n

Contributions

III.

By Karl

to the

Mathematical Theory of Evolution,

Pearson", University College, London.

Communicated by Professor Henrici, F.R.S.

Received October

18,

Read November 16,

"Plates

1893,

5.]

Contents.
Page.
I.

On the Dissection of Asymmetrical Frequency- Curves.

Example

General Theory, 1-8.

Professor Weldon's measurements of the " Forehead " of Crabs.

71-85

9-10
II.

85-90

On the Dissection

of

Symmetrical Frequency-Curves.

General Theory, 11-12

Crabs " No. 4," 13-15

Investigation of an Asymmetrical Frequency-, Curve representing Mr. H. Thomson's
Application.

III.

measurements
Table

I.

Table

II.

First Six

Carapace of Prawns.

Powers

16-18

of First Thirty Natural

100-106

Numbers

106

Ordinates of Normal Frequency- Curve

Note added February

I.

of the

90-100

On

107

10, 1894

the Dissection

107-110

of Asymmetrical Frequency-Curves.

(L) If measurements be made of the same part or organ in several hundred or

thousand specimens of the same type or family, and a curve be constructed of which
the abscissa x represents the size of the organ and the ordinate y the number of specimens falling within a definite small range Sx of organ, this curve may be termed a
frequency-curve.
please, be

The centre

or

origin for

taken at the mean of

frequency-curve

may

all

the

measurement of the organ may,

specimens measured.

we

if

In this case the

be looked upon as one in which the frequency per thousand or

per ten thousand, as the case

plotted up to the

may

mean

be of a given small range of deviations from the

Such frequency-curves play a large

part in the mathematical theory of evolution, and have been dealt with by
Mr. F. Galton, Professor Weldon, and others. In most cases, as in the case of
errors of observation, they have a fairly definite symmetrical shape 4 and one that
mean,

is

of that range.

'"

# Symmetrical shapes

Part

II.,

may

of course occur

which are not of the norma] or error-curve form.

See

11 of this paper.

9.5.94

ON THE MATHEMATICAL THEORY OF EVOLUTION.

PROF. K. PEARSON

72

approaches with a close degree of approximation to the well-known error or probabilitycurve.

frequency-curve, which, for practical purposes, can be represented by the

error curve, will for the remainder of this paper be

of measurements

series

gives

something approaching a
impartially round

economic

the mean.

measurements

normal shape, and

such deviation.

rise

it

to a normal

condition

stable

termed a normal

production and destruction

is

In the case of certain

there

biological,

however, a well-marked

is,

we may probably assume

curve,

there

When

curve.

sociological,

deviation

from

and
this

becomes important to determine the direction and amount of

The asymmetry may

arise

from the fact that the units grouped

It may happen that

together in the measured material are not really homogeneous.
we have a mixture of 2, 3, ... n homogeneous groups, each of which deviates about
its own mean symmetrically and in a manner represented with sufficient accuracy by
the normal curve. Thus an abnormal frequency- curve may be really built up of normal

curves having parallel but not necessarily coincident axes and different parameters.

Even where the material is really homogeneous, but gives an abnormal frequency-curve
the amount and direction of the abnormality will be indicated if this frequency-curve
can be split up into normal curves. The object of the present paper is to discuss the
The equations for the
dissection of abnormal frequency-curves into normal curves,
dissection of a frequency-curve into n normal curves can be written down in the same
manner as for the special case of n == 2 treated in this paper they require us only to
But the analytical difficulties, even for the case of n = 2,
calculate higher moments.
;

are so considerable, that

it

may be

questioned whether the general theory could ever

be applied in practice to any numerical

There are reasons, indeed,

family probably breaks up

why

first

case.

the resolution into two

into

two

species, rather

is

of special importance.

than three or more, owing to

the pressure at a given time of some particular form of natural selection

ing to procure an absolutely homogeneous material,

we

in attempt-

are less likely to have got a

mixture of three or more heterogeneous groups than of two only.

Lastly, even

where the heterogeneity may be threefold or more, the dissection, into two is likely
In the case of
to give us, at any rate, an approximation to the two chief groups.
homogeneous material, with an abnormal frequency-curve, dissection
curves will generally give us the

much, then,

may

amount and

into

two normal

direction of the chief abnormality.

So

be said of the value of the special case dealt with here.

must be made between the two cases which may theoretically occur.
If we have a real mixture of two normal groups represented by our abnormal frequencycurve, then, theoretically, it is possible to find the two components, and these two
components must be unique, If they were not unique, a relation of the following kind
must hold for every value of x :
distinction

e
<r

lX/(27r)

20Y*
_

_|_

o-

-*J*2or
2^8
e
'C\

>/(27r)

/-f- ^
lyo

a?jX

(2tt)

'J

2v.J>

_L

lit

t^

o 4V/ (2tt)

PBOF.

K.

PEABSON

Between the

ON*

THE MATHEMATICAL THEORY OF EVOLUTION.

six constants on either side of this equation

an

infinite variety of

relations can be reached by giving x an infinite variety of values, and

by the same

impossible to satisfy this series

example, let x be very great, and suppose

1
"j>

s>

~L-j
1

a*!

The
jl

first

<r3

cr

is

or

For

set of values of the constants.

(*-fti) 2

~*

= -A 6

*iV

^2

x very great we have

an(^ putting

l<x{"

<r L

*J

~|

P.

W~<riV

<T A

*s

whence, proceeding to the

unless

seems

it

to be the largest of all the quantities

o- x

ia

Dividing by -jj^~\

&& anc 0V

73

limit,

cr4 .

impossible by hypothesis, therefore the latter

This gives us at once c x

c3

true,

say

Returning to the original equation, and making x large in

two terms become equal on

must be

it,

we

see that the first

Hence, the second two terms must become

either side.

equal as x approaches infinity, or

X2

2o-/

C4

ultimately, to c 3

Our

=c

e""

25

?,

same manner

this leads in the

may now

as before to

cr2

<x4 ,

and,

be written
(x-b 3 V

<r

4.

original equation

<

x2
2or4

CT,

CTc

Dividing again by

= _*

20"! 2

_>-&

'2

2cr,

(3

(3

lV / (2tt)

^sV

Put x = ^ (6j + ^3)? then the left-hand

must vanish, but this involves either

h=

2)

2<tf

( 2?r )

fa)-

side vanishes and, accordingly, the right

or

h+h=
Similarly, putting

-J

(6 2

&*)>

we

b2

64

find that either

61

=&

&3

=h+

3>

or

h+

Thus, either the two sets of components are identical, or

MDCCCXCIV.

A.

()

6,

"(a) is true,

THEORY OF EVOLUTION.

PROF, K. PEARSON ON THE MATHEMATICAL

74

Multiply equation

(tj)

above by

x,

x and x B
?J

respectively between the limits a and

(6 l

{So-!

- h) + 6 -

(&!

reducing by aid of

ft}

and

(a)

63) c

in succession,

We

a #
c

= (h
=

*i

results

find

/>

4:

and integrate the

c3

2)

{3a-/ (6 4

68 )

+ 6 - hi)
4

-.

(/3) ?

c 2f

to

(/3)

and

{15^^ - 5 + ^(V - & + V - bi) c

= {15o/ (6, - 6 + 10<r (6 - 6 + 6 3
3 )

3)

reducing by aid of

80-/

36/

8<r 2

3/>

(a),

determine b l9

h%,

bg

&4

and

(8),

(/3),
,

we must have

cr 3 ,

<?

=5

61

and

two normal components can break up

Now

never be the absolutely true

c2

62

sum

4& x & 3

4& 2 &j,

3&/

<x l3 cr 2 ,

hence

64

in one

sum

it

of

it

is

We

must

c 1?

c.

3>

(8),

.....

and

c2

But

(e).

5 1? 5 2 are in

a curve which breaks

is,

up into

way, and one way only.

abnormal frequency-curve
;

indeed, if

it

will

be not a mixture,

not necessarily a very close approach to

two frequency-curves of normal type,-

asymmetrical binomiaht

bi}

follows that (a) cannot in general

That

of two normal-curves

but an asymmetrical frequency-curve,

the

2
2

clear that in actual statistical practice our

it is

are four independent equations, which suffice to

and

l5

8)

3
-f 3/> 3

as definite functions of

general independent of o^,

be true, or

(e)

two forms,

(y) to the

2o-/

2a"!

Equations

and

(a), (/3),

g)

it

may

the limit to an

be

not, therefore, be surprised if

more than one

by any method of dissection. A mathematical criterion for dismight easily be given. For example, in the method
of the present paper, we might define that as the " true," or at any rate the " best,"
solution which gave for the compound-curve a sixth moment nearest in value to that
Such a theoretical criterion, however, may not have much
of the observation-curve.
solution be given

criminating the " true " solution

The values

of the successive

moments

of

the normal-curve are given in

5 of this paper,

and

permit of these integrations being performed at once.

f The general form of the limit to asymmetrical binomials

where

0,

c,

and

slightly fuller

ft

form

are constants, and x

is

is

to

is

have positive values only.

given in the abstract of this paper,

'

Roy. Soc. Proc.,'

/3

is

always positive.

vol. 54, p. 381.]

[A

PROF. K. PEARSON

OX THE MATHEMATICAL THEORY OF EVOLUTION.

For after we have made the areas and

practical value.

moments

curves identical, their sixth

closer together

be

will in general

the practical statistician,

if

moment

on other organs-suffices

any other convenient mode

in

of

(like their contours)

Added

than either are to that of the curve of observations.

the great labour involved in the calculation of the sixth

e.g.,

is

two

much

to this

sufficient to deter

measurement

results of

the particular case to discriminate between the solutions

Thus, while the mathematical

found.

moments

first five

75

we have

should be unique, yet from the

solution

compound curve which fits the

observations closely, and more than one such compound curve may arise.
All we can
do is to adopt a method which minimizes the divergences of the actual statistics from
a mathematically true compound. The utilitarian problem is to find the most likely
components of a curve which is not the true curve, and would only be the true curve
As there are
had we an infinite number of absolutely accurate measurements.
utilitarian standpoint

different

methods of

whether we
or error of

mean

normal curve to a

fitting a

start from the

square,

to be content with a

mean

depending on
"
or the median, and proceed by
quartiles," mean error
series of observations,

and as these methods lead

in

some cases

to slightly different

normal-curves, so various methods for breaking up an abnormal frequency-curve

lead to different results.

As from the

utilitarian standpoint

normal curve are obtained by finding the mean from the

first

good results

may

for a simple

moment, and the

error of

mean square from the second moment, so it seems likely that the present investigation,
based on the first five or six moments of the frequency-curve, may also lead to good
results. While a method of equating chosen ordinates of the given curve and those of
the components leaves each equation based only on the measurements of organs of one
size,

the method of moments uses

all

the given data in the case of each equation for

the unknowns, and errors in measurement

At the same time

it

will, thus,

would be of great interest

individually have less influence.

to discover

whether other methods of

dissection lead to results identical or nearly identical with the

adopted by the present writer.

however, occurred to him

two

(2.)

other method analytically possible has not yet,

nor any criterion for distinguishing practically between

solutions so close as those of

Weldon when he appeals

Any

method of moments

to the

figs.

and

2,

other than that adopted by Professor

measurements of a correlated organ.

