]\gt ; tOIl [ ossr ?
OIlt I dopr puB ssod cJtai l ?
\mdash ; \mdash ; l UB OS auo SC IOH otuI ) ) 30 OSSR un rm eIIt .uu I ossc aq$ ?
se qmoo pue uIOD JJnooo 0.x amos .
El soddnS T puv $.InxANtIn 9 rlqlm ) ) poo6 se Normal of Frequency for Two Variables , etc. 325 Suppose that the sign of association in the elementary tetrads is positive , so that\mdash ; , ( 1 ) and similarly , .
Then , multiplying up and cancelling , we have .
( 3 ) That is to say , the association is still positive though the two -arrays are not adjacent .
3 .
An isotropic table remains isotropic in whatever way it may be condensed by grouping together adjacent rows or columns .
Thus , from ( 1 ) and ( 3 ) we have : adding , .
( 4 ) That is to say , the of the elementary association is unaffected by throwing the and into one .
4 .
As the extreme case of the preceding theolem , we may suppose both rows and columns grouped and regrouped until only a 2 -fold table is lefb ; we then have the theorem:\mdash ; If an isotropic distribution be reduced to a fourfold distribution in any way whatever , of the association in such fourfold table is the same as in the elementary tetrads of the original table .
5 .
Isotropy , therefore , is a quality that cannot be destroyed by any mode of grouping , or of extraction , of arrays .
If the smallness of the ntunber of observations in any array ( in a practical case ) render the discussion of approximate isotropy difficult , owing to the influence of " " errors of sampling we may either drop that array and treat those on either side of it adjacent , or we may group it with one or more adjacent arrays .
The latter process may of course conceal , but it cannot create , a departure from isotropy .
As a matter of practice no -table with ordinary fineness of grouping could be expected to exhibit st1ict isotropy ; th elementary associations are too small and the probable errors too large .
Some grouping is essential\mdash ; as an extreme grouping to -fold form .
In such a reduced table there will only four ssociations to inspect in order to determine the , viz. , those corresponding to iu the 1st , 2nd , ) and 4th ants of the table , taking the coDlpartments as numbered in the oldcr Mr. G. Udny Yule .
of [ Nov. 4 , Both and extraction of arl.ays have been freely used in Part II of this paper for the discussion of approximate isotropy .
6 .
The normal frequency-distribution for two variables is an isotropic distribution , to which the preceding theorems accordingly apply .
Let the frequency for an inberval of round deviations of ( he variables given by , where eing the stanrlard-deviations of arrays and the coefficient of corl e lation .
Writing down from ( o ) the frequencies of pairs of deviatio1Is , / , we ve for the cross-ratio of the frequencies , etc. , .
that is of the same ) as , the index on the right is of the same sign as .
The of the association in the elementary tetrad is therefore of the same as .
The distribution is therefore isotropic .
grouping of a normal ibution is therefore isotropic\mdash ; be the -intervals equal or unequal , or small\mdash ; and the sign of the association in a normal distribution reduced to -fold form is the same , whatever the axes of division .
* 7 .
In Professor Pearson 's earlier lnethod of calculating the of correlation unmeasured qualities , it is assumed that the contingencytable represents a by arbitrary intervals , of nornlally distributed frequency .
But the sumption of normality can only be valid for some one order of the classes\mdash ; at most\mdash ; and the question whether the right order has been assigned ; Professor Pearson 's later method of the " " cient \ldquo ; avoids all difficulty to class-order , and this is rightly claimed as one of its If , however , the contin -tables dealt with were really ements of llolmally distributed frequency , no difficulty as to class-orders should arise .
If the rows and colu1nns of an isotropic distribution be disin way whatevcl , they can be , in their order , by almost , the following perty : If , say , two columns in an -table be isolated , the Ill a previous memoir ( vide ) ibli r , 336 , 1 1900 ) I showed that this was .
the case for a ) normal distribution , did not recognise it as necessary consequence of Libl .
, ' ) ; the method enll ) loyed in numerous memoi1B of later date ( 1 , 2 , 3 , 4 , .
1905 .
] ofFrequency for riables , etc. ratios of the frequency in the one column to the sum of the frequencies in the two columns , for successive rows , must form a continuous series , ascending or descending .
8 .
Thus , suppose we isolate the and -arrays with frequencies , , , we have from the first tetrad , at the head of the column , supposing the fundamental association positive , .
From the second tetrad and so on , each atio bless ) than the one above it .
