]\gt ; [ 53 ] III .
Mathematicat Contribu to the Theory of Evolution . .
On a Generalised Theory of Alternative Inheritance , with special Reference to Laws .
KARL PEARSON , F.R. Received September ll , \mdash ; Read November 26 , 1903 .
( 1 .
) Introductory .
On Gene Alternative , with special Reference to MENDEL'S Laws .
* IT seems likely to be of interest at the present time to consider rather at length a fairly full mathematical theory of the pure gamete .
We do not venture to call this theory a generalised Mendelian theory ofinheritance , partly because it is not even the most general theory of the pure gamete conceiyable , partly because MENDEL'S original theory of heredity was perfectly clear and perfectly simple , and is not the theory here developed .
The pure and simple Mendelian theory seems to have been discarded in the light of recent experimental results by more than one Mendelian , both in this country and abroad .
The original Mendelian theory has been replaced by what are termed " " Mendelian Principles In this aspect of investigation the fundamental principles propounded by MENDEL are giyen up , and for each individual case a pure gamete formula of one kind or another is suggested as describing the facts .
This formula is then emphasised , modified or discarded , according as it fits well , badly , or not at all with the growing mass of experimental data .
It is quite clear that it is impossible while this process is going on to term anything whatever Mendelian as far as theory is concerned .
The present investigation is therefore not a generalised Mendelian theory of heredity : we speak of it merely as a generalised theory of alternative inheritance , and it is based on the conception that the gamete remains pure , and that the gametes of two groups , while they may link up to form a complete zygote , do not thereby absolutely fuse and lose their * I owe the incentive to this memoir to Professor W. F. R. WELDON , who had already worked at some of the simpler special cases who placed his results sntirely at my disposal .
See especially TSCHXRXAK , ' Zeitschrift .
landwirthsch .
Jahrg .
IV .
Ueber Zuchtung never Getreiderassen DE VRIES , 'Ber . .
deutsch .
botan .
vol. xviii .
( .1900 ) , pp. 435-443 ; BATESON , ' Proc. Camb .
Phil. Soc vol. 12 , p. 53 ; Nattlre , ' April and May , 1903 .
( 361 .
) 4.3.04 identity .
The analytical expression of this is represented by the fundamental formula : ( AA ' ) , where ( AA ' ) and are the parental zygotes , and the right-hand side of the equation represents the four possible constitutions of the offspring .
Such a formula as the above may be accepted without any hypothesis as to dominant an , recessive characters , but these terms were certainly essential to Mendelian theory as propounded by MENDEL himself , and it becomes very doubtful whether we ought to attach his name to any theory which discards these " " recognition marks It is very convenient , however , to have names for the alternative elements expressed by capital and small letters respectively .
I propose for the purpose of this paper to term an -element a protogene , and an -element an allogene .
Two protogenic elements will give rise to a protogenic zygote AA , two allogemc elements to an allogenic zygote , and a protogenic and allogenic element to what Mr. BATESON has termed a heterozygote .
We may thus class his homozygotes into protozygotes and allozygotes .
We reach pure Mendelianism by making our protozygotes " " dominants our allozygotes , " " recessives and our heterozygotes " " hybrids of dominant character In so far as our theory of pure gametes replaces protozygote , allozygote , and heterozygote by " " dominant " " recessive and " " hybrid with do1ninant character , \ldquo ; it becomes a generalised Mendelian theory , but only in case .
Otherwise we must look upon it as an attempt\mdash ; in one direction only of course\mdash ; to give a consistent mathematical basis to the various formulae which have been propounded for describing statistical data classed under Mendelian categories ; shortly we shall endeavour to develop a general pure gamete theory .
The results were worked out in a purely impartial frame of mind ; indeed , once state the hypotheses , and the analysis is far too complex to allow us to predict a priori what can possibly result from it , nor does the investigation admit of any but one solution .
If the hypotheses are admissible , then any narrower pure gamete formula must lead to results embraced under our general conclusions .
What we have to admit at the present time are the following conditions:\mdash ; The existence of a vast bulk of evidence that heredity , as far as measurable characters are concerned , follows within a population perfectly definite laws .
( ii .
) The existence of another mass of experiments , in which simple and pure Mendelianism is certainly inadmissible , but in which certain ratios undoubtedly approach the values they would have on such a simple and pure Mendelian theory .
It is possible , therefore , that a generalised theory of the pure gamete would account INHERITANCE , WITH SPECIAL REFERENCE TO MENDEL'S LAWS .
for ( ii .
) ; can only do so satisfactorily , howeve , if it.does not contradict the results of .
Hence arises the present attempt to develop in one direction a generalised theory of the action of the germ-cell .
As we have frequently had to assert , the laws investigated under have nothing whatever to do with any physiological hypothesis .
That a physiological hypothesis leads to them is not much test of its validity\mdash ; it is a necessary , but not criterion of its correctness .
If , however , it contradicts them , we are bound to discard it , and seek for its modification or replacement .
The present study is an attempt to see how far one generalised pure gamete theory leads to resuIts in accordance with the law of regression and the known nature of the distributions of offspring in populations .
( 2 .
) Nature of Hypothesis adopted .
We start with a zygote consisting not of a single protogenic pair AA , but built up of such pairs , We suppose this to produce gametes which unite with those of a similar allogenic zygote Any element of the protogenic gamete must unite with the corresponding element of the allogenic gamete , i.e. , with , and by the fundamental principle above , this gives rise to the four possibilities are all of the same constitution .
The result is the hybrid group , symbolised by the perfect multiple heterozygote .
The population will now be supposed to consist of any number of such perfect heterozygotes , which we shall suppose to again cross .
