]\gt ; Mr. F. Galton .
[ Jan. 21 , the author explain in any way the diflerence between his own result and that of Maxwell .
[ The method followed by Maxwell is liable to be vitiated to a very sensible degree by small errors of level of the movable disks , especially when they are closest to the fixed disks .
The final adjustmenl is tated to have been that of the fixed disks , and no special precantions seem to have been taken to secure the horizontality of the movable disks .
By a calculation fouuded on the equations of motion of a viscous fluid , I find that at the closest distance ( about the onesixth of an inch ) at which the fixed and movable disks were set , an error of leveI of only 1o 8 ' vould suffice to make the internal friction appear 8 per cent. too high .
In Mr. Tomlinson 's reductions no allowance has at present been made for the eitecb of the rotation of the spheres or cylinders about their own axes , which is not quite insensible , as it would bs in the case of a ball pendulum .
The introduction of a correction on this account would slightly diminish the values resulting from the experiments , especially in the case of the sphere , where it would come to about 4 per cent.\mdash ; G. G. S. ] January 21 , 1886 .
Professor STOKHiS , D.C.L. , President , in the Chair .
The presents received were laid on the table , and thanks ordered for them .
The following Papers were read:\mdash ; I. " " Family Likeness in Stature By FRANCIS GALTON , F.R.S. With an Appendix by J. D. HAMILTON DICKSON , Fellow and Tutol of St. Peter 's College , Cambridge .
Received January 1 , 1886 .
I propose to express by ] the relation that subsists between the statures of specified men and those of their kinsmen in any given degree , and to explain the processes through which family peculiarities of stature gradually diminish , until in every remote degree of kinship the group of kinsmen becomss undistinguishable from a group selected out of the general population at random .
I shall determine the constants in my formulae referring to kinship with a useful 1886 .
] Family Likeness in .
43 degree of precision .
These constants may provisionally and with some reservation be held applicable to other human peculiarities than stature , while the formulae themselves are , I presume , applicable to every one-dimensioned faculty that all men possess in some degree , but that drifferent men possess in different degrees .
I selected stature for the subject of this inquiry , for reasons fully set forth in two recent publica , tions , *which dealt with one small portion of the ground covered by the present memoir , and which it will be convenient that I should make as I proceed occasional short extracts , in order to complete the esent argument and to save cross-reference .
The reasons that combine to render stature an excellent subject for hereditary inquiry are , briefly , the ease and frequency of its measurement , its constancy during adult life , its inconsiderable influence on the death-rate , its dependence on a multiplicity of separate elements , and other points that I shall dwell on as I proceed , namely , the ease with which female statures transmuted to their male equivalents , and so enabled to be treated on equal terms with male statures , the tendency of the parental statures to blend in inheritance , and the disregard of stature in marliage selection .
Stature-schemes.\mdash ; It is an axiom of statistics that large samples taken out of the same population at random are statistically similar , and in such inquiries as these which do not aim at minute curacy , they may be considered identical .
Thus the statures in every group , say of 1000 male adults , when distributed in order of their magnitudes at equal distances apart and in a row , will form almost identical figures ; it being only towards either end of the long row that irregu will begin to show themselves .
These are unimportant in the present lnquiry and I disregard them .
The Diagram fig. 1 , shows the outline of such a group of statures .
It is drawn to scale , each of the statures bein supposed to have been represented by a vertical line of proportionate length , standing on a horizontal base , the lines being at equal distances apart , and the whole system being compressed into the space between two termini , which may be set at any convenient distance asunder .
The vertical lines in the figure do not indicate these statures , but they are divisions , ten in number , between each of which 100 stature lines are compressed .
The first and last stature will not touch the termini , but will be removed from them by a half-interval .
As it will be convenient to assign a name to this figure , I will call it a " " Presidential Addreae to the Anthropological Section of the British Association in 1885 ( 2 .
) " " Regrossion towards Mediocrity in Hereditary Stature " " Journ. Anthrop .
Institute 1885 , p. 246 .
The latter is a reprint of that portion of the former with which I am now concerned , together with some additional matter ; it contains tables and diagrams , and should be referred to in preference .
44 Mr. F. " " stature-scheme.\ldquo ; The nnmerous cases near mediocrity that differ little from one another , cause the middle portion of the upper boundary of the stature-scheme to assume a gentle slope , which increases rapidly towards either end , where the increasing rareness of more and more exceptional cases causes that boundary line to slope upwards , as an asymptote to one of the termini , and downwards as an asymptote to the other .
Now suppose that instead of compl.essing 1000 statures between the termini , I compressed , or a million of them , the stature-scheme would be unaltered , except that such small irregularities as might have been previously seen would become smoothed .
The height of the middlemost or median stature-line wonld remain the same as before , and so would the heights of the lines standing at each quarter , each tenth , and at every other proportionate distance between the termini .
Or again , instead of arranging the lines in a single scheme , we might arrange them in a thousand schemes , which as we have seen , would be practicalIy identical in shape , and we may place these schemes side by side , as is done in , fig. 1 , forming a 1886 .
] Family Likeness in Stature .
" " squadron\ldquo ; numbering 1000 statures each way , the whole standing upon a square base .
Our squadron may be considered as made up of ranks parallel to the plane of zx ) as in , or of files ( parallel to the plane of ) as in A. The ranks , as we have seen , are all similar stature-schemes , the files are all rectangles which have the same breadths but are of dissimilar heights .
It is mow easy to give a general idea , to be developed as we proceed , of the way in which any large sample of a population gives to group of k-insmen in any given degree , who are statistically in all respects except numbers ) undistinguishable as regards their statures from themselves .