If the rows have been , their right order will be given once by the order of the ratios ( subject , as always in practical cases , to difficulties created by ularities of the nature of el.rors of or .
It is assumed , of course , that the orders 1 , 2 , 3 .
3 , 2 , 1 are .
it is , , mere convention whether we to a series of eye-colours in .
from light to dark or from dark to 9 .
possibility of assigning the right order of arl.ays from the intrinsic charactel of the distribution , without reference to extraneous considerations , seemed so extremely curious that I judged it worth while to work out an illustration in a case where the order was priori , so as to provide a check .
The figures used are given in Table I .
The two columns with the greatest total frequencies are 4 and 6 , and these are best chosen for a first comparison as least affected by " " errors of sampling Working out the successive ratios , etc. , we have the series of tios ( below .
This gives as order for the , 3 , 5 , 2 , 4 , 1 , 6 , 7 , 8 , the ecl order se of 9 and ] uncertain .
Checking this esult by a similar use of columns 6 and we get the series ( 2 ) , which gives the salYle order eversed in .
Either and , or , and must therefore have been disarranged .
Assuming that are and that the fun damental association is positive , we may take the ordel as 8 , 7 , 6 , 1 , 4 , 2 , , Mr. G. Udny Yule .
Groupings of Normal [ Nov. 4 , As regards the last pair , a comparison of columns 4 and 8 or 4 and 9 sug , ests the order 9 , 10 whilst a comparison of 9 with 8 ( this pair being disarranged ) suggests 10 , 9 .
The preference lies slightly with the former order .
Table I.\mdash ; Anisotropic Contingency-Table for Two Qualities , A and Table II giving the Isotropic Rearrangement ) .
Quality A. Ratio 10 .
Next proceeding to the quality A and with rows precisely as we have done with columns , we find the following ratios for the pairs of rows 1 , , 4:\mdash ; 1905 .
] Distribution of Frequency for Two Variables , etc. Batio ( Here the position of 5 is indeterminate , but that and are disarranged , both series concur in giving the order for the remainder : 3 , 2 , 1 , 6 , 4 , 9 , 8 , 7 .
Our general knowledge of the attributes would certainly be called into play in any practically treated case in the position to ; the condition of isotropy alone can only offer a slight suggestion that it should lie near the upper end of the series , seeing that the combination is the only one occurring and that lies near the upper end .
The order may then be approximately 5 ?
, 3 , 2 , 1 , 6 , 4 , 9 , 8 , 7 .
11 .
If rows and columns be now in the orders determined we the comparatively orderly-looking Table II , which is , as a matter of fact , Table II.\mdash ; Showing Table I rearranged in ( approximately ) Isotropic Order .
The numbers at heads of columns and rows refer to Table I. Mr. G. Udny Yule .
Groupings of [ Nov. 4 , on ly a by 2-inch intervals of a table of Professor Pearson 's for inheritance of stature from father to son ( 8 , 1903 , Table XXII ) .
The columns should be headed " " Stature of Father " " ) \ldquo ; " " \ldquo ; etc. at left , the rows " " Stature of Son \ldquo ; " " \ldquo ; etc. ( commencing at the top ) .
Table I was a disarrangement of Table II , deliberately effected for purposes of experiment .
12 .
The success of the experiment ests that while errols will very likely be made as ards the of infrequent classes like , the method should be able to positions to the more frequent classes with a considerable degree of certainty , provided always that the assumption of isotropy is itimate .
For tables based on anthropometric measurements I think this assumption will in eneral hold .
I tested 14 of 1rofessor Pearson 's tables for stature , span , fole-al.m , and head measurements by the process of rouping to -fold form ( S 5 ) , in some cases two different .
The result was the same in every case , the grouped tables isotropic without exception .
The class-limits were so chosen as not to make any frequencies very small , but otherwise as much variety as possible was admitted .
* It is that more extensive trials on similar material will yield some anisotropic tables , but seems unlikely that nificant anisotropy will prove anything but exceptional .
II .
of th to the of The first -tables to which I endeavoured to apply the of the sections were the tables for inheritance of , based on Mr. Galton 's data , which are iven in Professor Pearsun 's Memoir ( 6 ) , 1900 .
In these tables the colours classified under blue , dnr blue ; ( 3 ) blue ; ( 4 ) dark 5 hazel ; ( 5 ) own ; ( 6 ) brown ; ( 7 ) dark brown : ' ( 8 ) very dark brown , black .
