Patty. Jo Watson
Steven A. Leblanc
A Comparative Statistical Analysis of Painted Pottery from
Seven Halafian Sites
In: Paléorient. 1973, Vol. 1 N°1. pp. 117-133.
Abstract
The painted pottery from seven Halaf sites spanning almost the entire known range of the culture was examined. Motif similarity
was compared by the Sj statistic of Sokal and Sneath. The results of this statistic were diagrammed by linking the most closely
related sites together as clusters, then overlaying the results into one figure. This figure indicates that the patterns of site
similarity were based primarily on geographical distance. In general, change is clinal in form, and dichotomous subgroups of sites
were not observed, negating the suggestion of basic eastern and western variants of Halaf.
Résumé
Examen de la poterie peinte de sept sites halafiens. La similitude des motifs a été étudiée selon la méthode statistique Sj de
Sokal et Sneath. Ces résultats statistiques ont donné lieu à des diagrammes rassemblant en faisceaux les sites dont la poterie
est la plus semblable puis en réunissant ces résultats en une seule figure. Celle-ci a montré que la similitude des motifs d'un site
à l'autre est dépendante de leur distance géographique. Il n'a pas été possible de distinguer de sous-groupes, ce qui permet de
nier la suggestion de variantes fondamentales entre la culture halafienne de l'est et celle de l'ouest.
Citer ce document / Cite this document :
Watson Patty. Jo, Leblanc Steven A. A Comparative Statistical Analysis of Painted Pottery from Seven Halafian Sites. In:
Paléorient. 1973, Vol. 1 N°1. pp. 117-133.
doi : 10.3406/paleo.1973.906
http://www.persee.fr/web/revues/home/prescript/article/paleo_0153-9345_1973_num_1_1_906
A COMPARATIVE
STATISTICAL ANALYSIS
OF PAINTED POTTERY FROM SEVEN
HALAFIAN SITES
Steven A. LeBlancPatty Jo Watson
ABSTRACT
The painted pottery from seven Halaf sites spanning almost the entire known range of the culture was examined. Motif similarity was
compared by the Sj statistic of Sokal and Sneath. The results of this statistic were diagrammed by linking the most closely related sites
together as clusters, then overlaying the results into one figure. This figure indicates that the patterns of site similarity were based
primarily on geographical distance. In general, change is clinal in form, and dichotomous subgroups of sites were not observed,
negating the suggestion of basic eastern and western variants of Halaf
RESUME
Examen de la poterie peinte de sept sites halafiens. La similitude des motifs a été étudiée selon la méthode statistique Sj de Sokal et
Sneath. Ces résultats statistiques ont donné lieu à des diagrammes rassemblant en faisceaux les sites dont la poterie est la plus semblable
puis en réunissant ces résultats en une seule figure. Celle-ci a montré que la similitude des motifs d'un site à l'autre est dépendante de
leur distance géographique. Il n'a pas été possible de distinguer de sous-groupes, ce qui permet de nier la suggestion de variantes
fondamentales entre la culture halafienne de l'est et celle de l'ouest.
INTRODUCTION
In the past ten years there has been increasing interest in
the nomothetic potentialities of the prehistoric archeological
record a This has been accompanied by a greater
emphasis on quantification procedures, often computerassisted.
However, there is a very large corpus of archeological
data collected at various times in the past and
reported with little or no attention to systematic quantif
ication. Often it is highly desirable to use such material
for comparative purposes, although this may be difficult
to do. In the present paper we give an example of a
procedure applied to ceramics which proved successful
and which may be widely applicable. As part of the analysis
of materials recovered from the Halafian site of Girikihaciyan
in Southeastern Turkey2 we wished to compare,
in a quantitative way, the painted pottery from Girikihaciyan
with that from the other known Halafian sites in
order to assess the relative similarities and dissimilarities.
Sites containing Halafian painted pottery, the hallmark
of the Halaf Culture, are spread across the northern part
of the Fertile Crescent (see fig. 1). In so far as is presently
known, these sites represent the first occurrence in South
westAsia of a widespread cultural horizon. Within a rela
tively short time, in the early 5th millenium B.C., this pot
tery style became common over an extremely large area,
and modified forms or influences of the style occur in
adjacent areas, especially to the west. This similarity in
painted pottery appears to be paralleled by common forms
of other artifacts, for example stamp seals, and by architec
ture(tholos type of construction) at the various Halaf sites.
