , clearer ; and further , the failure of Mr. Galton 's statement in the one case will not in the least affect its validity in the other .
( 2 ) Tfie Law of Reversion.\mdash ; Let us examine first what I take to be Mr. Galton 's view of this law .
Out of an array of offspring , 1/ 4 will follow each parent , 1/ 16 follow each of the four grandparents , and follow each of the grcat parents .
In this manner the total offspring is distributed by reversion among the Now I want to draw attention to one or two points here .
1/ 4 will not be all the children like , say , their father for ont of the 1/ 4 who are like members of his ancestry , those who are like ancestors like him\mdash ; and these ancestors will occur in certain proportions\mdash ; will thus also be like him .
This holds for each individual ancestor ; the number any ancestor will be considerably greater than the number who " " follow " " that special ancestor .
Now let ) and be the number of the like a parent , a parent , a great grandparent , and nth parent , \amp ; c. This brings me to my second point .
A special meaning is here given to the word like .
is not in the usual sense of the word all the number like the father .
If the offspring had the same distribution of character as we find in the general population , then undoubtedly some wonld have the same quantity or quality of the character as he has\mdash ; some , for instance , would be blue-eyed if he were blue-eyed\mdash ; but this is a random likeness and not like in the special sense in which we are using the word .
are like the father owing to the laws of heredity , the remainder have a random distribution so far as he is concerned , and we exclude any random likeness from our consideration .
How then are we in actual observation to distinguish hereditary from random likeness 1* The answer is simple out of pairs of parent and offspring will be absolutely correlated , i.e. , have a correlation equal to unity , but the remaining pairs will have zero correlation , although there may be random likenesses .
Hence , by the theorem given by me in the 'Phil .
Trans vol. 192 , p. 276 , the actual correlation will be perfect correlation reduced in the ratio of the * I exclude for tho present the influence of assortative mating .
A likeness to the mother , otherwise random so .
as the father is concerned , may thus become real likeness to the father .
Prof. Karl Pearsoo .
number of correlated pairs to the total number of pairs .
Thus the correlation of parent and offspring It thus follows that are the correlation coefficients to be expected between offspring and parent , grandparent , great parent , \amp ; c. Here we have assumed equal potency foe both sexes and all lines of descent , otherwise these coefficients must be looked upon as mean values of the correlations for different genera .
tions of ancestry .
Lastly , it seems to me that reversion may not be the proper word to apply to those who directly follow their parents , and that these may fairly considered direct inheritors and distinguished from reverters I shall accordingly assume no priori relation between these two classes , certainly not that direct inheritors and reverters are equally numerous , i.e. , and , as in Mr. Galton 's Law .
As for reversion itself I will only suppose it to diminish in geometrical progression we step backward to more and more distant ancestry .
I shall accordingly take offspring to follow either parent and , \amp ; c. , to follow grandparent , great grandparent , great great grandparent , \amp ; c. With these preliminaries arranged we can now proceed with the analysis .
( 3 ) The eralised Law of irsion.\mdash ; The total number of off spring is clearly the sum of all those that follow all the successive ancestors , i.e. , or ( i ) Now consider how the number of offspring " " like\ldquo ; or absolutely correlated with one parent are made up : they are in number ; they consist first of , the number directly inheriting from this parent ; also there will be like each of the parent 's parents , and the parent will be like one or other of the parent 's parents in proportion of cases ; similarly there will be like each of the parents ' grandparents , and the parent is like each of the parents ' grandparents in cases ; and so on .
Thus we have or Now note how the any one grandparent is made up .
We have directly reverting to this grandparent , to each of the grandparents ' parents , and in each case like the grandparent ; similarly out of those reverting to any grandparents ' grandIn this case there is every reason for supposing , and .
Thus , and since 4 .
Mathemat al Contributions to the Theory of Evolutio ) parent , there will be like the grandparent , and so on .
But beyond these contributions , certain of the .
who follow the parent will be like the grandparent , for the parent is like the grandparent in fraction of cases .
Hence we have finally : or . . .
( ui ) Proceeding in the same manner we find and so on .
