]\gt ; On Independenl in Measurements .
91 by a new function defined as the product of into that function of which has the value unity at all points of and is zero elsewhere , and , accordingly , , by the product of this function into a function of , which is defined to have the value unity at all points of the set which has for image and is zero elsewhere .
We then at once have the desired result , except that there remains a doubt as to the type of summability required for .
This must be such that , when multiplied by the auxiliary function just introduced , it is summable over the whole set of points which corresponds to the fundamental rectangle .
But this is clearly the same as demanding that should be summable over the set of points constituted by all the points in the -plane which correspond to the set , that is , over the set Hence our theorem has been proved , with all the generality given to it , in the enunciation of S2 .
On Independent Factors in rements .
By J. C. MAXWELL GAlNETT ( late Fellow of Trinity College , Cambridge ) .
( Communicated by Prof. A. .
Whitehead , F.R.S. Received March 26 , 1918 .
I. In a paper published in the ' American Journal of Psychology ' for 1904 , Prof. Spearman argued that " " all branches of intellectual activity have in common one fundamental function or group of functions ) even that this fundamental function enters into sensory discrimination no less truly than into more complicated intellectual activities .
In 1909 Mr. Cyril Burt experiments on two groups of Oxford schoolboys undertaken " " with a view to testing in practice the mathematical methods of Prof. Spearman Mr. Burt calculated the correlations between the performances of his subjects in 12 different tests , which he classified as sensory , motor , sensori-motor , and association tests , together with one ( the twelfth ) test of voluntary attention .
Mr. Burt 's results were consistent with the view that all the correlations were due to the operation of one and only one general factor .
Two years later Prof. Spearman and Dr. proposed a new test American Journal of Psychology , ' vol. 16 , p. 284 1904 ) .
'British Journal of Psychology , ' vol. 3 , pp. 1 'British Journal of Psychology , ' vol. 6 , VOL XCVI .
Mr. J. C. M. Garnett .
On Certain Independent for ( single ) general factor ; namely that , in such a correlation table as those published by Mr. Burt , in the form ( 27 ) below , the correlation between every pair of columns should , with proper allowance for sampling errors , be , or , as Dr. Webb* pointed out afterwards , .
Their formula for column correlation , corrected for errors of sampling , is in which the 's are the correlation coefficients measured from the mean of the column , and the 's are the probable errors of the 's divided by The bar indicates mean values .
The authors decided that , for the purpose of calculating a " " correlation between columns only those pairs of columns could be used in which ( is at least twice the correction ) .
With this limitation , they applied their formula to the five largest pairs and the five smallest pairs ( or as many as were up to the correctional standard ) of columns in 14 published tables of correlation coefficients dating from various periods up to 30 years back .
They found that\ldquo ; from beginning to end , the correlation between columns is positive and very high ; the mean is almost completel .
This is just the value demanded by the theory of a [ single ] General Factor.\ldquo ; Dr. Webb , in his essay on\ldquo ; Character and Intelligence to which reference has just been made , prepared a further correlation table of five columns ( and five rows ) the results of five examination tests set to 96 training college srudents .
Nine out of the ten pairs of columns were up to the correctional standard of Prof. Spearman and Dr. Hart , and the application of their formula to those nine pairs yielded as the average correlation between columns .
This result , says Dr. Webb , " " is an additional item of evidence in support of the theory of a general factor . .
It takes its place in the huge array of evidence collected by Prof. Spearman from experimental tests by many investigators\mdash ; the steadiness of results such as to rival the niceties which physical measurements reveal .
It should be remembered that the raw material for our own ( comparatively small ) contribution to this result consisted of test-papers numbering nearly Prof. Spearman 's mathematical argument for the existence of general ability has , however , been challenged .
Thus Mr. Thomson concludes a recent paperS on the subject with the words : " " It must not hastily and illogically be concluded by any one that general ability is a fiction .
Its existence or non*Character Intelligence \ldquo ; Monograph Supplement , .
cit. , p. 60 .
$ .
, .
37 .
S 'British Journal of Psychology , ' vol. 3 , .
