]\gt ; " " rent \ldquo ; to denote an individual compounded , in a simple from the two parental types , and giving oflspring of a definite ?
In ising , howevor , Mr. Galton 's conceptions in my " " Law Anccstral Heredity , \ldquo ; * I purposely placed before myself the of educing the theory to a purely statistical theory , and discardt , the conception that the type of offspring was settled by ompound ohethc genel.ised meneration bations f generation , and no assumption was made rTlS to any heritance i type ; the theory ) ecame purely a statistical theory of distribution in rious gtions of the deviatio11s from type .
' a somewhat later date the Mendelians gave up the conception thit tho type , of the offspring was known from the parental types .
actual effcct of crossing two compared to the of a componnd , the character of which couId not a be predicted from the known nature of the componsnts .
It was } matter to be determined by obse)'ttion or experiment only .
this wider view the original Mendelian thoory of ' ' dominant\ldquo ; " " recesve\ldquo ; characters has disappeared , and that eory has thns approximated itself to the " " Anccstral \ldquo ; In a to the Royal Society entitled " " of Reversion I endeavoured to work out a general theory alternave , on the hypothesis that a certain number of th offspring were for any character like one or other parent or lik some one or other ancestor , the proportions of offspring like types diminishing in number with ths distance of descent . ?
* ' Roy .
Soc. Proc vol. 62 , pp. ) 'Roy . .
Proc. , ' vol. 66 , p. 142 et seq. or by any other process .
A general theory of alternative heritance mtty cover Mendelian phenomena ; a theory of the individual domlnance of either parent or another group of offspring , a theory of what I have defined as intermittent unit prepotency , cannot do so .
Still , Dr. Boas considers that he has evidence for his theory in the inheritance of cephalic index in man , and it seems to me A his paper indicabes the ma1mer lin which it may be possible to still further generalise my resuIts of .
It is clear , however , that we badly need some criterion a between these competing theories in the case of measurements of the inheritance of any given character .
Since none of the three theories.referred to is essentially based on the determination of the type of the offspring from ths parental types , we are thrown back * ' The Anthropologist ' ( N.S. ) , vol. 5 , pp. 6$0\mdash ; 688 .
'Biometrika , ' vol. 2 , p. 389 . .
Dr. , I think , has not fully understood my theory of the midparent .
petttedly .peaks mid parental value of a character of the spring clustering round this value on th( , of ' GaIton and Peurson There is absolutely no onism between my and ths ture of Americo-European exceeding both parental types .
Ny midparcnt is based solely on irst investigation oubject Hdistrib these points w lally considered in the case of stature whether some of the offsprinffi followed the father and some the mother , and showed that did vary sensibly from array to array .
Subject , therefore , to a stration for each character that the frequency is approximately and the regression linear , we see that the purely statistical theory deviations from type ( ' Roy .
Soc. Proc vol. 62 , p. 387 ) , and the offspring typr { itself may be wholly different from types , or falling shorI ) of them .
that 2 is the mean of the standalrd deviation of all the arnys was { irst stnted by Yule , ' .
Soc. vol. 60 , p. 477 , for the case of xegression .
tural Inheritance , ' ) .
, and Table 10 , p. 207 .
This tigation ) scems to ]lave escaped Dr. BojlS , ) , 1 ' .
530 .
we to Test various Theories of redity summed uP in the " " Law of Heredity\ldquo ; would assert nt the offspring of all parents of a.given type would have a priability , whatever that type might be .
.\mdash ; If we take as a fair sample of this the Mendelian theory , discllssed by me in a recent { on to the Society , and now ) lished in the ' Phil. Trans. , ' we find that constancy of the dard deviation of the array no longer true .
: only becomes trne if ) of Iendelian couplets on hich Se cter depe1lds is indefinitely great .
In other cases , is still the mean of the standard deviations of the arrays , actual altle of the ndard d alters sensibly and connuously as we cross the correlation table , tending to crease one direction and reasc in the other .
Clearly we have , as I have out in the paper referred to , an excellent cl.iterion here between he theories .