We shall now have 56 SSOR K. PEARSON ON A GENERALISED THEORy OF ALTERNATIVI or , each couplet will give rise.to four possibilities , representing , however , only three constitutional differences , expressed by Since these four possibilities may occur with each of the couplets , we shall have when two perfect heterozygotes cross , resulting possibilities .
These form the resulting population of the second generation .
Our first problem will be to find the distribution of this population .
This , according to MENDEL , is the segregating generation .
We must inquire into the frequency of each constitutio ' difference in this segregating generation .
We now reach our second limiting hypothesis , which is needful if we are to apply our theory to sexual reproduction .
We suppose:\mdash ; There to be an absence of homogam ( including self-fertilisation ) , and the members of the second generation to cross absolutely at random and with equal fertility .
* We have then to ask what is the distribution of constitutional differences in the third generation .
Does the process of segregation begun in the second generation continue to the third , or does the population now remain stable Is the continual regation into pure protogenic and allogenic individuals a necessary result of any pure gamete theory , or does the belief in such necessity depend upon the first Mendelian experimenters working only with self-fertilising individuals ?
( 3 .
) PROBLEM I.\mdash ; To find the Distribution of the Offspring of the Perfect Heterozygotes .
We shall here use a symbolic form of analysis .
Let stand for for , and for AA ; then any corresponding couplets will give rise to and any one of these constitutions may be associated with one of the similar constitutions in any of the remaining couplets .
Hence the general distribution of the .
population will be given by the terms of the multinomial This equals * If one is to study heredity in populations with a view to the problem of evolution , the conditions as to fertilisation should approach as far as possible the conditions we suppose them to be under in a natural state ; we must fix our attention on the mass relations between successive generations of the population .
Throughout this memoir the symbol ITANCE , LAWS .
Thus , for example , there would be out of the total population of possibilities : 1 purely allogenic individual , individuals with allogenic couplets ; of these would have one heterogenic couplet , and would have one otogenic couplet .
Generally there would be individuals with allogenic couplets , and these individuals would be distributed according to the terms of the binomial We are thus able to write down at once the number of any class of individual that can appear in the segregating generation .
For example , how often do indiyiduals like appear , i. e. , individuals with allogenic , heterogenic , and protogenic couplets ?
To answer this problem all we have to do is to pick out the coefficient of in the aboye multinomial , and the result is We are thus fully able to predict how many individuals of each ought to occur when a population of perfect -couplet heterozygotes are crossed .
Corollary \mdash ; Let us consider only the number of allogenic couplets in the distribution of the segregating generation .
If we were " " pure Mendelians\ldquo ; we should for the purpose of character classification make , as the heterogenic couplet would then give the dominant character .
But without doing this we can assume and to be non-u 's .
Hence the distribution of allogenic characters in the population follows the simple binomial Thus we see that the distribution would be a skew closely to my skew curve of Type III , , * and becoming indefinitely close to a normal distribution of the form a when the number of couplets , , is indefinitely increased .
For any value of the mean of the skew binotnial as measured by the formula the memoir on " " Skew Variation and the standar deviation Thus the mean number of couplets the mbers of the segregating generation of the total number of couplets .
* , vol. 186 , p. 373 , Ibid. , p. 346 .
CCIII , 58 PROFESSOR K. PEARSON ON A GENERALiSED THEORY OF ALTERNATIVE Corollary ( ii.).\mdash ; The distribution of heterogenic couplets in the segregating generation is giyen by the symmetrical binomial The mean number is therefore , and the standard deviation This is a symmetrical binomial , and approaches extremely closely , even for a fairly small value of , to the normal curve .
We see that if any character depends solely upon heterogenic couplets , the distribution will be nearly normal , and the variability slightly greater than one depending on the allogenic ( or , of courae , the protogenic ) couplets only .
To sum up , then , so far as the distribution of characters depending upon allogenic or heterogenic couplets goes , we may say that a theory of the pure gamete leads us to those normal and skew distributions of frequency with which biometric studies of variability have made us already familiar .
It would not be possible to base a crucial experiment on the existence or non-existence of such frequency distributions .
The generalised pure gamete theory would , however , account for appearance , which , of course , a purely descriptive statistical theory cannot do .
On the other hand , distributions diverging much beyond the errors of random sampling from binomials of the aboye types would tell pro tatno against the pure gamete theory in its above form .
The presence of binomials of two types only , and , ought to be capable of detection , even if it would not already have been discovered , had it been the rule .
( 4 .
) PROPOSITION II.\mdash ; To determine the Distribution of the Offspring of the , supposing them to at Randorn and without Differential Fertility .
The solution of this problem may be reached as follows:\mdash ; Suppose any male , and any female , say each of couplets , producing an array of offspring , which we will denote by ; now suppose an additional couplet , the , added to both male and female zygote .
The male may be now : , or , or , or ; and the female may be , or , or , or ; that is , we get new mating individuals , with new offspring possibilities .
Now consider the first father ; the possibilities which arise from mating him with the four mothers are the array of offspring combined with any one of the INHERITANCE , WITH SPECIAL REFERENCE TO MENDEL'S LAWS .
16 possibilities , or this is the same thing as multiplying the array by the symbolic factor 8 , say .
The next pair of fathers 2 with the four mothers reproduce the array of offspring combined with 8 , or 32 possibilities .
But this is the same as multiplying the -array by the symbolic factor 8 say .
Lastly , the father with the four mothers gives 16 possibilities of the form to be combined with the -array of offspring , which is the same thing as multiplying the -array by the symbolic factor 8 , say .
We have at once the symbolic relation among the operators : ; and , further , the important result that the array of offspring due to any pair and of -couplet parents can be converted into the arrays of offspring due to the 16 pairs of,.parents formed by adding an additional couplet to and , by multiplying that array by the symbolic factor .