In the case of a frequency-curve whose components are two normal curves, the

complete solution depends in the method adopted in finding the roots of a numerical
equation of the ninth order.

It is possible that a simpler solution

the method adopted has only been chosen after

many

each component normal curve has three variables

(i.)

" standard-deviation" (Gauss's "

(iii.)

its area.

Mean

trials

and

may be

the position of

Error," Airy's "Error of

Mean

Square"), and
its

component

Innumerable relations

of this kind can be written down, but, unfortunately, the majority of

l 2

Clearly

its axis, (ii.) its

Six relations between the given frequency-curve and

curves would therefore suffice to determine the six unknowns.

found, but

failures.

them lead

to

PROF. K. PEARSON

76

ON THE MATHEMATICAL THEORY OF EVOLUTION,

exponential equations, the solution of which seems more beyond the wit of

man than

that of a numerical equation even of the ninth order,

In any given example the conditions will be sufficient to reduce the suitable

(3.)

roots of this equation very largely , possibly to

conditions will be considered later.

two or even

one.

These limiting

suitable root of this equation leads to a

quadratic for the areas of the two component normal curves.

This quadratic

mental, and appears to be highly suggestive for the problem of evolution.

two

cases

its

We

have

roots are positive.

In this case the given frequency-curve

units of the frequency-curve

two

funda-

Both

(i.)

is

species, each of

which

may be

is

is

the

sum

two normal

of

curves.

The

considered as composed of definite proportions of

The

stable about its mean.

process of differentiation

here appears complete.

One

(ii.)

root is positive

and

The given frequency-curve

the other negative.

now

The
birthupon
as
the
looked
now
be
may
population (unselectively diminished by death).
The negative probability-curve is a
selective diminution of units about a certain mean
that mean may, perhaps, be the
is

the difference of two probability-curves.

probability-curve, with positive area,

possibly

average of the less "fit."

It

possible that in

is

w ill be found
r

teristic
if

to exists

some numerical

cases solutions of both the types

(i.)

and

asymmetrical frequency-curve, either only one type of solution will exist,

two types do

exist,

(ii.)

but I imagine that in most cases of a well-marked and characor,

then one will give a much better agreement with the actual

That the two types of solutions should exist side

by side occasionally is, perhaps, to be expected. In such cases we have examples of
groups, which are, perhaps, in process of differentiation into separate species by the
shape of the curve than the other.

elimination of
(iii.)

members round a

From the nature

selected mean.

of the problem, the case of both roots negative does not

occur.

We now

pass to the solution of the problem

Given an asymmetrical frequency -curve

to

break

it

up, if possible, into Uvo com-

ponent probability -curves, or into Uvo normal curves.

(4.)

Preliminary Definitions and Problems.

(i.)

Given any curve ABC, and the

line

y y\

if

we take

the

sum

of the products of

every element of area by the wth power of the distance of the element from the line

we form the nth moment

y'y'\

of the area about the line yy\

Clearly, if y be the length of a strip parallel to y'y

n th

y'y' 9

then the

A to C in

our
y dx, the integration extending all over ABC,
where the curve is always bounded by a straight line, AC, perpendicular to yy.

moment

case,

and x its distance from

n= \x

or from

MATHEMATICAL THEORY OF EVOLUTION.

PROF. K. PEABSON Otf THE

or 100 units., then the nth.

lfh be any standard length, say 10

H

order h a

if

a be the area

We

numerical factor.

of ABC.

It therefore equals p' n ti%

moment

where

/i n is

is

77

of the

a purely

shall invariably represent it as the product of these three

factors.

n moments about yy\ or the coefficients

to find the nth moment about yy or the coefficient fi n
Let the distance between yy and yy be d = qh, then
(ii.)

Given the

first

/i

[/%,

/x'

3,

// 4

\i n ,

fl tt

h H 0L

'x

d) n y dx,

or

U (n

In particular, since ^'

Mi

Mi

Ms

m's

12

7 (?&
'

A1

^-s

M5

When

the line

symmetrical about
the right of

yy,

or

-2)

1,

Ms
M4

1)0*

+ 6gV8
5<?m'i + lOgVs

m's

yY
yy
d

"

42m's

m'*

+2
lOgVg + 5<2Vi -T

42
-

0).

m'i

passes through the centroid of the curve, and the curve

is

we take yy

to

\x!

x,

ju/

3,

negative,

ytt'

are

all zero.

Hence

if in this

case

PROF. K. PEARSON ON THE MATHEMATICAL THEORY OP EVOLUTION.

!?8

/*l

<1

H = P* + f
H = 3 2^2 + <f
l

Pi

= H-\ +

(2)-

6<zVa

</'

The distance of the centroid of ABC from yy is the ratio of its first moment
fi\ha to its area a, and = [i-Ji.
(iv.) To find the successive moments of a given curve about a given line.
For the purposes of the present problem we require only the first five moments of
(iii.)

a curve like

ABC

about a line yy passing through

The

its centroid.

solution

may be

obtained either analytically or graphically according to the accuracy or rapidity with

which we wish to work.

(a.)

Analytically.

Suppose the frequency-curve to be obtained by plotting up

equal change in
individual

the

Starting from the

deviation.

point

corresponding to an

beyond

C,

which

no

we may have in practice, perhaps, 20 to 30 equal ranges of

we reach the point A, which terminates the deviations on the left.

The equal range being taken as the unit of length,

1, 2, 3, 4, 5

units of distance from

Then the nth moment

1000, and h

let

the numbers in the groups at

be y l9 y 2 yZ9 y^ y 5

clearly equals very approximately

x y

may
1"

X2/ 3

y3

y4

be conveniently taken

100,

^xy^

x i,

x yx

2-

x y%

Sufficiently accurate values can then be found for // 1; // 2

know

the

occurs,

deviations before

or since a

AC

1000 measurements, each unit of length along

results of

jx/

.,

3,

jx/

4 , fi

&9

provided we

the 2nd, 3rd, 4th, and 5th powers of the natural numbers up to about 20 to 30.

The values of these powers up to 30 are given later in this paper.

Knowing the first five moments about the vertical through C, we can find the
centroid by aid of (iii.) above, and then the moments about the vertical through the
centroid by aid of equations (1).
= for the centroid [jl\ = q and therefore we have the following to
Since
determine the other moments
fjb }

PROF, K. PEARSON ON THE MATHEMATICAL

THEORY OF EVOLUTION.

79

^2

+ 6gV' - 3?*
5g/*' 4 + 10? V' lOg^/Zj +

4^'3

^4

fa

rs
HUMUf

g J

(4

4g

5
*P*r

The centroid having been found, it may be asked Why we should not calculate
The answer lies in the fact that the centroid will not generally
fa, fa, fa, fa directly ?
coincide with a unit division on the deviation axis, and the powers to be calculated,
instead of being those of two place figures, become in general powers of numbers
containing three or four figures.
Thus the labour of the arithmetic is much increased.
(b.) Graphically. If the figure be drawn on a large scale, the moments may he
found with a fair degree of accuracy by aid of the following process, which has. long
been of use in graphical statics for finding the first, second, and third moments, of
:

plane areas.*.

It

is

required to find the

straight line

PP

first to

OCA.

Take O'y"

O'y from

AC

to

O'y' of the curve

parallel to O'y

ABC]

ABC, bounded by

and at distance

h.

Take any

the
line

the perpendicular from P' on O'y" meet

it

let the

perpendicular from Q' on O'y" meet

it

let the

perpendicular from

it

let

and let 0'N meet PP' in Q L

N", and let ON" meet PP' in Q 2
N"', and let ON'" meet PP' in Q s
f

in N',
in

moments about

Q2

on O'y" meet

In this manner a series of points Q l9 Q2 Q.6

Let
these
points be determined for a series of positions of
Q& Q&
PP' taken at short intervals from C to A, then all the corresponding Q being joined,
we obtain curves termed respectively the first, second, third, fourth, and fifth moment-

in

are determined.

The

third

moment

spindle about its axis.

of a plane area

is

used in determining graphically the moment of inertia of a

is sometimes attributed to Collignon, but seems to have

The method described

been long in use to find " equivalent figures " in the case of beam

sections.

ON THE MATHEMATICAL THEORY OF EVOLUTION.

PROF. K. PEARSON"

80

^i* ^2*

AQ L C AQ

Let the areas

curves.
^3

J.

line

off

with a planimeter, and be

(4)

/V

= a i/0

^/

a6 / a

construct these curves with great readiness, and

sufficiently large scale, the results

about a

L 2 C, &c, be read

iien

A good draughtsman will

Equations

may

if

on a

be read to within the one per cent, error. *

then enable us to complete the problem of finding the moments

through the centroid.

so the centroid determined

Or, the

we may

first

shift O'y

moment being found about 0'y\ and

passes through the centroid,

till it

and then proceed to find /x 2

In this case care
/x
directly in the above manner.
5
will have to be taken in reading the areas of the moment-curves, which have now
.

pieces of their areas negative, to carry the planimeter point, in the proper sense, round
their contours.

(5.)

Properties of the probability -curve.

Let the equation to the probability-curve be

Then or will be termed its standard-deviation (error of mean square).

total number of units measured, or the area of the probability curve.
(i.)

To

find the second

c is the

and fourth moments of the probability-curve about the axis

of y\

Let them be

'

and

'.

Then
a

M/=2
M/ =
Clearly
[*

My

'

and

'

yx dx

=c

o-

yx 'dx

=c

3cr

9,

*a

2
J

2
.

are zero.

demonstrator, Mr. GL U. Yule, has graphically calculated the

of statistical frequency-curves, with the object of fitting

footnote, p. 74).

The method

is sufficiently

them

first

four

moments

of a

number

to the generalized probability- curve (see

accurate in practice, and I hope soon to have an instrument

to construct these curves mechanically, designed

by him. February

9,

1894.]

PROF. K. PEARSON ON THE MATHEMATICAL THEORY OF EVOLUTION.

Now

(ii.)

a be a standard area and h a standard length.

let

81

Let us use

IBr
Equations (2) of Art. 4

taking

(ii.),

curve, and yy at a distance b to the

as the axis of

yy
left,

symmetry of the

probability-

then

= be.

flJlO

+b
(36o- + V)
(3o- + 66V + 6)
(15cr 6+ 106V + 6 )c.
(cr

) C.

c.
2

4,

c.

Now

let

cJol

alb

z,

Then 2, -w, and y

moments round yy

= u,

and

b/'h

= y.

are purely numerical quantities,

M
M
M
M
M

= yzah,

==y%(l+^)aA3

(6.)