I must suppose for convenience of explanation , that tall , short , and mediocre men are equally fertile ( which is not , however , strictly the case , the tall being somewhat less fertile than the short* ) , and then on referring to fig. 2 , the FIG. 2 .
fortunes of the distant descendants of two of the rectangular files of squadron A will be seen traced .
As the number of kinsmen , in any remote degree we please to specify , of the men in each of the two files is about the same ; I take 1000 of them in each case .
Again , as the stature-schemes of those kinsmen are identical with those of equal numbers of men taken at random , as samples of the general population , it follows that they *Oddlenough , the shortest couple on my list have the largest family , nnmely sixteen children , of whom fourteen were measured .
Kinsmen in near degrees are represented by squadrons of intermediate form .
These will not have lost the whole of the curvature in rank of , nor will they have acquired the whole of the curvature in file of Z. Consequently they will be curved moderately in both ways .
*Also it will be found that the intersection of their surfaces by the horizontal plane of median height forms in each case approximately straight line that assumes diflerent and increasing inclinations , in the successive squadrons of intermediate shape between A and Z. These lines are indicated by straight lines on the squares below the squadrons in fig. 4 , which represent the square bases upon which the squadrons stand .
I shall now show how these curves in rank and file should be treated .
But before doing so , it is necessary to remark that female adult stature ( I speak throughout of adults ) may be safely transmuted to its male equivalent by multiplying it by a constant constant , which as regards my data is .
After this has been done , the transmuted female statures may be treated on equal terms with the male statures , and the word " " men ' ' or other masculine .
term will include both sexes , unless otherwise stated distinctly .
This procedure is udopted in the present memoir .
It is now genel.ally recognised that the statures in dinary population are distributed in ) proximate conformity with what might have been inferl.ed , if it were knolvn that their variations were governed by such conditions as those upon which the exponential law of frequency of error is based .
Therefore the upper boundary of the stature-scheme is approximately a curve I call it an " " ogive that admits of mathematical expl.ession .
The abscissae of the normal ogive ( fig. 3 ) are values of the probability integral .
and the ordinates are the corresponding values of .
These are given in column A of Table I. Column contains the same values divided by , by which means tlJey are expressed in units of the probable .
I find it enient to call the ordinates to an ogive ( drawn *A plaster model of one of these intermediate forms was exhibited at the meeting by Mr. J. D. H. Dickson , who stated that his recent mathematical investigation of ths properties of their surfaces , had shown that no strictly straight line could be drawn upon them .
\mdash ; F. G. 1886 .
] Family Likenpss in Stature .
from its axis ) by the name of " " deviates and to descl'ibe either of those two rical deviates of the normal ogive that stand at by the name of\ldquo ; quartile deviate or , more briefly , " " quartile I also give this name to the mean length of the upper and lower quartile , in those ogives which are drawn from observed data , and which are not strictly symmetrical .
The numerical value of the quartile is identical with that of the well-known but here inappropriate term of " " probable error Construction of Stature-Schemes and of Ogives Observations.\mdash ; The method of drawing an ogive from observations of stature is as follows .
The observations ( see Tables III , , and V , and compare with VI and ) are sorted into grades , such as " " . . .
cases of 60 inches andunder 61 " " . . .
cases of 61 inches and under \amp ; c. If we are constructing a stature-scheme , or desire to obtain the med-ian value of the series , we have to consider these values of inohes , but in con- structing no more than an ogive , which is only the upper boundary of a stature-scheme , it suffices to consider them as successive of 1 inch each , and I reckon the first rade not as , but as 1 .
This has been done in column , Table , for the sake of treating different groups on a uniform plan .
The numbel of cases in these grades are then summed from the inning , and the sum , up to each grade inclusive , is written down , as shown 48 Mr. F. Galton .
[ Jan. 21 , in column in Table .
The percentage values of these , taking the total number of observations as 100 , are written in column C. A series is there obtained which shows how many per cent. of the statures fall short of the parting value that separates each pair of adjacent grades .
Thus if per cent. of the statures fall within the first , that is to say , are less than the value of the parting line , then per cent. of them will exceed that value .
Consequently , if the observations read off and recorded to the utmost nicety , will be value of the ordinate representing the stature which has to be erected on a base line at per cent. of its length from one of its ends .
In short , a base line of any convenient length has to be divided into 100 parts , and an ordinate of a length proportionate to erected at the division As observations are never read off and recorded with perfect accuracy , a correction has here to be applied according to the circumstances of the particular case , whenever we are drawing a stature-scheme , and not merely an ogive .
If the records are kept to the nearest part of an inch , the phrase " " exceeding inches\ldquo ; would really mean exceeding inches This then is the true parting value corresponding to the nominal .
In drawing ogives , and not statureschemes , this correction may of course be disregarded .
Having erected ordinates corresponding to each value of , their tops are connected by straight lines forming a polygonal boundary that approximates to the curvature of an ogive , and would become one if it were corrected with a free hand , or otherwise smoothed .
The centre of the ogive lies at the intersection of the curve with the ordinate drawn from the base at the fiftieth division , and the horizontal axis of the ogive runs through that point of intersection ( see fig. 3 ) .
A half-ogive , whose dinates are the mean lengths of the symmetrically disposed ordinates of the complete ogive , is constructed on the same general principles , but more simply , because the base from which it is plotted coincides with the axis of the ogive , and the graduations run alike , , from to In Table , the entries in the first lines of each of the three groups it contains , are the lengths of the ordinates that have been measured from the bases of ogives constructed from the data in Table .
