J948] 243

THE INTERPRETATION OF STATISTICAL MAPS

By P. A. P. MORAN
(Institute of Statistics, Oxford)
[Received September 17, 1947]

1. It is sometimes necessary to consider the geographical distribution of some quality or

phenomenon in the counties or states of a country, and one of the questions we may then ask is
whether the factors causing the phenomenon may be regarded as statistically independent in
different counties, or, on the other hand, whether the presence of some quality in a county makes
its presence in neighbouring counties more or less likely. D. Barron Cruickshank (1, 2) has
considered a similar type of problem in relation to the incidence of tuberculosis in the United
Kingdom. For other work on this subject the references in (3) and a paper by Mahalanobis
(4) may be consulted.
2. We only consider the case where for each county or state a given event has or has not
occurred. If it has, we denote the state of the county by B (for "black"), and if not by W (for
"white"), To determine whether events in neighbouring counties are independent or not, we
consider the distribution of the number of contiguous counties which are both "black." We
represent the counties by points, and denote the fact that two counties touch by joining the
corresponding points by a line. Suppose there are N counties C1 • • . CN and n joins J1 • • • I n.
We may either regard the number of black points as fixed (= P) and the latter distributed at
random on the resulting lattice of points, * or we may suppose the individual counties to be B
and W independently with probabilities p and q = 1 - p. The former case, which we refer to
as the "fixed P" case, has the advantage that we do not have to estimate p, whereas in the latter
case, which we call the "fixed p" case, p usually has to be estimated from the data itself-a process
itself liable to error.
3. In a previous paper we considered the distribution of the number of BB and BW joins
for the fixed p case, and in the case where the points are arranged in the form of a rectangular
lattice of m x n points, we found the first and second moments and proved that the distribution
of the number of BW joins tended to normality as m and n increased. That the distribution of
the number of BB joins on a rectangular lattice in the fixed p case also tends to normality can be
shown in a similar manner. However, we did not obtain the third and fourth moments. In
the present paper we show how these may be obtained for the distributions of the numbers of
BB or BW joins in the fixed p case, for any type of lattice whatever. Because of the lengthy
algebra involved, we shall only carry out the calculation of the third and fourth moments for the
distribution of the number of BB joins, but we shall give the first and second moments of the
distributions of BB and BW joins in both fixed p and fixed P cases. In particular we apply the
results to a rectangular lattice, and give another and simpler proof of the tendency to normality
in the fixed pease.

Calculation 0/ First and Second Moments

4. Consider first the number of BB joins in the fixed p case. Write
Qi = I if the join Ji is a BB join.
= 0 otherwise.
Then Q = 2: Qi where the sum is taken over all joins, is the number of BB joins, and we mus
find the moments of the distribution of Q. The expectation of Q is 2:E (Qi), and equals the
expectation of a single Qi (that is, the probability that Ji is a BB join) multiplied by the number
of joins h Therefore [L'1 (Q) = E(Q) = np'.
* The object which we here call a "lattice of points" is known as a "linear graph" in topology, and it
should be noted that the theory developed in this paper does not require the graph to be planar.
244 MORAN-The Interpretation of Statistical Maps [No.2,

Now consider
To find this we must multiply out (~Qi)2 and calculate the expectation of each term. Three

0--0
(J () 0---0--0 0-0
FIG. I.-Configurations contributory to f1'2' The double bond indicates a squared term.

types of term result (Fig. 1). The first is of the form Qi 2 and its expectation is p2. There are
clearly n terms of this type and the total contribution is np".
Let each point, representing a county Cs , have joins with M. other points. The second type
of term will be. of the form QiQj where Ji and Jj have one point in common. The expectation
N
is p", and the number of such terms is ~ M 8(M8 - I).
8 = 1
The third type of term will be of the form QiQj, where Ji and Jj have no points in common.
Suppose that for each join Ji the sum of the number of points, to which the points at the end
of Ji are joined, is ki. Then the number of pairs of J's with no county in common is
n n
t~· (n - k j - 1) = tn(n - I) - t ~ ki•
j=l j=l