Factors in Mental Measurements .
existence is , as far as the mathematical argument goes , an entirely open question.\ldquo ; * It is with the mathematical argument that the present paper is principally concerned .
We shall consider what consequences necessarily follow when the correlation between every pair of columns in a correlation table is ; and we shall find that the correlation between every pair of columns will only be if there is one , and only one , general factor , either of the correlated variables themselves ( if Mr. Burt 's conditions\mdash ; equations ( 24 ) , below\mdash ; for a hierarchy are satisfied ) , or of the differences between these variables and a real multiple of a new variable that is independent of them all .
II .
Let us suppose that the mental qualities measured in any particular investigation are due to a number of independent factors , Let , measure the amounts of these factors ( or qualities ) possessed in excess of the average by any subject person investigated ) .
We shall assume that , among a very large number of subjects , each quality is distributed according to the normal law ; so that the probability that any particular subject will possess a factor\mdash ; say \mdash ; in a degree lying between and is .
Let us now choose our units of measurement for so that .
Some arbitrary choice has to be made ; how else are we to represent , say , " " degree of sense of humour\ldquo ; by a number ?
We , therefore , choose our units for measuring the independent factors so that the standard deviations of the normal distributions of each factor shall be equal to , p. added , March 24 , 1919.\mdash ; More precisely expressed , our result will be that whenever the correlation between variables as , for example , the measures of mental qualities ) , each of which is distributed according to the normal law with the same probable error , satisfy the " " correlation between columns\ldquo ; condition , these variables or , if their correlations do not satisfy Mr. Burt 's conditions for a hierarchy , the differences between them respectively and a real of an variable , independent of them all , but distributed according to the normal law , with still the same probable error ) , can be completely expressed as linear functions independent variables , each of which is distributed according to the normal law with the same probable error , and one of which is a single general factor , while the remainder are specffic factors .
] Dr. Webb , in his investigation of mental quahties , adopted such units for the measurement of each quality as would give the same constant ( standard deviation ) to the frequency distribution of each of the forty-eight qualities with which he was concerned , pp. 18 , 19 ; see especially the footnote on these two pages ) .
Mr. J. C. M. Garnett .
On Independent The probability that any particular subject will possess one of these factOrs ( qualities ) , say , in a lying between and is now , ( 1 ) and the probability that a subject chosen at random will possess all the independent factors in degrees lying respectively between XI and and and is the product of the independent probabilities or \ldquo ; The symmetry of this expression shows that the distribution of qualities is the same as would be due to other independent quantities connected with the first set by relations of the type , ( 3 ) . .
where , ( 4 ) , ( 5 ) and , ( 6 ) for all values of from 1 to , the probability that any subject will possess the quality in a degree lying between .
and being again It follows that , if any contain the independent factors in the proportions : l2 : : where , ( 7 ) so that a subject possessing the factors , , , in , , will possess in the degree , where , ( 8 ) then the probability that any particular subjecb will possess in a degree lying between and is .
( 9 ) Factors in Mental Measurements .
This result may also be obtained by noticing that any individual possessing the independent factors , .
, in degrees , .
may be represented in -dimensional space by a point whose co-ordinates are , .
; and that , when the number of individuals is very large , the density of distribution of such points , being given by equation ( 2 ) , is a function only of the distance from the origin , so that the density of distribution of the points representing individuals falls off according to the normal law ( with the same standard deviation ) in whatever , or Oq\mdash ; we proceed straight out from the origin .
No distribution of the quality , except such as is defined by a linear equation of the type ( 8 ) , will satisfy the relation ( 9 ) .
For consider variations of two independent variables at a time .
The probability that a subject chosen at random will possess these qualities in degrees intermediate between and and respectively , is where A is independent ' and .
We may represent this distribution by a surface of revolution about an axis , Oz , perpendicular to and Every section of this face by a plane perpendicular to that of and is a normal probability curve .
For example , the section of ( 10 ) by such a plane passing through the point and the origin has the equation , ( 11 ) while the section through the same point , but parallel to , has the equation where the constant is given by Similarly , sections by other planes perpendicular to , but having different traces on , would be probability curves with still other constants .
It follows that a section of the surface given by equation ( 10 ) by a surface generated by lines parallel to Oz but intersecting in any curve ) which is not a line , would not be a normal curve .