The rapid distribution of a common style over a great geo
graphic range —an horizon style — is a cultural phenomenon
known to have occurred prehistorically many times both
in the Old and New Worlds. However, the strong similar
ityamong far distant sites which characterizes the Halaf
ianperiod does not seem to have been equalled in South
westAsia in the Hassunan or earlier periods. Because the
nature of an horizon style is of general importance to
our knowledge of culture and cultural process, and the
Halafian is of particular importance in Southwestern
Asian prehistory, an understanding of the kind and degree
of similarity among these sites is of great interest.
Halafian painted pottery is relatively homogeneous if
one confines himself to sites within the Halaf area proper
and ignores the "influence" areas. This is seen most clearly
when one compares Halaf ware with other painted styles
(Hassunan, Samarran, Ubaidian, Jemdat Nasr, etc.) : the
distinction is invariably quite sharp.
However, internal variation within the Halafian has not
been intensively studied, although Perkins3
gested that the Halaf can be divided into eastern and wes
tern divisions. She bases her suggestion on three factors:
(1) Western Assemblages are notas consistent as Eastern ones. For instance,
pottery from Arpachiyah, Tepe Gaura Nineveh, and Hassunah shows less
variation from site to site than pottery from Shaghir Ba/.ar. lull al-Halal,
and Carchemish. (2) Eastern Assemblages show a clearer sequence; technical
qualities, forms and designs indicate a constant progression from earlier
strata to later ones. This might, of course, be accidental, since the manner
of excavation of the Eastern sites has given us on the whole better stratif
ication than that of the Western sites. (3) Eastern Assemblages are more
highly developed than Western ones. Nowhere in the West is there anything
comparable to the fine polychrome plates of Arpachiyah 6. (Perkins,
1949:44).
While this assessment cannot be lightly dismissed, it was
based on rather limited selections of sherds from many of
the sites, and was impressionistic and not quantified in any
way. Furthermore, large collections of sherds from several
Halaf sites not available to Perkins increases greatly the
material now available for comparison, and adds to our
knowledge of the areal distribution of the sites themselves.
Thus, the degree of internal variability within the Halaf
painted ware is worth re-examination.
Seven Halaf sites were chosen for study on the basis of their
1. BINFORD, 1962 : 217-225. FRITZ and PLOG, 1970
:
405-412. ISAAC, 1971
:
123-129. WATSON, LEBLANC
and REDMAN, 1971.
2. BRAIDWOOD, CAMBEL and WATSON, 1969 : 1275-
1276. BRAIDWOOD, ÇAMBEL, REDMAN and WATS
ON, 1971
:
1236-1240. LEBLANC, 1971.
3. PERKINS, 1949 : 44.
117
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
DISTRIBUTION MAP OF THE SEVEN HALAF SITES
O Halaf sites included in thepresent study
• Other Halaf sites
Fig. 1.
90
geographical positions and the amount of data available
on their painted pottery. The problem was, essentially,
to find the relative degrees of similarity among these sites
so that conclusions could be drawn concerning the follow ing
questions: ' Is there really an East-West division of the
Halafian style painted pottery? 2 Do other clusters of
sites exist? 3 What appear to be the main determinants
of similarity for these sites? Ideally, the inclusion of mater
ialother than the painted pottery in the study would have
been desirable; however, the painted pottery was the only
artifact category that could be used to make valid compar
isonsbecause of the highly selective recovery and descrip
tionof material from many of these sites. Further, for the
purposes of the present study and in the absence of clear
evidence one way or the other, we have assumed contem
poraneity among the chosen sites. As more information
accumulates, it may become apparent that this assumption
is incorrect, but until that time we believe it more fruitful
to make the operating assumption be that of contemporan
eityrather than noncontemporaneity.
In order to enable the reader to assess the nature and
amount of data employed in this study and to define the
nature of the sites themselves, we present a brief description
of each of them.
1. Arpachiyah. This mound, excavated by Mallowan in
the early thirties, is one of the best reported and most wellknown
of the Halafian sites. It is located northeast of
Mosul on the Mesopotamian plain, and covers an area
not much greater than 125 meters in diameter. Of its total
depth of 10.5 meters, apparently about 8 meters belongs
to the Halaf period, which is overlain by an Ubaid period
occupation. Mallowan divided the Halaf occupation into
at least five stages and felt that there was some evolution
in the painted pottery through this interval.
The present analysis is based on the painted pottery illus-
118
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
tra ted in the published report 4 and that made available
by the British Institute of Archeology in London: a total
of over 4.000 sherds. This latter howtver, is certainly a
selected sample of the total material recovered from this
site.
The general quality of workmanship of the artifacts from
Arpachiyah equals or exceeds that of any other Halaf site
known. Because of the occurrence of items displaying this
high quality of workmanship within an apparent workshop
area, one must assume that the best Halaf material could
be produced in sites no larger than modern villages in this
region, and housing a population probably no greater than
a few hundred people.