The abscissae corresponding to the measured ordinates , are in every case the same fractional lengths of the bases .
The entries in the second lines are the diflerences between these several ordinates and the median ordinate ; they are , therefore , the deviates .
The entries in the lines are the negative deviates written under the ing positive ones .
The entries in the fourth lines are means of the values of the positive and negative deviates , disregarding their signs .
Comparison of Ogives.\mdash ; The ogive being drawn according to the 1886 .
] Family Likeness in Stature .
observations , its axis is divided into 100 parts , the fiftieth division being reckoned as , then the deviates standing at the graduations of , and are measured .
The mean of each pair of lengths , not regarding si , has then to be divided by the mean lengths of the deviates at , that is by the deviate , and so is made to yield a series that is direct ] comparable with column in Table I. The closeness with which it conforms to that standard series is the test of the closeness with which the observations conform to the law of frequency of error .
Table II effects this comparison for all the series that I have to deal with in the present paper .
The values are entirely unsmoothed , except in two named instances , being ta , ken from measurements made to the above-mentioned polygonal boundary .
I thought it best to give these interpolated values in this , their rudest form , leaving it to be understood that with perfectly legitimate correctiou the accordance would become still closer .
I do not carry the comparison beyond partly because my cases are not numerous enough to admit of a fair comparison being made , and chiefly because I am well aware that conformity is not to be expected towards the end of any series .
I am content to deal with nine-tenths of the observations , namely , those veen 0 and , and to pay little heed to the remaining tenth , between and .
It will be seen that the conformity of more than one half of each series is oser than to the first decimal place , and that in absolute measurement it is closer than to one-tentb of an inch .
tlrithmetic and Geometric Means.\mdash ; I use throughout this inquiry the ordinary law of frequency of error , which being based on the assumption of entire ignorance of the conditions of variability , necessarily proceeds on the hypothesis that plus and minus deviations of equal amounts are equally probable .
In the present subject of discussion our ignorance is not so complete ; there is good reason to suppose that plus and minus deviations , of which the probability is equal , are so connected together that the ratio between the lower observed measurement and the truth is equal to that between the truth and the upper observed measurement .
My reasons for this were explained some years ago , and were accompanied by a memoir by .
Donald Macalister , showing how the law of frequency of error would be modified if based on the geometric , instesd of on the arithmetic mean .
* Though in the present instance the former process is undoubtedly the more correct of the , the smallness of the error here introduced by using the well known law is so insignificant that it is not worth regarding .
Thus the mean stature of the population is about 683 inches , and the quartile of the stature-scheme ( the } able error ) is inch , or only about one-fortieth of its amount , and the * ' ' Proc. Roy .
Soc vol. 29 ( 1879 ) , pp. 365 , 367 .
VOL. Mr. F. Galton .
[ Jan. 21 , difference between and 41 is that between about and , or only about 6 per thousand .
Regression.\mdash ; It is a universal rule that the unknown kinsmam in any deglee of any specified man , is probably more mediocre than he .
Let the relationship be what it may , it is safe to wager that the unknown kinsman of a person whose stature is inches , is of some height inches , where is less thau .
The reason of this can be shown to be dtle to the combined effect of two causes : ( 1 ) the statistical constancy during successive generations of the statures of the same population who live under , generally speaking , uniform conditions ; ( 2 ) to the reasonable presumption that a sample of the original population and a sample of their kinsmen in any specified degree are statistically similar in the distribution of their statures .
To fix the ideas , let us take an example , namely , that of the relation between men and their nepbews : A sample of men , and a sa , mple of the nephews of those men , are presnmed to be statistically alike in stature , that is to say , their mean heights and their quartile deviates of height will be of the same value .
I will call the value of this quartile Each family of nephews affords a series of statures that are distributed above and below the common mean of them .
They are deviations from a central family value , or , as we may phrase it , from a nepotal centre , and it will be found as we proceed ( it results from what appears in Tables III , IV , and V ) that deviations are in conformity with the law of error , and that the quartile values ( probable errors ) of these systems of deviations , which we will call , are practically uniform , whatever the value of the central 11epotal family stature may be .
) It will be found , as it is reasonable enough to anticipate , that the system of nepotal centres is distributed above and below the median stature the population , in conformity with the law of frequency of error , and with a quartile value that we will call .
It follows from ( a ) that we possess data for an equation betweell , and , which , from a wellknown property of the law of error , assumes the form Now the unknown nephew is more likely to be of the stature of his nepotal centre ' than any other stature that be named .
But the system of statures of nepotal centres is more concentrated than that of the general population ( is less than ) .
That is to say , the unknown nephew is likely to be more mediocre than the known man of whom he is nephew .
What I shall have to show is expressed in fig. 4 , where A and are side views of squadrons snch as A and in fig. 2 .
[ They are drawn than the stature-schemes in fig. 1 , and therefore out of scale , to save space , which is an unimpor- tant change , as it is only the val'iation in the ogives we are now concerned about .
] Let represent the level of mediocrity above the ground , and the heights of any two recta1lgular files in 1886 .
Family Likeness tature .
squadron of known men .
We have seen that becomes in remote degrees of kinship , and I shall show that in intermediate degrees the value of is constant for all statures in the same degree of kinship .
This fraction is what I call the atio of regression .
and I designate it by .
Consequently the above formula becomes , which is universally ) to all degrees of kinship belween man and man , so long as the statistics of height of the population remain unchanged .
Hence in the squadrons , the curvature in rank is an ogive witn the value of , and in file with one having the quartile value of , these two values being connected by the above formula .