We conclude that every quality that is distributed the normal curve must be related to the independent variables ( factors ) by a linear relation as in equation ( 8 ) .
For example , if " " degree of sense of humour\ldquo ; be distributed normally and depend on two and only two independent factors , say ' ability\ldquo ; and " " cleverness both normally distributed , then ' sense of humour\ldquo ; must be the sum of ( equal or different ) proportions of the two factors , but cannot be their product , quotient , or any other function of them .
Mr. J. C. M. Garnett .
On Independent Consider now the Bravais-Pearson correlation between and some other quality measured by ( 12 ) where , as before , .
( 13 ) It will be convenient to write ( 14 ) so that is a measure of the extent to which a subject is exceptional , regard being had to all the variables involved .
Then our above may be re-written , say ( 15 ) giving the probability that a subject will possess the various qualities in degrees lying between and , etc. Here is a measure of the density of distribution of subjects .
According to the definition of , we have on dividing numerator and denominator by Hence*the " " cosine law .
( 16 ) We note that , if there are only three independent factors and if we measure along axes at angles , , etc. , become direction-cosines , so that is measured in the .
direction , l3 , or , say , Oq .
Then the last equation beco1nes ( 17 ) affording us a very simple geometrical conception of the measure of correlation .
Moreover , however large may be the number of independent variables , [ Note adokd , March 24 , 1919 .
result equivalent to that given in equation 16 ) was obtained by Bravais in 1846 : see " " Analyses Mathematiques sur les Probabilites des Erreurs de Situation d'un Point ' Memoires de l'Institut de France , ' IX , pp. 260 et seq. ] Factors in JIental Measurements .
we can , by a linear transformation of the type shown in quations ( to ( 7 ) , choose new independent variables , such that and depend on two , and two only , among them ; so that equation ( 17 ) still holds when the number of independent variables three .
This equation may be interpreted to mean that represents the average deviation in ( or ) corresponding to unit deviation in ( or ) .
The condition that , with only three independent variables , three qualities , , and , should be dependent on two , and two only , follows at once .
The condition in question is that , and should lie in a plane , or that or that* .
( 18 ) It is easy to show that , however many independent variables there may be , equation ( 18 ) still gives the condition that three qualities , , and , should contain two , and only two , independent general factors , and that no one of them should contain any specific factor in addition .
If in equation ( 16 ) we make identical with , , , in turn we obtain the correlations between and the independent .
They are ; ; ; .
( 19 ) Again , let us take so as Go be independent of , but so as to added Jfarch 24 , 1919.\mdash ; This equation ( 18 ) may be written .
( 18 ) It may be shown , by employing the formulae of spherical trigonometry , that the corresponding condition that four qualities , and , should depend on three independent factors only ( or , in othex words , that , and , should lie in the same 3-dimensional is , if If three qualities , , contain two , and only two , independent general factors X and , and no specific factors , we may write so that equation ( 16 ) gives .
Since , therefore , we have Mr. J. C. M. Garnett .
On Certain Independent contain , , in the same proportions as these variables are contained in Q. Then we have ; in view of equation ( 7 ) .
If , now , we have some other quality related to as is related to , we have ( 20 ) which is Yule 's well-known equation .
One other result we may note in passing .
In dice-throwing experiments the score of each die is an independent variable .
If represent the total score obtained by throwing dice multiplied by a factor that will give to the same mean variation*as , , or , the scores obtained by the separate dice , we have where ) so that .
( 21 ) We note that the correlations of any distribution are the same as those of the scores , to which the 's bear a constant proportion .
If , as in Mr. Thomson 's .
experiment , we wish to know the correlation between a large number of successive pairs of throws of dice , such that dice are thrown for the first throw , , left lying , and thrown to form with the second throw of dice , and , if and be successive scores , we have ( 22 ) * The amount of this mean variation , , is easily determined .
If each die possesses faces marked , 1 , then , in a very large number of throws , say , each face will turn up times , so that giving .
' British Journal of Psychology , ' vol. 8 , p. 274 ( 1916 ) .