Hence we deduce from and ( ii ) or ( vi ) Similarly from ( iv ) and ( iii ) or ( vii ) Again from ( v ) and ( iv ) or ( viii ) Generally ( ix ) with , by ( vi ) .
To solve equation ( ix ) assume as usual , and we find Thus But by ( i ) .
Hence or ( x ) We have then ( xi ) where and A2 are constants .
146 .
Karl Pearson .
But , . .
Thus we may put .
( xii ) lvhere is a constant .
Let us substitute this in equation ; we find or ( xiii ) Write , then by ( i ) we have Hence from ( xiii ) and by ( xiv ) Suppose the parental and grandparental correlations observed , then .
( xvi ) will both be known .
These will give and ; then ( xv ) will give and ( xiv ) , while will determine , and the whole law of inheritance and reversion will have its constants fully determined .
We have , indeed , from ( xvi ) ( xviii ) ( xix ) Mathematical Contribntions to the of .
147 From where Lastly ( xiv ) Thus ( xviii ) and ( xix ) give and .
( xx ) then gives , taking the root less tham unit Finally ( xvii ) and ( xxi ) give and , completing the solution .
( 4 ) Cmparison with Law of jsfral \mdash ; Now let us compare these results with those I have obtained from the law of ancestral heredity .
On p. 390 of the memoir on that subject we have for the nth midparental correlation with the offspring , where the correlation of the offspring the individual great grandparent .
By p. 394 .
Hence .
( xxii ) or the correlations of the offspring with the ancestry follow a simple geometrical progression .
Comparing this with the result ( xii ) of this paper , or ( xii ) where now a different constant , we see that the two cannot possibly in agreement , unless one of the terms of the latter resnlt vanishes .
Thus there is in general a fundamental difference between the of ncestral heredity and the law of reversion ; they give expressions differing in character for the correlations between the offspring and individual ancestors .
Let us see when the t , wo laws agree .
There is unfortunately a bad slip in my memoir of 1898 .
The series at the top of p. 403leads to and not as there given Thus we have or .
Hence ( xxiii ) therefore by ( xii ) Thus by ( xxii ) of this paper . .
( xxiv ) $ ' Roy . .
Proc vol. 62 , pp. 148 Prof. Karl Pearson .
This shows us that the correlation with the individual ancestor halved at each backward step in the pedigree .
We see at once that can only be in agreement with ( xxiv)\mdash ; the letter being different in the two , and merely standing for a provided , or by ( i ) , provided .
Thus the condition that blended exclusive inheritance should lead to the same values for the correlations with the ancestry is : that reversion should form a series starting with the actual parent .
If this condition should hold , then , for example , the grandparental correlation must always be one-half the parental either for blended or for exclusive inheritance .
( 5 ) Correction of an in Memoir on ncestral \mdash ; It may be of value to insert here the modifications requirsd in my memoir on blended inheritance , owing to the slip just referred to : they apply to the results deduced from ( xvii ) on p. 403 ; these are the table on p. 403 , and the result immediately under ( xviii ) on p. 406 .
In the first place the law of ancestral heredity may now be written a very simple form .
In the second place we may replace the equation for 0R p. 394 by ( xxvi ) whence we find for the parental correlation or ( xxvii ) .
Each succeeding ancestral correlation will be obtained by repeated halving of this value .
Lastly , the result on p. 406 for the fraternal correlation becomes\mdash ; ( xxviii ) .
The following table indicates the effect of varying on the intensity of heredity , and should replace that on p. 403 of the memoir on the ancestral ( 6 ) arising when we apply thpse Results for Blended Inheritance .
\mdash ; Now the above table shows us that by varying ]ltly we can obtain a considerable range of values for the correlation of characters in kindred .
But these values are limited by two serious considerations , ' .
namely : ( i ) The ancestral correlation is halved at each stage .
( u ) The fraternal correlation appears to become perfect as we approach the upper limit of parental correlation , i.e. , Now actual dsterminations of grandparental correlation in the cases of eye-colour in man , of -colour in horses , and of coat-colour ill hounds , which I have recently made , do not as a rule seem to iustiiy the statement that the grandparental is half the parental corPelation .
