Journal of Archaeological Science 1987, 14, 1-11

An Analytical Technique for Investigating Spatial

Relationships
M. R. Attwell" and M. Fletcher b

(Received 23 January 1984, revised manuscript accepted 24 March 1986)

This paper presents the derivation of a new analytical technique that can be used to test
for the existence of a significant association between a point pattern distribution and a
variable which describes its environment. It has the major advantages that it indicates
the strength of the association, provides information concerning its direction and can
be used on small data sets. Its application is illustrated by analysing the relationship
between the distribution of chambered cairns on Rousay, Orkney Islands, and altitude.
The technique was developed following an examination of an earlier computer simu-
lation method and the relationship between the two is exposed. Finally, results pro-
duced by the two approaches are compared and the validity of the method of assessing
"goodness of fit" employed in the earlier study is questioned.

Keywords: STATISTICS, SPATIAL ANALYSIS, SIMULATION, MATHEMATI-

CAL METHODS, NEOLITHIC, CHAMBERED TOMBS, ORKNEY ISLANDS,
COMPUTERS.

Introduction
Since the pioneering work of Fox (1932) archaeologists have become increasingly inter-
ested in the relationship between the distribution of groups of archaeological entities and
selected attributes of the environment in which they subsist. In particular, in recent years
locational tomb studies have increasingly focused their attention upon the relationship
between the buildings and the landscape. The general approach adopted has been to
compile data for a range of characteristics which m a y have influenced the choice of site
(e.g. soil type, altitude). This information has then been analysed using either descriptive
statistics (e.g. Cooney, 1979), computer simulation (Davidson et al., 1976a) or univariate
and/or multivariate techniques (e.g. Fraser, 1983), to enable probabilistic statements to be
made about the behaviour of the builders, and to infer from these their locational objec-
tives and desires. Ultimately, it is hoped that by exposing the general principles underlying.
the choice of location it will be possible to obtain an insight into the nature of the society
which built the tombs (for further details of recent work in this field, see Fraser, 1983:
51-62).

aDepartment of Computing, North Staflbrdshire Polytechnic, Beaconside,

Stafford ST18 0AD.
bDepartment of Mathematics, North Staffordshire Polytechnic, Beaconside,
Staflbrd ST18 0AD.
1
0305-4403/87/010001 + 11 $03.00/0 © 1987AcademicPress Inc. (London) Limited
2 M.R. ATTWELL AND M. FLETCHER

This paper presents a new analytical technique which can be used by archaeologists to
investigate the possible relationships between a distribution pattern of archaeological sites
(or artifacts) and the land types (or categories) of an environmental variable. The method
was developed following an examination of the computer simulation procedure used by
Davidson to assess the relative importance of eight environmental characteristics in the
choice of cairn locations on the island of Rousay in Orkney (Davidson et al., 1976a;
Davidson, 1979). The new technique mirrors Davidson's simulation procedure analyti-
cally, and the similarities between the two approaches enable us, later in this paper, to
assess Davidson's results and to question the validity o f the methodology he employed.

