Spatial Analysis of Occupation Floors I: Application of Dimensional Analysis,
of Variance
Robert Whallon, Jr.
American Antiquity, Vol. 38, No. 3 (Jul., 1973), 266-278.
Stable URL
htp://links jstor-org/siisici=0002-73 16% 28 197307%2938%3A3% 3C266%3AS A OOFI%3E2.0.CO%3B2-Q
American Antiquity is currently published by Society for American Archaeology
‘Your use of the ISTOR archive indicates your acceptance of JSTOR’s Terms and Conditions of Use, available at
hup:/www,jstororglabout/terms.hml. ISTOR’s Terms and Conditions of Use provides, in part, that unless you
have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and
you may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
hutp:/wwww jstor.org/journals/sam. html
Each copy of any part of @ JSTOR transmission must contain the same copyright notice that appears on the sereen or
printed page of such transmission.
STOR is an independent not-for-profit organization dedicated to creating and preserving a digital archive of
scholarly journals. For more information regarding JSTOR, please contact support @jstor.org,
bupslwww jstor.org/
"Tue Jun 29 15:19:57 2008SPATIAL ANALYSIS OF OCCUPATION FLOORS I:
APPLICATION OF DIMENSIONAL ANALYSIS OF VARIANCE,
ROBERT WHALLON, JR.
ABSTRACT
A statistical procedure for the analysis of patterns of spatial distribution of artifacts over occupation
areas i outlined. This procedure envisions 3 steps: the testing for nonandomness of spatial concent
tions in the distribution of each class of artifacts, the reorganization of the data so as to best reflect such
concentrations in the analysis of intercomelation among those artifact classes which exhibit significant spatial
patterning, and finally the analis of spatial intercorelations among these artifact classes and the definition of
‘lsters of classes of artifacts based on ther similarly of distabution over the area concerned, A new statistical
‘method called dimensional analysis of variance is presented asa method with which to begin this series of steps.
Dimensional analysis of variance works with data inthe form of counts per gid unit, thus making
‘many archacologieal situations for which no other statistical methods are available for spatial
procedure of spatial analysis, bared on dimensional analysis of variance, is illustrated with the analysis of @
preceramic occupation floor ina small cave ia the valley of Oaxaca, Mexico,
Museum of Anthropology
University of Michigan
April, 1972
APPROACHES TO SPATIAL ANALYSIS.
INTEREST in the analysis of patterns of spatial distribution of artifacts and other items on
prehistoric occupation floors has been growing rapidly in recent years. This interest has been most
pronounced in Paleolithic studies, but examples can also be found in the analysis of sites of other
periods. The aim of such analyses is generally to define “tool kits,” or clusters of artifacts and
‘other items which occur together on occupation floors as a consequence of having been used
together in certain activities. It is hoped that inferences concerning patterns of prehistoric human
activity can be made by interpreting these “tool kits” in terms of their contents and their position
‘on the occupation floors. Spatial analysis would thus be of direct relevance to many of the current
questions in prehistory.
Virtually all archaeological examples of analysis of such spatial pattems have been based upon
inspection and impressionistic interpretation (For instance, Leroi-Gourhan and Brézillon 1966;
Freeman and Butzer 1966; de Lumley and others 1969), All artifacts or items of each class under
study are mapped on a plan of the excavated area of occupation. Each class is usually treated
separately, but as similarities in the patterns of distribution are noted among classes, joint maps are
often used to illustrate or confirm the similarities observed. In this standard procedure, the degree
of similarity or dissimilarity between distributions is judged by inspection. Whether or not a
distribution is significantly patterned is likewise decided by inspection. Since relatively important
inferences of prehistoric patterns of activity are frequently made through the interpretation of the
similarities of distribution among items and of the forms of these distributions in relation to
features within the occupation area, it would seem desirable, in many cases atleast, to attempt to
put the judgment of the significance of spatial aggregation or concentration in a distribution and of
similarity between distributions on a more objective, statistical basis,
In attempting to make some progress in this direction, we regard the problem of the analysis of
spatial pattern in terms of a series of 3 sequential steps, each step dependent upon the preceding
ine. The first step is to determine whether the distribution of artifacts or items of each class being
considered is essentially random or shows a significant tendency toward spatial aggregation or
‘concentration on a scale smaller than the entire area under analysis. The second step is to then
take those classes of artifacts or items which do show significant trends toward concentration and
to reorganize the original data so that similarities or correlations which are based on, or most
266Whatton) SPATIAL ANALYSIS 261
strongly reflect, these concentrations can be caleulated among the classes. As the final step, the
Similarities or correlations in spatial distribution are calculated and are organized or displayed in
some form, either as an ordered matrix or as a ree defined by cluster analysis, so that groups of
autifacts and items which are similarly distributed over the area considered are defined, along with
the relationships among these groups or clusters,
‘These 3 steps form the heart of the analysis of spatial patter in archaeological materials. The
crucial and distinctive archaeological task in utilizing such analyses lies then in the interpretation
of the results, in the formulating by inference of hypotheses conceming prehistoric patterns of
hhuman activity which would have let archaeological remains structured inthe manner revealed by
the analysis. Various aids may be profitably employed in this phase of research, particularly the
plotting of single classes or of entre clusters of clases of items on a plan of the area under analysis
to show the form of spatial distribution, especially in elation to various features in the area.