Further , in two of these cases , the average parental oorrelation is quite the fraternal correlation is , while largel than , still a good deal short of perfect .
Hence I am bound to nolude that : ( i ) These characters do not obey the laws of blended inheritance as deduced from the law of ancestral edity ; or , ( ii ) The laws of blended inheritance , as deduced from the law of ancestral heredity , would be largely modified if we considered the influence of assortative mating , or ( iu ) The fundamental assumption that if all the midparents right away back had the same amount of the character , the average offspring would have also the same amount , is not justified .
Thus the result in Equation ( xxiii ) , perhaps , is unnecessary , or there may be two independent constants of inheritance .
present .
but what if the contributions of certain ancestors by selection , artificial or natural , have been eliminated before reaching the existing generation ?
What if the coat-colours of certain ancestors were unfashionable , and only their unlike descendauts have been put to the stud 1 Our theory may be quite correct , but it may appear erroneous when tested by facts observed in the case of horse or dog breeding .
Let us investigate whether independent and in our expressions for parental and fraternal correlations would enable us in the case of blended inheritance to reach a value of the former as high as without the latter becoming perfect .
I find if bs the parental correlation , , from Equation of my memoir on the ancestral law ( p. 394 ) , and if be the fraternal correlation obtained from quation ( xviii ) of the same memoir : .
( xxx ) .
hence , eliminating , we have and ( xxxi ) .
Mathematical Contributions to the of Evolution .
These give and when the parenlal and fraternal correlations are known .
Now , since is will be imaginary , if be not .
Hence we should again need perfect fraternal correlation for to be as large as 0.5 .
Thus with blended inheritance and little or no assortative mating we cannot get a parental correlation as high as the value , which actually does occur in my data for both men and horses .
We must now consider how the problem will be affected , if we suppose exclusive and not blended inheritance .
( 7 ) Illustrations of thoe Law of Rev(Inheritance.\mdash ; ( i ) Let us first consider what happens if we take the chief feature .
of Mr. Galton 's view , i.e. , that the likeness to the parent is the beginnlng .
so to speak , of the reversion series .
Then ( in the notation the present memoir ) .
It follows from Equation ( i ) that : , or .
Thus by ( xviii ) and generally Equation ( xx ) to find now becomes while Thus as soon as we know , we can find all the ancestral correlations and the whole series of reversions .
For example : if should have , and and .
Thus in this case 35 per cent. of the oflspring take each parent , and 30 per cent. revert to higher { tncestry .
Of this 30 per cent. , or per cent. , revert to each of the four grandparents , leaving 9 per cent. , about , to revert to great grandparents and higher ancestry still .
( ii ) Next suppose Mr. Galton 's full view to be correct , and that 1 , / 4 of the offsp1ing follow each parent , 1/ 16 each grandparent , 1/ 64 each grandparent , and so .
Then we have\mdash ; Hence from , we find and Thus we should have : , \amp ; c. , or , prjcisdy the ancestral rdations in the case of exdusive that have the case of blended inheritance by the law of ancestral redity fthe special case of ( see table , p. 149 ) .
( iii ) Let us suppose the parental correlation to be , a value very far from what I have found for eye-colour in man and coat-colour , in horses .
Then by ( xvi ) .
' Putting in ( xx ) we deduce : , or which gives us But remembering the value of we have , using ( xxi ) , --S 3 and The first equation shows us that is impossible , for it gives negative .
Accordingly we conclude that and , while .
Thus reversion is totally excluded one-half the offspring after each parent .
In this case the grandparental correlation , , the great grandparental 0 , and so on .
The ancestry beyond the parents have no direct influence on the offspring , beyond the fqct that they have determined the parents .
We are dealing indeed a case like that investigated in my memoir on " " Regression , Heredity , and Panmixia.\ldquo ; So far our theory of exelusive inheritance with parental correlation agrees with that of blended inheritance with the same value of the parental correlation .
But we have seen that the latter leads to an impossible value for fraternal correlation , i.e. , one which does not fit the facts .
Does perfect fraternal correla tion necessarily flow from exclusive inheritance without reversion Certainly not , for this would connote that all the offspring of a given set of parents would be alike , or one parent in each family be absoIutely prepotent .