The Derivation of the New Analytical Technique

Assume that there are n sites distributed over an area for which various topographical
characteristics are known• In this section we define category weightings (A, B, C . . . ) which
represent the degree to which the land types (or categories) of a particular environmental
variable (~, 13, Y •. -) are favoured or avoided. Further, by the application of Bayesian
probability theory and simple algebra we will derive formulae for these weightings in
terms of easily obtainable probabilities.
Initially, we will assume that the variable is measured at three levels, so that any point
on the study area can be classified uniquely as either type a, type 13or type y. The general
case will be considered later. For these three categories we will let the proportion of the
total study area that can be classified as type a be a, type 13be b and type y be c. Thus, a
point chosen at random has a probability a o f being type ~t, b of being type 13and c of being
type y. Clearly a + b + c = 1. Similarly, let the proportion of the sites that can be classified
as type ct be a', type J3be b' and type y be c', so that the probability of a randomly chosen site
being of type ~zis a', type [3 is b' and type y is c'. Again a' + b' 4- c' =. 1,
These six proportions or probabilities can easily be found. O n e w a y of estimating a, b
and c is to superimpose a grid upon a map of the study area, and to count the number of
points of intersection falling into each of the three categories. Estimates for a, b and c are
then provided by the p r o p o r t i o n of the total number of points which fall into each
category.
Now, let A be the probability that a randomly chosen point in the study area is accepted
as a potential site given that the point is of type a. B and C are defined in a similar way. The
following algebra will produce estimates for A, B and C:
probability (point accepted[point type a) = A
probability (point type a) -- a
• •. probability (point accepted and type ~) -- aA
Hence probability (point accepted) = aA 4- b B + cC
Thus probability (type ~laccepted) = aA/(aA + bB + cC).
Out of n accepted simulated sites, we expect naA/(aA + bB + cC) to be of type a; whereas
we have observed na'. So for the best "closeness" or "fit", we need:
na' = naA/(aA + bB + cC).
By a similar argument for [3 and y we obtain the following three simultaneous equations:
a' = aA/(aA + bB + cC)
b' = bB/(aA 4- bB 4- cC) (1)
c' = cC/(aA + b B + cC).

Solving these equations for A, B and C gives:

 INVESTIGATING SPATIAL RELATIONSHIPS 3

A = a'bc/(a'bc + ab'c + abc')

B = ab'c/(a'bc + ab'c + abc') (2)
C = abc'/(a'bc + ab'c + abe').

If there are only two levels, a and [3, then equations (2) reduce to:

A = a'b/(a'b + ab')
B = ab'/(a'b + ab'). (3)

It can be shown that if a ' = a , b ' = b and c ' = c , then A = B = C = 0 . 3 3 ; or for two levels,
that if a ' = a and b ' = b, then A = B = 0.50. Thus, for the three-category case, a value of A
such that 0.00 ~<A ~<0.33 implies that type a has been avoided; whilst a value of A in the
range 0.33~<A~< 1.00 implies that type a has been favoured. Similar interpretations
apply to B and C. However, this does not necessarily mean that the category has been
significantly favoured or avoided, a subject we consider in the following section.
The necessary extension of equations (2) to cater for four or more categories is obvious,
e.g. for four categories:

A = a'bcd/(a'bcd+ ab'cd+ abc'd+ abcd').

Further, the approach can also be used to examine a number of variables simultaneously.
For example, if we wish to consider two variables, each divided into two or three
categories, they can be combined to make a new variable with either four, six or nine
categories.
For example, for two variables, each with two categories:

Categories
Variable 1 al [31 '
Variable 2 a2 ~2
New variable ala2 axe2 ~1a2 ~1132"

Significance Testing
It is possible to determine whether any category has been either significantly favoured or
significantly avoided by testing two sets of hypotheses. In the first instance we use:
Ho: No association vs Hi: Is association--at least one category is favoured, with a test
statistic of M = Max(A,B . . . . ) and critical region M > K so that p r o b ( M > K ) = c o . The
critical value K must be determined in advance to produce the correct Type I error co
(usually 5%). In the second case, the alternative hypothesis (H1) is that at least one
category is avoided; the test statistic becomes M = Min(A, B . . . . ); there is a different
critical value K; and a critical region M < K so that p r o b ( M < K) = 0~.
It should be noted that where only two categories are involved both sets of hypotheses
are equivalent; if one category is avoided this automatically implies that the other is
favoured. For three or more categories the situation is more complex. For example, if
one category (of three) is significantly favoured then there are a number of possibilities,
including: (1) both of the other two categories are equally avoided, or (2) one is
significantly avoided whilst the other shows neither avoidance nor preference.
We have shown elsewhere (Attwell & Fletcher, 1985) that for the two-category case
appropriate critical values can be obtained by adopting either a theoretical or a simulation
approach. Using these methods values of K w e r e produced for various numbers of sites (n)
and category distributions. F r o m these it was apparent that provided neither category is
rare (i.e. 0-3~a~<0.7) the critical value is independent of the category distribution.
4 M.R. ATTWELL AND M. FLETCHER