Although such procedures or aids ate usually extremely valuable and often indispensible for
interpretation and inference, the core of a rigorous, quantified approach to the study of spatial
patterns remains a statistical analysis as outlined above. It is therefore this sort of analysis with
‘hich we will be concerned inthis paper
There are at present 2 basically different statistical approaches to the kind of analysis we are
considering, One approach is the use of what we will call dimensional analysis of variance.
Dimensional analysis of variance is very little known in archaeology but it is the method which we
will discuss in this paper. The other approach is the use of nearest neighbor statistics, Ths method
is better known in archaeology and has been previously applied in a small way inthe field. We will
be concemmed here with nearest neighbor analysis only generally and in a comparative manner,
contrasting some of its advantages and disadvantages in archaeological applications to those of
dimensional analysis of variance.
These 2 methods require very different types of basic data from which to proceed and operate
fon quite different principles. They are applicable under quite different conditions; therefore,
although there isa fair range of overlap in the conditions under which each can be applied, and
each has its own particular series of advantages and disadvantages, these differences naturally
strongly affect the application ofthese methods in archaeology.
Dimensional analysis of variance utilizes data in the form of counts per grid square. Nearest
neighbor analysis requires 2-dimensional coordinates for the exact position of each individual
autifact or item. This basic difference in the type of data required is obviously of central
importance in considering the application of these 2 methods in archaeology. The use of data in
the form of counts per square gives the dimensional analysis of variance a great advantage in that a
substantial amount of previously gathered data which was excavated by ged squares and for which
exact coordinates for individual artifacts and items can never be obtained may be analyzed by this
method, while nearest neighbor analysis cannot be applied. In terms of the strategy of data
collection, also, it is often quicker, and sometimes necessary (as in salvage excavations for
example) to gather data in terms of counts per grid unit. Coordinates for each item are not only
more time-consuming to collect and record but ate often impossible to obtain
The recording of coordinate data and the application of nearest neighbor methods may
frequently be worthwhile, however. The constraints imposed by the use of a grid and by the
method of dimensional analysis of variance itself often pose severe problems concerning the shape
ofthe excavated area to be considered and the fineness or precision with which the scale and shape
of spatial concentrations can be defined. Nearest neighbor analysis generally eliminates these
problems of shape and scale.
Its, of course, always possible to reduce the more precise and particularstc data necessary for
nearest neighbor analysis to the more general form required for dimensional analysis of vatiance if
problems of shape and scale are not envisaged and the latter method of analysis is considered
advantageous for other reasons.