We

are

now

(^).

in a position to write

for the first five

= yh(l + 3u*)ah\
= y% (1 + 6w + 3tt aA
= y 2 (1 + 10^+1 5^ oA
s

and we have

down the

equations which give the general

Let the deviation-axis of the asymmetrical frequency-curve

solution of our problem.

be taken as axis of x and

let

the axis of y be a perpendicular on this axis through

Let

the centroid of the frequency-curve.

this centroid

and the

first five

about the axis of y of the frequency-curve, i.e., 0, 2 /x 3

either analytically or graphically by the methods suggested in Art. 4
coefficients

Then,

by the

if

/x,

5,

momentbe found

(iv.).

the position and magnitude of the component normal curves be given

quantities b l9 c l9

cr }

and

&2 c 2
,

7v

we

/x 4 , /x

have, since

MDCCCXCIV.

moments round the

A.

z i>

<x 2 ,

u i>

or the corresponding numerics

an(* 7z> z z>

Z9
i

vertical axis are clearly additive-

PROF. K. PEARSON ON THE MATHEMATICAL THEORY OF EVOLUTION.

82

+ c = a,
(Yl z + 7&) ah = ,
{y \ (1 + Mj + y \ (1 + w )} aW = ^ a/i
(! + 3m
+ y 2 (i + 3% )} ^ = A
{ri
{y^ (1 + 6< + 3<) + yA (1 + 6% + 3%*)} '** = /*M*.
(1 +
{yi
V + 15 V) + y/% (1 + 10 + 15?/,/)} A =

C]

2
2

3z i

fi 8

The
its

first

/x

a/*

6
.

equation here represents the equality of the areas of the resultant curve and

components.

to find the six

Reducing to the simplest terms, we have the following

unknowns, z x z3 y x ya u lt u%
,

yj *! (1

y l \(l
5

y,

(1

yi%( 1

"*""*"'

es*09se

|
.1-.

......

+ V) + ya%(! + V) = ^a

+ 3<) + y % (1 + 3^) =
3

+ 6u + Su^) + y,\(l +
"

+ 10< +

six equations

15V)

/x

=^

15V)

(1-0).

-(H).

6ui+3u^)

+ jth (1 + 0% +

/p\
\J Fe

/*s

(12).

13 )-

Equations (8) (13) give the complete solution of the problem.* After several trials,
I find that the elimination of z v %, u l9 u 2 from these equations, and the determination
of equations giving y xy3 and y x
y.z appear to lead to a resulting equation of the

lowest possible order.

(7.)

Eliminating

z2

between

(8)

and

<V"|

Similarly

we have

(9),

"~

solution.

attempts to obtain a simpler set have failed.

portions of area, or of

be found.

7i

7i

my

"*"

**

/*

* All

moments round the

It is possible,

axis of

a?,

all

/,

r\

7s

Equating of selected ordinates, or of selected

appear to lead to exponential equations defying

however, that some other six equations of a

less

complex kind

may

ultimately

PEOF. K. PEARSON

ON THE MATHEMATICAL THEORY OF EVOLUTION.

83

Equations (14) and (15) clearly give the numbers in the component groups so soon
as

yx and y3 are found.

Substituting these values of

determine

u^ and u^

in

and

zx

terms of y l9 y2

Solving them

we have two equations

(10) and (Ll),

z% in

we

=-

y Uf
x

to

find

(16).

7172

72

7173

These equations clearly give u x 2 and u%, and, therefore, the standard-deviations of

when y x and y3

the component groups

are known.

For brevity, put

3

V2

= Tl + 72.

2>L

(y*^)

jPa

Tir-2-

Then

% = /*2-l3Wy2- "SWl +^2

V2

2 (y x

-yV +
2

MsM

v,

^-

2)

Vc

^-

( yi

% - y %) + 3 - <) =
3

yiS

y*v = (yi y)
%

- yaX

(yi

- ys)

-y

t^ 2

(19):

2)

in (20)

and

V7l

"~*

72/

I
L

"72"

- Pi +
8

Ms

9P1

i'2

iJ 2

2^

-i> 3 -2/, 2
3

M2P1

9(^ 3

6p 2

_2

5/V^i
2

(21).

+ l^^

+ H
9

iJ 2

/**)

0.

y?2

2O/X3

(20),

tPiPa j>

whence,
a3

We find

(21).

72

^23

/V
-~ = (n- y )^-J
+ i^ + iPi

i/* 4/p 2 }

/^ te + P2

Pi
- *^^
+

/*2JPi

- y {|p^ - \p* - } H \p^

( 7l

7)

j/*a "~ 3

^-

|ip
2)

7i

+ P2

72

71

"We must now substitute (18) and (19)

7i v i

iPl72

"J

we have

while from (12) and (13)

2 (y x V x

/*2

(18),

2p x

4 PlJp 2

i^2

15 (2^ 2 ^3
^3

aO

0.

^3

3 JP1P2

84

PEARSON ON THE MATHEMATICAL THEORY OF EVOLUTION.

PROF. K.

Write

= 'W

\t

3 /^>

= 30 ^2/x

^5

3 /x 5

22 )>

6P2 8

24 )>

and put

then, multiplying up, the above equations

^ ~ ^Hlh
5/x 3

From

p3

- 2p + 4p
3

these equations let us

become

2^3 S

p2

- Va +

20/x 3

p2

- X p =0

first

find

p3

in terms of p%.

+ Ps (W +

hPz

2p 3

(25).

Multiply the

by p 3 and

first

subtract from the second

HPz
Multiply (24) by

W+

W+

(-

Ps

and add

2/x 3

to (26)

X4:p 2

we

2j9 3

3
)

20/iaPg

- X^* =

(26).

find

p -

2/x 3 X 4

X 5^2 3

X 5p/

8/x 3 p 2

0,

or

ql

^-

Hence, so soon as p 2 is known,

the two roots of the quadratic

p
7

Eeturning to
containing

jpa

24^ -

28X4 p/

+ (288/*

4
3

~Pi7

is

the following one

36/is

-12X,X 5/x 3

order, one real root

and X 5

= 30/x

3//,/,

2 //, 3

while

V-

-X^

jp a ,

as

//,

3/x 5 ,
3

3
3

97 \

can be found, and then y x and y3 will be

+ P2 =

.'

'

(28).

and we have an equation

Ws - 10A/)

we

and

/x

see

we should

JPs

- (148^% +

2X6 );p a*
(29).

may always be

ninth order vanish except the

Thus

which the whole solution of the problem now turns.

Some remarks may be made on

(8.)

^4 =

= p /p

(27), substitute this value of p<d in (24),

only, on

This equation

g^S^jPg

this equation.

found.

Since this equation

Further, remembering that X4

that in the case of a normal curve,

0, all

is

of an odd

9/x 2

for

3/x 4

which

the coefficients of the above equation of the

first.

naturally expect, will be zero.

increasing symmetry, the coefficients become small,

it

Accordingly, since, with

will be needful to

work

their

values out to a greater degree of exactness the slighter the degree of asymmetry.

ON THE MATHEMATICAL THEORY OP EVOLUTION.

PROF. K. PEARSON

Given that a frequency-curve

(29), (28), (27), (14), (15), (16),

We may
Stage

is

and

compounded of two normal

(17)

curves, equations

form the complete solution of the problem.

throw the whole solution into the following form

L Find

85

the centroid of the frequency-curve and calculate

/x

3,

/x

3,

/x

4,

/x

3,

X4

and X v
Stage

Stage

from

Solve (29) for y and find the corresponding values of p from (27).
III Find the positions of the axes of the component normal curves

II.-

(28).

Stage IV.-

The

and

fractions z Y

that the areas of the normal curves are of

the area of the frequency-curve are the roots of the quadratic

Pi

Stage V.

Since

crjh

= */v

at once on substituting in (18)

and crjh

and

(30),

= \/%

^he standard- deviations are given

Pz

z*

4p2

'

(19).

The whole method may be illustrated by the following numerical example

Breadth of "Forehead" of Crabs. Professor W. F. R. Weldon has very kindly
They are
given me the following statistics from among his measurements on crabs.
The abscissae of the curve are the ratio of "forefor 1000 individuals from Naples.
head " to body-length, and one unit of abscissa = *004 of body-length. No. 1 of the
The ordinates represent the
abscissae corresponds to 580 *588 of body-length.
:

(9.)

number of individual crabs corresponding to each set of ratios of forehead to bodylength.

Thus there was one crab fell into the range '580 *583, three fell into the
range 584 *587, five into the range 588 '591, and so on. The average length
#

of animals measured 35 millims,, and measurements were recorded to

Abscissae.

Ordinates.

2
3

5
2

5
6

10

13

19

20
25
40
31
60
62
54

10
11
12

13
14
15

Abscissas.

16
17
18
19
20
21
22
23
24
25
26
27
28
29

'1

Ordinates.

74

84
86
96
85
75

47
43
24
19
9
5
1

millim.

PROF. K. PEARSON ON THE MATHEMATICAL THEORT OF EVOLUTION.

86

This curve

is

plotted out as the dark continuous line in Plate

1,

fig.

1,

and

is

moments in the analytical

method suggested on p. 78 (a), each calculation being made twice independently.
I took h = 1, and clearly a = 1000.
The moments were taken about the vertical
and
through, the point 0,
were calculated by the aid of Table I. of the powers of the
first 30 natural numbers given at the end of this memoir.
The following results
I proceeded to calculate its first five

clearly asymmetrical.

:-

were obtained

=
=
^=
=
=

16799

fL{

304-923

'

/* B

/x

116,061*435

'

2,385,609-719

ix 5

fii,

since h

1, is

clearly the distance of the centroid vertical of the frequency-

curve from the origin O,

The moments about

There resulted

5,831759

'

i.e.

= q of p.

77

(ii.).

this centroid vertical

were now calculated by aid of

(1), p. 77.

=
=Pi =

1576-533,413

85'205 407

7920'604 761

22-716,599

1jl.

53-874,770

ix s

X5

where X4 X 5 are given

,

in terms of the fis

Turning now to the fundamental nonio

by

(22) of p 84.

(29), let it

be divided by 24, and written in

the form

p*

+ %p* + <*& + %Ps + 5^2 + a *pi + a i?i + a %v* + a

Then the

coefficients a s

a3

a3
a4

a6
a7

a8

a9

were calculated, and the following values found

=
=
=

a 6

o.

99-406
4,353-742

423,696

=
=
=
=

t5

t/

/ \J

uy) DO

jZii/Oji/

X1

1,232,409,400

957,080,900
24,451,990,000

ON THE MATHEMATICAL THEORY OP EVOLUTION.

PROP. K. PEARSON

87

pz =

9
10x and divide by 10 we then have for the fundamental nonic the
following equation, where only three decimal places are retained

Put

"

994X7

+ 4'354 x - 42-370x - 37-029 x +

9'571x - 24-452 =
4

fi

119-299x

123-241X

0.

After a somewhat laborious calculation, the values of Sturm's functions f{x)>

fx (x)> f% (x)> /s (x)> Mx)> /s (x)> /a (x)> fi (x). /s (x). /o (x) were ascertained
the following results
:

./(/l

/.(

oo

=+

oo

4-

oo

A
f.