2. Banahilk. This site, located in the hills of Iraqi Kurdistan
near modern Rawanduz, is only slightly smaller than Arpa
chiyah being 100 by 160 meters. It has very little later
occupation and the depth of the Halaf material is about
3 meters. The sounding by Watson " was limited in
extent, but the full range of Halafian artifacts was recovere
d.However, thev\ariety of painted pottery forms, and
the level of execution exhibited by them, is clearly less
than that of Arpachiyah. Thus, while the size of the site
compares favorably with Arpachiyah, Banahilk is in some
ways more peripheral. About 2900 painted ware sherds
from this site were available for study.
3. Girikihaciyan. Being situated at the northernmost part
of the Fertile Crescent sixty kilometers north of Diyarbakir,
Turkey, this site appears to be close to the north edge of the
Halaf range (recent work in the Keban area further north
possibly extends this range). Although it is near the head
waters of the Tigris drainage, it is almost directly north of
Tell Halaf. Surface debris covers an area greater than 250
meters by 150 meters; the maximum depth is about 3.5
meters. It was surface — collected and sounded in 1968,
and excavated in 1970 7 . Again the full Halafian artifact
assemblage was found, but what appears to be an atypical
pattern emerged. Of some 34,500 sherds recovered, only
12-13 percent were of the painted style (many of these
were in poor condition, and only slightly over 1,000 sherds
were usable for comparative purposes). This ratio
contrasts strikingly with the 50-60 % painted ware found
at Banahilk, and the preponderance of painted ware over
plainware reported for Arpachiyah.
A subjective impression of the pottery definitely suggests
inferior craftmanship in comparison with that of Arpa
chiyah or Tell Halaf. Here, then, is a case of variation from
at least some other Halaf sites, and an indication that the
site was probably a marginal one.
4. Tell Halaf. Located on the Khabur, a tributary of the
Euphrates, almost at the present Syrian-Turkish border,
this site was the first major Halaf site excavated. The
huge mound, the central portion alone being 350 by 300
meters, is composed of material from many different
periods. It is difficult to assess the extent of the Halaf
occupation, but it can be no less than 250 meters in max
imum diameter; and in places is at least over five meters in
depth. The Khabur appears to be an area of dense Halaf
occupation which includes the mounds of Chagar Bazar
and Tell Brak as well as others containing Halaf material
8
.
Unfortunately, little is known of the non-ceramic
material or the proportions of painted ware to plainware
at Tell Halaf. Thus while it appears to be a major Western
Halaf site, it is virtually impossible to assess its relative
size or sophistication with respect to the other sites.
Except for a few sherds seen at the British Institute of
Archeology in London, the entire sample was drawn from
the published account of the site 9
.
5. Tilki Tepe. (Samiramalti). The material from this site on
the edge of Lake Van in eastern Turkey has been considered
to represent an occurrence of Halaf material well outside
its assumed range 10 Although this view must be
modified with the discovery of Halaf material at Giriki
haciyan. in the Elazig region, and near Siirt which extends
the northern range of the culture, Tilki Tepe is certainly
at the northeastern margin of the Halaf area. Nothing is
known of either the extent of the area of occupation, the
depth of deposits, nor the total range of the assemblage
found at this site. Notes on some three hundred sherds
from this site, which are housed in the Archeological
Museum in Ankara, taken by Watson in 1955 were avai
lable.
6. Turlu. This recently excavated site on the Euphrates
near Ni/.ip in Turkey lies near the western boundary of the
Halafian culture area. Only a preliminary note has been
issued on the excavation
''
; thus it is impossible to assess
the nature of the deposits, although the total height of the
tell is approximately 30 meters and apparently only two
of the earlier levels were actually Halafian.
The assemblage appears to be typical, although no informat
ionis available on the painted ware to plainware ratio,
and pottery as finely executed as some of that from Arpa
chiyah or Tell Halaf does not seem to occur. Permission
was kindly granted by the excavator — Jean Perrot ■—
for us to examine over a thousand painted pottery sherds
from Turlu stored at the nearby museum in Gaziantep.
While this collection may not have contained all the pottery
that was excavated, it did probably include all sherds with
recognizable motifs.
7. Yunus (Carchemish). This site is also situated near the
Euphrates and lies on the Turkish-Syrian border. It was
excavated early in the century by Woolley 12 , and little
information is available on the non-pottery artifacts or on
the nature of the deposits, except that the material did
not seem to be overlain by later occupations. The pottery
4. MALLOWAN and ROSE, 1935
:
1-178.
5. WATSON, 1965 ; n.d.. BRAIDWOOD, HOWE et al.,
1960.