Dimensional analysis of variance requires a square or rectangular grid. The number of grid units
slong each side of this square or rectangle must be some power of 2, that i, 4, 8, 16, 32, 64, and
0 on, Ifthe grid is rectangular, one dimension must be just twice the other, as 8 x 16, 64 x 128,268 american antiqurcy [Vol 38, No. 3, 1973
and so on, These are obviously very severe restrictions indeed, considering the shapes of most
excavations and most known occupation floors. Is often possible to satisy the above conditions
of shape, however, by selecting the central portion of the excavated occupation, leaving aside a
few minor, peripheral areas, and/or adding a few imaginary or “dummy” grid units to the area
being analyzed as we will seein the example below. It is also frequently possible to subdivide a
large area into several smaller areas of the requisite shapes. These subsareas can be analyzed
separately by the method of dimensional analysis of variance and the results combined for the
final stop in spatial analysis, intercorrelating the various clases of artifacts and items. Ii further
possible in the same manner to treat Separate, unconnected areas of excavation independently by
this method, finally combining the results from each area for the lst step of analysis
The use of exact coordinates and the method of nearest neighbor analysis eliminates the
restrictions on the shape of the area to be considered, An atea shaped in virtually any regular or
inregular manner can be handled with nearest neighbor analysis, No portions of the area need be
eliminated from analysis, and no extra or “dummy” areas need be added. It is very difficult to
combine separate excavated areas, even of the same occupation, in a single analysis, however,
particularly inthe final step of defining the associations or intercortelations among clases.
Dimensional analysis of variance is limited also in the scale, or size, of spatial pattern which it
an detect and define. The minimum scale at which randomness or nonandomness of the
distribution of any class of items can be determined is, of course, the size of the minimal or
individual gid unit utilized. Beyond the simple determination of randommness/non-randomness,
hhowever, the minimal size of spatial concentration which can be defined effectively by this
method is generally at least twioe the size of the minimal grid unit utilized. This has been
demonstrated experimentally by Kershaw (1957) who thus illustrates the strong effect which
Initial grid unit size has upon the scale of patterning which will be detectable by this method.
‘The mechanics of dimensional analysis of variance require the calculation, as illustrated below,
of corected mean squares at different “block sizes” in which each block above the size of the
original grid units consists of double the area of the next smaller block. This necessitates
regrouping the data at each step into blocks with an area 2, 4, 8, 16, 32, and so on, times the
original grid square. This means on the one hand that areal biocks are alternately square and
rectangular and, on the other hand, that only those specific sizes of arca are tested for
randomness/non-randomness and degree of spatial concentration. The range in size of area of
possible concentrations between the sizes actualy tested obviously increases logarithmically with
the doubling of block sizes, with ¢ consequent progressive loss of fineness or prevsion in detecting
spatial patterns. The observation of a significant spatial concentration of items at block size 32
therefore does not necessarily mean that this isthe actual size of the concentration, but only that
the actual size wil lie somewhere in the range between an area 16 and 64 times the area of one of
the original grid squares Piclow 1969:105), The block size 32 is simply the best approximation of
the actual size of the concentration which can be reached by doubling and redoubling the orginal
grid units,
‘The shape of such a best approximation s ether square or rectangular due to the regrouping of
the original grid units as described above. Obviously, this may be a rather poor approximation of
the actual shape of any concentration. The rectangular block sizes in particular may give a poor
representation of the shape of a spatial concentration, especially if the concentration ise is
elongated but oriented differently from the rectangular block sizes. In square grids, a better
approximation may sometimes be made by doing 2 analyses, first grouping square blocks into
horizontal and then into vertical rectangles and using the results of the analysis which gives the
best or strongest patterning. The analysis of a rectangular grid requires the formation of oblong
blocks which ate oriented in the same way as the whole grid, but sucha grid can always be treated
‘as 2 separate square grids iit is thought that a rectangular grouping in the other orientation would
be preferable.
Nearest neighbor analysis avoids these problems of scale and shape of spatial pattern. Any shape
of concentration of virtually any size Is detectable end definable. A problem arises, however, if
there are multiple scales of patterning or concentration potentially present in the data. Nearest‘Whatlon} SPATIAL ANALYSIS 269
neighbor analysis is generally restricted to dealing with the smallest scale of significant
concentration. It is not particularly suited to handling the occurrence of larger-scale concentra:
tions, even if it is extended in the form of analysis of distances to second, third, or nth nearest
neighbor. Dimensional analysis of variance can isolate multiple scales of patterning if they are
present, however. Spatial concentrations are indicated by peaks in the graph of mean squares or
variances against block size, as illustrated below, and concentrations at more than 1 block size are
easily and clearly defined by multiple peaks in the graph.
