]\gt ; .
K. Pearson .
Mathematical Contributions [ Jan. 20 , " " Mathematical Contributions to the Theor } of in but Differentiated Organs KARL , University , London .
Received January February 19 , 1903 .
( 1 .
) In the on " " Homotyposis in the Kingdom , \ldquo ; defined homotypes : ' undiflerentiated like In the course of that , I endeavoured to indicate that I was not unconscious of the inllueltce , local environment , and position upon organism in homotypic correlation .
The obiect of my memoir , how , was to some general appreciation of the average intensity of individuality in living forms , and to see if it approached the average value of heredity in or animal life .
For this purpose I selected such material as was readily available , indicating the series where I thought differentiation of a sensible amount was present owing to the ge , the tion , or the environment factors .
From the standpoint of theory , however , we are not compelled to fldopt a mere indication of this kind .
As soon as we can correlate between : the quantitative character of the homologous organs , ( b ) on the orgaI ) and this same character , or local e the chal.acter , we can allow for the differentiation of homologous parts , or reduoe them to pure homotypes .
other ords , homotyposis be deduced from differentiated homologous arts , if we corrcct for the diflerentiation due to ) , ( b ) or .
The test for the existence of such differentiation simply the ) resence or bsence of the corresponding correlation .
ccordingly the following problems to find soIutions To find the correction to be made the apparent homotypic correlation , the .
of homologous are diffel'entiated from each other by their periods of ( ii .
) To the correction to be made to the pparent homotypic correlation , when e pair of homotypes is differentiated by a common period of growth from other pairs of homotypes .
( iii .
) To find the correction to be made to the pparent tion when the tirs of homologous parts arc differentiated from each other by tion on the organism .
( iv .
) To find correction to be made to the apparent homotypic tion pair of homotypes is differe1ttiated by the environment of its from other of homotypes .
It will be seen that in prol ) lems ( ii ) and iv ) are dealing with true homotypes , ) the homotypic factor requires modifying for the influcnce of or cnvironment on the organism .
In ( i ) and ( iii ) * .
Trans , pp. 1903 .
] to the Evolution .
we are not dealing with homotypes at all , but with homologous parts , and we wish to reduce them to homotypes by correcting for differences between them due to owth or to situation on the organism .
I propose at present to deal only with problems ( i ) to ( iii ) , not because ( iv ) does not admit of theoreticaI treatment , but because we have not thus far obtain ed data to illustrate satisfactorily the correIation between character and the immediate environment of the individual organism .
Experimental determinations of homotyposis in plants , when the individuals are subjected to a graduated environmental scale , e.g. , in depth of soil or quantity of moisture allowed would be fairly easy to carry out , and most interesting in resuIt .
I hope it may be possible to arrange experiments of this kind for the coming summer .
We can then illustrate the fourth proposition from actual observation , and the publication of its theoretical solution will be of greater value .
( 2 .
) To find the correction to be made to the apparent tian when the pair of homologous parts differentiated from each other by their periods of growth .
Let and denote the characters in the two homologous parts quantitatively determined , their respective periods of growth .
Then we have four variable quantities / ?
, , no one of which fixes absolutely any other , for individuals will have different characters even with the same period of growth .
The accordingly reduces to this : What is the correlation etweeIl x and for constant valtles of the variables , i.e. , selected values of , and This problem is answered in formulae ( lviii ) , ( lix ) , and ( 1x ) of my memoir : " " On the Influence of Natural Selection on the and Correlation of Organs.\ldquo ; * Let write in those formulae for the subscript 1 , for 2 , for 3 , and for 4 ; we have at once ( i ) , ( ii ) , ( iii ) .
Now if we deal with direct and not cross-homotyposis , i.e. , the of the same character in two homologous parts , can put these results more simply .
We in this case render our correIation tables symmetrical by entering each one of a of first as an and then as a .
We may then write * ' Phil. Trans , vol. 200 , VOL. LXXI .
Prof. K. Pearson .
Ilfathematicat [ Jan. 20 , and we find ( iv ) .
This is the full solution of the first problem .
We see that in order to solve it , it is necessary : To find the correIation of the homologous pairs as if they were simple homotypes .