6. REDMAN and WATSON, 1970
:
279-291. BRAIDWOOD,
ÇAMBEL and WATSON, 1969
:
1275-1276.
7. WATSON et al, in prep... BRAIDWOOD, ÇAMBEL,
REDMAN and WATSON, 1971
:
1236-1276.
8. MALLOWAN, 1936
:
91-177.
9. OPPENHEIM, 1943.
10. MELLAART. 1965 et 1967.
1
1. PERROT as reported by MELLINK, 1964
:
156.
12. WOOL LEY. 1934
:
146-162.
119
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
was found associated with what are called kilns by the
excavator, some of which may actually have been small
tholoi.
The material from these seven sites can be summarized in
terms of the quantities of sherds we were able to observe,
and the degree to which the available sample probably
represents the true range of the pottery that occurred at
the sites.
Arpachiyah: Largest sample, selected collection but
probably representative.
Banahilk: Large sample, unselected, very likely represent
ativeexcept for very rare occurrences.
Girikihaciyan: Large sample, unselected, very likely repre
sentative except for very rare occurrences.
Tell Halaf: Small sample, highly selected, does represent a
wide range of the pottery, but probably is not complete
.
Tilki Tepe: Very small sample, unlikely that it represents
the full range of pottery.
Turlu: Large sample, probably not highly selected, pro
bably does represent most of the variation.
Yunus: Small sample, highly selected, probably does not
represent the full range of pottery.
These seven sites provide a good sample of the geographical
distribution of the Halafian materials. However, there is
not sufficient information on most of them to enable one to
assess the variation of the entire Halaf assemblage among
them. In fact, there is not enough information even to
compare the pottery from these sites. To do this properly
one would need at least a knowledge of the painted ware
to plainware ratios and the percentage occurrence of the
various vessel shapes at these sites, but this information
is generally unavailable. The one item that did not appear
to be greatly biased by selective saving of artifacts was
the nature of the designs on the painted pottery (although
rare motifs may be overrepresented), and they more accu
rately reflect the true nature of the universe from which
they were removed than does any other available category
at these sites.
Thus, an initial assumption was made: the design elements
included in the data from each site were an unbiased sample
of the design elements present at that site. Although this
was probably not completely true it did appear close enough
to reality to make the analysis valid and worthwhile. It
was also felt that sherds might have been selectively retained
by excavators on the basis of sherd size, skill shown in the
painting, and how well the sherd had been preserved.
However, it appears unlikely that selection was based on the
location of design elements on the vessels. Therefore, it
was considered valid to include in the analysis data concern
inglocation of motifs as well as their presence or absence
from a collection.
The analysis then consists of comparing these seven sites
on the bases of the presence and absence of design motifs
and the location of these motifs on the vessels.
METHOD
The Halafian pottery was divided into five distinct vessel
forms. While others have classified this pottery differently
or more finely If , the forms defined here are easily di
stinguishable at all the sites, and are few enough in number
so that the sample size in each form class did not become
impossibly small. These five shapes were (1) Concave Sided
and Straight Sided Bowls , (2) Flare Rim Bowls — a general
class that includes the well-known "cream bowl" form ;
(3) Round Sided Bowls, with or without flat bases ; (4) Jars,
including a distinctive form of very squat jar or closed bowl
called by Schmidt a "Biichse" ; and (5) Hole Mouth Vess
els, (fig. 2).
The design elements found to occur for each vessel category
were then defined. A total of 78 motifs were included in the
final analysis, and these are listed in fig. 3. Some shapes
such as round sided bowls were not present in great quant
ities, so for these shapes several closely related motifs were
combined for purposes of the analysis. For example, four
different forms of cable motif were defined (designs 2, 3,
4 and 5 in fig. 3) but these were combined into one category
for the analysis of the Round Sided Bowls !4 .
Other motifs or particular variants of these 78 motifs did
occur on some vessels. However, in order to avoid includ
ingmotifs represented at only one site, or resulting from
idiosyncratic behavior, it was necessary to formulate a cut
offpoint for inclusion of motifs in the analysis. First, to be
of any use in comparison a motif must occur in at least two
sites. Secondly, in order to exclude rare variants and
accidents in painting, a motif had to occur twice at two
different sites before it could be included in the analysis.
Once a motif met this inclusion test—occuring at least
twice at two different sites—it needed to occur only once
at any of the remaining sites to be counted as present lor
that site. This inclusion test was made for each vessel shape
independently. Thus motifs were included in the jar analysis
which were excluded from the concave bowl analysis, etc.