We have thus by induction a means of finding the array of offsprin due to a population of parents of couplets from the series of arrays due to a population of \mdash ; 1 couplets .
Since all the arrays are to be multiplied by the same symbolic factor , we can multiply their total by this factor .
Or the distribution of offspring of -couplet parents being , that of -couplet parents J. Now consider parents of one couplet , their distribution is given by and they are to mate with the same series , But , , , Total symbolically .
60 SSOR K. PLARSON ON A RALISED THEORy OF ALTERNATIV Hence , by the above proposition , the distribution of offspring of parents of two couplets is and , by induction , the distribution of offspring for the random mating of parents of couplets is ' This , except for the constant factor , is absolute } identical with the distribution of the parental population , and accordingly if the next generation also mates at random , the mixed race will continue to reproduce itself without change .
We therefore reach the following result:\mdash ; However many couplets suppose the under investigation to depend upon , the of the hybrids\mdash ; or the segregating they breed at random inter se , will not further , but continue to reproduce themselves in the proportions as stable population .
It is thus clear that the apparent want of stability in a Mendelian population , the continued segregation and ultimate disappearance of the heterozygotes , is solely a result of self-fertilisation ; with random cross fertilisation there is no disappearance of any class whatever in the offspring of the )rids , but each class continues to be reproduced in the same proportions .
Thus our generalised theory lends no countenance to the appearance of any " " mutations\ldquo ; within a hybrid population under mating ; the only appearance of new constitutions is in the segregating generation , or the first generation of hybrid offspring .
Except at this stage , the appearance of the unfamiliar is only the chance occurrence of a very rare normal variation .
When we recollect that a purely allogenic individual is only to be expected once in a population of individuals , or if there be ten couplets , once in more than a million individuals , it will be clearly seen that the rarity of some of the more exceptional normal constitutions may easily lead to their being looked upon as " " mutations even if they appear in the offspring of a population many generations removed from hybridisation .
( 5 .
) PROPOSITION III.\mdash ; To find the Array of Offspring to a Parent of given Gametic ?
stitution at This can be again deduced by the method of induction adopted in the last proposition .
Supposing a male of couplets to mate with all possible females , and to be the array of offspring , then we have seen in the last proposition that if we add an couplet to , the array of offspring due to will be ; if we add a couplet , the array of offspring due to fathers of type , will RITANCE , WITH SPECIAL REFERENCE TO MENDEL'S LAWS .
, and if add a couplet of form , the array will be of the form Now start with a father of one couplet ; this must be , or , or , or in our symbolic notation , or ; the offspring array are respectively or , or , i.e. , , or .
These , therefore , are the possible values of .
Hence , by the principle just developed above , the array of offspring due to a father of type is ( 16U ) or remembering that such fathers occur with a frequency of , we have for the total distribution of offspring of all fathers of type the symbolic result Substituting , the following expression would give all offspring of fathers of the type , i.e. , with allogenic , heterogenic , and protogenic couplets Therefore , given and given and , it is merely a matter of expansion to find the array of offspring due to any special class of father .
Corollary .\mdash ; So far we have supposed our special class of father to be defined by the exact couplet distribution constitutional to him .
But it is of interest to consider the array of offspring we get supposing only the allogenic couplets fixed in number , for example , in a generalised Mendelian theory if the number of recessive couplets be fixed , but the heterogenic and dominant , as both exhibiting dominant characters , be considered as indifferent .
Let number of allogenic couplets , then we have to sum all arrays like subject to the condition that The result is clearly Tbis , we note , is not a pure binomial , or the arrays of offspring of a father with a G2 PROFESSOR K. PEARSON ON A GENERALISED THEORY OF ALTERNATIVE given allogenic constitution are not either symmetrical or skew bihomials , but of a much more complex character .
The only exception is the array of offspring of pure allogenic fathers , *which is given by This is a symmetrical binomial .
This result is , of course , of special interest , for it gives us the distribution of offspring if the hybrid offspring were at any time crossed with the pure allogenic race , which was one of the original factors of the hybridisation .
The deviation from binomial distribution in the above arrays ought to be further considered , for if this deviation should turn out to be very significant , it would form a convenient test for any generalised theory of pure gametes .
Corollary ( ii.).\mdash ; If we sum the above expressions for the array of offspring of all fathers of allogenic couplets for values of from to , we have the total offspring population a result we have already found in Proposition II .
as giving the distribution of the total offspring population .
( 6 .
) PROPOSITION IV.\mdash ; To find the Number of Allogenic Couplets in the Offspring of all Fathers in their Constitution -allogenic Couplets .
By the first corollary to the last proposition the distribution of such offspring is given by where is written for , a quantity which is unity so long as we consider not the distribution , but the total number of the non-allogenic couplets .
Now this is clearly the sum of a number of symmetrical binomials in , and may be put Now the means of each of these binomials can be found from the general theory of the binomial .
If we take our origin at allogenic ] , with a frequency zero , the mean of the first binomial , or * Or , of course , the array of sons from pure protogenic fathers , 'Pbil .
Trans , vol. 186 , , 373 .
INHERITANCE , WITH SPECIAL REFERENCE TO MENDEL'S LAWS .
is at , and its total frequency ; the mean of the second binomial , or .
the mean of the third binomial , or .
; the mean of the binomial 1 .
2 .
3 The total frequency is accordingly Hence if be the istance from the same origin of the mean of the above system of binomials Now Multiply by both sides- and divide by 2 , finally putting , and we find Hence we deduce or .
PROFFSSOR A NEIiALISED But the mean of the whole of offspring is at from our origin .
we the final results : Mean uumbel of llogenic couplets in pring of fatbers with allogenic couplets \mdash ; allogenic couplets .
Deviation from mean of population of this array of .
tiou of thers from mean of tion .