After a dimensional analysis of variance, itis very easy to regroup the original data at the block
size which best represents the scale of spatial patterning revealed by the analysis. Correlations
among the classes of artifacts or other items can then be calculated on this basis. There are still
some significant problems involved in determining spatial associations or correlations which are
derived from the results of nearest neighbor analysis. This is at the moment the weakest point of
nearest neighbor analysis and, in contrast, one of the greater advantages of dimensional analysis of
variance, This, and some of the other considerations necessary to the application of nearest
neighbor methods to archaeological data, will be treated in a subsequent paper.
It. might at first appear simpler and mote direct to proceed immediately to this step of
correlation analysis from the original counts per grid unit. An analysis of spatial concentration or
aggregation and the consequent reorganization of the data, our first and second steps, are generally
necessary, however. In dealing with a spatial distribution, the use of a grid and the size of the units
of that grid usually, if not inevitably, have a direct effect on the results of any analysis of
concentration, association, or correlation in the data. It is quite easy to show that significant
tendencies toward concentration or aggregation can be demonstrated or made to disappear and
that high positive correlations or associations can be turned into high negative ones (or vice versa)
simply by expanding or contracting the grid units in terms of which the data is expressed. Such
variations must be controlled in some way by the methods adopted for spatial analysis
Both dimensional analysis of variance and nearest neighbor analysis have their own distinctive
advantages, disadvantages, and possibilities for application to the analysis of spatial patterning of
archaeological materials on occupation floors. It can be seen that nearest neighbor analysis is in
some respects the more powerful method, particulatly in its ability to treat ateas of any shape and
to define spatial concentrations at any specific scale and of any configuration, The difficulties
encountered in proceeding to the third, final step of intercorrelation and clustering of classes of
artifacts and items is, however, a severe problem in its application. Its inability to handle in any
‘way data which has been gathered in terms of grid squares is particularly annoying in archaeology
where a large proportion of the data which one would like to analyze has been collected in this
form and where data in the form of coordinates for the position of each artifact or item is only
‘ecasionally and slowly becoming available from certain relatively recent excavations. Dimensional
analysis of variance should, therefore, have a useful and significant range of application in
archaeological research, and it seems worthwhile (0 outline this method here with an example of
hhow it can be successfully and informatively applied,
THE STATISTICAL METHOD
Dimensional analysis of variance was developed in the field of plant ecology (Greig:Smith 1952,
1961, 1964:86.93; Ketshaw 1957, 1964:104-113; Thompson 1958; GreigSmith, Kershaw, and
Andeison 1963; Pielou 1969:104-106), although in this field it is refereed to as the analysis of
pattern using a grid of contiguous quadrats. Its, in its general and most eommon form, simply a
method for detecting and defining spatial concentrations in data expressed as counts per grid unt.
Elaboration of the method has extended it into a dimensional analysis of covariance between
classes (Kershaw 1961), but this extension involves a great volume of calculations and has been
little applied. For archaeological applications it seems simpler and equally informative to use only
the dimensional analysis of variance as the first step in our sequence of 3 operations and
subsequently to apply an ordinary correlation analysis to determine spatial relations among classes270 american antiquity Vol. 38, No. 3, 1973
‘The method requires, as mentioned above, a grid whose sides contain a number of grid units
equal to some power of 2 and which is either square or rectangular with the greater dimension just,
twice the lesser dimension. The grid units are then grouped in the course of anlaysis into
successively larger “blocks,” each twice the area of the next smaller block, until the last and largest
block comprises the entire atea of the grid (Fig. 1.
Randomness of distribution of counts in the original grid units may be tested by use of the
Poisson distribution cither with the variance/mean ratio or by testing the fit of the observed
istribution to. the Poisson distribution with x7 (Greig-Smith 1964:61-64; Kershaw
1964:101-103), This test ignores the influence of larger scales of pattern which may in most cases
be expected to exist, however. A dimensional analysis of variance is more appropriate, therefore,
Which examines the tendency to concentration and allows a corrected variance/mean ratio test of
significant departure from random distribution at each block size
Fig. 1, A grid of size 8 x 8, having 2 block size of 64 for its total area, showing the progressive increase in
block size as grid unite are combined into 2, 4, 8, 16, and 32-unit blocks with horizontal grouping for oblong.