It can be validly argued that if a motif was considered
valid for the jar class, then it very likely represented a valid
motif and could have been used with the other form classes.
On the other hand, any motif that did not occur at least
twice in two different sites for a particular shape was cer
tainly a rare motif for that shape. Excluding these extre
mely rare motifs helped compensate for the small samples
of sherds from some of the sites. A motif that occurred only
once out of 4,000 sherds at Arpachiyah and once out of
1,000 sherds at Turlu was almost certain not to turn up in
13. PERKINS, 1949. MALLOWAN, SCHMIDT in
OPPIZNHEIM, 1943.
14. This shape is considered early Halaf by PERKINS, theo
retically occurring at Arpachiyah with the tholoi without
dromos entrances
;
however, at Girikinaci, they occur with
dromos tholoi, and therefore this distinction does not
appear universally valid.
120
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
BASIC HALAFIAN VESSEL SHAPES
Y / X_J
Flare Rim Bowls Concave Sided Bowls
Round Sided Bowls
Holemouth Vessels Jars
Fig. 2.
the samples from Yunus, Tilki Tepe, or Tell Halaf, even if
it was present on these sites at the same frequency as it
was for Turlu or Arpachiyah. By including very rare motifs
one would essentially be measuring the effects of varying
sample size, and not really getting at motif differences.
This bias was certainly not eliminated in this analysis, but
it was reduced as much as possible.
STATISTICS EMPLOYED
Once the presence and absence of motifs had been recorded
for each shape, it was necessary to quantify the degrees of
similarity represented by these data. It was decided to
employ the methods of cluster analysis or numeric taxo
nomy to assess this similarity. These techniques, popular
izedby Sokal and Sneath (1963), provide ways of finding
patterns in the morphological similarity of any collection of
items. The basic assumption underlying this approach is
that all morphological differences are equally valid for
assigning overall rankings of similarity; no traits are given
greater weight than any others. A correlate of this assump-
121
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
78 HALAFIAN PAINTED POTTERY MOTIFS
COCO
V
Fig. 3. Care has been taken by the artists to reproduce as closelyas possible the manner in which these motifs were executedby the prehistoric potters, hence the unevenness and lum-piness of many of the lines.122
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
35
37
40
41
42
о оооо
ооаоо43
© ©
0000©44
45
, . , .• - . >
47
48
шшшТТГ
49
0 0 8
О бее
51
52
\w\\\\\\
m62
64
72
ГТТ
73
///I/I/
74
75
76
77
78
123
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
Present
Absent
Present
A
С
Absent
В
D
tion is that the assessment of similarity must be based on
many different independent characters. In certain cases
the assumptions of this method are possibly questionable,
but it is ideal for this particular problem.
There is absolutely no basis for assigning some motifs grea
terimportance than others in defining overall similarity,
and the number of independent motifs was great enough
to permit the valid use of this technique.
Sokal and Sneath present a list of possible coefficients of
similarity for use with presence/ absence information. For
all these statistics, the data for each pair of items to be
compared is arranged in a two by two table; in our case
such tables were made for each pair of sites when analizing
shape class. These tables take the form:
Site 1
Number of Motifs
Site 2
In this diagram A represents the number of motifs that
were common to both sites 1 and 2; В represents the
number of motifs present at site 2, but absent at site 1.
С is the reverse of B, and represents the number of motifs
present at site 1 but absent at site 2; and D represents the
number of motifs that did not occur at either site 1 or site 2.
A represents the positive matches; В and С represent nonmatches,
and D gives the number of negative matches.
The major difference among the various statistics defined
by Sokal and Sneath is whether negative matches (D) are
included or excluded in their computation. The decision
whether to include or exclude negative matches is not tri
vial. One must decide whether the absence of trait at two
particular sites reflects real similarity between them. Traits
can be imagined for which this would or would not be the
case. Clearly the absence of steel from two Halaf sites does
not demonstrate a common similarity between them,
beyond the fact that they are both prehistoric. On the other
hand, the common lack of rim ticks on Round Sided Bowls
could well reflect a basic similarity between two sites. Even
in this case, however, the common lack of rim ticks does
not demonstrate similarity in as strong a way as the
common occurrence of tholos type house construction does.
Negative matches are valid measures of relationships in
some circumstances, but in general, they are probably never
so strong a clue as are positive matches.
In this particular case, because the samples might not
completely reflect the actual occurrences of motifs at these
sites and because of the general considerations just
ed, the use of negative matches in the assessment of simi
larity did not seem warranted.