Further, by fitting regression lines to the results produced by either the simulation or the
theoretical method it was shown that the critical values can be approximated by:
K=O.5+n-'~(O.3<~a<~0"7). : (4)
To obtain the critical values for situations in which there are more than two categories it is
necessary to use a simulation approach. For three categories, given the values of n, a, b and
c, this involves repeatedly selecting n points from a distribution in which the three categor-
ies are in the same overall proportions as a, b and c; and for these points calculating the
values of a', b', c', A, B, C, Min(A,B,C) and Max(A,B,C). The resultant distributions of
Min(A,B,C) and Max(A,B,C) can then be used to obtain estimates of the required 5th and
95th percentiles (assuming m -- 5%). This task can be performed quickly and easily using a
microcomputer. The generalization of the approach to four or more categories is obvious.
In summary, therefore, for the two-category case a number of methods can be used to
obtain the critical value K. For values of a not extreme (i.e. 0-3 ~ a ~ 0.7) acceptable results
are produced by using equation (4). However, if a microcomputer is available the simu-
lation method has several advantages. The constraints on a do not apply and the program
can also be used to calculate the values of the category weights. In situations where there
are more than two categories it is only possible to calculate the critical values by perform-
ing a simulation on a microcomputer. A program designed to produce critical values for
any given number of sites, number of categories and category distribution is available on
request from the authors.
Finally, it should be noted that although tests of this type may demonstrate an associ-
ation between an environmental variable and a site distribution, this does not necessarily
imply the existence of a causal link. Additional or intermediary variables may have been
involved in producing the observed relationship. In particular, it is essential t h a t o u r use of
the terms "favoured" and "avoided" is viewed in this context.

An Example of the Application of the Technique

To illustrate the application of the new technique to an archaeological data set we have
chosen to consider the relationship between altitude and the distribution of chambered
cairns on the island of Rousay in Orkney. The analysis is based on the 15 Rousay cairn
locations contained in Fraser (1983: appendix A), and the altitude variable has been
subdivided into five categories: 0 ~ 5 m ; 4 6 9 1 m; 92-137 m; 138-183 m~ > 183 m. The
distribution of the cairns and the contour intervals corresponding to the five categories are
shown in Figure 1.
Since there are five categories there will be five category weights A, B, C, D and E. To
calculate their values we need first to obtain the values of 10 probabilities: a, b, c, d, e, a', b',
c', d' and e'. The first five of these represent the proportions of land of each type. These
were estimated by superimposing a grid of 190 points onto the 1 : 50,000 Ordnance Survey
map of the island and counting the number of points falling into each category. The
computation of the remaining five probabilities (the proportions of sites on each category)
proved a relatively easy task since Fraser (1983: 4 4 4 4 4 5 ) provides the altitudes of the
sites. The relevant calculations and the values of the 10 probabilities are shown in Table 1.
For five categories the formulae for calculating the category weights are:
A = a'bcde/(a'bcde + ab'cde + abc'de + abcd'e + abcde')
B = ab'cde/(a'bcde + ab'cde + abc'de + abcd'e +abcde')
C = abc'de/(a'bcde + ab'cde + abc'de + abcd'e + abcde')
D = abcd'e/(a'bcde + ab'cde + abc'de + abcd'e + abcde')
E = abcde'/(a'bcde + ab'cde + abc'de + abcd'e +abcde').
 INVESTIGATING SPATIAL RELATIONSHIPS 5

Figure 1. Map of Rousay, Orkney Islands, showing the distribution of chambered

cairns (following Fraser, 1983) in relation to the five altitude categories: 0-45 m;
48 91 m; 92-137 m; 138-183 m; > 183 m. "

Table 1. Computation of the probabilities a, b, c, d, e, a', b', c', d' ande'