To find the correlation between the growth periods of each pair of homotypes .
( iii .
) To find the correlation between the character and the period of growth .
( iv .
) To find the correlation between the character of one homotype and the period of growth of its fellow .
Now these correIations can be found at once by the usual statistical processes , if the data are forthcoming .
( 3 .
) I propose to illustrate this on material , which , although not homotypic , is so analogous that it brings out all the important features .
We will determins the correlation between the head-length of rothers , such length being measured on school boys of all ages , from 4 to 19 .
It will be clear that we have here all the difficulties of the homotypic problem\mdash ; resemblance due to common origin obscured by differences in the period of growth of each individual .
Table I gives the correlation of pairs of brothers without regard to their differences of age .
Table II gives the oorrelation between age and length of head in the same individual .
Tables IIIA and IIIB gives the correIation between the age of one rother , and the length of head of the second .
Table gives the correlation between the ages of pairs of brothers .
These tables have been prepared by taking off from the brotherbrother data papers of my school measurement records all the available pairs of cases falling into each series .
Thus in some cases the ages of -both brothers were given , but not the head measurement of one or other ; in other cases the head measurements of both , but the age of one or other would fail , or again the age of one and the head measurement of the other might be all the information available .
Thus the total number of cases and the frequency distribution varies slightly from one table to a seoond .
measurements form part of the material obtained with the assistance of a grant from the Royal Society Government Grant Committee .
1903 .
] to the Theory of Evolution .
A few remarks must be made on these tables .
Table I gives following values of the constants:\mdash ; Iean length of head of elder brother .
7508 in mm. younger , , .
8296 , Standard deviation of elder brother 7 6027 , , younger , , The correlation is , then , found to be and the regression , younger on elder brother , .
These give the intensity of heredity , uncorrected , for the growth factor .
Now , the most noteworthy part of this result is , as we shall see later , that taking different teruds to exaggerate the apparent intensity of her .
If we were to take pairs of boys at ages from 4 to 19 , each pair having no hereditary relationship , but being , on the average , within a year or so of the same age , we should find a spurious correlation due to the mixture of material , each pair having approximately-like head-lengths because the members of it were , approximately , of like age .
On the other hand , if the boys were blood relations of very different ages , their apparent relationship would be weakened , because we should be correlating the same organ at different stages of its .
growth .
We have thus two factors : one tending to exaggerate , and the other to weaken the apparent strength of hereditary resemblanceIt is of great interest to note that the former factor in the present case is the more effective .
In Table II we have what I term a growth table , i.e. , a correlation table between period of growth and the quantitative measure of a character .
The constants of this table are as follows:\mdash ; Mean age of boy 13 0394 years .
Standard deviation of age 2 8207 Mean head-length .
4516 mm. Standard deviation of head-length 7 4991 Correlation of age and head-length .
453,496 The regression coefficient for head-length on age , and we have the probable head-length for observed age A given by Thus , on the average , boys ' heads grow in length mm. a year .
My results are based on 1637 cases entirely taken off the brotherbrother data papers .
Dr. Alice Lee at an earlier stage also worked out a growth table .
We had not then so many brother-brother data papers filled in .
She used in addition all the brother measurements on the brother-sister papers , and so reached 1856 boys , of which , I * Six figures have been kepl in the correlation coefficients , as we require to calculate the regression coefficients from the differences of products and powers .
292 Prof. K. Pearson .
Contributions [ Jan. 20 , think , we may safely assert that 400 at least are not included in my series .
She found : Mean age , ; mean head-length , ; and slope of regression line , , giving the formula .
2040 a result in substantial agreement with mine .
DIAGRAM 1 .
In Diagram 1 the formula is represented with the observed mean values at each year of life .
The results for the 4th , 5th , and 19th years of life ought not to be considered , for they are based on only 2 , 10 , and 12 observations respectively .
It will hardly be possible to express the growth curve better than by a straight The mean age is less , brother-sisters are obtained chiefly from primary , not secondary , schools .
1903 .
] to tlu Theory of Evolution .
line , until the range of data is very largely extended .
The regression is sensibly linear .
TabIe IIIA and Table IIIB give the following resuIts:\mdash ; Iean age of elder brother . .