blocks (after Thompson 1958).Whatton] SPATIAL ANALYSIS: am
A dimensional analysis of variance is based on the sums of squares Si, Sz, Se, Ss,
Si6y--+-++Sp for each block size. These sums of squares are calculated as,
where N; is the count of number of items in the ith block of j grid squares, and T is the total
number of grid squares in the original grid (Thompson 1958). In running through the entire
sequence of block sizes, j= 1, 2, 4,8, 16.....Tandi=1, 2, 3,4,5,.....T)j. At the largest
block size, that comprising the total area covered by the grid, Nis imply the total number of
items found in the area of the grid. For the sum of squares at each block size, in other words, the
count of items in each block is squared, these squared values are added up, and then divided by the
‘number of original grid units which are grouped together to form each block of that size,
‘The tendency toward spatial concentration and the scale at which such a tendency occurs are
then determined through the calculation of the variance in count per block at each block size. It
seems clear that as block size approaches the size of any actual spatial concentrations, the
tendency for the concentrations to be entirely contained within one or more blocks to the
exclusion of the other blocks will grow; that is, the variance in count between blocks will increase.
AAs block size passes beyond the size of concentration, the tendency for a concentration to “f
within a block will weaken, and the variance between blocks will consequently decrease.
This variance, or mean square, is calculated for each block size as
= S~Sjo 1D}
‘where D; isthe degrees of freedom appropriate to block size j and is defined as
pet
1a
‘The standard manner of displaying the results of this analysis of variance has been, besides the
usual table of block sizes, degrees of freedom, and mean squares, a graph of mean square versus
block size. Such a graph, which we illustrate below (Figs. 3 and 4), clearly shows tendencies to
concentration as peaks in the line of the graph at certain block sizes, This not only gives a visual
impression of the tendency to concentration but also allows a determination of the scale or size of
area at which this tendency most strongly expresses itself.
Testing the significance of observed peaks in the graph of mean square versus block size is
relatively difficult and has been the subject of some discussion (Thompson 1958; GreigSmith
1961, 1964:86-87). The first suggestion, to test the significance of variances (mean squares) at
larger block sizes against the variance at block size 1 (the original grid units) by means of an F, or
variance ratio, test, has been shown to be statistically unjustifiable. The necessary assumption of
normal and independent distribution of the counts in the blocks within the grid is shown to be
violated as soon as a “significant” tendency to concentration, that is, to non-randomness of
distribution, is demonstrated by the test (Thompson 1958:326; Greig Smith 1964:86), The
variance at the different block sizes cannot, in fact, be independent since they are calculated from
counts obtained by successively combining the counts in the original grid units rather than from
independent sampling of spatial distribution at different scales (Pielou 1969:105).
‘The generally accepted solution in ecological analysis is simply to use consistency of results
from several analyzed samples as the criterion for significance. This is not satisfactory for
archaeological application since most “samples” of occupation floors are unique, There are
relatively few comparable situations for which relevant data is available.
If the graph of mean squares is expressed in terms of variance/mean ratios, that is, as a
distribution with unit variance, however, significance bands for these ratio at different block sizesm american antiquity IVol. 38, No. 3, 1973
Fig. 2. Floor plan of the cave of Guila Naguitz showing the rid of 1m squares by which it was excavated.
The squares added to the central part to form a 8 x8 m grid for analyse ate indicated by dashed lines,
can readily be calculated (Thompton 1958; GreigSmith 1961, 1964:87), If a distribution is
random with unit variance, sums of squares with D, degrees of freedom are distributed as x? with
D, degrees of freedom and variance, or mean squate, is distributed as x*/D,, Significance bands
tten for, say, a confidence interval of 95%are calculated with the upper limit for sums of squares
U equal to the value of x? for p=.025 and D, degrees of freedom, The lower limit L is given by
the value of x? for p=.975 with D, degrees of freedom. The corresponding values for upper and
lower bounds for variance (in this case variance/mean ratio) ate given by U/D, and L/D). Tables for
these 95% confidence intervals afe given in Greig-Smith (1961, 1964:230). In this manner a
variance/mean ratio can be calculated for each block size by dividing the mean square at that block
size by the mean number of items per block. This process is applied to all block sizes, including
block size 1, whereby a “corrected” value for this ratio for the original grid units can be obtained.