If the squadron is resolved into its elements , and those elements are redistributed into an ordinary stature-scheme , the quartile of the latter will be Another way of explaining the universal tendency to regression may be followed by showing that this tendency necessarily exists in each of the three primary l.elationships , fraternal , filial , and ental , and therefore in all derivative kinships .
Fraternal regression may be ascribed to the compromise of two conflicfing tendencies on the part of the brother , the one to resemble the given man , the other to resemble the mean of the race , in other words to be mediocre .
It will seen that this compromise resuIts in a probable fraternal stature that is expressed by the formulae , in which is a constant as well as , therefore the ratio of fraternal regression is also a constant .
Filial regression is due ( as I explained more fully than 1 need do here , in the publications alluded to in the second ) to the concurrence of atavism with the tendency to resemble the parentremote ancestry in any mixed population resembles , as has beeu .52 .
F. Galton .
[ Jan. 21 , already said , any sample taken at random out of that population , therefore their mean stature is mediocre ; consequently the parental peculiarities are transmitted in a diluted amount .
Parental regression is shown to be the necessary converse of filial regression by mathematical considerations , kindly investigated for me by Mr. Dickson , in the Appendix to this memoir in Problem 1 .
It is easy in a general way to see that this would be the case , but I find it not easy otherwise to prove it .
Still less would it be easy to prove the connexion between filial and mid-parental regression , which depend on considerations that are thoroughly investigated in the Appendix .
Data.\mdash ; I now describe the data from which I obtain my conclusions .
They consist of two sets of practically independent observations , though they do in some small degree ] ( 1 .
) Special observations .
These concern variation in height among brothers .
I circulated cards of inqniry among trusted correspondents , stating that I wanted records of the heights of brothers who are more than 24 and less than 60 years of age ; not necessarily of all the brothers of the same family , but of as many of them as could be easily and accurately measured , the height of even two brothers being acceptable .
If more than one set of brothers were entered on the same card , the entries were of course to be kept separate .
The back of the card was ruled vertically in three parallel columns : ( a ) family name of each set of brothers ; ( b ) order of birth in each set ; ( c ) height , without shoes , in feet and inches .
A place was reserved at the bottom for the name and address of the sender .
The circle of inquiry widened , and I closed it when I had obtained returns of 295 families , containing in the aggregate 783 brothers .
I look upon these returns as quite as trustworthy as any such returns are likely to be .
They bear every internal test that I can apply to them very satisfactorily .
They are commonly recorded to and half inches .
( 2 .
) R.F.F. data .
By this abbreviation I lefer to the Records of Family Faculties that I obtained in the summer of 1884 , in reply to an offer of prizes .
I have been able to extract from these the heights of 205 couples of parents , with those of an aggregate of 930 adult children of both sexes .
I have transmuted all the female heights to their male equivalents , and have treated them thus transmuted on equal terms with the measurement of males , except where otherwise expressed .
These data have by no means the precision of the special observations .
There is in many cases considerable doubt whether measurements refer to the height with the shoes on or off ; many entries are , I fear , only estimates , and the heights are commonly given only to the nearest inch .
Still , speaking from a kuowledge of many of the ibutors , I am satisfied that a fair share of these 1886 .
] Family Likeness in .
53 returns are undoubtedly careful and thoroughly trustworthy , so that I have reason to place confidence in mean results .
They bear those internaI tests that Iapply to them better than I should have expected , and when taken in connexion with and checked by the special data , and used with statistical caution , they have proved very valuable to me .
I have discussed these materials in a great variety of ways to guard myself against rash concIusions , but I shall not present more than three primary tables , which contain sufficient materials for determining the constants of the formulae to be used .
The first of them ( Table III ) refers to the children of what I call " " mid-parents\ldquo ; of various statures .
A mid-parent is the imaginary mean of the two parents , after the female measurements have been transmuted to their male equivalents , so that a mid-parent of 70 inches in height refers to a couple whose mean stature under the above ations is inches .
I have given data in the " " Journ. Antbrop .
Inst loc. to show that we need not regard diflerences in stature between the parents , inasmuch as the distribution of heights among the children proves to be statistically the same , so long as the midparentages are alike , whether tho two parents are the same or of different statures .
This blending of paternal and maternal qualities in the stature of the offspring is one great advantage in selecting stature as a subject for the present inquiry .
General Population .
Its variability .
The value of the quartile deviate in the population ogive ( that is to say , the probable error ) may be deduced from the bottom lines of any one of the three Tables III , , and .
Those in III and IV refer to data that are in part but by no means wholly the same , that of V refers to almost totally distinct data .
The work is shown in Tables VI and ; in the former the ordinates are calculated whence the ogive is drawn , in the latter I have given the values of the measured ordinates at the same points along its axis as those to which the ordinates given in Table I refer .
The values of the quartile that I obtai in this way from the three cases are , and .
I should say that the more careful treatment that I originally adopted happened to make the first of these values also , so I have no hesitation in accepting as the proper value of for all my data .
( 2 .
) Variability of system of mid.parents .
I have published data in the memoir already alluded to , to show that marl'iage selection takes smaU account of stature , which is another great merit in stature as a subject for this inquiry .
Some further proof of this may be got by comparing the variability of the system of mid-parents with that of the general population .
If the married couples had paired together regardless of stature , their mean heights would be elements of a statistical system identical with one in which the pairs had been Mr. F. Galton .
[ Jan. 21 , selected at random .
In this latter case the quartile value of the system of mid-pa , rents would be inch .