This conclusion is much the same as that reached by Sokal
and Sneath, and the statistic that they prefer— one that
excludes negative matches from the computations—was
employed. This statistic, labeled Sj can be defined as
follows using the same symbols as above
:
S; = A
A+B+C
S: is found by taking the number of positive matches (A)
and dividing by the sum of the positive matches (A) and
the non-matches (B and C). Sj results in a coefficient that
ranges from 1 with perfect similarity between two sites, i.e.
В and С are both zero, to zero when no traits are common
to both sites.
It must be noted that all such similarity statistics are depen
denton the definition and number of traits used in making
the comparison, and one cannot assign meaning to the
absolute value of the coefficient. A Sj equal to. 5 represents
neither a high nor a low degree of similarity, but is high or
low only when compared with other sites in the same anal
ysis. (It is clear then that one cannot compare statistics
between sites based on different types of data.) It would
not be correct to assert that if known Halaf sites showed an
average similarity of. 6 and Ubaid sites only. 5, the Halaf
was more homogeneous than the Ubaid. However, even
with this limitation, a coefficient of similarity is still a very
useful tool.
The results of the analysis of each shape class will be given
separately, followed by an interpretation of the overall
results.
Jars
Of all the vessel shapes, jars provided the best data on the
relative similarity among the sites. This was because of the
large number of different motifs that occur on jars and
because they were a very common shape (approaching 50
percent of the vessels in most collections), so that a good
sample was available from all the sites. A total of 66 motifs
fit the inclusion criterion of two occurrences at two sites.
However, 12 of these motifs were found at all seven sites,
and thus could not provide us with any information on
differences among the sites. Therefore, motifs 1,2,9,12,31,
36,38,39,41,48,56 and 64 were excluded from this step in
the analysis. Fig.4 presents the data on occurrences of the
motifs at the seven sites. The coefficient, Sj , was then
computed for each pair of sites, and these values are given
in Fig.5. The matrix of Fig.5 represents each coefficient
twice; it is given once (below the diagonal) in fractional
form where the numerator is the number of positive
matches (A) and the denominator is the sum of A and the
124
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
3
4
5
6
7
8
10
11
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
32
33
34
35
37
40
42
43
44
45
46
47
49
50
51
52
53
54
55
57
58
59
60
61
62
63
65
66
FIG. 4
THE DISTRIBUTION OF DESIGN MOTIFS FOR JARS
AT SEVEN HALAF SITES
MOTIFS* SITES
Arpachiyah Banahilk Tilki Tepe Girikihaciyan Tell Halaf Turlu Yunus
* Numbers refer to the motifs in Fig. 3.
125
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
FIG. 5
THE VALUES OF Sj FOR SEVEN HALAF SITES
BASED ON ALL MOTIFS
Arpachiyah
Banahilk
Tilki Tepe
Girikihaciyan
Tell Halaf
Turlu
Yunus
Arpachiyah
25/47
7/46
27/51
41/54
31/54
18/50
Banahilk
.532
3/30
14/44
23/51
16/48
10/38
Tilki Tepe
.152
.100
5/34
6/48
4/41
2/27
Girikihaciyan
.529
.318
.147
29/51
23/47
16/37
Tell Halaf
.759
.451
.125
.569
37/49
19/51
Turlu
.574
.333
.098
.489
.755
17/43
Yunus
.360
.263
• 074
.432
.373
.395
FIG. 6
DISTRIBUTION OF DESIGN MOTIFS ON FLARE RIM
BOWLS FOR FIVE HALAF SITES
MOTIFS*
OUT:
12
21
73
36
6,7,8
74
13
1
38
2,3,4,5
24,25
7
23
57
41
33
IN:
75
72
12
14
76
6,7,8
77
2,3,4,5
28
1
43,44,45,
46,47
32
Arpachiyah
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Girikihaciyan
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
x
X
X
X
X
X
X
X
SITES
Tell Halaf
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Turlu Yunus
126
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
non-matches (B and C). By examining the coefficient as
presented in this form, one can determine the actual number
of common occurrences of these motifs and also the total
number of motifs that occurred at either of the sites being
compared. The other side of the diagonal of the. matrix
gives these fractions as decimals so that the relative degree
of similarity can be assessed.
Flare Rim Bowls
For this shape several motifs were combined in order to
increase the number of motifs that met the inclusion rule.
Motifs that occurred inside the vessels were recorded sepa
rately from those occurring on the outside. In this case not
only were motifs themselves being compared, but the loca
tion of the motifs was also considered. The numbers of
Flare Rim Bowls from Banahilk and Tilki Tepe were so
small that valid comparisons could not be made, so they
were excluded from this part of the analysis.
A total of 30 motifs, 16 outside and 14 inside, meet the
inclusion rule but two of these (36 and 38) occurred on the
inside at all five remaining sites and were therefore ignored.