/
/
/
/

fl

co )

oo )

7 ( )
8 (
9 (

<X)

CO

=+
=+
=+
= =:

)
)

3=

/.()
/

/ (_Q0)
/ ()
7

3 real roots.

Two

roots

One

root between

Xi

Xa

Xz

With

1,

-5,

'

sufficient accuracy

and

between

and

successive approximations, I found

To

=+
==+
=+

6 changes.

These three real roots were then localized as follows

As

=+

/9(

3 changes.

Thus there are

-65,

3rd

-870,
-670,
422,

-8757,
-6724,
4170.

for the values of p.z

= 8*757.
i>a= 6724.
4-170,
_p =

1st solution, jp 3

Xi an d Xa*

1, Xs

-89,

we may then take

2nd

1,

40,

5,

and gave

88

PROF. K. PEARSON ON THE MATHEMATICAL THEORY OF EVOLUTION.

Discussion of first solution.

p. 84,

and then^j

The quadratic

= p /p
3

p2

There resulted

found.

p 3 was

'8*757.

calculated from (27) on

first

1*027.

y l9 y2 which are here identical with b lf 6 2 (the distances of the

centroids of the component probability-curves from the centroid vertical of the
for

frequency-curve),

is

:
y

whence
y,

The values of z and

}

V027y

8757

3-517,

0,

y3

2-490.

were now found from (14) and (15) of

zl

%=

-4145,

p. 82.

'5855,

thus the numbers of individuals in either group are respectively

ex

414*5,

c2

585*5.

The values of the standard-deviations, or l and cr2 were now determined from
3
At the same time the
(18) and (19), where, since h = 1, v x = cr^, and % = cr3
maximum ordinates of the component probability-curves, y l and y25 were found from
,

ci

Vl

- ^(27T)

%
v/(27r)

O",

Org

There resulted
o-!

y1

Thus the

1st solution

1st

cr
2

uo

-*

as follows

= 4*4685,
= 37-008,

may

be

summed up

Component.

=
by =
=
yi =
Cr

o-y

3*1154.
4fc

Jomponent.

^jllCt v.

414-5,

ca

3-517,

63

4-4685,

er 2

37-008,

2fe

= 585-5.
= 2-490.
= 3-1154.
= 74-976.

These two normal curves were now drawn by aid of the Table
calculated afresh for this purpose from the exponential/*
in

fig. 1,

remarkably

close

curve, an agreement quite as

* I have always found

it

agreement with the

close as

more convenient

to

we

which was

These curves are plotted out

and their ordinates added together give the resultant curve.

this curve is in

II.,

It will be seen that

original asymmetrical frequency-

could reasonably expect from the com-

work with the standard- deviation than with the probable

error or the modulus, in terms of which the error-function

is

usually tabulated.

PEARSON ON THE MATHEMATICAL THEORY OF EVOLUTION.

PROF. K.

number

parative smallness of the

and the resulting

fact

can at best only be an approximation to the true

the observation-curve

that

of individuals dealt with,

89

resultant.

2nd

Solution.

p =
2

Precisely

and

6*724,

similar

undertaken

for

the value

accordingly, be sufficient to cite the final conclusions

will,

it

were

calculations

here.

Quadratic for y

1st

Ci

61
cr l

*3412y

0.

Component.

=
=
=
=

2nd Component.

= 532*8.
2*428.
b =
=
4*7702.
44*559.
ya =

467*2,

c2

2*769,

2*878,

cr

64*764,

These component-curves are drawn

We

6*724

in fig.

and their ordinates added together.

2,

we have again broken up our asymmetrical frequency-curve into two

probability-curves, whose sum is a very close approximation to the original curve.
see that

3rd Solution

While the
(p 2

=yy
{

same

i2

p%

first

of the

4*170.

two solutions have been additive,

same

this solution

makes y and y 2
{

component curves

sign, or the centroids of the

side of the centroid vertical of the frequency-curve.

fall

both on the

Accordingly the area of

one of them must be negative, and the solution promised to be a subtractive one,
to represent the frequency-curve as the difference of

Determining

p3

and then p from

L

The

(27),

we

3'605y

find

two normal

^=

4*170

we cannot

difference of

So

(10.)

biologist's

curves.
;

hence

0.

In the case of crabs' foreheads,

roots of this equation are, however, imaginary.

therefore,

3*605

represent the frequency-curve for their forehead lengths as the

two normal curves.

far as the nonic is concerned, our

work

measurements and assuming them

is

now

accomplished.

Taking the

to be the chance distribution of

unequal groups about two different means, then one or other of our solutions
correct answer.
/jl

Applying the test of the sixth moment, we find

177,004, while for the

192,446.

i.e.,

first

According to this

test,

solution

the

it is

first

188,099 and

solution

is

for

is

two
the

for the observations

the second solution

the required

one,*" but, as

we

have noticed, the two solutions are themselves much closer together than either to
*
the

The theory
first

of correlation will here, perhaps, confirm this result.

and not the second solution

MDCCCXCI V.

A.

is

Professor

Weldon

in good accordance with his other measurements.

tells

me

that

THEORY OF EVOLUTION.

PROF. K. PEARSON ON THE MATHEMATICAL

90

the observations (see

In

75).

p.

most normal curves

ments of

fact,

the contours of the compound-curve for both

and neither

solutions are very close together,

differs

more from the observations than

from symmetrical frequency-curves in

differ

statistical

measure-

this kind.

The contours are

we have demonstrated a theoretical

p, 72, et seq.), we see that, from the

so close that, notwithstanding

uniqueness for the solution of the problem (see

standpoint of practical statistics,

it is

possible for the given material to be broken

up

Thus the problem indeed becomes somethe asymmetry of the frequency-curve becomes much

into more than one pair of normal curves.

what arbitrary at any rate till

more marked than is the case with that of the foreheads of Naples crabs. Indeed,
although the method adopted leads to only two solutions, it is quite possible that
pairs of component normal curves might be tentatively found lying in the neighbourhood of those determined by the above solutions, which w ould give resultant-curves
r

fairly close to the frequency-curve.

trials

Professor

one such solution, but this solution

moments from the observations

Weldon

differs

widely in the third and higher

cannot, therefore, be considered to have the same

it

by the present theory.

justification as those given

had, indeed, found by repeated

Granted that the

vations represent a mixture of tw o species varying about their

original obser-

mean according to
Without corre-

exact normal curves, our method gives two solutions and two only.
9

lated measurements,

it

might be

difficult to discriminate

any rate from the standpoint of practical

test of the sixth

moment

decides for the

between these solutions at

The perhaps

statistics.

over-fine theoretical

first solution.

II.-The Dissection of Symmetrical Frequency-Curves.

(11.)

Another important case of the dissection of a frequency-curve can arise, when the

frequency-curve, without being asymmetrical,

two components,

i.e.,

is all

sum

the more interesting and important, as

to occur in statistical investigations,

then

consists of the

when the means about which the component groups

This case

are identical.

still

or difference of

are distributed

it is

not unlikely

and the symmetry of the frequency-curve

in itself likely to lead the statistician to believe that

he

is

is

dealing with an

example of the normal frequency-curve. It seems to me that without very strong

grounds for belief in the homogeneity of any statistical material, we ought not to be
satisfied

by

its

representation by the ordinary normal curve, simply because our

results are symmetrical

and

fit

the

to ascertain whether or not they

normal curves.

This, at

any

normal curve

would

r
of our material, although possibly our test for tw o

homogeneous, but because

*

Symmetry might arise

means of the components.

its

heterogeneity

in the case of

In

this case,

well.

sum

We

ought

first

two
stage to demonstrating the homogeneity

fit still

rate, is a first

fairly

is

better the

may

fail,

or difference of

not because our material

is

*
multiple rather than double.

compound frequency-curves, even without identity of the

for two components we should have for different means,

OE THE MATHEMATICAL THEORY OP EVOLUTION.

PROF. K. PEARSON

We

will

now modify the

91

results of our previous investigation to suit the case of an

asymmetrical frequency-curve which has arisen from the superposition of tw o normalr

curves having the same axis.

(= u^),

i\2

o-Jh

identically satisfied,

(=

In this case

we

if

unite,

we have

3 y3 ) in Equations (8) to (13)

and

(8), (10),

-and (12)

&3

0,

(11)

(9),

v{

<rjh

and (13)

become

g^ I

.-j

bx

..........

JL

(ol j,

equality of component group-totals and of their standard-deviations.

This equality seems less likely
than equality of means and divergence of totals and standard-deviations.
Should it exist, however, we
fall

back on a sub-case of the general case we have already dealt with.

(8)-(13), put z1

= %,

Yi

*i

Yg>

We

need only, in Equations

= % and we have

U\

=H= i

(1

7i

+"i 2 )

= /^

7i

&<i

+ 3 V) =

/t

4,

whence
7l

^- ITCW^)

>

or,
cl

^2

j1

= -j = i/w-ftU

rt

.-{#"f1U^The

depends on

possibility of the solution clearly

The following

an example

is

is

Book on Least Squares.

very far from a normal-curve.

H\=

6-482

^3

of

ju,

indicates general

We have

g1

=c

and

symmetry

53

He

2-486

assuming then that the shots were

fired in

two groups

accordingly

= - & = 1-082,
= 1*1'^>
a =
1000 shots as a whole a = 1*577.]
3

For the

*2

Allowing for a uniform error of defective sighting amonnting to

fitting closely

would

does not seem to have noticed that

almost exactly.

fy

divisions 5

^= -104
^= 15793.

The smallness

/*!=

with equal precision, I find

/i

I find that for these observations

= 44-502
/ = 320-582
y = 2405-094
/*'

being greater than

Mr. Merkiman gives some results for American

of this special case.

target practice, on page 14 of his Text

the resulting- curve

3/li^

and

arise

Mr. Mjerriman's
7,

and not at

from a change

figure,

that of 6.

'482,

we

find a

compound-curve

and indicating that the gun was aimed

at the centres nearly of

Six was possibly white, 5 and 7 black.

Like results of course

of sighting about midfiring.

1ST

iu

92

PROF.

PEARSON ON THE MATHEMATICAL

K.

<vi

^1^1

Clearly

we

require one

f .i-o^o

^i

%^2 ==

At

more equation.

.........

Wo

A1 4

3"

TILUO.UY OF EVOLUTION.

sight

first

'

it

is

the

sum

\^3).

might seem that a fourth

equation would come readily, from the fact that the mid-ordinate

curve

'

(oZ),

of the frequency-

component probability-curves.

of the mid-ordinates of the

This leads to

LI+ /;oV" = m
/v

or

7;;+7k= M

'

<

34 >>

if

m =

mA/a.