Now , I find the quartile of the series of the mid-parental system obtained the two columns in Table III , that are headed respectively " " Heights of the mid-parents " " and ' ' Total number of mid-parents , \ldquo ; to be inches , *which is an unexpectedly exact accordance .
3 .
) Median Stature .
I obtain the values , from the three series mentioned above , but the middle value , printed iri italics , is a smoothed value .
This is one of the only two smoothed values in the whole work , and has been justifiably corrected because the one ordinate that happens to accord closely with the median is out of harmony with all the rest of the curve .
This fortuitous discrepancy amounts to more than inch .
It does not affect the quartile value , because ' neither the upper nor the lower quartile is touched , and , efore , the half-interquartile remains unchanged .
It must be recollected that the series in question refers to R.F.F. brothers , which are a somewhal conditioned selection from the general R.F.F. popula , and could not be expected to aflord as regular an ogive as that made from observations of men selected from the population at hazard .
It is undoubtedly in this group the least accuracy was to have been expected .
Mean Ratios of Regress in of the stature of mid-parents of the same height , to the mean of the statures of all their children .
I already published the conclusions to which I arrived about this , but it is necessary to enter here into detail .
The data are contained in Table III , where each line exhibits the distribution of stature among the children of all the mid-parents in my list , who of the stature that forms the argument to that line .
The median stature in each successive line is the mean stature of all the childre1l , and is given at the side in the column headed ' ' Medians Their values are graphically represented in fig. 5 .
It will be there seen that these value are disposed about a straight line .
If the median statures of the children had been the same as those of their mid-parents this line would lJave accorded with the line AB , which , from the construction of the table , is inclined at an angle of to the line " " Mean Stature of Population which represents the level of mediocrity .
flowever , it does not do this , but its position is inclined at a smaller angle , , such that : : 2 : 3 .
This gives us the ratio of regression in the present case ; and , therefore , in the notation I adopt 2 .
) From the stature of men of the same height , to the mean of the statures of all their children .
We have just seen that when both * In all my measurements the second decimal is only approximately correct .
1886 .
] Family Likeness in tature .
FIOS .
5 and 6 .
parents have a deviate of , the mean of the deviates of all of family centres will be .
It follows that if one parent only has that deviate , and if the stature of the other parent is unknown , and , therefore , on the , mediocre , the mean of the statures of their children will be half the above amount , or .
I cannot test this conclusion very satisfactorily by direct observation , for my data are barely numerous enough to enable me to deal even with the midparentages .
They are consequently insufficient to deal with a question involving the additional large uncertainty of the stature of one of the parents .
I have , however , tabulated the data , but do not think it worth while to give them .
They yield a ratio of regression of instead of as above .
I disregard it , and adopt the latter , namely , ( 3 ) .
From the stature of mell oi the same height to the mean of he statures of their mid-parents .
By treating the vertical columns of in the same way as we have just dealt with the horizontal lines , we obtain results of the same general form as in the last paragraph but one , though of different values .
Taking the height of a group of men of the same stature , the " " Adult Children as given in the line that forms the heading to the table , we find tho median stature of all their mid-parents , whence I dednce in this case .
The apparent paradox that the same table should give results by no means converse in their values for converse degrees of kinship , will be more conveniently examined later on .
( 4 .
) From the stature of men of the snme height to the meall of the statures of all their brothers .
In seeking for this I shall at first confine myself to the more accurate special data , reserving to the end a comparison between results and those derived from the R.F.F. The entries in the column headed ' medians " " in Table are Mr. F. Galton .
[ Jan. 21 , graphically represented in fig. 6 , whence I deduce the value of Variability of Statures of " " -kinsmen\ldquo ; about their ( jommn mean \mdash ; By " " -kinsmen\ldquo ; I desire to express the gronp distributed in any one line of Tables III , IV , V , or of other tables constructed on a like principle .
They are the kinsmen in a specified degree , not of a single person , but of a group of like persons , who probably differ both in ancestry and nurture .
For example , the persons to whom the entries opposite in Table III refer are not brothers , bnt they are lvhat I call " " -fraternals , \ldquo ; or from another point of view , " " co-filials , \ldquo ; namely , the children of numerous mid-parentages , differing variously in their antecedents , and alike only in their personal statures .
ariability .
appears from Table III that the mean of the quartiles derived from the successive lines , and which I by , is inch ; also that the quartiles.are of nearly the same value in all of the lines , allowance being made for statistical irregularities .
A protraction on a large sheet of the individual observations in their several exact places , gave the result that the quartile was a trifle larger for the children of tall mid-parentages than for those of short ones .
This justifies what was said some time back about the use of the geometric mean ; it also justifies the neglect here of the method founded upon it , on the ground that it would lead to only an insignfficant improvement in the resuIts .
We have obtained the values of the three constants in tho general equation , when it is used to express the relation between mid-parentages and cofilials .
Thus the quartile of the population being , it was shown both by observation and by calcula- tion , that the quartile of the mid-parental system was , or It was shown that the ratio of regression in that case was , con- sequently the general equation becomes or , which is an exact accordance , satisfactorily cross.testing the various independent estimates .
Convexse Ratios of Regression.\mdash ; We are now sufficiently advanced to be able to examine more closely the apparent paradox that the ratio of ression from the stature of mid-parents of the same height to the mean of the statures of their sons should be 23 , while that of men of the same stature to the mean of the statures of their several mid-parents should be , not the numerical converse of this , but 13 .