The one disadvantage of these confidence intervals is that they rapidly become very large as the
number of degrees of freedom decreases, that is, as block sizes become larger (see Figs. 3 and 4)
This allows a wide range of random fluctuation of the variance/mean ratio at larger block sizes. It
seems that regularly recurring peaks of variance at these larger block sizes can be dismissed as,
random by the application of confidence limits to the variance/mean ratio. In ecological work it
appears that consistency of results is often preferable to confidence intervals as a criterion for
“significance” in cases like this, and it seems probable that such will also often be the case in
archaeological analysis. In short, subjective judgment of the form of the graphs will perhaps
frequently be desirable, but a rigorous statistical test for significance is available in dimensional
analysis of variance.
‘A dimensional analysis of variance can, therefore, be used as the first step in the analysis of,
spatial pattern of archaeological materials Counted by grid squares over an occupation area. It will
detect and define concentrations and the scale on which they occur. Further, it readily allows the
reorganization of the original data in a form appropriate for the best reflection of these
concentrations or aggregations in the study of associations or correlations between classes of items.Whallon} SPATIAL ANALYSIS: ey
25
20
15
° WEAN SOUKRE
™
Sox CONFIDENCE BANDS:
1204 8 16 32
hock site
Fig, 3. Graphs of mean square and variance/mean ratio versus block sie for Jatropha or sus nuts with the 95%
confidence interval indicated
WEAN SOUARE
10
8
1204 8 6 32
tLoce site
Fig. 4
raph of mean square versus block size for Mapighis or nanche fruitsm american antiquity [Vol. 38, No. 3, 1973
This reorganization is quite easily accomplished by expressing the data in the form of counts per
block for the block size at which the majority of the classes examined show the greatest significant
peak of variance. These counts are then used in a simple correlation analysis, and the matrix of
correlation coefficients is ordered to display the clusters of classes which theoretically represent
prehistoric “tool kits.” If there are obviously 2 major scales of pattern in the data as revealed by
the dimensional analysis of variance, the data can be grouped as counts per block for each block
size separately, 2 correlation analyses made, and the results compared.
AN EXAMPLE OF SPATIAL ANALYSIS
‘As an example of spatial analysis based on the use of dimensional analysis of variance we will
use the tiny, preceramic cave site of Guila Naquitz excavated under the direction of Kent Flannery
in the valley of Oaxaca, Mexico. Three early occupation levels represent the small, seasonal camps
of hunters and gatherers. These levels were excavated in 1 m squares (Fig. 2). Any sort of nearest
neighbor analysis was therefore impossible. Contour maps of density per 1 m? showed clear
tendencies for most of the classes of items found on these occupation floors to be spatially
aggregated. These concentrations were too confusing in their overlapping, however, to allow any
clear definition of clusters of spatially associated classes of items. The situation was thus ideal for
the application of dimensional analysis of variance. Only the results from one occupation level will
be briefly presented here. The full results and their interpretation will be published in the
excavation report.
‘The real grid of 1 m excavation units was neither square nor rectangular. It did not fit the
requitements outlined above for the application of dimensional analysis of variance. The-central
portion of the grid was not far from square, however, and through the addition of a few imaginary
fr “dummy” squares as previously mentioned, a square grid 8 m ona side was constructed which
covered the major portion of the occupation area and was of appropriate dimensions for analysis.
‘There were 18 classes of items which occurred in significant quantities on this occupation floor,
All but 1 of these classes, that of flint débitage, were either plant or animal remains from the
hhunting and gathering activities of the prehistoric inhabitants. Each of these classes was subjected
to dimensional analysis of variance.