\/(27r)

But besides the disadvantage of throwing our solution back on the correctness with
which we may have observed measurements of one size only, namely, the mean, the
result of eliminating between (31)-(34) leads to an equation of the eighth order. To
avoid this, it seems easier, as well as more accurate,* to take as the fourth equation
that obtained from the sixth moment.
Let n ah G be the sixth moment of the given frequency-curve about
symmetry, thent
C)

[jL
(

.ah G

Wi*
The solution of
Eliminating

have, writing

Zl

h
W
l

^=

and (35)

(31), (32), (33),

% we

K
K-

locr/V^

or,

=v
'%)
IV 2 )

is

2
,

l5or 2 c 2

(35).

easy.

w.2

= v/,

Hw
- /Xo^o,
=
*>

-l-fi,h

whence
__ T5
if;
j

-V

3 W>2
Mi - Ma'^s

Mfl

~ ^2^2
^2 ~ " 2

3 M-l

* Because our equation then depends on all the observations.

t Generally,

if

2r

be the 2r

moment

of a probability-curve about its axis

M =
2r

or,

2r

(2r

(2r

1) (2/-

I) ^M,,._,,

- 3)

axis of

T5

its

5.3.1ff 2r c.

PROF.

PEARSON OK THE MATHEMATICAL THEORY OF EVOLUTION.

-E.

93

Thus

The two

roots of this quadratic are clearly

w l and

w%, so that the

complete solution

is

CiL

Ot

w w
l

where

wa

and

(i.)

/x

the

positive,
(iii.)

is

lie

/x

of

between

cx

l9

we have

< w2

positive, or

Obviously
zero,

w*

w and w2
Y

is

(^ 2 /x 4

|/x

6)

w -

(^ liwa)

"

86 )

if

and

cY

c2

are both positive, or the frequency-curve

is

the greater component group

is

curves.

positive

and

c 2 negative, or

then a real difference solution.

c x is

negative and

we have
if

n\

note several general points about these equations.

two normal

>w

fx %

become

3^ 2

be the greater root, then

sum

(ii.)

Now we may

Let

Ow

are roots of

(fa

(12,)

Gn

^=

c 2 is positive, or

again the greater component group

a real difference solution.

3/x 2

2
,

and

/x

5/a 2 jlc 4 ,

the coefficients of the quadratic (36)

but these are just the conditions which would be

frequency-curve were a true normal curve.

This gives

for all

satisfied if

all

the

practical purposes a very

sufficient test of whether a given symmetrical frequency-curve

is

a true normal curve

If 4 be not equal to 3/x 2 and /x 6 be not equal to 5/x3 4 then we have no right to
assume that a symmetrical frequency-curve refers to homogeneous material. We must
then investigate whether a better result cannot be obtained by treating it as two
superposed normal curves having the same axis.
3

ju,

/z,

The quantities

_
I propose to call the excess

the excess of one-third

moment

/^j-3/V

and

,j

&

5 /^4

defect of the frequency-curve.

of the

fourth

moment over

the defect measures the defect of the fourth

ratio of the sixth

moment

to the second moment."

are used in the algebraic sense, and

5 '"

may

The excess measures

the square of

moment from

Here

take either sign.

The introduction of the factor l//* 3 3 into both excess and

tinguished from an absolute measure of divergence.
*

""-

defect

the second
one-fifth the

" excess" and "defect"

They appear

is to

to be a

good

preserve a relative as dis-

94

PROF. K.

measure

PEARSON ON THE MATHEMATICAL THEORY OF EVOLUTION.

for practical purposes of the divergence of a given symmetrical frequency-

curve from the normal type.

We

may now

express the quadratic (36) in terms of

and

results according to the character of the excess

and

e^

e2 ,

and analyze the

defect.

The quadratic becomes

w
W
%

+
fa/
fa
X

3ci(

2,

3 l (l

e2

el )

= 0.

This gives

_____

e3

^/{(e z

6 ei f

36e_ 8 }

^2

We have
x

(i.)

also be

the following cases

and

e2

Then the values of w are both real, but they must

and o-2 would not be real It is necessary, thereotherwise

both positive.

both positive,

that

fore,

of
e2

(ii.)

ex

and

e2

both negative.

<

3e 1 (1

Then w

Cl ).

be real

will

if,

when

v/(-*i)<l>
(

2)

does not

lie

between

6(-e

){l

+ /(-
v

)}

and

6(-i)

(l-y(-

ei )}.

If

then

we must have

(-ea )>6(Further, in order that

w may

+ a/(-i)}-

{1

ei )

have both values

positive,

we must have

(-e3 )> {-e2 -6(-e )r-S6(~, f

l

or

(-e a )>3(-e
This latter condition

is

){l

the other hand,

if

-(-<*)}.

clearly satisfied if

v/(-ej)

On

6e l

>

J.

PROF. K. PEARSON ON THE MATHEMATICAL THEORY OF EVOLUTION,

95

easy to see that

it is

(-0(1- (-e,)}

than

is less

6(-c ){l-v/ (-^.)}.

]

Hence, our

final conditions are

\/(-i)> h
then

(-e3 )>6(-6 ){l

1

but

+ y(-e )};
1

if

then either

(or

it

must

lie

6g

)>6(- ){l +
1

v/(-i)}.

between
3

(-0(1

-(-e,)}

and

6(-e
X

(iii.)

positive

and

negative;

e2

){l

if

v/ (-^)}-

the values of

are real, one

must be negative,

the values of

%o

are real, one

must be negative,

and therefore the solution impossible.

(iv.)

ex

negative and

e.

positive

if

and therefore the solution impossible.

Thus we conclude
If the excess and defect are not zero, the frequency-curve, although symmetrical, is
not normal. If the excess and defect are of opposite signs, then the frequency-curve
cannot be broken up into the sum or difference of two. normal curves with common
The frequency-curve, if compounded of normal-curves at all, is of a higher and,
axis.
more complex character. If the excess and defect are of the same sign, then,
provided certain relations hold between the numerical values of the excess and defect
:

given in

(i.)

and

(ii.)

above, there

is

a real solution of the equation

w hich
T

resolves the

frequency-curve into two components.

(13.) I propose to illustrate this

example.
details

crabs

at

Professor

Weldon

discussion

by the consideration

has kindly complied with

my

of a numerical

request for the numerical

of the most symmetrical curve deduced from his measurements of Naples

by placing the following statistics for a shell measurement No. 4 of his series
my disposal. The resultant-curve and the corresponding normal curve are

pictured in

fig.

3 (Plate

3).

Clearly, from the ordinary statistician's standpoint,

we

could not expect a more symmetrical result, or a closer graphical agreement, with the

normal curve.
as

we

But

is

this a real or

The answer is,

be put on Professor Weldon's results.

merely an apparent agreement?

shall see, vital for the interpretation to

96

PROF, K.

PEARSON OK THE MATHEMATICAL THEORY OF EVOLUTION.

Crab Measurements.

No.

Number

(Total

4.

_.,..._,..-.

of Crabs
~

'".
.

Ordi nates

xv 0dC1SSo6

(1 unit

5
11

4
5
6

40

98
121
152
147

55-

8
9

10

The

first six

of Table

I.,

moments were

except that a

h equals unity as

Pi
P$

20

now

equals 999, and

we go a

Pl
[l

Pg,

=
=

by aid
/x,

Pf,

for the position of the centroid

= y^ = 27403.

7-5092

==

Pi
Pa ~

2,150,845-6867

<r

9,

stage further to // 6 and

have

101-3022

These results give

standard-deviation

1 crab).

13,334-0710

"

72
41
28
8

165,488-8438
p6

126
82

1,129-9971

(i unit

11
12
13
14
15
16
17
18
19

9-684,684

"

calculated exactly as in the previous case of

We

before.

999.)
"

Ordinates

Abscissae.

1 crab).

=
'

3-4751

176-7280

271-6007
7,919-2781

= /x/ =

9*6847, and for the

This gives the modulus 3*874 and the central

?

The modulus, as calculated from the mean

error, is 3*8634, so that the agreement is very close. The normal curve in fig. 3 is constructed from the values d = 9*6847, or = 2*7403, and y = 145*44 by aid of Table IL
The following additional quantities were now calculated :
ordinate of the normal curve 145 44.
#

/*4

"~"

3/^2

7*5637
044 712
?

10/x^ 8

22*6911
10*6485

X5
l

31*9455

1283*8486
*606 45

62

If
zero.

we had

a perfect probability-curve,

This, of course,

/x 3 , /x

we should not expect

comparative smallness of

/x

3,

ju,

5,

X4 X 6
,

e3 ,

5,

in

and

/x

3/x/,

any actual
es

and

/x

5/x^ 4 ought to be

set of observations,

shows a very

the symmetry of the normal curve in these results,

fair

but the

approximation to

PROF.
Since

e3

PEARSON ON THE MATHEMATICAL THEORY OF EVOLUTION.

K.

>

3c! (1

and accordingly

e x ),

we

see that the roots (37) of our p. 94 are both positive,

possible to break

it is

97

up the observation-curve into two normal

curves with coincident axes.

Calculating the two values of

w we

have

^= 3-50971,
whence from

p.

93

1-01148,

=a

Cl

W%

0-!==

-0046,

v^ X

c2

3-50971),'

o~ 2

=aX
=y
7

1*0046,

(/x

1*01148)

or
cx
o-j

For

all

= - 5, #
= 5*134,

c2
cr 2

= 1004,
= 2756.

group gives the normal curve

practical purposes the second

(c

999,

that a half per cent, of Crabs have been

removed by selection about the same mean is not large enough to be significant in
measurements of the kind we are here dealing with. So far, then, we may say that
o-

2*740) of the set of observations

No. 4 of Professor Weldon's measurements cannot be treated as the sum or difference

of two normal curves having their axes coincident with any substantial improvement

on the normal curve peculiar to the

we have used

(14.) Hitherto

original group.

" Crab Measurements No. 4" to illustrate the dis-

section of symmetrical frequency-curves, but a little consideration shows at once that

this
after

judging of symmetry by the eye

all,

heads," then
in all

very likely to be fallacious, and No. 4 may,

break up into two normal curves with non- coincident axes,

two curves correspond

is

is

we

same groups as

to practically the

in the case of the " Fore-

have demonstrated that the asymmetry of that frequency-curve

probability due to a mixture of two families in the Naples Crabs and not a
shall

result of differentiation going on in one

homogeneous

of No. 4 weighs nothing in the balance, as

may be

two normal curves with not widely divergent axes

What we
in

have been investigating, therefore, in

which the method of our

first

The apparent symmetry

species.

readily tested

investigation would

13

is

fail,

really only the special case

owing to the coincidence of

therefore, proceeded to form the nonic for No< 4, a result

/x
X 4 and X5 already
a
The nonic being

values of

Pi

by adding together

or totals.

the axes of the component normal curves- a coincidence which

I,

Should these

is

improbable a priori.

which requires only the

given.

+ %^2 + aPs + <*>% + a hVz + a&z +

7

^ 7iV

+ aPa +

The nearest whole number is here taken for the Crabs in each group,
t The arithmetic throughout was of course of a most laborious character.

MDCCCXCIV.

A,

0>

MATHEMATICAL THEORY OF EVOLUTION.

PROF. K. PE ARSON ON THE

98

the coefficients were26-47295.

a3

18-11448.

=
=
=
=
=
-

<%
Cl~

a6

7
a8
ag

Writing

p.2

x>

we have

f r

325-54964639.
1604-777825,114,
977-342,6614.