The tints , 1 , 5 , and are relatively infrequent , at least in the tables for direct inheritance ; as a first rough trial I dropped these tints endeavoured by tlIe methods of SS to 11 to deternino the true order itr of the tints 2 , 3 , 4 , 6 , !
* The the tables cited from Pearson ( 8 ) , with the used us given ) the limits to the central class ; two are given , two distinct were teste ( 8):\mdash ; XXII , stature , and ; XIII , XXIV , -arm , 18\mdash ; 19 ; , stature , CI\mdash ; C3 nd 62\mdash ; 64 ; XXII , } ) 62\mdash ; 64 ; XXXJII fore-arnl , illdex , and , ditto , ) .
: ) , ditto , : , head-b ] ) ) ditto , -height , ) , ditto , 1905 .
] Distribution of Frequency for Two , etc. only the four tables I to for inheritance in the first degree .
To afford a check on the results , I worked out the ratios not for one or two pairs of columns only , but for every adjacent pair 2 : 3 , 3 : 4 , 4 : 6 , 6 : 7 .
} done this I did not think it necessary to evaluate the similar ratios for pairs of rows .
14 .
The results for Professor Pearson 's Table I ( inheritance from fathel to son ) were most chaotic , but I persevered with the others and give the results , as ards the order assigned to the colours , in Table III .
It will be seen that the orders deduced from the different comparisons in one and the same table were utterly discordant .
Whatever the mathematical form of the frequency-distribution of these tables , it was clear that it could not be normal nor even a disarrangement of normally distributed frequency .
A little further inspection suggested , moreover , a certain peculiar order in the chaos : the results obtained from comparisons of the same pair of columns in different tables were much more alike than the results obtained by comparisons of different pairs of columns on the same table .
Further , in 12 cases out of 16 the first place assigned to the eye-colour of the whose frequencies were treated as numerators , in 13 cases out of 16 the last place fell to the : of the denominator-column .
Table III.\mdash ; Result of Testing for the Order of Eye-colours , 2 , , 4 , 6 and 7 , from Tables I , II , III and of Pearson ( 6 ) , 1900 .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Table .
Ratio of frequencies in columns .
Order igned .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
Table .
Ratio of frequencies Order igned .
in columns .
I. Father.son daughter.erpo .
II .
Father-daughter I. Fathel..son . .
daughter .
II .
Father-daughter I. Fathel..son . .
daughter .
II .
Father-daughter I. Fathel..son . .
daughter .
II .
Father-daughter I. Fathel..son . .
daughter .
II .
Father-daughter I. Fathel..son . .
daughter .
II .
Father-daughter I. Fathel..son . .
daughter .
II .
Father-daughter Mr. G. Udny Yule .
Groupings of [ Nov. 4 , 15 .
When the four tables were pooled together the empirical rule thus ested held good without exception .
When the ratios of column 2 to 23 were taken , was placed first and 3 last .
When the ratios of column 3 to 34 were taken , 3 was placed first and 4 last and so on ( cf. the last division of Table III ) .
There seemed to be only one simple explanation of these results , vi a relative excess of frequency in the diagonal compartments 11 , 22 , etc. , corresponding to the same eyecolour in parent and offspring , as compared with the frequencies in a normal or other isotropic distribution .
16 .
This is in striking contrast to the result obtained for stature and other anthropometric measurements .
But , as the reader will probably remember , Mr. Galton affirms that , judging from his material , * " " stature and eye-colour are not only different as qualities , but they are more contrasted in hereditary behaviour than perhaps any other common qualities .
Parents of different statures usually transmit a blended to their children , but parents of different -colours usually transmit an alternative . .
if one parent a iight eye-colour and the other a dark eye-colour , some of the children will , as a rule , be and the rest dark ; they will seldom be medium eye-coloured , like the childlen of medium eye-coloured parents Professor Pearson , working on Mr. Galton 's data , concurred , with some lifications , in this conclusion ( ( 6 ) , 1900 , p. 117 and p. 120 ) .
Now if the simplest possible form of such alternative inheritance hold , , that the offspring resemble identically ( in about equal numbers ) either the one parent or the other , the ibution in the contingency-table for parent and must be giveu by a very simple rule .
One-half of the offspring of parents of any one type of character must be of the same type ; the other half\mdash ; if we neglect the small of actually existing\mdash ; must be distributed similarly to parents of the other sex .
A like rule , mutcmdis , would hold for brother-brother or sister-sister tables .