S.D. of elder brother 's age . .
Mean head-length , younger brother. . . . . . . . . . . . . .
S.D. of head-length , you1lger brother . . . . . . . . . . . . . .
Correlationofage of elder and head-length of younger . .
Mean age of younger brother S. D. of younger broffier 's age Mean head-length , elder brother . . . . . . . . . . . . . . .
S.D. of head-length , elder brother. . . . . . . . . . . . . . . .
7 5005 Correlation of age of younger and head-length of elder .
see accordingly that within the limits of the probable error , the correlation between younger brother 's head-length and elder brother 's ags is the same as that between elder brother 's head-length and ounger brother 's age .
This resuIt might , to some extent , have been anticipated , } ) actual proof of this type of cross-relation is of value .
In Table we have the correlation between ages of brothers giving the constants:\mdash ; lIean age of elder brother .
1508 Mean age of younger brother .
7487 S.D. of elder brother 's age 2 5080 S.D. of younger brother 's age 2 7220 Correlation of brothers ' ages .884,186 The first four results are in good agreement with those of Tables IIIA and IIIB .
The last result shows how nearly there is an approximation to a constant difference in age between brothers in schools .
Very closely we havs\mdash ; Probable age of younger brother ( age of elder brother ) When the elder brother is 6 , his younger brother is probably years younger than he is ; when the elder brother is 12 , the younger brother is probably years youn , and when he is 18 , years mger .
The explanation of this is that when the elder other is very young only his near or second brother will , as a rule , be at the same but in the secondary schools , which he reaches at a much later age , it is possible for a much younger brother to be at the same school .
Now let us substitute the correlation values , found in equations to ( iii ) , of page 290 .
We have , .
379,326 , .
396,598 .
Prof. K. Pearson .
Jlathematical ontributions [ Jan , 20 , Whence we find .
891,209 , and .
5446 .
This is a very reasonable value of fraternal correlation , agreeing quite well with results obtained for horse , man and dog .
It is worth noting that , and , therefore , either equals or fairly closely ; in fact , within the probable error of their difference .
Hence , it lvould appear highly probable that the cross-relation between one brother 's head length and a second brother 's age is due to the correlation of the ages between the two brothers .
If such a resuIt as should be verified on the reduction of further data , it will enable us to much simplify our formulae .
Thus we easily find for this case and .
Or , we require to find only the uncorrected correlation the growth correlation , and the correlation between periods of growth ( r ) .
The correction to to the apparent correlation is then the subtraction from it of I hope shortly to ascertain whether relations like above hoId also for other head-measurements on growing children .
1103 .
] to the TTeeor of Evolution .
to the of Prof. K. Pearson .
Jfathematical Contributions [ Jan. 20 , 1903 .
] to Theory of Evolution .
5 5 300 Prof. K. Pearson .
Contributions [ Jan. 20 , ( 4 .
) Tofimd correchon to be made to the homotypic correlation , each of is erentiated ba rommon period of growth ofher of The soIution of this problem may be deduced at ones from equation ( iii ) of the preceding problem by simply putting .
In this case , and we find ( v ) .
This equation was given by me in a note in metrika , vol. 1 , p. 404 , and its use illustrated on Dr. Simpson 's data for Para ; mec(5 .
) To Jind the correction to be made to the ppnrent homotypic correlation the pair .
of homologous pnrts are cntiated from each other situafion on the organism .
have only to put in formula iii ) on p. 290 , and the positional -ordinates of the first and second homologous parts , to make that formula available for position instead of age differsntiation .
If we denote by and the characters of the parts in the positions respectively , our solution takes the form below , where we have confined our attention to the same character , - ( vi ) .
This follows since , and We have again , therefore , to find four correlation coefficients .
But this formuln simplifies immensely if we observe the following conditions : Take the same number of homotypes or homologous parts from the same positions in each organism .
Enter each o1te of these homotypes or homologous parts with each other on the same organism , so as to obtain a symmetrical table , i.e. , is first entered with and then with These conditions are or can be usually satisfied in any homotyposis investigation .
( 6 .
) Further , the positions will , as a rule , be arranged in series and may be ered 1 , 2 , 3 , 4 , . .