3154-2006888.
4412-284,2437.

1761-180374.

the nonic f(x) and

its first

derived function*

_/j

(x)

the following expressions

/( x )

=x +

18-114,48x

+ 325-549,646x 1604-777,825^
+ 977-342,661x + 3154-200,689x'
+ 1761-180,374 = 0,
9

26-472,95x

5
3

4412-284,244x

and

= X + 20-590,07 X - 12-076,32 X
+ 180-860,915x 713-234 589x
+ 325-780,887x + 700-933,486x
490-253,805.
5

/i (X)

The Sturm's

functions were

/(x)=

fM=

f(x) =
/(x)=
/*(x)=

now

Mx) =
fs (x) =
fa (x)

Totals

X=
+
+

co

o.

Mx)=
fa (x)

formed, and with the following results

+
+

4-

+
+

+
miJn"ia

-j

-.

change

4 changes

5 changes.

and oo. and no

Thus the nonic has one root of x between
oo
and
In other words it has 8 imaginary roots and only 1 real

00

# Divided by the factor

9<

roots between
one.

PROF. K. PEARSON
This root was

now

ON THE MATHEMATICAL THEORY OF EVOLUTION.

99

Putting p 3
to/x' ]n ^ ie original nonic, I easily
and 1, then between '15 and '16, and by a succession of

localized.

found x to lie between

approximations to be *1533, and

finally '15326.

Thus

p%

1-5326.

p s was

then ascertained from equation (27) of p. 84, and finally

found to be 2*17245. The quadratic (28) for y was then

= pjpo,

was

/
which has both

its

is

1-5326

0,

roots imaginary.

Thus, considerably to
that there

2'17245y

my surprise,

but greatly to

my satisfaction, it

was demonstrated

no solution whatever of the problem of breaking up the curve of No. 4

measurements into two normal components.

imaginary solutions of the problem.
The best and most accurate representation of No. 4 is the normal curve of fig. 3.
All nine roots of the fundamental nonic lead

The

result of this investigation

material

is

seems to me most important.

homogeneous, and the asymmetry

differentiation in that organ,

to

Professor

Weldon's

of the " forehead " curve points to a real

and not to a mixture of two

families

having been

dredged up.

On

the other hand, I cannot think that for the problem of evolution the dissection

most symmetrical curve given by the measurements is unnecessary. There

always be the problem Is the material homogeneous and a true evolution going

of the
will

on, or is the material a mixture

in

To throw the solution on the judgment of the eye

examining the graphical results

is,

I feel certain, quite futile.

Whenever in measuring a series of organs the

we must accordingly proceed as follows :

results give

an asymmetrical curve,

Stage (i). Break up this asymmetrical curve into components

solutions, the theory of correlation or the test of the sixth

enable us to say which

Stage

(ii).

is

there are several

if

moment

will,

perhaps,

the most satisfactory.

Endeavour

to break

up the most symmetrical curve

if it

cannot be

broken up, either into normal components with non-coincident axes or normal components with coincident axes, the material

is

homogeneous and the asymmetrical curve

If, on the other hand,
it refers.

points to a true differentiation in the organ to which

the most symmetrical frequency-curve does break up, then

component groups be the same

the asymmetrical curve,

and we

shall

we

(or practically the

if

the numbers in

its

same) as in those corresponding to

are really dealing with a mixture of heterogeneous material,

have ascertained the proportions of the mixture.

If the

numbers

we have a mixture, but we have

found a case of differentiation in both organs at the same time.*
should not be the same, then we cannot assert that

* Beetillon has found a double-humped frequency- curve for the height of the inhabitants of the

o 2

MATHEMATICAL THEORY Of EVOLUTION.

PEOF. K. PEARSON ON THE

100

These stages seem to represent the mathematical treatment of this portion of the

problem of evolution.

(15.)

on tracing

I found

root,

p3

Although the nonic corresponding to "Crabs No.

minimum

'82 it reached a

< 1254

value for values of

its

at

to be

'8497.

and

between

_p 2

2.

tendency

" in

This led to

p
y

from equation (27) being

2,

that near

any w ay corresponded
7

5*2521y

'8497

y2

5-40915.

to

I therefore investigated

derived function of the nonic lying about

first

were, a tendency towards a root, and

it

"

the groups into which the "foreheads" were differentiated.

the root of the

has no real negative

value of about 199 as compared with about 1761

Here then was, as

the question occurred to me whether this

at

4,"

'82,

and found

it

5*2521, whence

0,

or
7l

Whence

-15705,

nearly
zl

or the

numbers

-972,

-028,

c2

two groups are

in the

cx

971

and

"tendency to a root"

Clearly even this

in

28.

no way

fits

either solution of the

"forehead" case, and No. 4 measurements neither break up, nor have they even
a tendency to break up, in the same manner as the " foreheads." Since the nonic

must always have a "tendency"

other root to

between

may

*9

which

and

may

it

1,

say that not only

is

and

to

two

real roots at a time,

we may note

be said to tend, or for which f(p%)

is

is

that the

a minimum,

just as insignificant as that investigated above.

the material of No. 4 homogeneous, but

it

lies

We

has not even a

" tendency " towards heterogeneity.

III.

(16.)
for
it

The

object of the present paper being solely to illustrate a general

method

the reduction frequency-curves to normal types, and not a biological investigation,

might

suffice to stop at this point,

when

the rules for the reduction of symmetrical

and asymmetrical curves have been given and

But it must be remembered

and that the variety presented

illustrated.

that the method depends upon the solution of a nonic,

lias found a double-humped curve for the elaspers of Earwigs.

measurements of another organ, it seems impossible to say whether the
inhabitants of the Doubs, as Beetillon supposes, are a mixture of races, or Mr. Bateson's earwigs were
In either case our methods of investigation would show the proportions belonging
really homogeneous.

department of the Doubs.

Mr. Bateson

Without the investigation

of

to each

group of the mixture, or to each group of the differentiating species.

PROF. K. PEARSON ON THE

by the roots of

this equation suggests

as likely to arise,

The

MATHEMATICAL THEORY OF EVOLUTION.

when

101

very considerable divergences and peculiarities

number of frequency-curves are dealt with.

must not be taken as indicating that the

a considerable

discussion of the case of Crabs

incidents of this case will be generally true for other groups of biological measure-

ments, until a very great variety of

such groups of

measurements have been

mathematically analyzed.

In order to throw more light on the general question, I have added the following
analysis for the case of Prawns, the measurements for which were kindly placed at

my

disposal

by Mr. H. Thompson, who has been making elaborate measurements of

1,000 specimens in the Zoological Laboratory of University College, London.

Palcemon serratus.

Measurements

998

in

specimens (adult) from penultimate

to hindmost tooth on the carapace.

Measurements reduced
to thousandths of body

Number of specimens.

length.

3
3

4
11

24
38
56
80
105
121
117
108
77
69
62
48

The novel and somewhat remarkable

To neglect these giants, as
65 and 69.
be convenient, so far as the analysis
of the group.

may

Number of specimens.

length.

27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48

and

Measurements reduced
to thousandths of body

is

49
50
51
52

25
17
11
8

53

54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69

1
1

feature in these results are the " giants " at

in

some degree anomalous, would, no doubt

concerned, and would lead to a simpler reduction

They have, however, been retained

as

among the data given

their presence affords an interesting illustration of the various singularities

arise in the solution of the

(17.)

The curve

(see

fig.

to me,

which

fundamental nonic.

4) given

by the observed numbers

will

be at once seen to

PROF. K. PEARSON

102

OIST

be distinctly asymmetrical.
coordinates, and using the

= d (=
=
fi\ =
fi\

Adopting the carapace length 31 as the

same notation

q)

THE MATHEMATICAL THEORY OF EVOLUTION".

4,963-876,753,5

;jl\=

94,386*734,469

1,920,725-520,040
X4
X5

=
=

The standard-deviation of the group

14116,678,13

fi 3

33*424,02673

/x

of

1,288*640,094,26

== 16,752*563,9961

2072-394,903
36,102-605,1706.
as a whole

errort obtained

the following results

fi 2

/x

o-=
The mean

we have

^=
=
=
=

16-191,382,8

276-277,555

fi/ s

//5

as before,

origin

given by

= \/^

or

3*7572.

from

<x

directly

is

=2*9978

2*8776.

(In the case of the " foreheads " of Crabs, the

mean error from <x was 3*8028, and

This divergence between the mean error, as found practically from

directly 4*4087.

first moments, is a very good test of the asymmetry of the frequencyIn the very symmetrical measurements of " Crabs No. 4," the modulus, as

second and
curve.

calculated from the standard-deviation

and from the mean

error,

had the near values

3*874 and 3*863.)

The curve obtained from the observations

3*7572)

is

given in

fig.

4 (Plate

as a single group

(i.e.,

16*1914 and

4).

Taking ^ = ygp we have for the fundamental nonic and its first differential
/' (X) = 9X
AX) = X
+ 169-245,583,743 x
+ 24-177,940,535 x
+ 10-054,490,066 x
+ 1-675,748, 344 x
+ 1498'101,518,851 x
+ 299-620,303,770 X
943-393,909,962 x
- 3773-575,639,850 x
864-540 147,350 x
- 2593-620,442,052 x
274'750,163,918 x
- 549-500,327,835 x
34-486,278,563 X
34-486,278,563
1-394,286,418 =
2

0.

* These results were calculated to a higher degree of accuracy than in the case of the Crabs, a result
rendered necessary by the apparent sensitiveness of the roots in this case to a slight change in the value
of the coefficients of the nonic.

Mean

error

is

here used, not in Gauss's sense, but in the sense of arithmetically

-7979 a theoretically.

mean

error,

PROF.

K PEARSON ON THE MATHEMATICAL THEORY OF EVOLUTION.

Clearly there

is

only one positive root.

103

This was found to be

= 2*5868658.

This gave

whence

p2

25-868,658,

Pl

9-669,970.

I found

Consequently the roots of

-Ptf+P* =

were imaginary and no solution involving the difference of two normal components

was possible.
The next stage was to find the negative roots. These were easily demonstrated to
lie between
and 1, and then it was shown that the value of f{\) on ^J changed sign
twice between these values. Thus the nonic was proved, without calculating Sturm's
The two negative roots are
functions, to have only three real roots.
:

Xi=

-154,481,14

-078,262,95.

and
Xa

These roots lead to the following solutions

Carapace of Prawns.

(A.) First additive Solution for

jP2=
Pl

zx

1st

1-544,8114,

26-758,0108,

-057,6086,

y2

-997,856,

=
=

b
=
y =
x

o-i

26-815,6194,
-002,144.

2nd Component.

Component.

cl

(B.)

995,860,
-057,6086;

3-5595,
.111-6142.

ca

6a

= 2-140,
= 26-815,6194,
= 5-7626 v
= imaginary.
7

cr

y2

Second additive Solution for Carapace of Prawns.

=
p =
p.z
x

-782,6295,

5-163,5907,

7i=

-147,3614,

y%

-973,0024,

23

%x

= 5-310,9521,
= -026,9976.