17 .
Here then we seem , at first , to have the clue to the over- weighting of the frequencies ; it may be due simply to the exclusive or alternative character of the tried tively , however , the theory breaks down .
It gives far too great an .
Thus take the oures of Table ( condensed from Table of Pearson 's Memoir , with the addition of figu1es for mothers his Table III ) .
There were thers of sons with eye-colours 1 or 2 , and 269 mothers .
Assuming random and .
1905 .
] stribution of equency for Two Variables , etc. alternative inheritance , the.number of sons with eye-colours 1 or 2 to be expected is therefore , whilst only 194 observed .
Table \mdash ; Eye-colour of Father .
Table Frequencies Calculated on the Assumption of iSimple Alternative Inheritance .
The number to be expected with eye-colour 3 from the same fathers is ainst 8 actually observed .
in this way we obtain the figures of Table , from which the observed frequencies diverge largely .
It will be seen that the calculated frequencies in the diagonal compartments , corresponding to identity of eye-colour in father and son , are in every case much too great , whilst the calculated frequencies of the diagonal-borders are too small .
The two or three other cases which I tried ( though less completely ) at random , gave similar results , and it would seem therefore that the theory of simple alternative inheritance or " " exclusive inheritance without reversion\ldquo ; must be rejected or modified .
It ives , as pointed out by Professor Pearson , a value of the correlation-coefficient approximately coincident with that observed , but the distribution of frequency it implies is not in accordance with fact .
If the inheritance of eye-colour be exclusive or alternative in some way , if it involve for instance Mr. G. Udny Yule .
Groupings of [ Nov. 4 , a Mendelian separation of characters in the formation of the germ cells , the results of such process are not visible in their simplest form , but are somchow masked .
18 .
The anisotropy of eye-colour tables is , however , raordinarily well marked .
I have examined the 24 tables of Professor Pearson 's Memoir by the short -fold method ( S5 ) , taking the colours in the Groups , and the three eye-colour tables of his Huxley Lecture which are of -fold form as published .
Of all the 27 tables not a single one is isotropic ; in 17 of the 27 the association is positive in the first and third quadrants , negative in the second and fourth .
This , and not isotropy , appears to be the characteristic form ; it is the distribution of signs that would be given by an excess of frequency in the diagonal compartments , though ences from normality may , and indeed do , occur .
19 .
Further , this is not the elistic distribution of eye-colour alone .
An extensive series of tables for inheritance of coat-colour in horses has been cbtTiven by Professor Pearson and his collaborators in the ' Philosophical Transactions ' and in ' Biometrika ' .
These I reduced to form by extlacting the for " " brown " " bay and chestnnt three headings ( out of 16 ) under which 90 per cent. or mole of all the entries occur , antl examined for isotropy .
Of the 20 tables seven onJy are isotropic , whilst nine exhibit the alternate distribution of .
Of the seven tables for relationships of the first degree , only one is iso01opic whilst ve are of the alternate 20 .
Professor 's Huxley Lecture , a number of taDles were iven for the reseml ) lance of qualities brothers and sisters , 21 of which , for 7 characters , were similarly examined .
Alternacion of signs was the form ibited by the three tables ooiven for each of the characters , curliness of hair , athletics and telnper , and by two tables of the three for health .
* Of the distributions ether , only 6 were isotropic , 13 altelnation of .
Taking the eye-colour tables , -colour tables , and the misc.cllaneous tables of the ' Huxley Lecture 68 distributions wel'eexamilled , of only were isotropic , for the used , whilsf , 39 exhibited alternation of .
These tables include certainly the leat ltajority of the tables for inheritance of unmeasured orouping the tables in the tion was higher than threefold , the following were as the central ] asses : , normally healthy ; Hair colour , brown : , slow , ligent ; , moderate .
Professor Pearson regards the tl , table as unreliable for special reason , but it appears to me to be only a larked c of the common charactelistic .
1905 .
] Distribution of Frequency for Two Va } , etc. characters , with a three-fold or higher classification , and based on a considerable number of observations , that have been recently published .
21 .
With the exception of the tables for handwriting , the distributions for all the characters exhibit a common peculiarity , viz. , tendency to departure from isotropy due apparently to excess of frequency for identically-named characters in the two individuals .
Tables for measured characters , on the other hand , exhibit no peculiarity so as I ated them by the same rough vest .
Now this result seems curious .