( 4 ) Lastly , let us turn to the theory of individual parental domiJlance .
I will give the analysis for this case , extending and generalising .
Boas 's formulae .
I suppose the total offspring of a pair of trents be diyided into two groups and in nnmber .
In the first } ) iith a mean the fathers are considered as predominant without the Jlnothers being supposed at presenG entirely without influence ; in the tecond group with mean the mothers are supposed to have the jredominating influence .
We may speak of these two groups , .
for onvenience only , as " " father 's offspring\ldquo ; and ' ' mother 's \ldquo ; and be the standard deviations of " " father 's offspring\ldquo ; and 1t 'mother 's offspring\ldquo ; for a given cter let be the Ytttandard deviations of the fathers for the same character , of the nothers , and of the offspring as a whole .
The mean 7 of the ffspriug as a whole will be given by .
Fnrther be the paternal offspring correlations for " " father 's md ' ' mother 's offspring and the maternal offsprlng corredtions for the same two groups respectively ; shall be the coefiieient assortative mating between parents , are the characters in 4 father , mother , and child , being measured from the parenttd means and from some other origin .
Then , if stands for a ation of all the pring of the first and of the seco1ld have .
, or ( i ) .
1 Let and be the total paternal and maternal correlations ; } have , * ' Phil. !
rans , vol. p. 66 .
: Here we have supposed that although the fathers of ' ' offspring\ldquo ; and of ' ' mother 's offspring\ldquo ; will , when weighted with thoill offspring , be unequal in 1lumber ( i.e. and ) , yet are the same , and similarly for the mothers .
This is equivalent supposing no correlation between the dominant effect of a parent an his or her deviation from type .
Otherwise we cannot put in the , sum the same as in the sum .
We are now able to write down the general regression equation -parental inheritance , i.e. , is the probable value of the character in offspring of characters and .
Hence , if we remember that , we on substituting from ( ii ) and ( iii):\mdash ; are the -parental co-efficients and , of the " " father 's ' offspring\ldquo ; and similar quantities for the " " mother 's offspring Now , fixing our atten tion for moment on " " father 's we should expect parents of characters and to produce an array father 's offspring with a mean : ( v ) , and with a standard deviation given by ( vi ) , Similarly the arrays of mother. 's offspring\ldquo ; for parents of the characters would have a mean : and a standard given by ( viii ) : to eories of .
267 , if and be the llumbers of children iu the array of arents ) elonging to either group , and , we shall have for ' .
he standard deviation of the total offspring of parents : ( ix ) , here is given by ( iv ) is the mean of the whole array .
If the relative influences of mother and .
there depended tlpon their , we could go no further with ( ix ) until this had been etermined .
If , , we pose this influence on the average to not sensibly dependent the characters , we writs bstitutc from ( iv ) , ( v ) and ( vii ) , we ( x ) .
To verify this equation I tl.ansferred to the mean , snmmed for } ery possible array , , for all values of and , and divided by the total umber of arrays .
The left-hand side shou ld be , the right-hand after some ) rednctions , identical with the ight-hand side of ( i ) , as it should be .
We have in equation ( x ) tccordingly , the variability of .
array of offspring on the hypothesis that t , he children may be divided into ( two groups , differently related to the two parents .
We see at o1lce a that the variability of an trray will depend on the actnal of the parental characters , unless : But this is asserting that the ) -pare1ttal for the two groups are the same , or , as we may it , hat there no distinction between the infiuences the two gloups .
the case usually assumed under the " " Law of Ancestral Hcredity\ldquo ; its consbant bility within the limits of random sampling the 1 arrays .
In every other case the arrays in oility the magnitude of the tal charitcter .
Let us look at the ttel .
from nothel .
light every value of only , or we can obtain lihe same esult by direct 1 investigation , namely , we find if be the of au array of due to fathers of ( xi ) .
Now parental correlations in my experience of extensive masses of data are rarely as low as , generally over , or nearer even .
But even we see that there ought to be about 20 per .
increase in the standard deviation of an array as we pass from mean father to a father with a deviation equal to } the Rternal deviation , , to an array which will be given with least moderate number of cases in any parental correlation table .