104

PROF. K. PEARSOIST
1st

cx

bx
o-j

yx

2nd Component.

Component.

= 971-0564,
-147,3614,
= 3-389,672,
114-28698.

c2

b2
er2

y%

we may add :
Parameters of Normal Curve deduced from

To these
(C.)

ON THE MATHEMATICAL THEORY OF EVOLUTION.

solutions

d=
=
=
y =
o

entire

group of observations.

16*191,383,
998,

3-7572,

<r

(D.)

= 26-9436,
= 5-310,9521,
= 8-932,996,
1-203,280.

105*968,04.

Parameters of Normal Curve deduced by excluding

tioo

"giants" from

observations.

=
c =

16*14357

=-

-04781),

996,

<r=

3*6051,

y
The curves corresponding
curve are given in figs. 4 and

(6

110-21786.

to (A), (B), (C),

5.

and

I shall

and (D) as well as the observation-

now proceed

to discuss several important

points with regard to them.

(18.)

The

first

point to be noted

is

the existence of the dwarf, carapace 27, and

the giants, carapaces 65 and 69.

The normal curve has a standard-deviation 3*7572, and the mean carapace being
about 43, we have no less than three measurements deviating by more than four
times the standard- deviation from the mean two of them, indeed, differ by nearly
;

six

times the standard-deviation from the mean.

We

might expect three such

deviations of over four times the standard-deviation to occur in the measurement of

50,000 Prawns, but they are extremely improbable in the measurement of 1000

That two should occur

prawns.

deviation six times the standard,

is

in

the measurement of 1000 Prawns, with a

so improbable that it

ought to lead us to reject

the normal curve as a representation of the measurements.

We

are either dealing

with a mixed population of Prawns, or possibly there are a few deformed individuals

amid a .normal
There

is

population."'

another point, however, in which the normal curve, based on the total

* I exclude the possibility of any serious error of measurement, having reason to "believe in the great
care with which the determinations were made,

PROF.

K.

PEARSON ON THE MATHEMATICAL THEORY OF EVOLUTION.

observations, diverges considerably from the observational result,

in

the defect

of

carapaces

asymmetrical appearance

about

of the

This defect

45.

curve.

I felt

namely

(see

largely contributes

105

fig. 4),

the

to

very confident that by neglecting

the eccentric group of " giants" I could find two components, whose resultant would
fit

the curve of observation as closely as the resultant-curves found for the similar

case of the forehead of Crabs.

was peculiarly

interested, however, in ascertaining

whether the method of resolution by aid of the nonic would pay more attention to
the outlying giants or to the less improbable defect of individuals about 45.

imagined that out of the nine possible solutions some might be solutions

As

giants and some for the 45 defect.

even

for the

a matter of fact, the two solutions which

have any meaning are entirely taken up with the very improbable outlying eccentricities

These eccentricities must

of the observations.

first

be removed from the

observations before the method will be of service in resolving the asymmetry of the

bulk of the observation-curve.

The method in which the nonic deals with the abnormalities

and I venture to think highly suggestive.
In

fig.

two giants

4 the normal curve excluding the

is

given.

is

very characteristic,

It fits the observa-

tion-curve, as far as appearances go, slightly better than the true normal curve.

But the
gives us

solution of the nonic tells us not to absolutely reject the giants.

first

two components, the

the normal curve

first

of

which

fits

the observations slightly better than

It has practically the

(giants excluded).

It

same area (995*86 as

compared with 996), a slightly less standard-deviation (3*5595 as compared with

3 6051), and consequently an increased maximum ordinate.
This, with a slightly
shifted axis, gives a somewhat better fit.
In addition to this first component we have
k

a second component with an area of 2*140, and a

component corresponds
ever,

distributed really, could

if

of 70 for the carapace.

closely to the two giants with a

an imaginary standard-deviation.

component,

mean

and we may then sum up the

first

mean

This

It has,

how-

Clearly the addition of two to the

first

make no

of 67.

sensible change in its appearance,

solution of the nonic in the following

words :

two giants, but places an imaginary distribution of

neighbourhood, and thus obtains for the other component and the

It does not absolutely reject the

2*14

in

their

resultant-curve (which

must be

observation-curve than

if

practically identical with

it)

a better approach to the

the giants had been rejected.

would appear, therefore, that our method of dissection offers, by means of

small components with imaginary distributions, a means of obtaining better results
than by simply rejecting (or, perhaps, even weighting) anomalous observations.
The second method by which the nonic attempts to account for the eccentricities of
It

these carapace measurements,

is

by mixing a small population of about

giants with the normal population.

2*7 per cent, of

These giants have a mean carapace of 48*5, while

the rest of the population has a mean of only 43. This population of giants, however,
has a very large standard-deviation, i.e., 8'9330 as compared with the 3*3897 of the
MDCCOXCIV.

PROF. K. PEARSON

106

ON THE MATHEMATICAL THEORY OF EVOLUTION.

rest of the population.

population,

i.e.,

It is

clear

that this population of giants

a very small disturbance would largely change

accounts for and covers the dwarf and two giant anomalies
curve, based on the addition of the

two components,

is

an unstable

is

its centre.

is clear,

That

it

and the resultant-

a fairly close approach to the

observation-curvefar closer indeed than that provided by the

first solution,

and a

great advance on the normal-curve C, resulting from the observations as a whole (see
fig.

5).

am

inclined,

accordingly, to suspect that the family of

Prawns was not

homogeneous, but contained between 2 and 3 per cent, of a giant population with a
large standard deviation.
is

may

Possibly the theory of correlations

settle

whether

this

the real state of the case, or whether the anomalies referred to ought to be rejected

and a new investigation made to

when

dissect the asymmetrical curve for the carapaces

the outlying parts, which control the nonic at present, are removed.

The

investigation of this case, however, with

the great variety of solutions which

may

all

the observations included, shows

be suggested by the dissection of various

anomalous and asymmetrical frequency-curves.

Table

I.

Powers of the Natural Numbers.

Powers.

First.

Second.

Third.

Fourth.

Sixth.

Fifth.

2
3
4

8
27
64
125
216

16
81

32

64
729
4,096

5
6
1-7

8
9

10
11
12
13
14
15
16
17
18
19
20
21
22
23

24
25
26
27
28

29
30

16
25
36
49
64
81
100
121
144
169
196
225
256
289
324
361

400
441
484
529
576
625
676
729
784
841
900

343
512
729
1,000
1,331
],728
2,197
2,744
3,375

4 096
;

4,913
5,832
6,859
8,000
9,261
10,648
12,167
13,824
15,625
17,576
19,683
21,952
24,389
27,000

256
625
1,296
2,401
4,096
6,561
10,000
14,641
20,736
28,561
38,416
50,625
65,536
83,521
104,976
130,321
160,000
194,481
234,256
279,841
331,776
390,625
456,976
531,441
614,656
707,281
810,000

243
1,024
3,125
7,776
16,807
32,768
59,049
100,000
161,051
248,832
371,293
537,824
759,375
1,048,576
1,419,857
1,889,568
2,476,099
3,200,000
4,084,101
5,153,632
6,436,343
7,962,624
9,765,625
11,881,376
14,348,907
17,210,368
20,511,149
24,300,000

15,625
46,656
117,649
262,144
531,441
1,000,000
1,771,561
2,985,984
4,826,809
7,529,536
11,390,625
16,777,216
24,137,569
34,012.224
47,045,881
64,000,000
85,766,121
113,379,904
148,035,889
191,102,976
244,140,625
308,915,776
387,420,489
481,890,304
594,823,321
729,000,000

PROF. K.

PEARSON ON THE MATHEMATICAL THEORY OF EVOLUTION.

Table

D=

IT.

Ordinates of Normal Curve.

= Standard Deviation.
P = Maximum Frequency 7?r0
S

Deviation.

Frequency.

F/P.

o-i
0-2

9950
9802
9560
9231
8825
8353
7827
7262
6670
6065
5467
4868
4286
3753
3246

0-5

0-6
0-7
0-8
0*9
1
1-1
"

1-2
1-3

1-4
1-5

[Note, added February

rical

10,

D/S.

F/P.

1-6
1-7

2780
2357
1979

1-8
1-9

1645

1353

2
2-2

0889

2-4
2-6

0340

2-8

0198

0111
0060

0561

3-2
3-4

0031
0015

3-6

3-8

0007

0003
000,004

1894. (1.) The importance

of breaking

up asymmet-

frequency-curves into normal components has been recognized for a long time

by anthropologists and
R. Livi,

biologists.

vol. 8,

Attempts at a solution have been made by

Firenze, 1883 (see also
Archivio per

'Sulla statura degli Italiani,'

V Antropologia e VEtnologia? vol.

1883, pp. 119-56).

13,

Also by O.

Auslese heim Menschenj Jena, 1893.

as

D/S.

0-3
0-4

107

Professor

serious.

Lexis and Dr.

'

Firenze,

Ammon

1883,

and

'

Annali di Statistical

work 'Die naturliche

These attempts can hardly be looked upon
in his recent

Venn have

pointed out that the curve of

deaths for each year for 1000 persons born in the same year
curve is also in all probability a compound curve.

the

true mortality-

Since writing the above memoir I have succeeded in resolving this mortality-curve
into components which are not, however,

all of the normal type, but become, as

approach infantile mortality, of the skew form (see p. 74 above).

O.

Ammon,

in the

volume cited above, endeavours

we

an evolution in
the length-breadth index of the skull of South-Germans since primitive times.
He
does this by comparison of the index as obtained from measurements on skulls from
the Row-Graves and on modern skulls.
frequency-curve

for

up

into

breaking

it

He

to demonstrate

has not, however, noticed that the

Row-Grave skulls is asymmetrical.

I have succeeded in
two components, one of which practically coincides in mean

and standard-deviation with the frequency-curve

P 2

for the skulls of

modern South-

108

PROF. K. PEARSON ON THE

MATHEMATICAL THEORY OF EVOLUTION.

Germans. In other words, the Row-Graves contain a mixed population, one element of
which corresponds closely to the modern South-German population.

Ammon's

state-

ment, therefore, that an evolution has taken place in this particular skull index appears

The whole problem of the compound nature of skull frequencycurves, both in England and Germany, is a very interesting and difficult one, and I
do not wish at present to anticipate results, which I hope when my investigations
The above may suffice to indicate the range of
are complete to publish as a whole.
to fall to the ground.

problems

to

which a resolution

asymmetrical

of

frequency-curves into

normal

components may be applied.

(2.)

With regard

to the

method adopted

of the defects under which

it

in the

memoir

itself, I

am

very conscious

suffersthe laborious character of the arithmetic

what may be the probable error of the solution obtained

by the method of higher moments. But I had to deal with the fact that the problem
is one which urgently needed a solution in the case of both economic and biological

involved, and the question of

statistics.

may be

Better solutions than mine

than one mathematically trained

statistician has for

am

tance of the problem, no solution, so far as I

With regard to
word may be said.
to

me by

some time recognized the impor-

aware, has hitherto been forthcoming.

the amount of error introduced by the use of higher moments, a

I

have not been able to work out the general problem suggested

George Darwin

Professor

of a frequency-curve,

curves into which

ultimately found, but although more

it

"

Given the probable error of every ordinate

what are the probable

may be

errors of the elements of the

dissected

which are likely to occur

I can, however, indicate the sort of differences

based on high or on low moments.