, if homotypes or homologous be taken from each individual organism .
The position scale is , of course , perfectly arbitrary , and has nothing to do , for exampls , with the lctual distances betwcen on the organism .
We can make it a uniform numerical scale , which for convenience we can take to be the same serial order as that of positions on the organism .
1903 .
] to the of Bvolution , 301 Let mean position , deviation of positions on the arbitrary position scale .
Let there be organisms , and suppose that denotes a summation of all homologous parts on an organism , and a summation for all organisms .
Then , if , hence , since or ( vii ) .
there But Hence ( viii ) , a relation precisely bimilar to that discovered in the case of growth periods for brother 's head-lengths from the actual numbers on p. 294 .
Substituting we find Then substituting in ( vi ) and using ( vii ) we determine the simple formula for homotyposis corrected for positional differentation ( ix ) .
where stands for the correlation of character and position on the organism .
An exactly similar formla might be found for the correction for the age or growth factor , if the homologous parts dealt with had the same distribution of ages or growths in each organism .
( 7 .
) Now the equation just found has the serious disadvantage that it is based on the linearity of the regression reIation between position * The reader should note that this condition does not involve any assumption of normal frequency , or the Gaussian law .
The latter applies only to a very speclal case of linear regression .
302 Prof. K. Pearson .
JIathematieal Contributions [ Jan. 20 , and racter .
But while organic and homotypic correIations give for a surprising variety of cases sensibly linear regression relations , the reIation between position and mean character is far more rarely linear .
obtain , as a rule , remarkably smooth curves .
, therefore , require some modification of equation ( ix ) .
Still the regression of character and position linear , we houId have , if be the mean standard deviation of an array of organs the same position , .
But if be the standard tion of the means of the arrays , we have from first principles ' Hence We can now write equation ) in the form ( x ) .
This is quite free from , and , what is more , although we have deduced it from ( ix ) and the relation peculiar to linear regression , it is now free of any limitation as to the nature of the relation between position and mean character .
Thus ( x ) is a far more important formula than ( ix ) , and should always be used , until we have shown that the relation between position and mean racter is sensibly linear .
If anything , it involves less arithmetic than ( ix ) .
We can show this ab initio as follows : the individuality of the organism in any homologous part be measured by its excess above ( respectively defect below ) the mean value of the character for the homologous part in that position .
Then , if element of character due to individuality , and be the mean character in any position for the individuals deaIt with , , and Hence we easily find , might be considered better , if the standard deviations of the homologous parts very considerably with position , to measure the individuality by the ratio of this excess to the corresponding standard deviation .
Not only , however , does the use of such a ratio immensely increase the labour , which is a , which of course , we could face , but there is also a question as to whether the ratio is really a truer measure of individunlity .
A full discussion of this important point must for the present be deferred .
1903 .
] to the TheoJ'y of Evolution .
Or , noting that , we have where is ths standard deviation of the character-individualities free from the position factor .
We see that it is precisely the same quantity as we have previously used for the mean standard deviation of the arrays for given positions .
Next taking the correlation of characters and in positions and we have To get this resuIt we have multiplied every quantity like by every other quantity like and by itself , and then added such quantities together for every position on the one organism .
Thus on the left hand side there are terms in the first , terms in the second summation ; on the right hand side there are terms in the first , terms in the second and terms in the third summation .
Now sum for each of the organisms , and we have Whence or , as before ( x ) .
Now while this proof is independent of the theory of partial correlation coefficients , only simple algebra , and is further independent of any consideration of linear regression , it yet wants something of the width of the former theory , whioh allows us at once , for example , to correct for a combination of factors , such for example as for both growth and position influences simultaneously .
The difficulty hes entirely in the extent within which it is legitimate assume the relation between position or age , and the mean value of the character at that position or age to be linear .
It is therefore clearly advisable to start by plotting this relationship , and fitting , if possible , such position or growth appropriate curves .
If , for the series of positions deaIt with or the period of growth taken , we find that a straight line is a cIose approximation to the relationship , then * In the case of some animals and many plants the relationship is in itseIf of much interest , for it expresses a law of development or growth in serial parts .
The analytical consideration of this point is very simple .
If the regression Prof. K. Pearson .