} We have then the following results , which could certainly not have been foreseen :\mdash ; The ession is constant all arrays , or the regression curve is a straight line .
slope of this straight line is , or , since we have seen that the population is , the parental correlation is also .
Now these results are of very singular importance .
A general theory of the pure gamete type leads to linearity of the regression curve , a result amply verified observations on inheritance in populations ; and this result is quite independent of the number of couplets supposed to form the total character of the parent , or of the fact that in this case the arrays of offspring are skew and do not obey the normal law .
Further , the of the correlation reached is numerically identical with the value obtained by FRANCIS GALTON in his original investigations on the inheritance of stature !
The generalised theory of the pure gamete is thus shown , whatever the number of couplets taken , to lead to precisely the chief results already obtained by those who have studied heredity statistically .
So then it might appear that a generalised theory of the pure gamete was capable of being brought into accordance with the chief results of biometric experience in heredity .
This would undoubtedly be a great step as linking up perfectly definite inheritance results with a physiological theory of heredity .
Unfortunately the whole drift of recent biometric observations on heredity emphasises three points First.\mdash ; That the parental oorrelation appears to be markedly greater than to 45 to GALTON , ' Natural Inheritance , ' 1 ) .
, pp. 216 This is furthel demonstration that linearity of regression has nothing whatever to do with GaussLaplacian law of errors , , normal curVes or surfaces ) INHERITANCE , WITH SPECIAL REFERhNCE TO MENDEL'S LAWS , Secondly.\mdash ; That this correlation appears to vary slightly from character to oharaoter .
Thirdly.\mdash ; That it does not appear to be absolutely the same for all species .
It is most unfortunate for this general theory of the pure gamete , that it throws the Mendelian back into the position , of the biometrician of One might have hoped that the generality involved in couplets would have led to the requisite elasticity , or , failing this , to a numerical value of parental correlation nearer the cluster point of existing measurements than .
We can only say , at present , that a generalised theory of the pure gamete leads to precisely the same general features of regression as have been observed by the biometricians , but it appears numerically too narrow to describe the observed facts .
( 7 .
) PROPOSITION V.\mdash ; To find the Standard Deviation of the of Offspring to with -allogenic Couplets .
We have to find the standard deviation of the combination of binomials dealt with in the previouq proposition .
Each component standard deviation must , of course , be weighted with the total frequency of the component , and there must be the proper reduction to the mean of the array as a whole .
The binomial has for its standard deviation , and the distance of its mean from the mean of the arTay Further , the frequency of this component is We thus see that it contributes to the total second moment about the mean of the array .
This gives us ' * ' GALTON'S law makes the amount of inheritance an absolute constant for each pair of relatives .
It would thus appear not to be a character of race or species , or one capable of modification by natural selection More ample statistical experience of populations since 1885 shows that absolute constanoy of the heredity coefficients is not consonant with actual measurements.\mdash ; ' Roy .
Soc. Proc vol. 62 , p. 411 , 'Phil .
Trans , vol. 185 , p. 373 .
VOL. CCIII .
or Now and by differentiating Repeating the process Hence , putting , we have the required expressions on the right of the above result , or Therefore .
Now the standard deviation of the whole population , as far as allogenic units are .
concerned , is Thus This result is of singular interest .
The variability of an array of .
offspring corresponding to a father of given allogenic constitution is not independent of the father , but increases steadily from a minimum of , when there are no aUogenic couplets in the father , to a maximum of when there are only allogenic couplets .
In other words , fixing our attention on the same character , let the offspring of the hybrids inter se be crossed first with one race , and then with the second pure race , i.e. , first with pure allogenic and then with pure protogenic individuals , there ought to be a marked difference in the variability of the resulting in the two cases .
Corollary .\mdash ; In the theory of linear regression as apart from the theory of normal correlation on the basis of the Gauss-Laplacian distribution , * if be the standard deviation of any character , and its correlation with a second character , then YULE , ' Journal of Royal Statistical ' vol. 60 , December , l89 TO MENDEL'S LAWS .
is the mean standard deviation of all the arrays of the first character for a given value of the second .
This expression is no longer the actual standard deviation of each array .
It is of interest to see that this general law of linear regression is verified in the present case .
We have and .
Hence if be the mean standard deviation of the arrays , we should expect Remembering the weight of each array , $ we have .
whence , as anticipated .
Corollary ( ii.).\mdash ; It is clear that some arrays of offspring will be more , others less variable than the general population .
The standard deviation of an array will be equal to that of the general population when is found from , or Now the mean number of allogenic couplets in the general population is .
Thus the offspring array equally variable with the general population is at distance \amp ; om the mean .
But for the general population Hence , if we take fathers iatin from the population mean by , we should expect their offspring to be equally variable with the general population .
Supposing , therefore , the theory under discussion were true , we should have a means of finding , at least approximately , the number of couplets corresponding to the character under consideration .
All we should have to do would be to find the standard deviation of each array of offspring corresponding to a given father ; these standard deviations ought to increase or decrease steadily across the table , their squares giying a straight line when otted .
Smooth the results , interpolate a value equal to , and we shall have the character of the father whose offspring are equally variable with the general population ; but the deviation of this father from the mean ought to be Hence , since can be found , we have at once an approximat6 value of mate only , of course , because our arrays are classed by units of measurement , inches , * The number of offsprlng in the array due to the fathers with allogenic couplets is found at once by putting in the formula of Corollary on p. 61 , and equals PROFESSOR PEARSO A GENERALISED centimetres , \amp ; c. , and each such unit will not , as a rule , represent one allogeni6 couplet ; but interpolation ought to give a result not widely divergent from the truth .
The method would of course fail practically if were very large .