The results were quite clear, and the tendency to spatial concentration was easily seen in
virtually all the analyses. The graphs of mean squares generally showed a striking major peak as in
the case of Jatropha or susf nuts (Fig. 3). In this example, as generally, the mean squares were
VARINNCE/RERK ATIC
s
2
1
°
1204 8 6 2
af " (al ti
cc
HUGH SIZE AND DEGREES OF FREEDOM)
Fig, 5. Graph of variance/mean ratio versus block size for Maighia with 98% confidence bands for this ratio,
showing the significance ofthe peaks a Block sizes 1,2, and 8.Whatton} SPATIAL ANALYSIS: 2s
high, and the variance/mean ratios were also high, all far above the upper confidence limit for
random values of this ratio at the various block sizes, In one instance, that of Malpighia, the West
Indian cherry or nanche fruit, the mean squares were extremely small due to the very low density
of this item on the occupation floor (Fig. 4). The variance/mean ratios were also low, but the
application of 95% confidence intervals as a test of significance at each block size clearly
demonstrated the reality of spatial pattern revealed by the peaks in the graph (Fig. 5).
Such results were gratifying confirmation of the tendency to strong spatial concentration
already observed on the contour maps of density. They also demonstrated a relatively high
frequency of strongest spatial patterning at block size 8 (Table 1). This successfully completed the
first step in the analysis of spatial pattern on this floor. The data were accordingly regrouped in
units of block size 8, all classes of items were intercorrelated using the standard Pearson's ¢
coefficient, and the resultant correlation matrix was ordered with Craytor's program SERIATE,
(Craytor and Johnson 1968).
‘Table 1. Block sizes at which the peak variance was observed for each ofthe 18 clases of items on the oceu-
pation floor of Gua Naquitz eave discussed inthe text,
Block size 1248 16 32
[Number of variables with maximum 10.0.9 3° $
Variance at this block size
Several clusters of highly intercorrelated classes of items can be defined from this ordered
correlation matrix (Table 2). Consideration of the members of each cluster in terms of the
environmental zone from which they come, their time of availability, and the type of human
activity which they reflect strongly suggests that the analysis has isolated true clusters of archae-
ological significance, They represent different sets of prehistoric human activities which were
carried out at different times and places and possibly by somewhat different work groups. The
following environmental data has been provided by Kent Flannery (personal communication),
In group 1 we find a series of plant items which come from the Lower Thorn Forest of the
piedmont. The 2 tree legumes Acacia and Prosopis or mesquite, the leaf (or stem section) of
Opuntia the prickly pear (nopal), and the leaves and chewed quids of Agave the century plant or
‘maguey are all items which are well known to occur in the Lower Thor Forest and which ean
easily be gathered at the same time on a collecting trip. They are highly associated spatially within
the occupation area and seem to have been processed together as the products of one type of
gathering activity. Chewing maguey while working seems to have been common, judging by the
Quids and their strong association with this group. All these items show little or no cortelation
with other items.
Included in this group we find also a wild Cucurbita. It was previously suspected that this plant
‘was collected from the Lower Thorn Forest, but there was no proof that it actually did come from
this environment zone. Now, on the basis of the strong association of this curcubit with the other
plants collected strictly from this zone, this suspicion is strengthened, if not substantiated,
Group 2is composed of the fruit of Maipighia, the West Indian cherry or nanche, Nanches come
from the Upper Oak Woodland, but fruit somewhat earlier than the other items which are
gathered from this zone and which form group 3b, Group 2 pethaps represents earlier collecting
trips into this zone.
‘The plants in group 3a, Lucaena/Lysiloma (pods of 2 tree legumes, both known as guajes) and
Jatropha or susf nuts, come from the Lower Thorn Forest. They tend to cluster a bit upslope on
the piedmont, however, and may be best collected on special gathering forays,
Group 3b comprises acoms, pine nuts (pinyon), the small fruit Celtis or hackberry, and
Phaseolus, wild runner beans. Ail these items are from the Upper Oak Woodlands, the wild beans
being found in the underbrush.[Vol. 38, No. 3, 1973
amenican antiquity
216
2ynay undo-s dor “any s200 “ageyi~y dnosp ouors paddy (0)
“mossy ‘su}29 swHODY ‘siNU woKuIA (Q)‘oydodey ‘auoney/ouavonT (#)~€ nord MYBdNc Uno%e ‘Idosauy *HIguRonD ‘AEH SN OMG "BOY 1 40H)
PEPEPE PPS PEPEPP PPE?