Mathematical Contributions [ Jan. 20 , may use the general theory of partial correlation , wise we must fall back on results like ( x ) .
For example , in head growth in boys , we cannot much improve on a straight line ; in positional influence on the branches in the whorls of Equisetum arvense we need at least a third order parabola .
( 8 .
) Although material for several investigations on the homotyposis of serial homologous parts has been collected , the progress in some of these cases is slow , as it involves rather laborious nlicroscopic measurement .
I content myself at present with an illustration from the vegetable kingdom .
I collected in the autumn of last year , 126 plants of Equisetum arvense in Raydale Side , an offshoot of Wensley Dale ; the plant was growing on a lane side high up above Semmerwatsr .
This Equisetum grolvs the top with a single stem , and I counted the number of branches to the lvhorl from the root upwards .
As a rule , there will be one or two whorls close to the soil which have never developed any branches at all ; then we have what I shall term the first whorl in which some branches have developed , but the number is irregular and obviously subject to some cause of variation , other than the growth law of the plant .
The number of branches to the whorl theri increases uniformly and steadily up to the 4th whorl , after which it falls almost equally steadily to the whorl .
Beyond this the results becomes somewhat egular again : for very few plants will be found\mdash ; at any rate in the locality considered\mdash ; with more than 12 or 13 whorls , and even in these whorls there is a certain amount of forking or irregularity which it is difficuIt to deal with .
The plants were certainly fully developed be linear , the means of the arrays all lie on the Qsion line , and the mean standard deviation of the arrays about their means is .
If regression be not linear , the means of the arrays will have a mean sqnare deviation from the regression .
The mean square tion of the from the regression line , but not from their means , is still .
The mean standard deviation ( deviation of mean square from means ) is now given , since But we easily find Hence is a good measurement of the deviation of regression from linearity , or of from .
If we take , we have Clearly must lie between and 1 .
Further , can only vanish when the correlation is zero , or become when the correlation is perfect .
Between these values it gives the mean reduction in variability of an array as with the lvhole population .
Further , the deviation of from is a good measure of the deviation of the system from linearity .
Thus is a useful constant which ought always to be given for non-linear stems .
It measures the approach of the system not only to but to a single valuod relationship , i.e. , to a causal nexus .
1903 .
] to the Theory of Bvolution .
gathered at least as far as the 12th or 13th whorl , and I doubt whether even beyond this so late in the season , any further branching old have taken place .
A few branches were broken off , and these of course counted ; there was no difficulty , however , in easily ascertaining whether a branch had in any case been developed or not , and the peculiarity of the 1st whorl was certainly not due to missing , but to undeveloped branches .
Table V gives the relation between branches to the whorl position for the whole of the 126 plants .
In two columns to the right are eriven the means and variabilities of the brancbes for each whorl .
No , whether we judge by mean or standard deviation , we see a perfectly gradual change from to whorl , which absolutely preeludss us from considering the number of branches to the whorl as a pure homotypic character .
We see a marked differenti , tion due to position of the whorl on the plant ; the whorls are homologous but not homotypic parts .
Suppose , however , that we disregard our test for differentiation , and proceed to find a correlation tabls for the whole material as homotypic .
We have Table , for which I have to heartily thank Dr. Alice Lee .
The value found for the homotypic correlation from this table is .
0185 , or , there is no sensible homotyposis at all .
But we might have gone to the other extreme and taken only the 3rd , 4th , and 5th whorls , which have more nearly the same means and standard deviations as homotypes .
The result is Table VII , giving .
0129 .
It lvill be perfectly clear , therefore , as these two results ought to be the same , if the whorls were true homotypes , that we may get any result at all if we neglect differentiation .
The anslver to this is that lto trained biometrician would call these whorls " " undifferentiated like organs\ldquo ; with the two right hand columns of Table V before him .
* On the test for diiferentiation , see ' Biometrika , ' vol. 1 , p. } Bateson , ' Roy .
Soc. PIOC vol. 69 , p. 200 .
VOL. LXXI .
Prof. K. Pearson Malhernatical 308 Prof. K. Pearson .
Mathematical Contributions [ Jan. 20 , Table VII.\mdash ; Whorls , 3rd , 4th , and 5th only .