For example , if were 48 , the deviation of the required group of fathers would be , and hence such a father would only occur once in 1000 individuals .
In a manageable population , therefore , we are very unlikely to have enough such fathers to form any reliable measure of the variability of their offspring .
At the same time , the squares of the variabilities of the arrays of sons due to quite frequent fathers ought to give a straight line , and if this line be determined properly , there should be no difficulty in finding the theoretical position of the above father , and so finding Many other physiological theories besides the present might give this peculiarity of the diminishing of the arrays of offspring as we pass from one side of the correlation table to the other .
Such changes in variability are familiar to those ho have had to deal with skew correlation .
But , as far as we are aware , they have not hitherto been noticed in inheritance tables .
The existence of this changing variabilit would not affect in any way the general theory regression applied to heredity in populations , It would , however , lead to an immediate extension of that theory consisting in the tabling of the standard deyiations of the arrays .
Should the standard deviations of these arrays show no bias towards a linear distribution , but only the fluctuation to be expected from random sampling about the mean value we should have a strong argument against the present general theory of alternative inheritance .
We seem here , therefore , to have a crucial test of the validity of the theory , which may be quite as easy to apply as the previous test of the numerical value of the parental co1Telation .
Of course the results now reached are not consistent theoretically with normal correlation surfaces with their elliptic contour lines .
The fact that Mr. GALTON came to his elliptic contours in the first instance on the basis of his observations , and not from any theory , that they must in the case he was dealing with be approximately correct .
Further , there is no doubt that in other statistics for characters in man there is within the limits of random sampling a close approxlmation to normal distribution , It might be hard to consider that such a deviation as would arise with a continuously increasing variability of the arrays from one side to the other of the table could exist and escape notice , had we not in physics had evidence that theory has often led to the discovery of an obvious relation which time after time must have been overlooked by previous observers the theoretical hint of what to seek for .
Hence while we may say that the parental correlation giVen by the theory is too rigid for the facts , we must leave this second test until ore careful examination hoc has been made of the ample existing data : ' Natural Inheritance , ' p. 101 .
INHERITANCE , WITH SPECIAL REFERENCE TO ) ( 8 .
) PROPOSITION the of due to a of Couplets , snpposing Complete Mating the The general distribution of the population is The fathers with -allogenic units in their correlation given by The array of offspring due to these fathel .
S is simply obtained by writing for V for , and for , and multiplying by .
This is a genel.al rule for gettin the offspring from any father if he mates at random .
It gives us , as on p. 61 , for the offspring distribution Us To get the offspring of this array treated as fathers and mating at random , we have only to repeat the process , and we find Offspring of grandfather of -allogenic couplets and is equal to if we identify and as not allogenic .
This can be dealt with exactly as Proposition IV .
we dealt with the array of offspring due to a father of -allogenic couplets , i.e. , by.analysing the array into the sum of a number of weighted nomials ; in this case all skew .
Writing as before , , we have to expand The general term is This hag a total frequency , and its meau is a distance from the origin which is taken at allogenic couplets .
70 ESSO .
PHiARSON ON A THEORY OF The total frequency of the array is .
Hence , if , be the mean of the grandchildren measured from the same origin , we have or .
Thus Mean of grandchildren \mdash ; .
Deviation from general population mean .
Deviation of grandparent from general population mean .
Hence This atio is the same whatever be the allogenic constitution of the grandparent .
( 9 .
) PROPOSITION .\mdash ; To find the Arrciy of due to an -grandfather of llogenic Couplets , supposing Complete Mating in each Generation .
The array due to a father of -allogenic couplets is and , as we have already seen , we multiply by and put for for , and for to get the array due to the grandparent of allogemc units .
This process must be repeated times if we wish to obtain the array due to the great-grandparent .
We must first investigate what happens to if this interchange be made times .
Suppose that it has been done times , and let the answer be .
Repeat the operation , and the expression becomes INHERITANCE , WITH SPECIAL REFERENCE TO MENDEL'S LAWS .
or Therefore Hence Hence , finally , Thus the result of changes on is known .
Similarly the result of changes on is We can now write down the array of offspring due to an great-grandparent of genic couplets .
It is We must now find the mean of this array .
For brevity let us write , where Hence we have to find the mean of the system The component binomial of this sum of binomials is It therefore has its mean at a distance from allogenic couplets , and a frequency given by The total frequency of the whole array Hence , taking moments round the origin at allogenic couplets , we have , if be the mean of the array , Summing and dividing by we find Hence the mean number of allogenic couplets in the members of the array .
Deviation of offspring from mean of general population We now substitute for and in terms of and , and find INHERITANCE , WITH SPECIAL REFERENCE TO MENDEL'S LAWS .
Hence : Deviation of offspring but the deviation of great-grandparent Thus we have This result is independent of , and of Thus we conclude : The regression of offspring on any individual ancestor is linear ; ( ii .
) The correlation coefficient is halved at each stage in ; ( iii .
) The result is perfectly independent of the number of couplets introduced into the formula .
The first two results are very familiar to biometric workers in heredity .
The actual numerical values of the grandparental , great-grandparental , great-greatgrandparental correlations are , \amp ; c. These are distinctly less than the values so far reached for ancestral correlation , the grandparental correlations , for instance , lying between2 and 3 .
The results show , however , that a general of the pure gamete , embracing the simpler forms of the Mendelian principle , leads us directly to a series of ancestral correlations decreasing in a geometrical progression .
Thus , when we suppose a population arising from hybridisation to cross at random , we find that it obeys the second fundamental assumption of the biometric theory of heredity .
In other words , ancestry is of the utmost importance , and the population follows laws identical in form with those propounded in the biometrical theory on the basis of a linear regression multiple correlation .