Site Overall
Categories Description No. of sites proportions No. of points proportions

a (weight - A ) 0-45 m 5 a'= 5/15 =0.33 66 a = 66/190 = 0-35

13(weight =B) 46-91 m 7 b' =7/15 =0.47 40 b = 40/190 = 0.21
y (weight = C) 92-137 m 2 c' = 2/15 - 0.13 44 c = 44/190 = 0.23
6 (weight - D) 138-183 m 1 d' = l/15 = 0.06 28 d= 28/190 = 0.15
(weight=E) > 183 m 0 e' =0 12 e= 12/190=0.06
Totals 15 1.00 190 1.00

S u b s t i t u t i n g t h e p r o b a b i l i t i e s a, b, c, d, e, a', b', c', d' a n d e' p r e s e n t e d in T a b l e l i n t o these

e q u a t i o n s gives the f o l l o w i n g values: A = 0 . 2 3 , B = 0 . 5 3 , C = 0 . 1 4 , D = 0 . 1 1 , E = 0 . 0 0 . I f
there is n o a s s o c i a t i o n b e t w e e n t h e sites a n d the five c a t e g o r i e s t h e n the m o s t p r o b a b l e
v a l u e s f o r t h e s e c a t e g o r y w e i g h t s a r e 0-2. I n this case, since B is c o n s i d e r a b l y g r e a t e r t h a n
0.2 it w o u l d a p p e a r t h a t the s e c o n d c a t e g o r y (46-91 m ) is f a v o u r e d . S i m i l a r l y , since E is
c o n s i d e r a b l y less t h a n 0-2 it w o u l d a p p e a r t h a t the fifth c a t e g o r y ( > 183 m ) has b e e n
avoided.
6 M.R. ATTWELL AND M. FLETCHER

However, to test the significance of these results it is necessary to use the simulation
program. As was stated earlier, in addition to producing the required critical values, this
program has been designed to perform most of the calculations above, thus making the
application of the technique relatively simple. To illustrate this the full listing of the run is
presented below:
> RUN
INPUT THE NUMBER OF CATEGORIES
?g
N O W I N P U T T H E N U M B E R O F SITES
?15
INPUT THE EXPECTED PROPORTIONS FOR EACH OF THE 5 CATEGORIES
C A T E G O R Y 1 ?.35
C A T E G O R Y 2 ?.21
C A T E G O R Y 3 ?.23
C A T E G O R Y 4 ?.15
C A T E G O R Y 5 ?.06
N O W I N P U T T H E O B S E R V E D N U M B E R O F SITES IN E A C H C A T E G O R Y
C A T E G O R Y 1 ?5
C A T E G O R Y 2 ?7
C A T E G O R Y 3 ?2
C A T E G O R Y 4 ?1
C A T E G O R Y 5 ?0
T H E O B S E R V E D C A T E G O R Y W E I G H T S ARE:
C A T E G O R Y 1 W E I G H T 0.23
C A T E G O R Y 2 W E I G H T 0.53
C A T E G O R Y 3 W E I G H T 0.14
C A T E G O R Y 4 W E I G H T 0.11
C A T E G O R Y 5 W E I G H T 0.00
95TH P E R C E N T I L E F O R T H E M A X W E I G H T IS 0.52
5TH P E R C E N T I L E F O R M I N W E I G H T IS 0.00
BOTH T H E A B O V E A S S U M E A N U L L H Y P O T H E S I S O F N O A S S O C I A T I O N

To determine whether any category is significantly favoured we test the null hypothesis
that there is no association (H~) against the alternative that at least one category is
favoured (H1). Since we have
M a x ( A , B , C , D , E ) = B = 0.53 > 0.52 = K,
we reject H o at the 5% level and we are 95% confident that the second category (i.e. land
between 46 and 91 m) is favoured. To determine whether any category has been signifi-
cantly avoided we test the null hypothesis (H0) against the alternative that at least one
category is avoided (H1). Since we have
Min(A,B, C ,D ,E) = E = 0.00 ~ 0.00 = K,
we accept H 0 at the 5 % level. Thus, whilst we have some evidence that the fifth category is
avoided it is not statistically significant.
We have, therefore, shown that a significant relationship exists between altitude and the
distribution of chambered cairns on the island of Rousay, with the altitude band 46-91 m
being favoured. It is not our intention here to place any interpretation upon this result
since this would have to form part of a much more extensive analysis.
Davidson's Simulation Approach
The new technique was developed following an analytical examination of the simulation
procedure used by Davidson in his study o f the environmental "factors" influencing the
location of 13 Rousay chambered cairns (Davidson et al., 1976; see also Davidson et al.,
1976b; Davidson, 1979). Davidson's overall aim was to assess the relative importance of
eight selected environmental characteristics in the choice of cairn location. In order to
 INVESTIGATING SPATIAL RELATIONSHIPS 7