This is of course not the fact .
Supposing all families to consist of members , and that both parents were equipotent in the family , there would be pairs of brethren alike , out of a total of pairs , or the fraternal correlation would be .
The average size of a human * See Natural Inheritance , ' chapter viii , \amp ; c. Mr. Galton thers uses th ' correlation coefficients corresponding to blended inheritance for , an exclusive inheritance .
But , directly investigated , such values are far from holding for eye-colour .
'Phil .
Trans vol. 187 , p. 803 .
Mathematical to Theory of Evolution .
153 family is about ; but if we confine ourselves to one sex , we must exclude all sterile marriages and all not leading to two brothers , or sisters .
We might then very well take , which gives for the fraternal correlation .
Thus we might expect in the case of exclusive inheritance that the fraternal correlation would lie between and 1 , according as to the degree of prepotency of one or other parent in the individual marriage .
Thus our theory of exclusive inheritance is not , like that of blended ritance , incompatible with observed facts , i.e. , high values of parental correlation and values substantially less than unity of the fraternal correlation .
But for such cases we must deny the existencs of any regular and continuous law of reversion .
We should have to look upon reversion , if it occurred at all , as merely an irregular and infrequent phenomenon .
On the other hand , if we differentiate the taking after parents from the reversion to ancestry as phenomena of a quite distinct nature , our theory will enable us to surmount , for some cases at least , difficulties in ancestral correlation , which arise when we take Mr. Galton 's Law in its iginal form to cover both blended and exclusive inheritance .
I illustrate this from data for the coat-colour of Basset Hounds in the following section .
( 8 ) Apphcation to Bassef Hounds.\mdash ; Understanding that I was desirous of testing my theory on a character which was definitely exdusive , Mr. Galton , with his invariable kindness , at once placed at my disposal his material on Basset Oounds .
The reader will remember from the statements in Mr. own memoir ( ' Roy .
Soc. Proc vol. 61 , p. 403 ) that these dwarf bloodhounds are either lemon and white , or black , lemon , and whits ; and here , as in Mr. Galton 's work , they will be classified as non-tricolour and tricolour , or by the symbols and T. In dealing with the offspring I was in many cases unable to determine the sex of the dog , as that information is not in the stnd book , and all individuals are not again recorded as sires or dams , nor do they possess obviously male or female names .
Thus in my principal tables all the offspring of both sexes are clubbed together .
To measure the legitimacy of this , I have formed separate tables of the sexes in the case of sires and dams .
Further , in dealing with great grandparents there were so few of each of the eight individual types alone , that I have formed merely one table , that of great grandparent and offspring , disregarding the line of descent .
mother and father may be equipotent on the averag6 , but in the individual family one or other be markedly prepotent , It is to this prepotency of the individual .
ardless of sex , that the increase of fraternal correlation beyond is very probably due .
ir Everett Millais , ' Th6 asset Hounds Club Rules and Stud Book , ' 1874\mdash ; 0-4 Prof. Karl Pearson .
Mathematical to the Theory of )ofution .
saJI saJIsSm(r ' ributions to the Theory of Brolnfion .
l57 The extraction of eleven tables from Mr. Galton 's datit papers I owe to Miss Alice Lee , D.Sc .
A twelfth is due to Mr. K. Tressler .
For the other three I am responsible .
Of the determinations of the corrclation coefficients , I owe five to Miss Alice Lee , no less than six to .
L. N. G. Filon , M.A. , and the remaining four only are my own work .
The method by which the correlation coefficients have been calculated be explained and justified in another memoir ; it is it novel process , which we believe to be of considerable importance , and which we have already applied to a variety of attributes not capable of exact quantitative measurement .
The )robable error of this method of determining the correlation has also been ascertained , and may be taken in the present cases to from to , so differences which are significant oan be appreciated .
Coefficients of Correlation .
Sire and offspring:\mdash ; Sire and all offspring , , Sire and 8 offspring Sire and ?
offspring 1104 Dam offspring Dam and all offspring . .
, .