This data is given as Fig. 6. S, was computed and presented
as in the case of jars. These results are shown in fig. 8.
Round Sided Bowls
The number of vessels in this shape class was not very great
for most sites, and even though it presents a limited amount
of information, the analysis of this class was performed.
Only four sites had enough round sided bowls to be consi
dered and for these 22 motifs met the inclusion rule, 8
occurring on the outside and 14 on the inside (one inside
motif, a combination of motifs 2,3,4 and 5, occurred at all
sites and was excluded). The tabulation of these data is
given in fig. 7 and the resulting coefficients are found in
fig.9.
Hole Mouth Vessels and Concave Sided Bowls
The numbers of vessels in these shape classes were quite
small for several sites, and at the same time the motifs found
on these vessels were either very common or very rare. Thus,
when the motifs were found that fit the inclusion rule and
did not occur at all sites, their numbers were too small to
permit the valid computation of Sj. Therefore, these two
classes could not be used in the comparison of the sites.
All Motifs Combined
A final tally of the seven sites was made combining all
vessel shapes and ignoring the location of the motifs on the
vessels. These results are given in fig. 10 and the values of
S j are given in fig. 1 1 . It can be observed that there is little
difference between these data and those for the jars. Only a
few additional motifs could be included and there were only
small differences in the resulting coefficients.
RESULTS
There are several ways of analyzing the coefficient matrices
that are given in figs.5,8, 9 and 11. The traditional approach
in numeric taxonomy is to use these coefficients to produce
a dendrograph or hierarchial ordering of the sites. Such
diagrams reproduce some but not all the information in the
original matrix. They are made by finding the units that are
most similar, then continuously adding new units to them.
When these additional units are added they are compared
to the clusters already formed and not to each unit as a
distinct entity. Thus these methods result in greater and
greater generalizations about the units (sites) in question.
For many problems this procedure is desirable, and when
there are great numbers of units there is little else that is
feasible to do. However, in this particular case, because
there are so few units involved, we can employ a graphic
method that reproduces more of the information of the
matrices and results in less generalization.
The method employed was to connect diagrammatically
each site with the sites closely related to it; in all cases the
number of closely related sites was limited to one or two.
Thus the data on the jars showed that Tell Halaf was most
closely related to Turlu and Arpachiyah, and the rela
tionships were almost equal for both sites. Banahilk,
however, was much more similar to Arpachiyah than to any
other site, so only one connecting line was made for Banah
ilk. This procedure was continued until all the sites were
connected. The resulting plot of the jar data is given in
fig. 12. While this figure does not carry all the information
of the matrix, it does carry more of it than would a cluster
diagram, and it clearly establishes the basic form of the
relations among these sites. In this figure the values of the
coefficients between the sites are represented by the type of
line connecting the sites. Solid lines represent relatively high
coefficients of about .7; heavy dashes represent the next
highest group of coefficients .5, the light dashes are for
weaker links of about .4; and the dotted lines give the
weakest links.
The most obvious characteristic of this diagram is that the
sites do not form two clusters. The most closely related sites
are Turlu, Tell Halaf, and Arpachiyah; the other four sites
relate primarily to them and do not form any other close
groupings. As mentioned, Banahilk is more closely related
to Arpachiyah than to any other site. Girikihaciyan,
however, was equally related to the main cluster sites of
Tell Halaf and Arpachiyah. Yunus has its closest affinities
(but relatively not very close) with Turlu and Girikihaciyan.
Tilki Tepe is unlike the other sites, but its strongest links
are with Arpachiyah and Girikihaciyan.
These very interesting results do not necessarily fit with
generally held ideas about the Halaf. The main cluster of
sites —Turlu, Arpachiyah and Tell Halaf—includes both
Western and Eastern sites, and this strongly indicates that
there is not a major dichotomous division of the Halaf
culture. Both Girikihaciyan and Banahilk were initially
interpreted as being culturally peripheral in some fashion,
this hypothesis is strengthened for Banahilk, and the single
close bond with the geographically nearest site of Arpa
chiyah fits our intuitive expectation of the manner in which
this marginal site should relate to the rest of the culture.
127
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
FIG. 7
DISTRIBUTION OF DESIGN MOTIFS ON ROUND SIDED
BOWLS FOR FOUR HALAF SITES
MOTIFS*
Arpachiyah
SITES
Banahilk Tell Halaf Yunus
OUT:
10
13
36
6,7,8
2,3,4,5
38,39
33
IN:
78
12,21
1
10
75
13
36
9
76
77
38,39
29
70
Numbers refer to motifs in Fig. 3.