The corresponding number of terms in the expansion is therefore

n
nin - 1) - ~ kj and the expectation is p", Adding we obtain
j = 1
N N
[I..'(Q) = np2 + pa ~ M.(M8 - 1) + p4{n(n - 1) - ~ kj }
8=1 j=l
and so
N N
[l.2(Q) = np2(l - pi) + p3 ~ M.(M8 - I) - p4 ~ kj.
8=1 j=1
If we know p we can therefore find (J./l(Q) and [l.2(Q), and so make a test of significance either by
assuming normality in cases where this seems plausible or in other cases by using Tchebycheff's
inequality. It has been proved (3) that the distribution tends to normality for a large enough
rectangular lattice, and for an arbitrary lattice the assumption is probably justifiable if ~1 and
~2 calculated from the third and fourth moments (which we show how to find for any lattice
in the next section) are near the values 0 and 3 which they take for a normal distribution. If
we do calculate the moments and normality is not strongly indicated we could fit a Pearson type
curve with some assurance that the resulting test of significance is not likely to be far out.
N n
For an irregular lattice where ~ MiM8-1) and ~ kj cannot be calculated theoretically,
8=1 j=1
they can usually be found by actual counting. For n can be found by simply finding the number
of joins of points. Then for each point it is easy to calculate M 8(M8 - 1) and add these together.
n
Similarly ~ kj can easily be found. We shall show in the next section that the numerical
j = 1
calculation of the third and fourth moments is, in practice, hot much more difficult than that
of the second, and so we can, in general, find out how far our assumption of normality is justified.
5. If we wanted to discuss the distribution of the number, R, of BW joins, we simply replace
the expectations p2, p3 and p4 of the individual types of terms in the above argument by the
corresponding ones in this case which are 2pq, pq, and 4p 2q2, thus obtaining
[1./1 (R) = 2npq
N n
[.L/. (R) = 2npq + pq ~ M 8(M8 -1) + 4p 2q2{n(n - 1) - ~ kj}
8=1 j=l
1948] MORAN-The Interpretation of Statistical Maps 245
and so
N n
[.L2 (R) = pq S Ms(Ms - 1) - 4 p 2q 2 ~ kj + 2npq - 4np2q . 2
8=1 j=l

6. The above method with p fixed suffers from the disadvantage that it assumes p to be known.
We may therefore consider thedistribution of BB or BW joins in the fixed P case. We then write
Q* for the number of BB joins, and Qi* = 1 if Ji is a BB join, = 0 otherwise. Then Q* =
~ Qi* and E(Q*) is n times the expectation of a single join. Now the number of ways of arranging
P black points on N positions is (!j,), whilst the number of ways of doing this so that Ji is
N-2
always BB is (p _ 2)' It follows that
E (Qi) = pcp - 1)/N(N - 1),
and
[l'lQ*) = nP(P - 1)/N(N - 1).

Carrying out a similar procedure for evaluating the expectations of the types of terms which
arise when we consider the second moment we get
, * _ pcp - 1) (P -- 2) N
[l2(Q ) - nP(P - 1)/N(N - 1) + N(N -1) (N ~~ s:: 1 Ms(Ms -1)
pcp - I)(P - 2)(P ~ 3) ( £
+ N(N _ 1) (N -2)(N _ 3) {n n - 1) -j = ~j}
and we can find [l2(Q*) from

Similarly we may consider the distribution of BW joins. Write R* for the number of such
joins and consider their distribution in the fixed P case. Write Ri* = 1 if Ji, is BW and = 0
otherwise. Then R* = :E Ri* and we find
, * * peN - P)
[ll(R ) = E(R ) = 2n N(N - 1)

, *) - 2 P(N-P) P(N-P] ~ M(M -I'

and ll., (R - nN(N _ 1) + N(N - 1) 8 = 1 s, )
+
4P(P - 1) (N - P) (N-P - 1)1~
N(N - 1) (N - i)(N ,=-3)' { n(n - 1)~-
)
1 kj s- j::
But although the fixed P distribution has the advantage of not assuming a knowledge of p, we
know less about it than the other case, for our methods will not prove it tends to normality for
a large rectangular lattice, and its third and fourth moments, either for a rectangular or an
arbitrary lattice, seem more difficult to calculate.