, Dam and 8 offspring Dam and ?
offspring Mean parental correlation ' Grandparent and offspring:\mdash ; Sire 's sire and all offspring Sire 's dam and all offspring Dam 's sire and all offspring Dam 's dam and all offspring Mean grandparental correlation Great rent and oftspring : All great grandparents and all offspring Whole siblings from same litter Whole siblings from different litters Half siblings , dam 's side Half siblings , sire 's side . . . .
Every writer on heredity must have felt the inconvenienoe of our nguage having preserved no word for either member of a pair of offipring of either or both sexes from the S{m parent .
After some hesitation , I 1 ave ventured to duoe a good Anglo-Saxon word with this sense .
VOL. LXVI .
158 Prof. I(arl Pearson .
Now several noteworthy conclusions follow at once from theu numbers :\mdash ; ( 1 ) Inheritance .
The dam has a great prepotency in the matter of coat-colour . .
Galton has already remarked on this .
We here see that , quantita- tively , the dam is , on the average of both sexes , thrice as highly with the offspring as the sire .
While she has reached the high value , he has fallen below , and the theoretical value of lmmodified law of ancestral heredity is neither satisfactory for individual cases nor for their average .
Offspring take more after the dam than the sire , but 8 offspring more than 9 after the sire , and 2 offspring more than after the dam .
In other words , the parent hands down its characteristics more strongly to its own than to the opposite sex .
Curiously enough , the sire 's parents seem to have more influence than the dam 's .
In particular the dam 's sire has , within the probable error of our determinations , no influence at all .
In the unchanging line of descent , the dam 's dam has more influence than the sire 's sire , which is what we should expect from ; but ( a ) also makes the male element of much less importance than the female , and so the dam 's sire insignificant as compared with the sire 's dam .
The final result is thus to give a slight preponderance to the sire 's over the dam 's parents .
( 2 ) Collateral Inheritance .
( a ) The degree of resemblance between puppies of the same parents is greater when they are of the same than when they are of different litters .
It is clear , , that we have only been able to find comparatively few pairs of whole siblings from different litters , and the differenco between and is of the order of the probable error of the differences .
With greater numbers , possibly a more sensible difference might be found for the correlation of siblings from the same and different litters .
At present there seems nothing to warrant the idea that puppies from the same litter have the high degree of resemblance which we find between twins in the case of mankind .
( b ) A comparison of the correlations for half siblings on the dam 's side and on the sire 's side again emphasises , if the breeding be straightforward , the great prepotency of the dam in the matter of coat-colour .
The fact that we have upwards of five times as many pairs of half siblings on the sire 's side as on the dam 's side shows how large a fashion there is in selecting sires .
It is possible that largely used and 'Roy .
Soc. Proc vol. 61 , p. 404 .
The Tabl ' , II , p. 410 , requires hango of eadings , as already pointed out by .
Galton .
Mathenvatieat to the of Bvotution .
159 possibly overworked sires lose some of their hereditary influence , while not losing their power of fe.rtilising the dam .
( c ) The great reduction in the degeee of fraternal correlation when we turn from whole to half siblings is very remarkable , and is , at any rate for half siblings on the dam 's side , not very explicable .
Had we assumed the parental correlation to be , and found 7 from ( xxvii ) , , we should have deduced from ( xxviii ) for the correlation the value , which is in fair ance with the observed result for whole , iblings .
But , as we have seen , xxvii ) belong to a theory which gives very poor values for the and correlations , and of and .
Further , we should theory have expected the average colrel tion for half siblings to be half the value above , si1lce one-half of the common ncestry 1 cut off , , and not , as it actually is .
Thus the fra- correlation does not appear to be in accord with the theory of blended inheritance .
Its determination ill the general case of exclnsive inheritance seems a of considerable difficulty , which in this case is rendsred much greater by the immense prepotency of the dam , so that it would seem very desirable to differentiate the sexes when dealing with the resemblance of siblings reverting to ancestral } ) ( 9 ) Application of of eversion to \mdash ; We have for mean values .
Now these correlations certainly do not obey the tion ) required , when we ( xxiii ) to gover1l the of ancestral heredity ( .
S6 ( iii ) ) .
A glance at the table on p. 149 will show that such a series of 's as the above cannot fit into it .