Only the values of the constants deduced for the law of ancestral heredity from the present theory of the pure gamete ( which appears to cover the bulk of Mendelian formulae hitherto propounded ) are sensibly too small to satisfy the best recent observations on inheritance .
It is of interest to find ' ' Mendelian Principles\ldquo ; when given a wide analytical expression leading up to the very laws of linear regression , of distribution of frequency , and of ancestral inheritance in populations , which have been called into question as exhibiting only a blurred and confused picture of what actually takes place .
It would be an immense advantage if we could accept such a theory of the pure gamete as has been here analysed as a physiological basis for the theory of heredity .
We should then have a physiological origin for the ideas of regression and of ancestral inheritance which statistics of heredity in populations have made familiar to biometric workers .
Unfortunately , even such a general pure gamete theory as we have here dealt with , while leading to results which form a special case of the law of ancestral heredity , is not sufficiently elastic to cover the observed facts .
The lesson * Biometrika , ' vol. 2 , p. 220 .
VOL. CCIII.\mdash ; A. to be learnt from the present investigation is , however , that there is no essential repugnance between any of the main results of the biometric school and a theory of the pure gamete , but , on the contrary , it is perfectly possible to test such theories by biometric methods .
We may fairly ask anyone who propounds in future a Mendelian or pure gamete formula as a general theory of heredity , to remember that it involves in itself definite laws regulating the reproduction of a population mating at random , and that it is incumbent on the propounder to test whether or not such laws are consistent with what we already know of the inheritance statistics of such populations .
When we remember that deducing the effects of such a formula within the whole field of inheritance will almost always form a very laborious piece of mathematical analysis , there seems a touch of scientific irresponsibility in propounding an immense variety of formulae to suit one or other special case , and the modifying or withdrawing them when they are found to fail in another .
( 10 .
) PROPOSITION VIII.\mdash ; To find the and Correlation of Brethren on Theory developed We shall suppose the of brethren to consist of members , or any pair of parents to have a family of Consider first parents of one couplet only , the offspring of the 16 possible pairs are given in the table below:\mdash ; Now let us take every ir of brothers out of each of these 16 families and form a correlation table of brothers .
The following table gives the distribution of the various types:\mdash ; INHERITANCE , WITH SPECIAL REFERENCE TO MENDEL'S LAWS .
This gives a total of pairs of brothers , as it should , every brother in 16 families having brethren , and there being in each family .
Clearly we can divide by , and we have the simplified form : where the subscripts 1 and 2 refer to the first and second brothers .
Let us simplify this by considering only the allogenic elements , denoting either a heterogenic or a protogenic element .
Then we have : Thus the distribution of pairs of brothers in the case of the character being fixed by a single couplet is This is to be read as follows : there are cases of both brothers being allogenic , to cases of neither brother allogenic , to cases of one only of the two brothers being allogenic .
Now when we pass from a character fixed by one couplet to a character fixed by two , the above distribution can occur in either couplet , and very possible pair of brothers will be got by squaring the above expression .
Proceeding in this way to couplet characters , we have the following symbolic expression for the distribution of brothers the omitted factor being restored .
This , I think , represents the distribution of a character measured by allogenic couplets in a population of pairs of brothers , i.e. , is the correlation table for brothers in the population\mdash ; any term involving representing when we put and equal to unity the number of pairs in which the first brother has and the second allogenic couplets in their constitutions .
Hence if we find the coefficient of in terms of , we shall have the array or brothers to be found associated with a brother of allogenic couplets .
The above expression may be written The term involving is Neglecting the constant factor , the distribution in is given by We require to find the mean of this array .
Put Then again , but for a factor independent of the power of , the array may be read The general term is therefore and we must sum from , to The general term has therefore its mean at the distance from allogenic couplets , or its mean , allogenic couplets , and its total frequency This gives a total frequency of the array proportional to Hence , taking moments , we have for the mean of the array given by Therefore Hence we see that The regression between brothers is linear .
( ii .
) The fraternal correlation which is equal to the regression ' INHERITANCE , WITH SPECIAL REFERENCE TO LAWS .
and is quite independent of the nber of couplets .
It is , however , a function of the size of the family used in forming the table .
We have the following values:\mdash ; The value of fraternal correlation thus varies with the size of bhe family dealt with from to 4 .
Probably the more correct way of looking at any fraternal correlation table would be to suppose it a random sample of all the pairs of brothers which would be obtained by giving a large , or eyen indefinitely large , fertility to each pair , for what we actually do is to take families of varying size and take as many pairs of brethren as they proyide .
In this case we ought to reach a fraternal correlation precisely the value reached by the ancestral law when we take FRANCIS GALTON'S original series of ancestral correlations .
* Thus we conclude that on the general theory of the pure gamete here dealt with , the fraternal correlation is slightly larger than the parental .
This is in accordance with the general result of biometric investigations on populations .
But the value , as in the case of the parental correlation , is very sensibly lower than the value\mdash ; about \mdash ; found from recent investigations on man .
It is further very inelastic even if we allow for some variation in the size of families dealt with .
There can be little doubt that .fraternal correlation varies from character to character and species to species in a manner sensibly beyond what can be accounted for by differences in the size of the family dealt with .
Corollary.\mdash ; We can exhibit the regression in the form : Mean of array \mdash ; mean of general population by observing that , whence * .
Soc. Proc vol. 62 , p. 410 .
There is sensible variation even for different characters , when we take the same series of pairs of brothers , and only one pair from each family .
or .
( 11 .
) PROPOSITION IX.\mdash ; To find the Formula for Biparental Regression on the Theory of the Pure the to be to the " " Midparent If we applied without further consideration the general formula for biparental regression to this case , we should have , if be the mean of the offspring due to fathers of -allogenic couplets , mated with mothers of -allogenic couplets , .