Then

random and measuring

in

a size near the

mean than

in a size far

it is

obvious that in

their organs, an error of the

the frequency of an organ of a given size

same magnitude

in results

Suppose the distribution of an organ in a group

of animals actually does follow a normal frequency-curve.

selecting 1000 of these animals at

two normal

"

from the mean.

is

more

Now

likely to occur in

a low

moment pays

to an error in the frequency near the mean

moment
In other words, a frequencyto one far off.
moment
and less attention than a high
curve calculated from low moments fits best near the centre one calculated from

greater attention than a high

high moments

fits

accordingly the following

an error in frequency near the

more attention

pays more attention

tail is

The problem

is

not as probable as an

moment pays
mean
on the other hand, the low moment
to it than a low moment
than the high moment to more probable errors in frequency.

equal error in frequency near the

much

observation-curve.

best near the tails of the

but

does occur a high

if it
;

Which tendency on the whole

will prevail

Turning to the result in the foot-note,

Ms,

(2r

p. 92,

1) (2r

and
M.

= S (x

2r

3)

we have
.

8a?).

th

for the 2r

5.3.1 o^c,

moment-

GENERAL EQUATION OF A CUBIC SURFACE.

45

Then the equation (A) takes the form

xyzu

and

it

(x

aT) (y

6T)

(z

cT) (u

cT)

....

(0)5

represents, besides the plane T, the cubic surface passing through the twelve

straight lines, which are represented in the annexed figure, as well as three other
straight lines which are not represented in the figure.

C
The equations

of the lines

may
~~~

be written as follows

01/ X.

aT

x = aT]

=
=cT

H
,

8)

(10)

u
z

u
I

J-(ll)

= dT
,

2=0^

|>(12)

anci

m
*

d
(13).

""

e
*-

which meets

(3), (4), (9),

and (10)

y.= 6T

>(9)

2/

(4)

= dT 1
x = 01
u

= cT
y =
u = dT
=^

2)

6T

=
z

:-

H3

110

PROF. K.

moments

PEARSON ON THE MATHEMATICAL THEORY OF EVOLUTION.

respectively, I notice tbe following values for the standard-deviation of

" Crabs No. 4," as calculated from the second, fourth, and sixth

a-2
cr 4

cr 6

Practically, it

would be

theoretical curve.

For

difficult to

moments

= 274,
= 277,
= 2*84.

say which of these results gives the best fitting

statistics of this

kind they are sensibly the same.

Thus,

till

another method of attacking the problem of the resolution of asymmetrical frequencycurves

is

moments

propounded, I think there

to lead us to discard a

to large errors, K. P.]

is

not sufficient evidence against the use of higher

method based upon them

as essentially likely to lead

Pear>$6rV:

PhttJrcmsAQ3^A.*jPlate, 1.

ScaZ&: fs"-004-of bo cfy IUivg th,

1

West.HewmaiL

sc.

Pexvrson.

Phil.Trans.\W)4>.K.Platz

Wesfc,Hewma,"a.' sc,

i
!

PhiLTrans.l8d4fAJ}lat& 3

Pearson^

V-

~*

--

f -"

'

j"

"

"
t

/i

1
i

vt

-*

f*>

f-

-7

II
It

'

r
I

iV.
1

Sj

SJ
{..

&

^
^

j /
At

NO

te5

'J
^0^

C*

E?

s~
s^iT*

**

S"'
.

^*

^^ ^^

^^ -"

-- ~-

3
.._

""

;t

__

/-' r-

^r.

^ - 7^

-&}-

'

f"

-"

fc/b

^BH-rr

-*"

"^ ^^"

**
=.

7*7

"p5 onq.* v0^pyOuq:

>...

W)
"A

V
v.
1

-^^Si

^*^~

^C

5^*^:. *-*

-.

--:
"-~
*-.
'**

~-

"".
'*"**">.

'

~-^

"

"

*-

^^ *~^

fs

^>
**^*^w.

---

-05

7^.

CO

--

**>*^.

~1

--

>

.-x

i
i

I
.

\f-

-~ -

..__

i
!

_._,

S"

&Hi

'

""

^
'S

'

i
I

i
i

!
j
I

r^t^i"

i
i

_,.

""
i
j

<<

02'

^U

^l Knr,tnArrru
I J}) '&ixy
"""
T

'^T'-

'1

J^i^

i"

>k

i
1

]
j

--

-.

_1

1-

.1

1
.

_L

Wis.t,lf<-jsvTa.%;tt.

sc.

Peccrsorv.

\>

'/
/

k\

/ t

\
i

If

7
f

\
4
//

7
i

<*

Fig.

//
IV.

Length/ of CarccpcLce<

of

Prawns

l!
jI

'

ii1

n 1
p%

R/da R11A
O^tHf,
OII4f'

ft

//
//
/

/
If
*l

i
//

|S

ft

If

.11

it

<h

'

ii

CO

^
1
^

In
^

In

'-1

Oj

//

'Si
i^j-

X'
i

tad..

N
A

>7

^1

^-1

3o

'

$\

//
[

/i
ji
,**1

-^>

-jt&z1
\

J=-^2-

ifis

rig tru"'

Aj xs

of! .Me:xsx* r&n uen\ :s

T/?

*T

"*

^r

^t

<?,.,-.7.

Phil.

\
\
\
I

\*

\
\
w
i

\\

\v

V
<

\
\\

998

OoservcubLorue

Normal/

Cur\/

/x

x X

X-_x

1 st Component;

5 -<36
2 i4

2'^OympoTheivb

4
\

5.96

ff JWt? "qixxruts" cub

38

anxL

42

&jccLLuled/)

StccnxLarcb DeMiaJtixmjs

i 5 * CompoTberub

3.

2^ Comporverub

5 -1626.

NorrwaJb Curve

3-6051

ft

5595

/J

\
1

\V
*

^
1

\H

^
VV

^^
Y'

\^

NV

\V

^=

^s.

r^-Y

Sf*,T 7fl

-?

7-*^

~f

&

An T/1
7

ifZ.;
i

L,

/ o-t~ r-i+V.

'

PhiLTrans. 18 04 .jV.Plate

H
XsSS
1\\

^1
ire-

Ms |\
HIv\W

a "UJ'TU

Adzls

of I/f&OL&ijr&iiverute

16

3377

W- i&im
Origin* cav

Wf.lOOq o^

booby -Z&'vgtiv

4-

Scxxh.

^^

^V^v

SocLLe>: ZtWiits

1V1000 of bo dy

-A

43-00700

~ ler> jgth /.

^^^
West Newman, iith
,

'

Pearsoru

>

/
//

yr**

;'/

'

.7

/
7^
jl
.

/
/

//
.'y

//
h5

/.
/

:7
I

:7
/

Ui /
/

V
/

''

Fig.
Length
2.

V.

17
-

Prawns

of Coorapcuoe of

SdbuddjDTu of Noruuo:

?*=-

//J

-782, 6295.

/.if

JM

if

hi?

',,

:i

I
f
ii

'~
t

I
1
1
'7/
.

hi
Si
ill
ii

'

'/'

rll

"

I fl

I
,

-M

1*

iff
*

jl
i

'

/1

si
Oil

i
.

'

<0

&r

^*-j

i
)
.

'.

...

Hi
t

oi

i?

f:

!
|

'

i
;

I
I

1
i

I
Jr
I

ii

'

41
"f

f\//

'1

~ipw

"

1
i

ii!

1
1

j
i

..

-4
ji

4i

J
i

...

6j

1
'

''//

Fq

t/

w,
/ v/

j.
|

_-

1
i

i
.

...

t
t
i

\
i

l>

1
i

L_

i
1.
1

W.

I
I
i

\
!

j
!

U,

-t

-i

i-

--i.-.

v?

.";

*--f

.,-&>

"*-*
!

Cr

^
t

^rZi

&&sk

(.

.1

J __t

Aip.<s of jMeo

i-

1
!

ii

>-

! U-J

'

1
;

_ x

\-\~-

-j

'

|-

\-

...|

I:....

*\

/"

J<\-

V'\\
//

\
Xr

/
/

s*\

.-

<

vV

X
i

... fc

..-,...,..

X
w

i
\

%
\

t\

re

1
I

iiAJ
!

V
!

No
1

Obs ejrvaiirm.jR

of JridiMukials

998

ft

st

'

911

Comporuenl

2 nr CoTnporueruO

27

V 1

pLesuLtajut/

Normal/ Curv&-*

998
998

...

x *_^

V
'-.-

V.

-,

\
'

l siy

T)pz/i n+i nri a

.9//7T7 /i srr*ri\

v. \

RQ 7

Crnrrnmn^i-j

3-

.1

^ Comporusrvb

8-

9330

77

3 ^5 7 2

Normxil; Guurv&

II
.1

'

Oj

i-

ffiH

ii

1
1

oi

\\

JtH

1
!

1
$

^N v

IJ

^ "-TV

Lj_
.

tea

i.

=iMWO

II

_, I

*
1

Ii

i
1

__

i
|

!
i

1
l

"1

'(

<

-7!N

o|

..in.

hScdlZ&:

-v

'i

vs
N

{A

J.

.....j

<

!
i

...
1

i'jl

|\\

1-

c3

zLi)

\\\! V

...q]
-^1

'

\-\\.
!

i,

\~

!
j

!
1

k\ &l
4-*

i
:!
I

\\

fi

V
i

ij

"i

1
1

""'?

*,
i

....

,i

^^
i^^*

^(j

*^fe= rJ^U-

II

1
i

1
i

h od^ -L VTLckth

"n

1
i

j
i

Phil.IrcmJsim^K.PlaJb& 5

dzoals

i
i

i
i

i
1

!
.

1
.1

'

......J......

i
i

'
i

i
i

<

i
j

!
1

1
1

i
1

i
!

^
.

,.,.i

i
i

i
i

'

i
i

'!'

I
i

j
1
i

--^H^

^j-"-**T^~**.

1
1

i
i

4
1

'

1
i

___

_j

>> r-

sj.

i
I

..

Sec JL&

zjn iZ$

. yn 100

07

c h ody

.1

!
"

H^

1
,

1
i

\
l

^v^

-^fe= *

..

^^

.J

i
1

......

*^~

^ _ m

-1 7Uf th

i
'

:l

-*

I
i

i
i

f
i
i

- -

i
!

!
i

.,

il

i
i
1

f -j
i

i
i

I
1

1
i
1

i
!

"

""

I
1

t
I

..

>-*---

'

'

i
1
!

!
1

1
j

_j

\\

I
1

!
i

i
j
,

4-

-!

t r~

ill
III!
i

r
i

i
,

t.

..~

.-,

'!

J.
West/Kfe-'-v-rr: &n, Jj+h.