expose the relationship between the two approaches a brief summary of Davidson's
method is presented below.
It should be noted that certain key words in the following description differ from the
terminology employed by Davidson. In particular, "factors", "sub-classes" and "factor
weightings"/"factor loadings" are referred to throughout this paper as "variables",
"categories" and "category weightings", respectively. This is to prevent any possible
confusion which may result from the use of words more usually associated with the
technique of factor analysis. The likelihood of this occurring is evidence by Martlew (1981:
45), who erroneously describes Davidson's approach as being based on a multivariable
simulation strategy.
Davidson's method is best summarized by considering initially his treatment of a single
variable. The selected variable was first subdivided into either two or three mutually
exclusive categories (e.g. in the case of altitude: < 76 m; 76-152 m; > 152 m). It is assumed
in the following description that the variable has been divided into three categories: a,
and 7. The two-category case is similar. A regular grid was superimposed on the island of
Rousay and the value of the variable (~, 13or 7) recorded for the 166 points of intersection.
Before each simulation run three integer weightings, A, B and C (A + B + C = 10), were
assigned to the three categories.
Simulated distributions consisting of 13 points were then generated using the following
procedure. A point was chosen at random from the grid and accepted with a probability
A/10 if it was of type a, B/10 if it was type f3and C/10 if it was type y. This selection process
was repeated until 13 points (equal to the number of tombs) had been accepted. The values
of the weightings were varied and eventually distributions of tombs simulated for every
combination of integer values A, B and C such that A + B + C = 10.
By generating large numbers of distributions and comparing the "closeness" of the
simulated patterns to the locations of the actual sites the "best" combination of weight-
ings was identified. "Closeness" was assessed by the following method. For each simu-
lated pattern "the distance between the actual and simulated pattern" was measured
and the "average distance between [the] 13 simulated and 13 actual sites" calculated
(Davidson, 1979: 17). A more precise definition of these distance measures is not given in
either published version of this work (Davidson et al., 1976a; Davidson, 1979), although
from the published computer program (Davidson et al., 1976b) it would appear that the 13
distances were produced by measuring from each simulated site to the nearest actual site
(i.e. the simulated and actual sites were not necessarily paired one-to-one).
The most successful patterns were considered to be those which produced the smallest
such distances and the "best" combination of weightings was interpreted as reflecting the
relative importance of the different categories. The ultimate aim of assessing the relative
importance of the eight selected variables was achieved by comparing the "best" results
obtained during the individual analyses--the variable whose "best" combination of
weightings produced the "closest" simulated pattern to the actual distribution being
regarded as the most important, etc.
From this description it can be seen that Davidson's approach is composed of three
elements: the production of the simulated distributions; a method of determining the
"best" combination of weightings, and the interpretation of the results. The first of these
was mirrored analytically in the derivation of the new technique, the only major difference
between the definitions of the category weightings being that one set the sum to 10, whilst
the other set it to 1. Due to this similarity we expected the two approaches to produce
similar results, with the new technique having the major advantages that it was easier to
apply and did not require extensive use of computer time. To assess whether this was the
case we decided to compare results produced by Davidson with a set produced by the new
technique.
8 M.R. ATTWELL AND M. FLETCHER

Figure 2. Map of Rousayshowingthe distribution of chambered cairns [following

Henshall, 1963;exceptfor Kierfea Hill (A) whichhas been derivedfrom Davidson
et al., 1976a: fig. 7[ in relation Davidson's three altitude categories: <76m;
7(~152 m; > 152m.