We may look upon the entries in Table III as the values of ( vertical ) ordinates in to be erected upon it at the points where those entries lie , and which are specified arguments of ' ' heights of midparents\ldquo ; ritten along the side , as values of ordinates in , and of " " heights of adult chiJdren\ldquo ; written along the top , as values of ordinates in .
The smoothed rssult would form a curved surface of frequency .
I accordingly oothed the table by writing at each 1886 .
] Family Likeness in Stature .
intersection of the lines that separated the vertical columns with those that separated the horizontal lines , the sums of the four adjacent entries .
Then I drew lines with a free hand through all entries , or ationS between entries , that were of the same value .
These lines formed a concentric series of elliptical figures , passing through values of that diminished , going outwards .
Their common centre at which was the greatest , and which therefore was the portion of maximum frequency , lay at the point where both and were of the same value of inches , that is , of the value of the mean stature of the population .
The line in which the major axes of the ellipses lay was inclined nearer to the axis of than that of .
It was evident from the construction that the median value of the entries , whether in each line or in each column of the table , must lie at the point where that line or column was touched by the projection of one of these ellipses .
It was easy also to believe that the equation to the surface of frequency and the lines of loci of the above-mentioned points of contact , admitted of mathematical expression .
Also that the problem to be solved might be expressed in a form that had no reference to heredity .
In such a form I submitted it to Mr. J. Hamilton Dickson , who very kindly undertook its solution , which appears as an Appendix to this paper , and which helps in various ways to test and confirm the apploximate and uncertain conclusions suggested by the statisticaI treatment of the observations themselves .
I shall make frequent use of his mathematical results , both in respect to this problem and to another one ( also given in the Appendix ) , in the course of my further remarks .
As regards the present subject of the connexion between the regression in direct and in converse kinships , it appears that it wholly depends on the relation between the quartiles of the two series of " " arguments and is expressed by the formula .
In this case , and ; also ; It will be observed that in all cases of converse kinship , from man to man\mdash ; as from man to brother , and conversely ; from man to nephew , and conversely ; from father to son , and conversely ; , therefore in these the ratio of regression is the same in the converse as in direct kinship .
Brotherly Variability.\mdash ; The size of human families is much too small to admit of the quartile of brotherly variability being determined in the same way as that of the ] ation , namely , by finding the quartiles in single families , but there are four indirect ways of finding its value , which I will call A collection of differences ( see Table VIII ) between the statures of individual brothers , in families of ?
brothers , and the mean of all the statures in the same family , ives a quartile value , I will FIG. 7 .
who are not brothers , and MC represents the mean of the mid-deviates of their respective families of brothers .
It can be shown ( see Appendix , Problem 2 ) that if the position of varies with respect to with a quartile , and if varies with respect to with a quartile , then , when only is observed , the most probable value of CM is such that or .
Snbstituting for , and for inch .
( 3 .
) It can also be shown ( see Appendix , Problem that the variability of particular mid-brotherly deviates , , about , the 1886 .
] Family Likeness in Stature .
mean of all them , is such that its quartile .
Now the distribution of values in each line of Table , whose quartile , is due to the combination of two variables .
The one is the variability of , about ; the other is the variability of the individual brothers in each family , about , \amp ; c. , respectively .
Therefore .
Substituting for its value , we obtain .
The observed value of in Table V is , whence we obtain ( 4 .
) Pairs of brothers may be taken at random , and the differences noted between their statures ; then under the following reservation , as regards the differences to be taken , we should expect the observed quartile of the diiferences to be .
The reservation is , that only as many differences should be taken of each family as are independent .
A family of brothers admits of pairs , but no more than of these are independent and only these should be taken .
I did not appreciate this necessity at first , and selected pairs of brothers on an arbitrary system , which had at all events the merit of not taking more than four of differences from any family , however numerous .
It was faulty in taking differences instead of only two from a family of three , and four dffierences instead of only three from a family of four : and therefore giving an increased weight to those families , but in other respects the system was hardly objectionable .
On the whole the roduced error would be so slight as scarcely to make it worth while now to go over the work again .
By the system adopted I fonnd a quartile value of , which divided by gives inch .
Thus far we have dealt with the special data only .
The less worthy R.F.F. larger values of .
An epitome of all the lesults appears in the following table .
From fFrom pairs oobtaine.byfrom ddiff .
Mr. F. Galton .
[ Jan. 21 , R.F.F. resuIts refer to brothers only and not to transmuted , except in method ( 2 ) , where the paucity of the data compelled me to include them .
I should point out that the data used in these four methods differ .
In ( 1 ) I did not use families under four .
In ( 2 ) and ( 3 ) I did not use large families .
In 4 ) the method of selection was as we have seen , again different .
This makes the accordance of the results still more gratifying .
I gather from the above that we may securely consider the value of to be than , and allowing for some want of precision in the special data , the very convenient value of inch may reasonably be adopted .
We are now able to deal completely with the distribution of statures in every degree of kinship of the kinsmen of those whose statures we know , but whose ancestral statures we are ignorant of or do not take into account .
We are , in short , able to construct tables on the form of III , , and V , for every degree of kinship , and to reconstruct those tables in a way that shall be free from irrsgularities .
The fraternal relation as distinguished from the -fraternal has also been clearly explained .
In constructing a table of the of III , , and V , we tirst find the value of for the degree of kinship in question , thence we deduce by means of the general equation is supposed to be known , or for the general purpose of comparing the relative nearness of degrees of kinship as tested by family likeness iu stature , it may be taken as unity ) .
The entries to be made in the several lines are then to be calculated from the ordinary tables of the " " probability integral As an example of the first part of the process , suppose we are constructing a table of men and their nephews .