Factors in Wental Mxsurements .
where are variables independent of so that equation ( 16 ) gives ( terms ) , ( 23 ) a result already obtained by Prof. Spearman and by Mr. Thomson .
We observe that , according to equation ( 22 ) , and ( or and ) possess not independent common factors , but one only .
For by a linear transformation of dependent variables such as is iven by equations ( 3 ) to ( 7 ) above , we might take , thus replacing , by one variable , without affecting the other independent variables , III .
We shall next show that , when the correlations between variables , , .
, each of which is distributed according to the normal law with the same probable error , satisfy Mr. Burt 's conditions , namely , * , ( 24 ) where have any different values from 1 inclusive , the 's may be expressed linear functions of independent variables , each of which is distributed according to the normal law with the same probable error as that of the , and one of which is a single general , while the remainder are specific factors .
Following Mr. Burt , we shall speak of the coefficients that satisfy equations ( 24 ) as forming a hierarchy .
From equations ( 24 ) it follows that , so that , say , ( 25 ) where depends upon above .
Now suppose that the absolute magnitude of is greater than that of any of the other correlations , and that .
Then , in general , equations ( 24 ) give so that BIJItT , Mr. J. C. M. Garnett .
On Independent Moreover , from equation ( 24 ) .
It follows that the absolute magnitude of lies between zero and unity , and that the sign of is the same as that of for all values of and from 1 to inclusive .
Since every suffix occurs twice in equations ( 24 ) , these equations would not be affected if we were to change the sign of the unit in terms of which any one ( or more ) of the 's in measured .
We may therefore so choose the signs of the units in terms of which , , are measured that , , are all positive .
Then , since and every has the same sign , will have the same sign for all values of and from 2 to inclusive , and this sign will be the same as that of for all values of from 1 to .
It follows that , if the signs of the units in terms of which , , are measured be so chosen that all the 's in the first row or column of the correlation table '/ 23 ( 26 ) : : are positive , will be positive for all values of , unless wtside t first and cotumn is negative , a condition that is not fulfilled , so far as the present writer has been able to ascertain , by any table of correlations between mental tests that has hitherto been published .
We shall therefore take it that is positive for all values of when equations ( 24 ) are satisfied b.y the correlations between mental tests .
Since , then , , we may write , ( 27 ) on the an .
of equation ( 17 ) , where is the correlation between and some variable , is also distributed according to the normal law with the same probable error as before .
Now , it may be shown that correlated , each of which is distributed according to the normal law with the same probable error , can always be expressed with degrees of freedom , in terms of independent variables , each of which is distributed according to the normal law with the same probable error as before , by means of equations ( 8 ) .
Thus ( 28 ) Factors in Measurements .
We may therefore regard , , as direction-cosines ( or as " " correlation-cosines , \ldquo ; is the coefficient of correlation between and ) of a line through the origin in -dimensional space , and making with the axis an equal to .
We can , therefore , draw , in dimensional space , a iine Oq making given angles with each of the lines .
And since ( unless the italicised condition in the previous paragraph were satisfied ) is positive and less than unity , we may take these angles to be , etc. , say , where , 2 , , ; so that , according to equation ( 17 ) , is the correlation between and a new quality , that distributed according to the normal law with still the same probable error .
* By means of a transformation of independent variables such as is given in equations ( 3 ) to ( 7 ) let us now express our qualities , , in terms of new independent variables , , , distributed as before , and let us choose Then , in place of equations ( 28 ) we have , ( 29 ) where ( 30 ) for all values of from 1 to inclusive .
Equation ( 16 ) then gives .
( 31 ) But ; ( 32 ) so that equation ( 31 ) gives If , therefore , we write .
( 33 ) ' , ( 34 ) equation shows that will be variables distributed as before , since , according to equation ( 30 ) , the sum of the squares of the coefficients of the 's is unity .
Moreover , from equations ( 33 ) it follows that the 's are independent of one another ; and , since they depend only on the 's which * In order that it may be possible to draw , making given angles , with times Omust n ' must be ; or must be , since , as we about to see , ; and this last in equality is always satisfied , since .
Mr. J. C. M. Garnett .
On Independent are independent of , they also are independent of .