Still less do they appear consonant ( except to the first roughest approximation ) with Mr. Galton 's form of that law , i.e. , .
Nor do they satisfy for the reasons the law of reversion when we start the reversion series from the parents , , put as in S7 ( ii ) .
Accordingly , I have to find what would be the of , if and had the values given above , our ised 1 of reversion were correct .
Turmng to S3 , substituting in ( xix ) for itnd we : * There is to be considcrod tho possibilit.v of of in sire 's case .
Giveil a large stud of and scrv , lllts of average suess , lllld a may easily go astray evcn is lined by the l.cquired .
Everett Millais , in a lecture on Telegony , delivered at St. ' Hospital , in 1895 , stated that he knew of " " quite two dozen such mples resulting in supposed telegony " " The master is the person to whom such little lapscs of duty are confided.\ldquo ; But if doze1l be } off as telegony , .
many alliances within the blood may be oonvoniontly matical Contributions to the Theory of Evolution .
161 .
not be denied that the difference is of extreme biological interest .
In the former case we have a comparatively small total reversion widely spread ; in the latter a much larger total reversion concentrated on the first stages of ancestry .
Which system is more accordant with facts 1 It needs far wider observation and experiment than are yet available to settle this .
So far we can only say that the former covers Mr. Galton 's as a special sub-case , and that the data for !
Basset Hounds appear capable of treatment under the wider rule , but can o1lly be fitted with some straining to the special oase .
Are there or are there not physiological reasons for supposing resemblance to a parent arises from a differenlt source from reversion to an ancestor ?
Here reversion to an ancestor must not be measured by cases of resemblance to an anoestor , for a portion of this resemblance is due to common likeness to the parent ; we must approach the matter from the standpoint of cases in which the offspring inherits a character like 4 ' that of the grandparent or higher ancestor , and unlihe that of the parent .
Current use of the term " " reversion\ldquo ; at least iustifies us for the time being in not speaking of all inheritance , including like1less to : parents reversion , and in our theory we may be permitted to differentiate parental and reversional inheI'itance of exclusive characters , we find it needed by our numerical data .
Here , elsewhere , we sadly need a widely extended range of tion and experiment .
( 10 ) the having to Scx and \mdash ; We have already indicated that our instification in applying the methods of normal correlatio11 to coat-colour is considered in another memoir .
We merely suppose at present that there is some variable , approximately the law of normal variation , on which the coat-colour can be thrown back .
This being so , let in terms of this variable be the distance of its mean value from the division line between tricolour and non-tricolour , and let be the devittion of this variable for the same group .
Then if , we easily deduce by aid of tables of the probability integral from the correlation tables in S8 the following results:\mdash ; Prof. Karl Pearson .
Table of Values of ire , , , , his \ldquo ; , , , , her , , their Whole siblings , same littcrs different littelB , Half siblings , sire 's side dam 's side \ldquo ; Now , if there were no seoular change due to or other selec tion , lvould remain the same in each generation , and therefore would give a proper measure of the variability due to sex or to relative position in ancestry .
The table at once suggests a number of interesting points , which I proceed to note .
Turning first to the offspring , as given in the second column , we serve that the dogs with the gest pedigree the largest We have , in fact , the values of for pedigrees stretching to great grandparents , grandparents , parents , and merely to brethren , respec .
tively , as or , oughly , we have a geometrical series with the ratio of about Now this result may be reached in more than one way , either decreasing , or ( ii ) increasing .
Increase of would signify that , for the longer pedigree , the msan of the quantity on which the coat-colour depends is being thrust further into the tricolour section ; decrease of vould signify greater concentration in the tricolour section within which the mean lies .
Whether the longer pedigree signffies the more modern hounds , the more careful preservation of ancestors ' names in the more fashionable hounds , we reach practically the same conclusion : tho process of breeding is emphasising melanism .
urther , by aring the 8 and 9 offspring in the case of both sire and dam , we conclude that this process is sensibly more significant for the male than offspring .
As very often only one or two puppies out of a litter recorded in the 'Stud Book , ' this apparently artificial selecMathematical Contributions to the volufion .
tion , which is more stringent for the malcs than the femnles , to be , theoretically at least , a mistake when we reme1nber that the potency of the femaIe is thrice that of the male in coat-col.ourdff .