A nephew is the son of a brother , therefore in his case we have ; and Form of Data for calculating Tables of Distribution of Stature among Kinsmen .
UMid parents Trustworthiness of the Constants.\mdash ; There is difficulty in correcting the esults obtained solely from the R.F.F. data , by help of the knowledge of their general inaccuracy as compared with the 1886 .
] Family Likeness in Stature .
special data .
The reason is that this inaccuracy callnot be ascribed an uncertainty of equal amount in every entry , such as might be due to a doubt of " " shoes off\ldquo ; or " " shoes on If it were so , the quartile deviate of the R.F.F. would be greater than that of the specials , whel.eas it proves to be the same .
It is likely that the inaccuracy is a result of the uncertainty above mentioned , which would increase the value of the quartile deviate , combined with a tendency on the part of my correspondents to ecord medium statures when they were in doubt , and which would reduce the quartile deviate .
What the effect of all this might be on the value of in Table which is a datum of primary importance , I am not prepared to except that it cannot be great .
While sincerely desirous of obtaining a revised ] of from new and more accurate data , the provisional value I have adopted may be accepted as quite accurate enough for the present .
Separate Contribution of each Ancestor to the Heritage of the Child.\mdash ; I here insert a short extract from my paper in the " " Journ. Anthrop .
Inst with slight revision , as this .
would be incomplete withont it .
When we say that the mid-parent contributes two-thirds of his peculiarity of height to the offspring , it is supposed that nothing is known about the previous ancestor .
But though nothing is something is implied , and this must be eliminated before we ] earn what the parental bequest .
pure and simple , may amount to .
Let , the deviate of the mid-parent be ( including the sign ) , then the implied deviate of the mid-grandparent will be , of the mid-ancestor in the next generation , and so on .
Hence the sum of the deviates of the mid-generations that contribute to the tage of the offspring is Do they contribute on equal terms , or otherwise I have nob snfficient data to yield a direct reply , and must , therefore , try the effects of limiting suppositions .
First , suppose the generations to contribute in proportion to the values of their tive mid-deviates ; then as an accumulation of ancestral deviates whose sum amounts to , yields an effiective heritage of only , it follows that each piece of heritable property must be reduced , as it were , by a succession tax , to of its original amount , because Another supposition is that of successive portionate diminutions , the property being taxed afresh in each transmission to of its amount , so that the effective heritage would be\mdash ; and this must , as before , be equaI to , whencc F. alton .
[ Jan. 21 , A third possible supposition of the mid-ancestral deviate in any one remote generation contributing more than would be done by an equal mid-parental deviate , is notoriously incorrect .
Thus the descendants of " " pedigree wheat\ldquo ; in the ( say ) twentieth generation show no sign of the remarkable size of their mid-ancestors in that degree , but the offspring in the first generation do so unmistakably .
The resuIts of our only two valid limiting suppositions are therefore 1 ) that the mid-parental deviate , pure and simple , influences the offspring to of its amount ; ( 2 ) that it influences it to the of its amount .
These values differ but slightly from , and heir mean is closely , so lve may fairly accept that resuIt .
Hence the influence , pure and simple , of the mid-parent may be taken as , of the mid , of the mid-great-grandparent , and so on .
That of the individual parent would therefore , of the individual grandparent , of an individual in the next generation , and so on .
[ I do not plopose here to discuss the reason why the effective of the child should be less than the accumulated deviates of his ancestors .
It is obviously connected with considerations that bear on stability of type .
] breed.\mdash ; In a perfectly pure breed , maintained during an indefinitely long period by careful selection , would become , and the value of would be changed , but apparently only a little .
Callits new value .
It may be estimated as follows .
In mix , breeds tlJe value of includes the probable uncertainty of the implied value of the contributions inherited from the mid-grandparents , and from the mid-ancestry of each preceding generation .
This can be but a trifle .
Suppose the quartile of the uncel.tainty in the implied stature of each grandparent to be even as much as inch ( we need not wait to discuss its precise value ) , then the quartile of uncertainty as regards the implied ental stature would be that amount , or say .
The proportion of this , which would on the average be transmitted to the child , would be only as much , or From all the higher ancestry put together , the contribution would be much less than this , and we may disregard it .
The result then is .
Taking , this gives inch .
Probable Stature of the when Statures of several of are known .
we have to add their several contributions as assessed in the lasb paragraph but one , and to these we have to add whatever else may be implied .
A just estimate of the latter requires the soIution of a very complex problem .
Thus : tall son has a short father ; this piece of knowledge makes us suspect that the lnother was tall , and we should do to set down her unknown stature as mediocre .
Onr revised estimate lvould be further modified if we knew the stature of one of her brothers , and so ou .
Moreover , the general equation may cease to hold good .
The pos1886 .
] Family Likeness in Stature .
sible problems are evidently very various and complicated , I do not propose to speak further about them now .
It is some consolation to know that in the commoner questions of hereditary interest , the genealogy is fully known for two generations , and that the average influence of the preceding ones is small .
In c. , onclusion , it must be borne in mind that 1 have spoken throughout of heredity in respect to a quality that blends freely in inheritance .
I reserve for a future inquiry ( as yet incomplete ) the inheritance of a quality that refuses to blend freely , namely , the colour of the eyes .
These may be looked upon as extreme cases , between which all ordinary phenomena of heredity lie .
Appendix .
By J. D. IIAITILTO DICKSON .
Problem 1 .