Substituting from equation ( 34 ) in equation ( 39 ) , and writing for in that equation , we obtain , ( 35 ) in which , , are independent variables , each of which is distributed according to the normal law with the same probable error as that of the and among which is a facto while all the other 's are specific factors .
There are no group factors .
IV .
We have now to examine the consequences of the fulfilment of Prof. Spearman 's conditions that the correlation between every pair of columns ( or rows ) in the table ( 26 ) : : is , where is the Bravais-Pearson coefficient of correlation between and , being any two of correlated variables each of which is distributed according to the normal law with the same probable error .
We first observe that , if the correlation between any two sets of quantities where corresponding values ) is , then .
For let bars denote mean values and put .
Then we have so or whence ; that is .
It follows that , where A and are constants independent of * It will be remembered that if the italicised condition on p. 100 were not satisfied would be imaginary .
Factors in Mental Consider then the four coefficients , in table ( 26 ) , where have any different values from 1 to inclusive .
Equation ( 36 ) applied first to the and columns and then to the and th columns ( between each which two pairs of columns the correlation is , gives us , first , , ( 37 ) , ( 38 ) where depend on and alone , and then , ( 39 ) , ( 40 ) where depend on and alone .
From the second and third of these equations we obtain while the first- and fourth equations give Subtracting one of the last two equations from the other , ; so that , say , where is a constant independent of .
Equation ( 37 ) now becomes ; so that the four equations ( 37 ) to ( 40 ) may be replaced by , ( 41 ) relations resembling Mr. Burt 's conditions for a hierarchy .
Equation ( 41 ) gives .
( 42 ) If we alter the signs of the units in terms of which and are measured , without altering the signs of the units in terms of which and are measured , we shall alter the sign of without affecting the fulfilment of the condition that the correlation between every pair of columns in the table ( 26 ) is We may therefore assume that is positive and equal to , say .
Now suppose that ; and , ( 43 ) where are independent variables , each distributed according to the normal law , with the same probable error as that of the .
Let us .
J. C. M. Garnett .
On Certain Independent now introduce a new set of independent variables , , defined by , ( 44 ) in which , as will appear from equation ( 48 ) below , is a real quantity , mediate in value between ; and let us put , ( 46 ) where is a variable , independent of , , , but distributed according to the normal law , with the same probable error as , , .
Then it follows from equation ( 45 ) , in which the sum of the squares of the coefficients of the is equal to unity , that ( where has any value from 1 to inclusive ) is distributed according to the normal law , with the same probable error as the , but a different probable error from that of the 's and Using equations ( 45 ) , we obtain from equation ( 16 ) an expression for the correlation , , between and , namely , ( 47 ) .
, ( 48 ) if , from which it follows that is positive and less than unity .
From equations ( 48 ) combined with equations ( 41 ) we now obtain , so that the correlations between the satisfy Mr. Burt 's conditions expressed in equations ( 24 ) .
It follows from equations that where and where is a single general factor of , , while is a specific factor of .
Here and are distributed according to the normal law , with the same probable error as that of the .
Substituting from equations ( 46 ) , we now obtain ; which , if ( 51 ) becomes , ( 52 ) *The words in italics on p. 100 indicate that , under certain conditions , might become imaginary .
But these conditions do not appear to be fulfllled in the case of correlations between mental tests .
Factors in Mental for values of from 1 to inclusive .
A comparison of equations ( 44 ) and ( 51 ) shows that , and are variables distributed according to the normal ] , with the same probable error as that of the .
Moreover , is independent of the , while the 's in equations ( 52 ) and ( 53 ) are independent variables , of which is a single general factor of the expressions on the left-hand sides of equations ( 53 ) , while is a specific factor of the expression on the left of the equation .
Finally , in equations ( 53 ) is proportional to given by , so that the correlations of the are the same as of the ; and the are distributed according to the normal law , with the same probable error as that of the We conclucie that if the correlation between every pair of columns in a table of coefficients of correlation between variables , , ( each of which has a normal probability distribution , with the same probable error ) is , then the difference between the 's and a constant multiple ( which becomes zero when Mr. Burt 's conditions for a hierarchy are fulfilled by the correlations of the ) of an variable ( that is independent of the , but distributed according to the normal law with the same probable error ) may be expressed in terms of independent factors ( distributed according to the normal law with still the same probable error ) , of which one is a single general factor , while the remaining are specific factors .