Number of Branches to 1st Whorl of Pair .
Now let us consider how to handle the material , allowing for the differentiation of the whorls .
To begin with , our formula the use of the same number of homologous parts for each organism , and it is , on account of the value of the probable error of the random sample , undesirable to use fewer than 100 individuaIs .
This leads to our cutting off Table at the 10th whorl .
In this way we get rid also of the forking , which certainly begins in many individuals at the 11th or 12th whorl .
VIII gives us the data of Table reconstituted for 110 plants , with ten whorls apiece .
The only serious difficulty now remaining is that which I have referred to as arising from heterogenity in the first whorl .
A glance at the mean and standard deviation of the branches in the first whorl given in Table V DIAOBAX 2 .
03 .
] to the Theory of Bvolution .
1903 .
] to the Theory of Evolution .
3,938,354 .
Further , if be the standard deviation of the frequency distribution of branches , as found from the bottom row of Table VIII , we have Henoe for use in formula ( x ) we have , since ( .
) .
Table IX gives the uncorrected homotyposis for the nine whorIs treated as simple homotypes .
From this I find , 131,258 ( xii ) .
Substituting ( xi ) and ( xii ) in ( x ) , we find for the homotyposis of the number of branches in the whorls in Equisetum arvense , when corrected for differentiation due to position , .
4939 .
This result it must be admitted is emely satisfactory , and indicates how it is quite possible to correct a resuIt like by allowing for the differentiation of the homologous parts due to serial position .
I hope before long to other results dealing with homotyposis in serial parts , where the differentiation has every variety of intensity .
I think they will suffice to show that differentiation is not a subtle and evasive quality beyond the appreciation of the naturalist who is provided with the training requisite for modern biometric researoh .
( 9 .
) The values of as given by ( ix ) and ) may be illustrated from the actual numbers for Equisetum arvense .
We have seen in the footnote , p. 304 , that This in our case gives by direct calculation on VIII , using whorls 2 to ] .
Lee finds .
Hence with the notation of the footnote referred to * The value obtained for crude homotyposis he members of the whorJs in Asperula odorata in my first memoir ( ' Phil. Trans , vol. 197 , p. 326 ) .
I have little doubt that when we are able next summor to lats the correction for ferentiation in position of whorl , we shall find for woodruff in good )rdance lvith other homotypic results .
My remarks about it were : " " In counting the members on the whorls I soon foumd evidences of ntiation in position , the whorls towards the top of the spray having , as a rule , fewer members than those lower down " " .
Unfortunately I have not kept my records of position .
Prof. K. 1earson .
Jfathemat Contributions 1903 .
] to the Theory of Thus diverges much from and from zero .
Indeed , a glance at the diagram shows how far we are from true linear regression .
If we use as above instead of , i.e. , ( ix ) instead of ( x ) , we have a very much below the actual value .
This illustration wilI suffice to emphasise the importance of testing the actual curve of regression before we assume it to be linear and use equation ) .
( 10 .
) The subjeot of differentiation due either to position or age is , of course , a difficuIt one , but it does not seem at beyond biometric treatment .
The greatest difficulty which it seems to me will have to be encountered is not that of overing and allowing for differentiation due to serial position , ) in ensuring that this has been allowed for , there is not ining an organic correlation due to the necessily of adjacent parts " " fitting On this account it is most desirable that as large a lumbe of homotypes as possible shall be taken , so that the part of the colrelation due to the homologous parts having to fit , or , indeed , to serve a common end , should be reduced to as quantity as possible .
For example , if ppose adjacent whorls to have their ) of branches influenced ] ) an organic relationship , this result ] ] only bias nine of the fort.-five pairs we should form in dealing with ten whorls .
The question of separating organic from homotypic correlation is one that I hope to return to at a later date .
Meanwhile the present paper will suffice to indicate how partial correlation coefficients enable the biometrician to free himseIf from the diiferentiation between individuals due to different periods of growth , or to differenl positions on the organism .
In conclusion I should like to thank Dr. Alice Lee and Mr. F. E. Lutz for aid readily tnted one or other stage of this investigation .
1903 .
] to the of Ofolutioe .