A point is capable of moving along a straight line , making an angle with the axis of , which is drawn through the mean position of ; the probable of the projection of on Oy is inch : another point , whose mean pof ; ition ab any time is is capable of moving from parallel to the axis of ( rectangular -ordinates ) wibh a probable errol of inch .
To discuss the " " surface of frequency\ldquo ; of 1 .
Expressing the " " surface of frequency\ldquo ; by an equation in the exponent , with its sign changed , of the exponential which appears in the value of in the equation of the surface is , save as to a factor , ( 1 ) hence all sections of the " " surface of frequency\ldquo ; by planes parallel the plane of are ellipses , whose equations may be written in the form , , a constant . . . .
( 2 ) 2 .
Tangents to these ellipses parallel to axis of are found , diiferentiating ( 2 ) putting the coefficient of equaI to zero , to meet the ellipses on the line , ( 3 ) that is or , approximately , on the line .
Let this be the line OM .
1Ir .
J. D. milton Dickson .
[ Jan. 21 , From the nature of conjugate diameters , and because is the mean position of , it is evident that tangents to these ellipses parallel to the axis of meet them on the line , on OP .
3 .
Sections of bhe " " surface of frequency\ldquo ; parallel to the plane of , are , from the nature of the question , evidently curves of frequency with a probable error , and the locus of their vertices lies in the plane Sections of the same surface parallel to the plane of yz got from the exponential factor ( 1 ) by making constant .
The result is simplified by taking the origin on the line OM .
Thus putting and , where by ( 3 ) the exponential takes the form ( 4 ) whence , if be the probable error of this section , or [ on referring to ] ( 5 ) that is , the probable error of sections parallel to the plane of is nearly times that of those parallel to the plane of , and .the locus of their vertices lies in the plane It is important to notice that all sections parallel to the same coordinate plane have the probable error .
4 .
The ellipses ( 2 ) when referred to their principal axes become , after some arithmetical simplification , constant , ( 6 ) the major axis being inclined to the axis of at an angle whose tangent is .
[ ln the approximato case the ellipses are const .
, and the major axis is inclined to the axis of at an angle ] 5 .
The question may be solved in general terms by putting YON , and replacing probable errors and by and respectively : then the ellipses ( 2 ) are 1886 .
] Family Likeness in Stature .
, ( 7 ) equation ( 3 ) becomes or and ( 5 ) becomes whence If be the probable error of the projection of 's whole motion on the plane of , then which is independent of the distance of of motion from the axis of .
Hence also . . . . .
11 Prob lem 2 .
An index moves under some restraint up and down a bar AQB , its mean position for any given position of the being ; the bar , always carrying the index with it , moves under some restraint up and down a fixed frame YMY ' , the mean position of being : the movements of the index relatively to the bar and of the bar relatively to the frame being quite independent .
For any given observed position of , required the most probable position of ( which cannot be observed ) ; it being known that the probable errol of relatively to VOL. 66 Mr. J. D. Hamilton Dickson .
[ Jan. 21 , in all tions is , and that of relatively to is .
The ordinary law of error is to be assumed .
If in any one observation , , then the law of error requires . . . .
( 12 ) to be a minimum , subject to the condition , a constant .
ence we have at once , to determine the most probablo values of , . . . . .
( 13 ) and the most probable position of , measured from , when 's observed distance from is , is It also follows at once that the probable error of ( which may be obtained by substituting for in 12 ) ) is given by , or ( 14 ) which , it is important to notice , is the same for values of Throughout this discussion the technical term " " probable error\ldquo ; has been used ; it may in every instance be replaced by Mr. Galton 's very apt name " " quartile in which case the resuIts of these problems may be read in conjunction with .
Galton 's papers .
Table I. Ogive , or Normal Curve of Distribution of Error .
Table II .
Comparison of observed ives with the Normal .
parentages \ldquo ; , ,Brothers iandom pairs , speci.unit ininchesof t Note.\mdash ; The second decimal is only approxinlate .
Mr. J. D. Hamilton Dickson .
[ Jan. 21 , 1886 .
] Family Likeness in J. D. Hamilton Dickson .
an .
21 1886 .
] Family Likeness in Stature . .
ni : $ S : : : : : : : : : : : : : 72 Family Likeness in Stature .
[ Jan. 21 , 1886 .
] Early Development of Julus terrestris .
Table VIII .
( Special Data .
Number of cases in which the Stature of individual Brothers was found to deviate to various amounts from the lvIean Stature of their respective families .
' The ] Development of By F. G. HEATHCOTE , M.A. , Trin .
Coll. Cam .
Communicated by Professor M. FOSTER , Sec. R.S. Received January 6 , 1886 .
The following are the principal resuIts of my investigations on the early development of Julus terrestris since June , 1882 .
When laid the eggs are oval in shape , white , and covered with a thick chitinous chorion .
The nucleus is situated in a mass of protoplasm in the centre of the ovum .
This mass of protoplasm is of irregular shape , but its long axis corresponds with that of the ovum .
From it , anastomosing processes radiate all directions , forming a network throughout the egg .
The yolk-spherules are contained within the meshes of this network .
The nucleus is not a distinct vesicle but its position is marked by chromatin granules .
There is no nucleolus .
Early on the second day the nucleus and the central mass of protoplasm apparently divide into two parts .
But this division is llot complete , the two resulting masses with their nuclei remaining connected by a network of protoplasm .
Each of these divides in the same incomplete manner , so that we now have segments all connected together .
This process is continued until there are a considerable number of segmentation masses present , and early on the * Mr. J. D. Gibson Carmichael , , has kindly identified species for me as Julus terrestris , Leach , 1814 .