There are no group factors .
When equations ( 52 ) or ( 53 ) become the same as equations ( 35 ) , in which case the 's will be expressible in terms of independent variables , each of which is distributed according to the normal law , and one of which is a single general factor , while the others are specific factors .
Equation ( 42 ) shows that and will be zero whenever Mr. Burt 's conditions for a hierarchy are satisfied , unless indeed for all values of from 1 to inclusive .
It is evident that , if the correlations table ( 26 ) represent the correlations between a sufficiently large number of sufficiently dissimilar mental tests , the average correlation between any one test and all others will be zero , and and will consequently vanish .
It follows that , if the correlation between every pair of columns in a correlation table representing the results of a large number of sufficiently similar mental tests be , the measures of the correlated qualities can be expressed by means of equations ( 35 ) , in terms of a single general factor and specific factors only .
There will be no group factors .
Mr. J. C. M. Garnett .
On Independent We now return to the consideration of equation ( 18 ) , which gives the condition that three qualities , and should depend upon two independent variables and upon two only .
Our equation was : , ( 18 ) where , if X and are the two independent variables in question , the measurements of , Q2 , and are $(}iven by ( 54 ) Dr. Webb , in his paper on " " Character and Intelligence\ldquo ; describes an investigation of 48 mental qualities .
The subjects of his enquiry were 98 men students ( average age 21 ) at a training during the last six months of their second year of training ( January to July , 1912 ) , and a similar group of 96 students during the corresponding period of the following year ( January to July .
1913 ) .
Of the students ' 48 mental qualities investigated , 43 were estimated by pairs of prefects who acted as judges ; to each pair a group of 20 ( or 19 ) students was assigned .
The measurements of the qualities were so chosen as to the same constant to the frequency distribution of each .
* The 43 mental qualities estimated by the prefects included the following , described in Dr. Webb 's schedule under the head ' Intellect ' No. 3 Quickness of apprehension .
36 .
Profoundness of apprehension .
37 .
Soundness of common-sense .
38 .
Originality of ideas .
Among the five qualities not estimated by the prefects , two were objectively measured , namely , Examinational Ability and the general factor\mdash ; General Ability\mdash ; whose measure is The manner in which was measured is fully described in Dr. Webb 's paper .
Following Dr. Hart and Prof. Spearman , Dr. Webb described ' ' as a " " General Factor of Intellective Energy\ldquo ; ; S but Dr. Webb produces no additional evidence for regarding as a measure of intellective energy rather than as , for example , a measure of power of * See footnote , p. 93 .
, pp. 18 and 19 .
It will be remembered that , like the measures of the other qualities with which we are concerned , is distributed according to the normal law , and may have any value to , its mean value being zero .
For brevity we shall , for the future , commonly write ' ' instead of ' ' the quality whose measure is , p. 37 .
Factors in Mental rements .
voluntarily concentrating nervous energy , or , .
what amounts to the same thing , voluntarily concentrating attention .
The correlations of with the four intellectual qualities*just enumerated lie between and .
The qualities in question are not , therefore , functions of alone .
Dr. Webb observes that these correlations\ldquo ; are certainly high and significant , and among the highest correlations of with any of the estimates , but the question naturally arises why they are not higher\mdash ; the inter-correlations among the estimates themselves reaching in all cases over \ldquo ; Dr. Webb , noticing that the correlation of quality No. 8\mdash ; " " Degree of Sense of Humour\ldquo ; \mdash ; with was very small , while its correlations with the four qualities enumerated above are respectively and , tried the effect of eliminating the influence of " " Humour\ldquo ; from the correlations between and fvhe prefects ' estimates of the four intellectual qualities enumerated .
For this purpose Yule 's formula , obtained in equation above , was used .
The four partial correlations thus calculated were respectively , and .
It follows that not only ' ' but some other quality or qualities independent of ' ' one or more of which is connected with Humour , enter into the prefects ' estimates of the four intellectual qualities .