This follows at once , since the mean of the general population , the regression coefficient for either parent , and there is no assortative mating .
Hence we should have .
Now suppose both parents of pure allogenic race , then , and all the offspring will be of pure allogenic race , or we must have But the above formula gives which is not correct .
In other words , while the above formula gives the best plane to fit the array of points determined by the parental constitutions , that array of points does not truly lie in a plane .
Or , although the simple regressions are linear , the compound regression is not truly planar .
We have therefore to find its true and measure the amount of deviation from the truth involved in using a biparental formula of the type indicated .
Given a character resulting from couplets , we require individuals , male and female , to form the whole possible system of random matings .
In such a population there would be fathers of -allogenic couplets , and mothers of -allogenic couplets .
The chance therefore of a mating of a -allogenic father and a -allogenic mother is and when / ?
be even moderately large , this gets very small if and at all nearly approach .
For example , if were 5 , and father and mother were both pure allogenes , the chance of such a pure allogenic mating would be only 1/ 1,048,576 , or in a population of a million would hardly occur once .
INHERITANCE , WITH SPECIAL REFERENCE TO MENDEL'S LAWS .
Still keeping , take .
Then we have the chance of such a mating 1/ 4215 still extremely improbable .
Thus when is eyen moderately large , pure allogenic matings are so that they have vanishingly small influence in the population at large .
Even if were 2 , the chance of a pure allogenic mating is only .
These points must be borne in mind in what follows .
Consider first a father and a mother of one couplet their zygotes either or , involving a gametic constitution .
We have the following scheme:\mdash ; Zygote of father .
Zygoto of mother .
Number of matings .
Offspring .
Hence if there be 1 allogenic couplet in father and 1 in mother , offspring . . . . .
1 . . . . .
, . . . . . . . . . . .
Let us write Then consider the relation , where and are mere symbols , and , 1 , etc. , denote their powers .
refer respectively to father and mother , and their powers denote the number of allogenic couplets in the zygotes of father and mother .
Then the above is a symbolical relation which gives , by equating any power or product of and on either side , the offspring of a pair of parents of definite constitution .
Now suppose the parents not to consist of a single couplet , but of couplets , then the total distribution of offspring that we have given above for any couplet may occur in each couplet , and each such distribution must be combined with every other couplet distribution .
We then reach , dropping unnecessary indices , the general symbolic relation , the lrray of offspring due to parents of zygotes with and allogenic couplets resI ) , to \mdash ; is the coefficient of on the right-hand side , or in the expansion of This may be written the coefficient of is We require to pick the coefficient of out of this in order to get the array pring due to fathers of , and to mothers of , allogenic couplets .
But this is clearly \ldquo ; , 1- .
, or , more briefly , We shall first find the mean and frequency corresponding to the term as given above of this series .
What we have to deal with is We may write this \ldquo ; , where and , and both may be put unity when we are merely finding the distribution of allogenlc couplets .
Now the general term in the above expression is INHERITANCE , WITH SPECIAL REFERENCE TO MENDEL'S LAWS .
and must be taken from to : The frequency of this term is and its mean is allogenic couplets .
The total frequency of the term is therefore which of course must ultimately be multiplied by the factorials in omitted above .
If be the mean number of allogenic couplets in the we have Thus : , This is the mean of the term , and its total frequency is Hence , if we shall have where is the mean of the array of offspring due to fathers of and mothers of allogenic couplets .
The only difficulty here is summing the series But this may be written , or multipled by , it represents the total offspring of fathers of and VOL. CCIII .
mothers of allogenic units a population with couplets in their constitution Hence , -l , Now Thus or , if denote the above series , we have This leads us to .
This is a most remarkable result , for it shows that the regression surface is not a plane but a hyperboloid .
Let us measure all the quantities in deviations from the mean of the general population , i.e. , put We find This formula reconciles at once the Mendelian and Galtonian positions .
When the number of couplets is large , parents having a number of allogenic couplets comparable with are vanishingly small in number .
The standard deviation of the population is ( see p. 57 ) .
Hence we may write But can only once per thousand cases be as big as , and accordingly when is large , both terms of the product will be small .
In this case the surface becomes practically planar , and we have the Galtonian result of that the offspring from the Galtonian " " midparent\ldquo ; are one-third nearer the general mean of the population .
On the other hand , when is small we see that for midparents not differing too widely from the population mean , Galtonian regression of the value holds , but that 'Natural Inheritance , ' p. 97 .
INHERITANCE , WITH SPECIAL REFERENCE TO MENDEL'S .
LAWS .
as we pass away from the mean of the population the regression gets less and less , becoming absolutely zero when we take two pure allogenic parents .
Just as the regression is reduced when we emphasise the allogenic constitution of the parents , it is increased when we emphasise the non-allogenic elements .
To illustrate these points I have drawn a diagram for of the contour lines of Midparental contours .
the regression surface as a plane and a hyperboloid .
The area ABCD contains all possible parents ; 999 out of 1000 random matings fall in the loop out of 100 random mating within the loop , which has been carried right round to mark that it excludes matings in which both parents have no allogenic couplets .
The means of the array of oflspring due to any pair of parents are marked in slim figures along the Mendelian hyperbolic contours , and in heavy figures along the Galtonian straight lines .
We see comparatively small differences as long as we deal with matings within the 99 per cent. loop , somewhat greater differences as we approach the per cent. loop , and very marked differences as we go beyond the latter boundary towards pure allogenic .
A study of the diagram illustrates at once how the Mendelian theory exhibits for the bulk of the population Galtonian regression , such regression bscoming , however , less and less as we proceed to indiyiduals , the frequency of whose matings may be less than one in a million .
It will be seen that in this proposition we have no fundamental antagonism between the Mendelian and biometric standpoints .