pr ePpe PETE Leiba
i : ppg’ bige
ovenppur nor? Bupynses ya's 221s 39019 201g JOM SluNe> uodn paseq suaISI}209 woNETANOD Jo MINEW PA!PIO 7 =IMELWhatton] SPATIAL ANALYSIS: an
Chipped stone waste flakes form group 3c. They are corelated both withthe items in groups 3a
and b and with those in group 4. These simple flakes clearly have multiple functions and are used
in both plant processing and in butchering game. Their low correlation with groups 1 and 2,
however, shows that they are used for only some types of plant processing
The 3 animals hunted, (der, urtle, and rabbit) are not naturally found together but
nevertheless appear to have been butchered or processed together. The deer come from the Upper
Oak Woodlands, the turtle from ponds on the valley floor, and rabbits are ubiquitous, We
have here perhaps an indication of an activity or activities differentiated from others not so much
by the environmental zone being exploited or the time of exploitation, but by the composition of
the work group involved. Hunting and butchering would seem to be, on comparative grounds,
primarily an activity of the men of the group, while the gathering of plant products would more
likely have been done by a group of both sexes or by a women's group,
Finally, the fruit of the eactus Opuntia, the tuna or prickly pear, stands alone as group S,
uncortelated or negatively cortelated with all other items. This fruit obviously comes from the
Lower Thorn Forest, andthe leaves or stems ofthis same plant are found associated with group 1.
This i, however, an economically important fruit. Either it was simply processed separately from
the other plant products gathered from the same zone, or it might be hypothesized that it was
such an important food that it was gathered in large quantities on trips on which whole days were
devoted to nothing else. This would perhaps explain its independent spatial distribution and
frequent negative comelation with other items.
The above is just a brief overview of the application of dimensional analysis of variance to the
occupation floors in this cave, of the results, and of their interpretation. The emphasis in this
paper has been upon the presentation of 1 rigorous, quantitative approach to the analysis of spatial
pattern on occupation floors. The example presented above hopefully wll give some idea of the
actual use of the method and what can be expected from it.
Spatial analysis could be and should be a useful tool in future archaeological research, Many of
the important problems of interpretation of certain archaeological data revolve around the
question of the functional reality of certain hypothetical “tool kits." This question can often be
easily answered by the application of the methods of spatial analysis to the study of the
distribution of artifacts and other items on occupation floors as has been illustrated above, We
hhope that with a method now available some progress can be made inthis ditection.
(Cayor, Wiliam Bert, and LeRoy Johnson, J.
1968 Refinemenis in computerized item seriation. University of Oregon Museum of Natural History,
Bulletin 10,
de Lumiey, Henry, and others
1969. Une cabane Acheuléenne dans la Grote du Lazatet (Nie), Société Prélstorique Frangate, Mémoires
7,
Freeman, L. G., Jt, and K. W. Butzer
1966" The Acheulean Station of Torralba (Spain). A progres report. Quoternaria 8:9-22,
GreigSmith, P.
1982 The use of random and contiguous quadrats in the study of the structure of plant communities.
Annals of Botany 8. 16:293-316.
1961” Data on pattern within plant communities . The analysis of patter. Journal of Eeology 49:695-702.
1964. Quantiative plant ecology, 2nd. edition. Plenum Press, New Yo
GreigSmith, P, KA. Kershaw, and DJ. Anderson
1963. The analysis of pattern in vegetation: comment on a paper by D. W, Goodall, Jounal of Ecology
51:225.229,
Kershaw, KA.
1957" The use of cover and frequency in the detection of pattern in plant communities. Ecology
38:291.299,
1961 Associaton and co-ariance analysis of plant communities. Journal of Eeology 49:643-684,
1964 Quantitative and Dynamic Ecology, American Elsevier, New York.
LeroiGourhan, Ande, and Michel Brézillon
1966. Lihabitation Magdalenienne No. 1 de Pincevent prés Montereau (Seine-*t- Marne). Galle Préhiszore
9:263-385
Plow, F.
1969" An introduction ro mathematica ecology. Wily Interscience, New Yorkx8 amenican antiquity [Vol 38, No. 3, 1973
‘Thompson, H.R.
1958. The statistic
Botany 6:322-342
study of plant distribution patterns using a grid of quadrats, Australion Journal of