Moreover , the fact that , when the effect of variation of Humour is eliminated , the partial correlations between ' ' and Quickness , as well as between ' ' and Originality , are approximately , suggests that each of the four qualities\mdash ; ' ' Humour , Originality and Quickness\mdash ; depends upon two independent variables and upon two only .
Let us denote these four qualities by , and respectively .
According to equation ( 18 ) , if the first three of these depend upon two independent factors , and upon two only , .
( 55 ) Dr. Webb 's table of corrected coefficients between all the 48 qualities involved in his investigation ives ; ; so that ; ; If we take the same sign for the first two and the other sign for the last of these angles , we have , ( 56 ) which differs from zero by much less than the probable error .
We conclude * Here , and often in the sequel , the word ' ' quahty\ldquo ; is used for brevity , to denote the measure of the same quality .
, p. 43 .
VOL. XCYI .
I Mr. J. C. M. Garnett .
On Certain Independent that ' as determined experimentally , and HumoUl .
and Originality , as estimated in a very large number of cases by Dr. Webb 's collaborators , are compounded two , and two only , independent factors .
Let us now construct a by drawing lines Og , Oh , Oo in one plane , the following angles with each other:\mdash ; If we would see whether the axis of Quickness , Oq , lies in the same plane as the axes of ' ' Humour and Originality , we obtain from Dr. Webb 's table ; ; ; so that ; ; We obtain from equation ( 18 ) , if , as before , we take the two smaller angles with one sign and the larger angle with the other sign , We may take it , then , that the axis of Quickness lies very nearly , if not quite , in the same plane as the three axes of ' ' Humour and Originality .
Or , in other words , Quickness consists almost entirely , if not quite , of some combination of the same two independent factors as constitute the other three qualities .
We might therefore add to the another , Oq , coincident , or almost coincident , with Oo .
The angles would then be very nearly equal to , and , as obtained from Dr. Webb 's table .
If the Training College students who formed the subjects of Dr. Webb 's investigation constitute a fair sample of adult Englishmen , the , Sense of Humour , Originality , and ( probably ) Quickness of Apprehension of every Englishman can be represented by a single point on the plane of the diagram\mdash ; the intellectual plane , as we may perhaps facetiously call it .
And the proportion of Englishmen whose intellectual qualities ( if for the moment Factors in Mental Measurements .
we may confine this term to the four qualities just named ) are represented by points lying within any small area , would be equal to the volume of a cylinder having its axis perpendicular to , its base on , and bounded its other extremity by the surface , ( 57 ) where and are the measurements of the two independent variables and whose axes lie in the plane of the diagram , and where , as in equation ( 14 ) above , so that measures the degree in which subjects represented by points in are exceptional .
Since 'equation ( 57 ) is independent of the particular axes chosen , the number of individuals represented by points in is independent of the orientation of the radius joining to Since the equation ( 57 ) is independent of the particular axes chosen in the " " intellectual\ldquo ; plane , we may choose Og as one axis .
What , then , is the other axis , represented by a line Oc at right angles to Og ?
Evidently , since the Humour axis makes with it an angle of only some , it is very nearly identical with Humour .
Now Mr. McDougall has suggested that the process of reproduction by similars , or , as William James*called it , " " association by similarity is due to\ldquo ; a partial identity of the complex neural systems involved in the perception of two objects .
Each system consists of many sub-systems , and one or more of these sub-systems is common to the two .
When the one system is excited , its excitement spreads , not , as is most commonly the case , through some association-path previously established by some temporal contiguity , but from the sub.system , which rms also a part of another system , radiates itself through that other system .
In the commonplace type of mind this process comparatively rarely occurs .
It would seem that , in the brains of such persons , neural systems tend to become circumscribed and individualised , whereas in a higher type of brain the neural systems are more complexly interwoven , sub-systems becoming freely associated with many principal systems .
In a brain so constituted , reproduction of similars will frequently occur , causing the dull chain of simple redintegration , the serial reproduction of impressions associated by temporal contiguity , to be broken across .