We reach a single formula which approaches and more closely to the biometric standpoint when we deal with characters depending on many ] ] ogenic couplets .
* On the other hand , it gives the absence of regression which is obtained when pure allogenic parents are mated .
We see , however , quite clearly that it is totally erroneous to argue from this single case against regression in general .
Such regression actually exists on " " Mendelian Principles\ldquo ; when any population breeding at random is taken , and involves in itself the whole conception of ancestral correlation and the influence of ancestry .
Of the formula for the midparent now reached , however , we can only say that on the basis of our experience in populations the factor seems too inelastic to work .
Here again the data must be especially investigated from the standpoint of a midparent given by before judgment can be final on this test .
Theoretically , by assuming the midparent to be we should have a means of finding by ayerages , and therefore , the number of couplets involved .
( 12 .
) Generat In this paper we have dealt with a general theory of the pure gamete\mdash ; possibly not the widest that could be conceived\mdash ; but sufficiently wide to indicate the real bearing of Mendel.ian formulae when applied to a population mating at random .
We see that under such circumstances : The population which results from the offspring of hybrids remains stable ; eyery variation which appears , appears with a certain definite and predicable * For or 100 the population the 1 in 1000 line sensibly obeys ordinary linear midparental regression .
INHERITANCE , WITH SPECIAL REFERENCE TO MENDEL'S LAWS .
frequency ; there is no room for the appearance of " " mutations although certain variations with very small frequency would be extremely rare in a limited population .
A mutation\mdash ; a variation not hitherto observed\mdash ; would only appear in the offspring of the hybrids between two pure races ; after this with random the mixed race would remain perfectly stable until disturbed by sexual or natural selection .
These are the only mutations which arise on the generalised theory of the pure gamete , i.e. , two pure races form one mixed race , breeding true to itself ; it is difficult under these circumstances to account for the origin of the two pure races by a mutation-theory of the differentiation of species ( ii .
) Between any two relations\mdash ; if we measure character by the number of allogenic or protogenic couplets in the zygote of the individual\mdash ; we have a linear regression .
The frequency distributton of any character is skew , approaching closely the normal distribution as the number of couplets which determines the constitution of the zygote is increased .
( iii .
) The correlations between pairs of blood relations take definite numerical values absolutely independent of the number of couplets , and the same for all characters and races .
The ancestral correlations form a geometrical series of common ratio one-half Fraternal correlation is fixed between narrow limits depending on the number of brothers per family dealt with , and is very slightly larger than parental correlation .
The theory of the midparent for a considerable number of couplets approaches closely that originally given by GALTON , except for extreme values of the character , when the regression becomes rapidly smaller and ultimately vanishes .
We thus see that a generalised theory of the pure gamete would be of very great advantage if it could be accepted .
It would lead to a system of inheritance in randormly mating populations with non-differential fertility , which in its broad features would be essentially the same as that which has been biometrically developed not from theoretical hypotheses , but from the statistical description of observed facts in populations .
Unfortunately , however , when we come to the actual numerical values for the coefficients of heredity deducible from such a theory of the pure gamete , they do not accord with observation .
They diverge in two ways : First , they give a rigid value for these coefficients for all races and characters\mdash ; a result not in easonable accordance with observation .
Secondly , they give values distinctly too small , as compared with the average values , or with the modal values of large series of population observations .
We thus reach the point we have so often had to insist upon : that the biometric or statistical theory of heredity does not involve a denial of any physiological theory of heredity , but it serves in itself to confirm or refute such a theory .
Mendelian formulae analytically developed for randomly mating populations are either consistent or not with the biometric observations on such populations . .
If they are consistent , it shows their possibility , but does not prove their necessity .
If they are not , it shows they are inadequate .
The present investigation shows that in the theory of the pure gamete there is nothing in essential opposition to the broad features .
of linear regression , skew distribution , the geometric law of ancestral correlation , etc. , of the biometric description of inheritance in populations .
But it does show that the generalised theory here dealt with is not elastic enough to account for the numerical values of the constants of heredity hitherto observed .
It will be time enough to consider other more or less general Mendelian formulae when there is far better evidence than exists at present that they cover a real range of observation , and have not been solely invented to describe isolated experiences , the numerical results of which are not in complete accordance with simple Mendelianism .
Given such neo-Mendelian formulae , there is a perfectly straightforward mathematical method of applying them to randomly mating populations , but that method is excessively laborious , and the biometrician may well hesitate to undertake the task of their investigation .
A few minutes suffice to invent a Mendelian formula , but weeks of labour may be involved in testing whether it leads to legitimate results when applied to sexually crossing races .
Let us therefore have a few simple general principles stated which embrace the facts deducible from the hybridisation experiments of the lVIendelians ; these can form the basis of a new mathematical investigation , but it is idle to undertake such an investigation so long as Mendelian Principles remain in a state of flux .
Any combination of the theory of pure gametes here discussed with homogamy , or with fertility correlated with homogamy , or again with prepotency of individual or of type , would emphasise the correlations which we have found above to be too low ; but such hypotheses would involye a fundamental ] teration in the formula Such a formula would then give the possibilities of the cross , but the proportions of these possibilities actually occurring would be quite different .
* Such loading of the possibilities\mdash ; not only of the individual couplet\mdash ; but very probably of associated couplets in the constitution\mdash ; might conceivably enable us to deduce better values for the ancestral and collateral correlations .
But it would abolish not only the simplicity of the fundamental Mendelian formula , it would also involve lengthy preliminary studies on homogamy , fertility , and prepotency before any effective formula could be propounded .
Toss two pennies , and the result of tossings will closely approximate to the distribution .
Load one or both coins , and the possible variations will still be HH , HT or , but their proportions will be far from