As we approach a parental correlation of , this increase of the \amp ; standard deviation amounts to 40\mdash ; 72 per cent. , and should be still more conspicuous .
We conclude , therefore , that if the parents are respectively dominant in two separate gronps of plot the standard deviations of the arrays of ofispring to the adeviations of the parent from the mean , we ought to get a very nsibly hyperbolic curve .
Or , if we plot the squares of the one to those of the other , we ought to get a sloping straight line of very sensible slope , about , instead of the horizontal of the ' Ancestral Law If we suppose the same conditions to apply to a -parental , , we find reduces to Or , using ( ii ) and ( iii ) , which give ) ) : ' ( xiii ) .
We find : of the arrays of offspring due to mean parents .
Generally deduce\mdash ; Values of for Arrays .
Hence we have the impossible value for the azardous arrors , therefore , arents tathers oothers ovith , rements oupwards o the general population .
there , for any glven male there would a female relatively only very slightly taller than he is , who would have spring with him of sensibly no variability .
In view of this result we may safely assert that the hypothesis of ' ' father 's \ldquo ; " " mother 's oflspring\ldquo ; apply to stature in man under an conditions in the approximating to Dr. Boas 's assumptions in tho 4 case of cephalic index .
( 6 ) Our conclusions may be snmmed up as follows : -The variability of the array of pring due to a group of parents of definite character il can be used as a criterion between various theories of inheritance .
In particular if the variability of the array be plotted to the character of the parent\mdash ; ( a ) On the hypothesis of the ' Ancestral La a horizontal straight line is the resuIting curve .
( b ) On the generalised Mendelian theory the result is a parabola 1 with horizontal * ' vol. 2 , pp. 373 and 378 .
axis .
: ( 7 ) I have applied the criterion here developed to my measurements father and son in more than 1000 families .
The three characters ' ritature , span and forearm are dealt with .
The correlation tables for paternal inheritance of these characters be found 111 a recent 11 emoir by Dr. Lee and myself on the " " Inheritance of the Physical haracters in We excluded all arrays with less than eight ndividualsl in them , deeming it absolutely untrustworthy to find a ean and standard deviation from fewer than eight cases .
The standard -i eviation of each array for a given paternal character was found by Dr. Alice Lee .
These were then plotted to the paternal character by Mr. W. L. Atcherley , and results are shown in the accompanylng Diagram 1 .
The zigzag polygons in each case give the plotted variabilities of the arrays , the vertical mlbers being the total on which !
the variability is based .
The horizontal line AA gives the value , , of the standard tions of the arrays according to the tatistical theory .
The broke1l lines glve twice the probable error of the tion of an rray from .
Thus if tho deviation of an array of individuals out of a total of , we have plotted up and do } from the quantity Now unless a difference is at least its error we certainly cannot assert it to be significant .
Now , if the diagram be examined , it wi 'll be seen that almost without exception the zigzag fall well within the on-sigYlificant ) ded by \ldquo ; There are , indeed , three exceptions , } ) ) ] ] three occur iu with less than twenty individuals , or arrays where some eccentricity of individual or measureme.nt might sily m itself felt .
But there really no nsed to appeal even to this tion , we thirty-me observations on arrays , and three of these only exceed the donble of rl their probable error ; this is less than the excesses might have expected on the theory of probability .
There is clearly absolutely nothing in the observed results opposed to that constancy of variability the arrays which is crgested in the usual treatment of the " " Law of Heredity ( 8 ) I next look at the theory of alternative , or the hypothesis that some children follow the one , some the other parcnt .
1 An examination of the diagrams show that there is not the ' vol. 2 , pp. 415\mdash ; 417 .
\mdash ; gram I of sons .
cRse , namely , Dr. Boas 's hypothesis of equal numbers of olfspring with equal variability and of like mean following each parent .
Tho equation to this hyperbola may be found from , by putting , here It will be seen at once that the variability of the array is too small the mean father and far too large for the exceptional father .