Comparison of Results Produced by the Two Approaches

In the example presented earlier, we investigated the relationship between altitude and the
distribution of the Rousay chambered cairns. It is not possible, however, for us to com-
pare these results with those produced by Davidson for the altitude variable, since his
analysis was based on a different cairn distribution and the variable was subdivided into
different categories ( < 76 m; 76-152 m; > 152 m). However, it is a relatively simple task to
apply the analytical technique to Davidson's data set.
The 13 cairn locations included in Davidson's analysis are shown in Figure 2 (following
Henshall, 1963; except for Kierfea Hill which has been derived from Davidson et al.,
1976a: fig. 7), along with the relevant contour intervals. If we are going to use the computer
program to calculate the values of the three category weights, A, B and C, then we simply
need to know the number of sites located on each category and the proportions of land of
each type. This information is shown in Table 2.
Running the computer program for these data produced the following output:
> RUN
INPUT THE NUMBER OF CATEGORIES
?5
NOW INPUT THE NUMBER OF SITES
?13
INPUT THE EXPECTEDPROPORTIONSFOR EACH OF THE 3 CATEGORIES
CATEGORY 1?.49
CATEGORY2 ?.36
CATEGORY 3 ?.15
 INVESTIGATING SPATIAL RELATIONSHIPS 9

Table 2. Number of sites on each category and the proportions of land of each type

Overall
Categories Description No. of sites No. of points proportions

c~(weight = A) < 76 m 11 66 a = 66/190 = 0.35

[3(weight = B) 76-152 m 2 40 b = 40/190 = 0.21
y (weight= C) > 152 rn 0 44 c = 44/290 =0-23
Totals 13 190 1.00

N O W I N P U T T H E OBSERVED N U M B E R OF SITES I N EACH C A T E G O R Y

CATEGORY 1 ?ll
C A T E G O R Y 2 ?2
C A T E G O R Y 3 ?0
THE OBSERVED C A T E G O R Y W E I G H T S ARE:
C A T E G O R Y 1 W E I G H T 0.80
C A T E G O R Y 2 W E I G H T 0.20
C A T E G O R Y 3 W E I G H T 0.00
95TH P E R C E N T I L E F O R THE MAX W E I G H T IS 0.65
5TH P E R C E N T I L E F O R T H E M I N W E I G H T IS 0.00
BOTH THE ABOVE ASSUME A N U L L HYPOTHESIS OF NO A S S O C I A T I O N

Thus, for the 13 locations analysed by Davidson the technique produces category
weightings of 0.80 (A), 0.20 (B) and 0.00 (C). Further, since Max(A,B,C) = A = 0.80 > 0-65
we are 95% confident that land below 76 m has been favoured.
As was stated earlier, due to the similarities between the two approaches we expected a
broad agreement between these results and those produced by Davidson. This was not the
case. We present below the average results of all six of Davidson's computer simulation
"runs" for the altitude variable (Davidson, 1979: 19, table 3). To facilitate a direct com-
parison we have converted the results so that they total to 1 instead of 10. They were as
follows: A =0.06, B= 0'08, C = 0.86. These results are clearly in conflict not only with
those produced by the analytical technique, but also with the observed distribution--they
suggest that land over 152 m was favoured, whilst 11 of the 13 sites lie below 76 m. From
this it would appear that Davidson's optimum weightings do not always accurately reflect
the relative importance of the different categories. This view is reinforced by a comparison
between the results produced by Davidson for the variable "distance from the coast" and
a visual inspection of the distribution map (see Figure 3). Davidson divided this variable
into three categories, < 1 km; 1-2 km; >2 km (Davidson, 1979: 18, table 2), and obtained
weightings (converted as above) of 0-29, 0.31 and 0.40, respectively. This result is again at
total variance with the observed distribution.
The major difference between the two approaches is the method of assessing "goodness
of fit". In the derivation of the analytical technique we adopted a frequency method which
differs fundamentally from the Euclidean distance measure employed by Davidson. In the
light of the above, we suggest that the latter is capable of producing misleading results
and believe that Davidson would have produced similar results to those produced by
the analytical technique if he had employed a more appropriate method of assessing
"goodness of fit". Finally, it is clear that the results of Davidson's analysis of the variables
influencing the location of the Rousay tombs should be approached with caution.