If we had to deal with two distinct lnodes of inheritance , we would expect a contrast between some characters and others in the case of measuled variables , and between } characters others in the case of unmeasured ities .
The contrast actually appears to lie betweell measured characters and un measured as a whole .
The result is such as to that the xcess of homony1nous pairs , observed only in the latter case , 1nay be of subjecve , arising in some way from the nlode in which diHerent observers assign to pairs of qualities that are more or less cruely defined .
22 .
As the truth of this hvpothesis would , ravely affect the inter tation of many recent statistics , I decided to put it to the test of experiment , and hope to be able to publish shortly a full account of the results .
Here it may be briefly stated that the effect of bias and personal equation in the expel.imental case was of the same kind as that noted nbove , tables of observers ' returns marked excess of homonymons pairs with , in some cases , an excess of contrasts ) .
A live ] ) isoIJ , however , between the ence of observers ' returns from ClltCh and the divergence of some of the -tables from normality showed that the latter was eonsiderably the reater .
) of this divergence demands , , further elucidation .
At present it certainly cannot , owing to its nitude , be ascribed with confidence entirely to a ; ubjective origin , whilst on the other , the absence of any coniil tion statistics of measurements , and the qualitative similarity to )tive effects , compel some eseryc in a biological nificance .
BIBLIOGRAPHY .
1 .
N. .
" " On the Inheritance of Colour in ] Thoroughbred Holses , ' ' Biometrika , ' vol. 1 , p. 361 ( 1902 ) .
2 .
W. R. Macdonell .
" " On Authropometry and the Identi{ication of ' Bionletrika , ' vol. 1 , p. 177 ( 1902 ) .
3 .
\mdash ; " " On the Influence of Vaccinatio in of ' ' ' vol. 1 , p. ( lJ02 ) .
336 Normal Distribution of Frequency for Two Variables , etc. 4 .
W. B. Macdonell .
" " A Further Study of Statistics relating to Vaccination and Small-pox , \ldquo ; ' Biometrika , ' vol. 2 , p. 13 ( 1903 ) .
5 .
Karl Pearson .
" " On the Correlation of Characters not Quantitatively Measurable ' Phil. Trans A , vol. 195 , p. 1 ( 1900 ) .
6 .
\mdash ; with Alioe Lee .
" " On the Inheritanoe of Characters not capable of exact Quantitative Measnrement ' Phil. Trans A , vol. 195 , p. 79 ( IJ00 ) .
7 .
\mdash ; " " The Law of Ancestral Heredity\ldquo ; ( wiih Appendices on Inheritance in Racehorses by N. Blanchard and A. Lee ) , ' Biometrika , ' vol. 2 , p. 211 ( 1903 ) .
8 .
\mdash ; with Alice Lee .
" " On the Laws of Inheritance in Man ' Biometrika , ' vol. 2 , p. 357 ( 1903 ) .
9 .
\mdash ; " " On the Inheritance of the Mental and Moral Characters in Mau and its Comparison with the Inheritance of the Physical Characters\ldquo ; ( Huxley Lecture ) , ' .
Inst. Joul.n .
, ' vol. 33 ( 1903 ) , and ' Biometrika , ' vol. 3 , p. 131 ( 1904 ) .
10 .
\mdash ; with A. Barrington and A. Lee .
" " On the Inheritance of Coat Colour in the Greyhound ' Biometrika , ' vol. 3 , p. ) .
11 .
\mdash ; " " the Tbeory of Contingency and its Relation to Association and to Normal Correlation ( ' Drapers ' Company 's Research Memoirs , ' Biometric Series , vol. 1 , Dulau , 1904 ) .
12 .
W. F. Sheppard .
" " On the Application of the Theory of Error to Cases of Normnl Distribution tion , \ldquo ; ' Phil. Trans A , vol. 192 , p. 101 ( 1898 ) .
13 .
\mdash ; " " On the Calculation of the Double Integral expressing Normal Correlation Phil. Tl ans vo ] .
19 , p. 1 ( 1900 ) .
14 .
\mdash ; " " New Tables of the Probability Integral , ' ' Biometrika , ' vol. 2 , p. 174 ( 1903 ) .
15 . .
U. Yule .
" " On the A ssociation of Attributes in Statistics , etc ' Phil. Trans A , vol. 194 , p. 25 ( 1900 ) .
16 .
\mdash ; " " Notes on the Theory of Association of Attributes in Statistics ' Biometrika , ' 2 , p. 121 ( 1903 ) .