Within the grollps of grandoffspring and ) differences arc hardly enough for special conclusions to be drawn .
( b ) Turning next to the parental groups , we see that ( i ) the sires nd dams , neither of which can be considered to form a more moderIt group than the other , have yet remarkably different of ' for the sire being about seven times as great as that for the dam .
The sires are thus far more stringently selected than the dams , and a deal of this difference must nndoubtedly be due to the lesser variability of the sires .
Here again the breeders , if they are selecting at coat-colour , would appear , at least theoretically , to be in error .
( ii ) Grandsires , 8 and 2 , appear to be less variable than the sircs , and granddams , 8 and 9 , less variable than the dams .
This may be due to the original paucity of the breed , or be all instance of the general rule to which I have elsewhere referred , i.e. , that parents are selection out of the general population , and so less than their offspring .
( c ) But this rule meets with a remarkable exception in the case of 9 parentage ; both granddams and dams are more variable than theix oiispring , and very significantly so .
An examination of dam and female offspring shows that the 9 offspring have a value of double as great as that of their dams .
With few original dams , it is difficult to understand how they could be more variable than their offspring .
Considering the great prepotency of the dam , it is difficult to attribute this increase of entirely to the nction of the less riable sire ; one more or less forced to believe that there is a process of selection of the offspring which are entel'ed on the record going 011 , and that thus a group of dams possibly fairly variable , { not very marked tendency to melanism , is represented ill the ncxt generation by offspring of a more stringently selected charactcr ; the stringent selectio1l of sires may have contril ) uted , but can hardly ) the sole source of this change .
A further conclusion is worth noting : Parents , whether male or female , whe11 they have male are apparcntly more variable thaIl whezl they have female offspring .
( 10 ) General esults .
( a ) The laws hitherto propourlded for )lended inheritance do not $ appear to cover the cases of exclusive inheritance , e.g. , such cases as eye-colour in man , coat-colour in horses or hounds , \amp ; c. ( b ) The law of ancestral heredity must be guished from a law of reversion .
Neither seem to fit the facts if we adopt the amounts of [ heritage , 14 , \amp ; c. , from parent , grandparent , great grandp rent , \amp ; c , , originally taken as a first approximation by Mr. Galton .
Testing theory by the case of Basset Hounds , we find much difficulty , owing partly to the great prepotency of the dam , and partly to the large amount of artificial selection which is evidenced at every tnrn , and obscures what may be termed the atural laws of inheritance .
There is an urgent need to widely extend our knowledge of hcredity by new experiments and observations on other organs in different races .
Facts are of the first necessity at the present time , facts collected on large scale for a wide range .
* may be of service to indicate to would-be investigators what has already been done or is now in hand:\mdash ; In ( a ) Stature first degree collateral , fraternal ) .
( b ) Head index .
( c ) Span and forearm ( direct to first degree and collateral , fraternal ) .
( d ) Eye-colour ( direct to second degree , collateral , fraternal and avuncular ) .
Shape of head , physique , intellectual capacities , tastes ( collateral only ) .
( fl Fertility ( direct to second degree ) .
( g ) Longevity ( direct and collateral , fraternal ) .
In horses : ( h ) Coat-colour direct to second degree collateral ) .
Fecundity ( direct to second gree and collaternl , fraternal and avun cular ) .
In hounds:\mdash ; Coat-colour ( direct to third degree and collateral ) .
In moths : \mdash ; ( k ) Wing.markings ( direct and collateral ) .
In daphnia:\mdash ; ( l ) Shape of spine ( direct and collateral ) .
In all these cases the coeifioients of correlation have already been worked out , or material is being collected to detcrmine them , by Mr. Francis Galton , Professor W. F. R. Weldon , Dr. Warren , or by my collaboratol.s and myseIf at University College .
Hence I would impress upon others to take as far as possible widely different cbaracters in widely different races .
Above all , cases in which artificial selection plays a great part , i. e. , dogs , fancy pigeons , \amp ; c. , ought to be avoided .