Conclusion
The analytical test presented here should prove a useful addition to the range of quantitive
techniques available for investigating the relationship between a point pattern, whether of
10 M.R. ATTWELL AND M. FLETCHER

N
t

~lkm •
• • f
0 KM 2

Figure 3. Map of Rousay showing the distribution of chambered cairns [following

Henshall, 1963; except Kierfea Hill (A) which has been derived from Davidson et
al., 1976a: fig. 7] in relation to Davidson's "distance from the coast" categories:
< 1 km; 1-2 km; > 2 km.

sites or artifacts, and variables which describe the environment of the pattern. The tech-
nique indicates not only the strength of any association, but also provides information
concerning its direction. It is quick and easy to apply and significance levels can be
obtained readily by computer simulation.
In their important review of spatial analysis in archaeology Hodder & Orton (1976)
suggest the use of two analytical tests to aid the investigation of the relationship between
sites and their environment. These are the X2 test in the case of essentially non-numeric
variables (i.e. nominal data such as soil type) and the K o l m o g o r o v - S m i r n o v test for
variables whose categories can be ranked in a definite order (i.e. ordinal data such ~s
altitude bands). Both these tests, however, can only demonstrate the existence of a
relationship and do not provide any indication of its direction. Furthermore, m a n y
archaeological situations preclude the use of)~ 2, the most widely applicable of the two, due
to the test's inability to handle small data sets or proportionally small categories. An
initial investigation of the effect of both these factors upon the significance levels
associated with our test suggests that similar constraints do not apply.

Acknowledgements
We would like to express our thanks to our colleagues (past and present) at N o r t h
Staffordshire Polytechnic for both their encouragement and criticisms during the develop-
ment of the technique; and to Paul Reilly and Cheryl Turner for their criticisms of drafts of
this paper.
 INVESTIGATING SPATIAL RELATIONSHIPS 11

Copies of the simulation p r o g r a m can be obtained from the authors by writing to:
M. Fletcher, Department of Mathematics, N o r t h Staffordshire Polytechnic, Beaconside,
Stafford ST18 0AD, England,

References
Attwell, M. R. & Fletcher, M. (in press). A new technique for investigating spatial relationships:
significance testing. In (A. Voorrips & S. Loving, Eds) To Pattern the Past: Mathematical
Methods in Archaeology. Journal of the European Study Group on Physical Chemical and
Mathematical Methods" in Archaeology 11.
Cooney, G. (1979). Some aspects of the siting of megalithic tombs in County Leitrim. Journal of
the Royal Society of Antiquaries of Ireland 109, 74-91.
Davidson, D. A. (1979). The Orcadian environment and cairn location. In (C. Renfrew)
Investigations in Orkney. Report of the Research Committee of the Society of Antiquaries,
London 38. London: Society of Antiquaries, pp. 7-20.
Davidson, D. A., Jones, R. L. & Renfrew, C. (1976a). Palaeoenvironmental reconstruction and
evaluation: a case study from Orkney. Transactions of the Institute of British Geographers, n.s.,
1, 346-361.
Davidson, D. A., Richer, S. & Rogers, A. (1976a). A simulation program for the analysis of
archaeological distributions. Newsletter of Computer Archaeology 11(4), 1-10.
Fox, C. (1932). The Personality of Britain. Cardiff: National Museum and University of Wales.
Fraser, D. (1983). Land and Society in Neolithic Orkney. British Archaeological Reports, British
Series, No. 117.
Henshall, A. A. (1963). The Chambered Tombs of Scotland, Vol. I. Edinburgh: Edinburgh
University Press.
Hodder, I. & Orton, C. (1976). Spatial Analysis in Archaeology. Edinburgh University Press.
Martlew, R. (1981). The analysis of prehistoric settlement location. Leicester University
Geography Department Occasional Paper 4.