The possessor of a brain so constituted will never be a commonplace person ; he may be a crank or an original thinker , or merely a wit.\ldquo ; 'Principles of Psychology , ' , p. 578 .
' Physiological Psychology , ' p. 139 .
I 2 Mr. J. C. M. Garnett .
On Certain Independent There is no evident reason why such a constitution of brain should have anything to do with capacity voluntarily to concentrate attention , the capacity which Binet and others have identified with the quality whose measure is .
Now the quality independent of ' ' for which we are seeking is very closely connected with Wit or Humour , and also closely connected with Originality .
Its correlation with Humour is , as we see from our diagram or 0.98 , and its correlation with Originality is or Let us call it Cleverness .
Then Cleverness is defined as a quality , independent of ' which , combined 'g ' in different proportions , wholly constitutes Humour or Originality , and wholly or mainly constitutes Quickness of Apprehension .
Being closely connected with Humour and with Originality , Cleverness as thus defined is also closely connected with the form of brain constitution described by Mr. McDougall in the passage we have quoted .
Moreover , so far as we can see , this form of brain constitution , like Cleverness as just defined , is independent of ' ' We have , therefore , grounds for identifying it with CleveIness .
If further investigation confirms whese conclusions and shows , as appears probable , that , of the two independent measured by and respectively and here described as Ability and Cleverness , the former alone is educable ( although its educability may be innate ) , the distinction between ' ' and ' c ' should have important consequences for education .
* The correlations of Cleverness ( ' with the 48 qualities investigated by Dr. Webb may be calculated as follows : Let us take the general factor measured by and Cleverness measured by as two of the independent variables in terms of which any of Dr. Webb 's qualities may be expressed .
Then measure of any of his qualities is given by equation ( 8 ) which now becomes , ( 58 ) where ate the remaining independent variables .
of the qualities represented in our , say Humour , is measured by , ( 59 ) .
Dr. Mercier 's essay on " " Cleverness and Capability 'Human 1917 .
added , Jfarch 2 , measured by , is a " " group factor\ldquo ; which , along with the general factor , , enters into the constitution of a group of " " intellectual\ldquo ; qualities .
From amy set of sufficiently dissimilar mental tests the correlation-producing effect of will tend to disappear , leaving the single general factor as the only common factor .
The same is true of the " " group factor closely connected with Purpose , discussed in the next section : it is an importanl common factor in a certain group of ' ' purpose\ldquo ; qualities , but its effect in any set of sufficiently mental tests tends to be negligible .
These questions are more fully discussed in a further paper just completed .
] Factors in Mental Measurements .
so that , according to our cosine , equation ( 16 ) , , from which we obtain .
( 60 ) All the 's on the -hand side of this equation are iven in Dr. Webb 's lable of corrected coefficients for the students .
If we substitute Dr. Webb 's values and calculate for all values of , except and , we obtain the required series of coefficients of correlation .
The ighest correlations of erness include those with Cheerfulness and Quickness of Apprehension .
It follows that Cleverness may be nised in practice\mdash ; as , for example , when interviewing for an appointment a candidate , to whose general ability ( g ) testimonials or examination tests bear witness\mdash ; by noting his Sense of Humour , eneral tendency to Cheerfulness ( which is perhaps difficult to judge on the occasion of such an interview !
) , or Quickness of Apprehension .
VI .
It be shown that the group*factor discovered by Dr. Webb to be independent of 'g ' and in some close relation to " " persistence of motives\ldquo ; or " " Purpose\ldquo ; is independent of Cleverness as well as of ' By adding a third axis at right angles to those of ' ' and ' c ' in the diagram on p. 108 , we may represent the extent to which a given individual possesses all three independent factors , and any other quality wholly dependent upon those three , by a single point in a three-dimensional diagram .
It will be found that , when a person 's " " representative point\ldquo ; in this diagram has been determined , surprisingly large proportion , but of course by no means the whole , of his moral as well as of his intellectual qualities have also been defined .
It is even possible to classify certain important types of character according to the region occupied in such a three-dimensional diagram by the representative point that corresponds to each character in question in so far as that character depends upon the three independent factors , two of which we have called Ability and Cleverness , and of which the third is closely related to Purpose .
preceding footnote .