I bhink we may safely conclude that for man the case of the three haracters under investigation no such theory of alternative applies .
( 9 ) I now turn to the third or Iendelian hypothesis .
Are distributions better represented by parabolas of horlzontal than by horizontal straight lines We have a series of points :ach case and the question is : What is the best fitting parabola [ The equation to the Mendelian parabola , as glven in my memo ailready referred to , in the notation of the present paper .
, .
, where q is the number of Mendelian couplets , is measured front mean .
Writing , this is of the form , being numerics : Applying the method of least squares , since is bsolute constant , we find for the best value of being the total in the array : ' or , ,1 the number of Iendelian couplets is to be found from Working this out for the case of stature.first , I found , aftel some rather laborious arithmetic , that .
Thus the best fittiug Mendelian parabola needs no less than 48 plets .
We rite the parabola in the form * ' Phil. Trans , vol. 203 , pp. 66-67 .
To reduce to the result of the above memoir we have ' , , where is an mlned constnnt pcuding on bef , ween actual scale und number of Mendelian couplets .
This , substituting the values of and , is a straight line of .
Now the best fitting line to the observations , , has $ : slope of , or we conclude that the Mendelian paral ) is when number of couplets is as large as in the present case sensibly to the line which best represents the variability of the arrays ) to the parents ' character .
The Mendelian line of Diagram is not as good a fit as ths line , because the theory constrains the actual mean point bservations aitting 1when targe aation differs feptahsat thoertn theory gives reonstant aHence , The parabola thus sensibly coincides in direction with the best fitting straight line , but is raised above it in position .
I give the best fitting straight lines for the three characters we been considering For .
Probable error of the slope , equals .
Probttble error of the slope , equals : For For .
Probable error of the slope , equals Thus , of the three slopes all differ by less than twice and two ) them by less than once their probable error from zero .
We may accordingly conclude that neither in the best fitting straight lines , nor consequently in the Mendelian parabolas for the measurable rang6 , ' is there any sensible deviation from horizontality .
In fact we have couplets in the case of stature , and roundly 150 for span and 2000 for forearm .
With such numbers the Mendelian theory cannot on the problem of variability of arrays give any other sensible answer : .
for the range available for investigation than the constant variability * If the Mendelinn theory discussed were correct , it would be to grasp how the forearm-inheritance could be detel.mined by .
more couplets than tho span is , or why slope should be negative and the other positive .
who are ' father 's oflspring , \ldquo ; their moan , and their ndard deviation ; let and ) be the correspollding constants for the brethren who are " " mother 's ffsl ) ring Let and 1 be the number of brethren of each class .
Then if an individual have deviation from the mean of all brethren , he will if a " " father 's offspring\ldquo ; a -fraternity of standard deviation and from the mean of offspring given by .
He will have of such brethretl , but he will also havo brethren of mother 's offspring type with a standard deviation and at distance from mean of given by .
Similarly , if the brother of deviation were a " " mother 's olfspring , \ldquo ; his brethren would be made up of ) of standard deviation and mean , and of brethren of ndard dtion and mean where Prof. K. Pearson .
On a Criterion which may .
Hellae , generally for the variability of the brethren of a given of deviation , we have ( xix ) .
Clearly this gives a hyperbola for in terms of , with its real xtXiS perpendicular to Similarly we have : xx , and there ' and In any actual case we may reasonably suppose our pairs of brothers to be a random sample from families with indefinitely great numbers , or put and ) infinite in the ratio of to .
Hence : .
( xxi ) , * ' Phil. Trans , vol. 20$ , p. 77 .
( xxii ) . !
: To simplify still further , assume no distinction in the totals of .
' mothers ' offspring\ldquo ; and ' ' fathers ' offspring or take .
Hence Hence we see that if the variability of the arrays of brethren is constant , it is absolutely necessary that , or the legree of likeness between brothers whether they belong to " " mothers ' ffspring\ldquo ; or " " fathers ' offspring\ldquo ; must be identical .