The Third and Fourth Moments

7. We now confine ourselves to considering the distribution of BB joins in the fixed pease,
and we proceed to calculate its third and fourth moments. The third and fourth moments of
the distribution of the number of BW joins in the fixed p case can be found by very similar
methods. Instead of calculating the third and fourth moments about the origin we find them
about the mean, and this simplifies the results as then those terms whose frequencies on the
lattice are most difficult to calculate, vanish.
Consider first the third moment [l3(Q) = E[(~Qi - E(~Qi»3]

Write Si = Qi - E(Qi) = Qi - p'

Then [l3(Q) = E{ (:ESi )3 }
246 MORAN-The Interpretation 0/ Statistical Maps [No.2,

Consider the various types of terms which occur in the expansion of this. There are eight in all.
(1) Terms corresponding to three joins without common points. Such terms clearly
have zero expectation.
(2) Terms corresponding to two joins without common points, such that one of these
joins occurs twice because the corresponding S, is squared. This has zero expectation.
(3) Terms corresponding to three joins, two of which have one point in common,
but no point in common with the third. The expectation is again zero.
The remaining terms have non-zero expectation (Fig. 2).

(E3:J 0-0-0-0

FIG. 2.-Configurations corresponding to non-zero contributions to P,2'

(4) Terms corresponding to a single join occurring three times. To find the expectation
is therefore to find E(Si 3) , and two methods of doing this will be used for this and subsequent
terms, thus providing a check on the algebra. In the first we write
E(Si 3) = E(Qi3 - 3p 2Qi 2 + 3p 4Qi - p6)
and calculate the expectation of each term separately, obtaining
p2 _ 3p 4 + 3p 6 _ p6 = p2(l_p2) (l-2 p2).

This is the simplest method to use on subsequent terms, but as a check we may write down
each possible combination of B's and W's, and after multiplying the resulting value of the
term by the probability of that particular combination of B's and W's, sum the resulting
products. This method is much longer, but provides a useful check on the algebra.
(5) Terms corresponding to two joins having a single point in common and such that
one of the joins occurs twice. The expectation is found to be p3(l - p) (l - 2p2).
(6) Terms correspqnding to three joins occurring in a chain so that one join has a
point in common with each of the other two who have no points in common with each
other. The expectation is p4(l _ p)2.
(7) Terms corresponding to three joins which have only one point in common and
that common to all three. The expectation is p4(l '- p) (l - 2p).
(8) Terms corresponding to three joins which join three points, there being, therefore,
one point common to each pair of joins. This type of term does not occur in a
rectangular or cubic lattice. Its expectation is p3(l - p)2 (l + 2p).
This completes the enumeration of the types of terms in fL3(Q). To find how often each occurs
in the expansion; we must enumerate the number of possible ways in which each configuration
can be placed on the lattice. We calculate these numbers explicitly for a rectangular lattice,
but for an arbitrary irregular lattice such as results from a geographical map, it is necessary to
count the number of ways of arranging such groups of joins on the lattice. This does not require
excessive labour for the average map, provided it be carried out systematically. This is best
illustrated by an example. Suppose we wanted to find the number of terms of type (6) above.
Consider each join of the lattice in turn, and associate with each join the number of distinct ways
of putting a chain of three joins on the lattice such that the given join is the middle join of the
chain. Adding all these numbers will give the number of ways in which a term of type (6) can
occur. Having found the frequencies of occurrence of terms (4) to (8), denote these frequencies
by 11,,, • . . 11 8 and then multiply them by the corresponding expectation and by the multi-nomial
1948] MORAN- The Interpretation of Statistical Maps 247

coefficientarising from the expansion of (:ESiP. For the terms (4), . .. (8) these are respectively
1, 3, 6, 6 and 6. Notice in particular that in calculating n5 we have to take account that two
joins on the lattice with a point in common contribute two terms to n5 because either of the two
joins can be chosen as the' one corresponding to the squared Si in the term. Finally we have
1J-3(Q) = /14p 2(l - p2)(l - 2p2) + 3n 5 p3(l - p) (l - 2p2) + 6n6p4(l - p)2 + 6n7p4(l - p)
(l - 2p) + 6n8 p3(1 - p)2(1 + 2p).
8. Consider now the values of /14' • • • /18 in the case of a rectangular lattice of m x n points.
A detailed enumeration of the number of ways in which terms of each type can lie on such a
lattice shows that
/14 = 2mn - m - n. n 5 = 2(6mn - 6m - 6n 4).+
n 6 = 18mn - 25m - 25n +28. n 7 = 4mn - 6m - 6n 8. +
n8 = O.
These formulae hold for m and n greater than, or equal to, three. It follows that
IJ-lQ) = (2mn - m - n)p2(1 - p2) (l - 2p2) 6(6mn - 6m - 6n + 4)p3(l - p) +
(1 - 2p2) +
6(l8mn - 25m - 25n 28)p4(l - p)2 +
12(2mn - 3m - 3n + 4) +
p4(1 - p) (l - 2p).

9. Now consider the fourth moment IJ-.(Q) = E(:ESi)4. There are 23 distinct types of term in
its expansion.
(9) Terms corresponding to four joins without common points.
(10) Terms corresponding to three joins, without common points, one of which occurs
twice:
(11) Terms corresponding to two joins without common points, such that one of the
joins corresponds to an Si raised to the third power.
(12) Terms corresponding to two groups of joins, one of which is a single join without
points common to the others, and the other group consists of three joins which share a
single point in common, that is, the term consists of a single join together with a group
of joins of the type of term (7) for IJ-lQ).
(13) Terms corresponding to two groups of joins, one consisting of a single join, and
the other of a group of three joins of the type considered in (5).
(14) Terms corresponding to three groups of joins, two of which are single joins, and
the third consists of two joins with a common point.
(l5) Terms corresponding to two groups of joins, one of which is a single join and the
other of the type considered in (6).
(16) Terms corresponding to two groups of joins, one of which is a single join and the
other of three joins, joining three points. Terms of this type do not occur on a rectangular
lattice.
The terms of types (9)-(l6) above all have expectation zero, because of the fact that they
contain at least one single join separated from the others. Moreover, amongst their number
they contain all terms corresponding to more than two groups of joins. It is very awkward to
calculate the number of ways such terms occur on the lattice, and this is why the calculation of
IJ-,(Q) and 1J-4(Q) is possible, whereas the direct calculation of lJ-'lQ) and 1J-'4(Q) which was the
method used in a previous paper (except that we there considered the distribution of BW joins)
would be very difficult. Using the present methods, we can not only calculate 1J-3(Q) and 1J-4(Q)
explicitly for a given rectangular lattice, but also, in practice, find the third and fourth moments
for an arbitrary lattice such as that resulting from a geographical map of counties or states. We
next consider the types of terms which do contribute something non-zero to IJ-.(Q) (Fig. 3).
(17) Terms corresponding to a single join on the lattice, occurring four times. The
expectation is p2(1 - p2) (l - 3p2 + 3p4).
(18) Terms corresponding to two joins on the lattice, having a single point in common
and such that one of them occurs three times. The expectation is p3(l - p) (l - 3p2
+ 3p 4).
248 MORAN- The Interpretation of Statistical Maps [No.2,

(19) Terms corresponding to three joins with a single point common to all three and
such that one of them occurs twice. The expectation is p4(l - p) (l - p - 2p2 + 3p 3)
(20) Terms corresponding to three joins arranged in a chain (i.e, joins Jr, J 2 and J 3
such that J 1 and J 2 have one point in common and J 2 and J 3 another), and such that one
of the end joins (i.e. .J,. or J 3) occurs twice. The expectation is p4(l - p) (1 - 2p2 + p3).
(21) Terms similar to (20) but such that it is the middle join which appears twice. The
expectation is p4(l - p)2(l - 2 p2).
(22) Terms corresponding to two joins with a single common point and such that
both occur twice. The expectation is p3(l - p) (l + p - 3p 2 - p3 + 3p 4).
(23) Terms corresponding to two joins without common points and' such that each
occurs twice. The expectation is p4(l _ p2)2.

0:::=:<) <E3>-O 0 ,< cr-o---< ) 0

0-0 (>--0 (J ( ) ()
()
()
()
() W
0---0--0
0---0---0

() ()

0--0--0
FIG. 3.-Configurations corresponding to non-zero contributions to ""4'
(24) Terms corresponding to four joins, each of which has a single point in common
with two of the others, but not with the fourth, the set of four being thus related like the
sides of a square. The expectation is p4(l - p)2 (l + 2p - p2).
(25) Terms corresponding to two groups of joins, each consisting of two joins having
a single point in common. The expectation is p6(l - p)2.
(26) Terms corresponding to four joins which have one point common to all four,
but no other point belonging to more than one join. The expectation is p5(l - p) (l -
3p + 3p 2).
(27) Terms corresponding to four joins forming a chain, i.e. Jr, J 2 , J 3 and J 4 such that
J1 and J 2, J 2 and J 3, and J 3 and J 4 have a point in common. The expectation is p5 (l -r-r- p)2.
(28) Terms corresponding to four joins, Ji, J 2 , J 3 and J 4 , such that J1 and J 2 have one
point in common and J 2 , J 3 and J 4 have another, but there are no other common points.
The expectation is p5(l - p)3.
(29) Terms corresponding to three joins such that J1 occurs twice and has no points
in common with J 2 and J 3 , which have one point in common. The expectation is
p5(1 - p) (l - p2).
1948] MORAN-1'he Interpretation of Statistical Maps 249