If this be 1lot , the variability of the array of brethren of a brother of given haracter must obey a hyperbolic law , being least for a brother of mean character .
If we adopt Dr. Boas 's theory of complete alternate inheritance we have and Whence and ( xxiii ) .
This is a very easy result to test .
On the whole , however , it is better to ask the general question : Are the varial)ility of arrays of brethren hyperbolically distributed ?
( 11 .
) I propose to answer this by appeal to my data for nearly pairs of brothers measured for their cephalic index .
Selecting the arrays of 20 brothers and upwards the results plotted in Diagram .
were reached .
Here and and the mean cephalic index .
Hence the ' Law of Ancestral Heredity ' gives for the mean value of .
On Dr. Boas 's hypothesis These results are shown in the same manner as in the first diagram , the broken lines marking the limits of twice the probable error of and the actually observed We may , I think , safely conclude from this resuIt that the VOL , LXXIII .
X disturbing in these small arrays .
Even then we only find four CA$iJ in seventeen which deviate by more than twice the probable error the mean line .
In this case the mean error of the " " Ancestral Law\ldquo ; the seventeen arrays is , and of Dr. Boas 's hyperbola Diagram II .
( 12 .
) It may be asked how Dr. Boas has reached the conclusion ) that in the case of cephalic index the " " ] does not apply and there the children break up into " " mothers ' offspring\ldquo ; and ' ' offspring\ldquo ; He has used the measurements of Dr. Maurice of New York , on forty-eight families of ' ' East European Jews.\ldquo ; Now , I think it may be reasonably questioned whether a populationo defined as " " East European Jews\ldquo ; can be considered as homogeneous. .
I mean by this : Would any such category breed true to itself ?
Is it to be looked upon as a " " race\ldquo ; in the sense used by me of a the large values often found for the bility of extreme arrays of a tion table have real significance , which I doubt , I rsonally am inclined } think it is due to deviation from linear regression , the regression curve being a cubic curve with its inflexional portion representiug the regression line of the { ordinary range .
Trans.ompare wendelian p the value used by Dr. Boas , but the much longer expression iven by the general formula for " " Ancestral Hel.edity.\ldquo ; It is only we take the avsrage results for all milies of , that the other ancestral terms will disappear , we the ) resnlt His only method of reaching pproximntion to the ) eflll of indinidnal family , would be to actually find it from the mily itself , and then investigate the deviation from this This he has not done , and I think we must , even if he done it , the validity of means and variabilities based upon to individuals .
Still his memoir is very suggestive , and it seems to me the lnl.estigation of a iarge series of head surements in rents , , if possible , adult children , say , in 1000 families forming homogeneous population , would be of great value .
*I have sholvn in 'Biometrika , ' vol. 1 , p. 399 , thal values of stalldard deviation found from two individuals will , on the average , ouly bc of its true .
Similar large reductions occul .
if it be found three to foul individuals , and so on .
This is quite ftpart from the ltarge probable introduced by paucity of numbers .
'Roy .
Soc. Proc vol. 62 , p. 394 .
There would , I take it , certainly be correlation eell } llncestral terms the family mean .
ii ) Variability plotted to parental character will give ( a ) straight line on the Theory of Ancestral Heredity , ( b ) a parabola withrN axis along the axis of paren tal character on Mendelian Principles , ( a ) ) hyperbola with its real axis perpendicular to the axis of parenta } haracter on the Theory of Alternate Inheritance which is summed up 1 the idea of parsntally differentiated groups lvithin the family .
As far as my own measurements on stature , span , forearm , ; cephalic index in man go , there is nothing to support the view the variation curve of ( ii ) is effectively represented by either a parabola sampling it appears to be a horizontal straight line .
or Further aications oriterion woubt bWithin timits orrors o comlng , but it is essential they should be made with a full standing of what the various theories amount to , and how they must .
be applied to observations .
In particular it is very needful thai we should distinguish between the mean of an individual family as determined from all its ancestry , and when mean of an array of offspring of ) given parentage-type\mdash ; but with all varieties of earlier ancestry\mdash ; determined from that parentage only .