(30) Terms corresponding to four joins, J" J 2 , J 3 and J 4 , such that they form the sides
of a triangle whose vertices are the points they join, and one end of J 4 is one of these
vertices. The expectation is p4(l - p)3 (1 + 2p). This type of term does not occur on a
rectangular lattice.
(31) Terms corresponding to three joins J" J. and J 3 forming the sides of a triangle as
in (8) but such that one of them occurs twice. The expectation is p3(l - p)2 (l + 2p
- p2 _ 3p 3). This term also does not occur on a rectangular lattice.
Calling the frequencies with which the above terms occur on the lattice n9 , • • • n3b and
taking into account the corresponding multinomial coefficients, we have
[J.lQ) = n,7p2(1 - p2) (l - 3p3+ + +
3p4) 4nl8P3(l - p) (l - 3p2 3p4)
+ +
12nu p4(l - p) (l - p - 2p2 3p3) + +
12n2op4(l - p) (l - 2pa p3)
+ 12n2,p4(l - p)a (1 - 2p2) + 6n22p3(l - p) (l + p - 3p2 - p3 + 3p4)
+ 6n23p4(l - p2)2+ 24n24p4(l - p)2 (l+ +
2p - p2) 24nasp6(l - p)2
+ 24n26pS(l - p) (l - 3p + + +
3p2) 24n27pS(1 - p)2 24n28pS(l - p)3
+ 12n2ops(l - p) (l - p2) + 24n30p4(1 - p)3 (l + 2p)
+ +
12n3,p3 (l - p)2 (l 2p - p2 - 3p3).

10. It can be seen that, in practice, the amount of labour required in finding nl • • • n3l
for an arbitrary lattice should not be excessive. The method may be illustrated by considering
as an example the determination of n26' We have already denoted by M, the number of counties
to which the county C. is joined, and na6 is clearly ~ (~8) taken over all s such that M. ;> 4. This
sum can be quickly found by considering in turn the counties of the map. Similar methods can
be used to evaluate the other cases. For a rectangular lattice we can calculate nl7 ' • • n3l
explicitly and find (m, n ;> 4):
nl 7 2mn·- m - n.
=
nl 8 12mn - 12m - 12n + 8.
=
nu = 12mn - 18m - 18n + 24.
n20 = 36mn - 50m - 50n + 56.
n2l = 18mn - 25m - 25n + 28.
n22 = 6mn - 6m - 6n + 4.
na3 = !(4m 2n2 - 4m 2n - 4mn 2 + m a + n 2 - 12mn + 13m + 13n - 8).
na4 = mn - m - n + 1.
n2S = 18m2n2 - 36m 2n - 36mn 2 - 64mn + 18m 2 + 18n2 + 184m + 184n - 300.
n26 = mn - 2m - 2n -+
4.
n27 = 50mn - 90m - 90n + 136.
n28 = 36mn - 64m - 64n + 104.
Jl29= 12m2Jl2 -18m 2n - 18mn2 - 40mn + 6m 2 + 6n 2 + 76m + 76n - 88.
n 30 = O.
n 3l = O.

In a manner similar to the above we can find the third and fourth moments of the distribution
of the number of BW joins in the fixed p case. For the fixed P case the calculations would be
much more complicated.

The Approach to Normality

11. For a rectangular lattice of m x n points and the fixed p case, the distribution when
reduced to zero mean and unit standard deviation tends to normality when m and n increase.
This was proved for the distribution of BW joins in (3) by using methods developed by Bernstein,
and similar methods lead to the same conclusion for BB joins. But in fact in both cases it is
much simpler to argue in another manner, namely, to consider the asymptotic behaviour of the
higher moments. We do this here for the distribution of BB joins. The tendency to normality
distributions for P fixed do not seem to be easily amenable to either of these methods.
Consider a rectangular lattice of m x n points and denote the number of BB joins by Q.
250 MORAN-The Interpretation of Statistical Maps [No.2,

Then we have (3)

[L'l(Q) = p2 (2mn - m - n)

1.l.2(Q) = p2(2mn - m - n) + 2p3(6mn - 6m - 6n + 4) + p4(l3m + 13n - 1411111 - 8).

By a well known theorem, in order to show that the distribution of Q tends to normality, it is
sufficient to show that
[L2k (2k)!
[L2 k --+ 2k k!
[L2k+l 0
k+t ---+
[L2

as m and 11 tend to infinity. It is clear from the method of obtaining [L3(Q) and [LiQ) given
above that [Lp(Q) is a polynomial in m and n whose coefficients are fixed when p is given, provided
that the lattice is sufficiently large (m, n :> 4 for [LiQ)) for all typical terms in the expansion to
appear. To prove the above asymptotic formulae it is therefore sufficient to consider the terms
of highest order in m and n.
Consider first the various types of terms which arise when we expand [L2k+l(Q) = ECEQi)2k+ 1.
Each of these terms consists of a product of order 2k + 1 in the Qi. We classify the terms in
accordance with the manner in which the corresponding joins on the lattice have (or have not)
common points. Each product will consist of a set of groups (p say, in number) such that within
each group the corresponding joins have common points and form a connected whole, whilst
different groups have no points in common. If, in any term, there is a group corresponding to
a single join with the corresponding Qi raised only to the first power, the expectation of that term
will clearly be zero. It follows from this that any term containing more than k groups will have
expectation zero because in that case there will be at least one group corresponding to a single
join. Now, for given k, the number of different types of group is finite and so the total expectation
of all terms with p ( <; k) groups will be a polynomial in m and n of highest power (ml1)P at
most. Therefore there is a constant A independent of m and n such that

[L2k+tCQ) < A(mn)p.

But [L2(Q) is of order 11111 and so

as m and n - - + 00.

12. Now consider [L2k(Q) = (I:Q2 i)k. We expand this and consider the types of term which
result. As before, any product consisting of more than k groups will have expectation zero.
Any product consisting of less than k groups will contribute a polynomial in m and n of order
not greater than k - 1. Consider products with exactly k groups, none of which correspond to
a single join taken by itself. Each group must therefore correspond, either to a single join taken
twice, or to two joins with a common point. Moreover, the number of ways of arranging k such
terms, of definite type and orientation, on a lattice will be asymptotically equal to mknk, and
would be equal to this if we neglected edge and interference effects. The two types of group
have expectations p2(l - p2) and p3(l - p), whilst their frequencies on the lattice are (2mn
- m - n) and (6mn - 6m - 611 + 4) respectively. The total expectation, omitting terms of
order less than mknk, will be asymptotically equal to

{p2(l - p2) (2ml1 - m -r-- n) + 2p3(l - p) (6111n - 6m - 611 + 4)}k = {[L2(Q)}k

multiplied by a multinomial coefficient which represents the number of ways of choosing k pairs of
joins to form separate groups in the expansion of (I:Qt)2k. This coefficient is
(2k)!
2k k!
1948] MORAN-The Interpretation of Statistical Maps 251

and so

(2k)!
---+
2k k!
as m and n tend to infinity, and the distribution tends to normality. This method of argument
can be applied to three dimensional lattices in exactly the same way. Knowing that the distri-
bution is approximately normal for a large rectangular lattice we may have a good deal of confi-
dence that the same will be true of any fairly large lattice of irregular pattern, and as further
evidence, we may, if necessary, calculate the third and fourth moments.

SUMMARY.
Tests of significance for the random distribution of some qualitative determination on the
counties or states of a map are discussed. These depend on assuming either that what happens
in different counties is independent, or that the total number of counties possessing the quality
considered is fixed but their arrangement is random. First and second moments of the distribution
of the number of contiguous counties, both possessing the quality, are given in both cases, and a
method of finding the third and fourth moments in the former case. Explicit formulae are given
for the third and fourth moments when the lattice is rectangular, and in this case the standardized
distribution is shown to tend to normality as the size of the lattice increases.

References.
(1) Cruickshank, D. Barron (1940), Papworth Research Bulletin.
(2) - - (1947), Brit. Journ. Cancer, 1, 109-128.
(3) Moran, P. A. P. (1947), Proc. Cam. Phil. Soc., 43, 321-328.
(4) Mahalanobis, P. C., Phil. Trans. Roy. Soc. (B), 231,329.