FIGURE 8
THE VALUES OF S j FOR FIVE HALAF SITES
BASED ON FLARE RIM BOWL MOTIFS
Arpachiyah
Girikihaciyan
Tell Halaf
Turlu
Yunus
Arpachiyah
21/28
17/28
7/25
12/26
Girikihaciyan
.750
18/28
/26
13/25
Tell Halaf
.607
.643
6/23
11/23
Turlu
.280
.269
.261
6/16
Yunus
.462
.520
.478
.375
128
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
FIGURE 9
THE VALUES OF Sj FOR FOUR HALAF SITES
BASED ON ROUND SIDED BOWL MOTIFS
Arpachiyah Banahilk Tell Halaf Yunus
Arpachiyah
Banahilk
Tell Halaf
Yunus
18/22
14/22
7/22
.818
10/22
5/20
.636
.455
6/15
.318
.250
.400
Girkihaciyan does not show a similar pattern. Besides being
most closely related to both Tell Halaf and Arpachiyah, it
is linked with Tilki Tepe and Yunus. While Girikihaciyan
is culturally peripheral in the sense that it does not have a
high frequency of painted pottery nor is the pottery crafted
in the best fashion, it does not have a particularly strong
relationship to any one site in the present sample. It is
geographically closest to Tell Halaf, and it is next closest
to Arpachiyah (Girikihaciyan and Arpachiyah are both in
the Tigris drainage); Girikihaciyan motifs are equally
similar to both sites. Tilki Tepe is physically closest to
Arpachiyah and Banahilk in terms of straight-line dis
tance, but from the viewpoint of the actual terrain, it would
be easier to move between Tilki Tepe and Girikihaciyan.
Hence it is not surprising that Tilki Tepe is most similar to
Girikihaciyan. Thus Girikihaciyan is not dyadically related
to one site, but is variously related to several.
Yunus is predictably similar to Turlu, but is unexpectedly
close to Girikihaciyan. It must be noted that all the coeffi
cients for Yunus are relatively low. The same is true for
Tilki Tepe, where they are extremely low, although the
particular sites to which Tilki Tepe is related do not seem
surprising. In these two cases the effects of small sample
sizes are probably evident. An examination of the matrix
of fig.9 shows that there is a tendency for the number of
positive matches to decrease very rapidly as the total
number of motifs decreases. Because the number of posi
tive matches falls off so much more rapidly than does the
total of motifs for these two sites, it is probably the case
that sites with small samples and thus fewer total motifs
had lower coefficients than would have been the case if
bigger samples had been available.
What is not clear from these results is whether the relative
order of similarities among these sites would have been
affected by sample size. That is, would Tilki Tepe have been
most similar to different sites if its sample had been doubled
or would it still be closest to Girkihaciyan and Arpachiyah
and only the magnitudes of the coefficients have been
increased? A similar question might be asked for Yunus.
Because it is not possible to answer this question, the posi
tions of Tilki Tepe and especially Yunus must be viewed
with some caution.
The above discussion has been based on the data given in
the jar analysis. However, an examination of the results of
the Flare Rim Bowl, Round Sided Bowl, and all motifs
combined analyses show that exactly the same general
results were obtained, although some variation in the size
of the coefficients occurred. Each separate analysis placed
the sites in the same relative positions, and thus the conclu
sionsdrawn concerning these relationships were conside
rablyreinforced.
INTERPRETATION
From these results it appears that there is not a single
major partition of all known Halaf sites. However, we may
still ask if there are any other patterns underlying these
similarity relationships. Are there any other facets of he
pottery itself, or the sites in general that correlate well with
the values of the Sj ? One obvious relationship is that Arpa
chiyah and Tell Halaf not only have high coefficients, but
they also have been considered to have produced pottery of
the finest quality. Although subjective assessments on
the quality of craftsmanship are difficult to substantiate
and the results of such assessments vary widely, the
quality of design execution does not seem to be well corre
lated with patterns of motif similarity. Turlu is as similar
to Tell Halaf as is Arpachiyah, but the quality of pottery
manufacturing does not seem significantly superior to that
of the remaining four sites. Furthermore, the range of varia
tionat any one site is quite great. At Girikihaciyan the
average level of expertise was probably lower than for the
main cluster sites. However, a few vessels were found that
very closely approximate what must be considered the
better workmanship from the central cluster sites. In gener
al,then, not only is the determination of quality somewhat
difficult for these sites, but no pattern is seen that correlates
with their motif similarities.
However, while collecting these data several interesting
aspects of the painted pottery were noted. While no
definite conclusion can be drawn, there is some indication
that the relative proportions of the various vessel shapes
vary remarkably from site to site. For example, Round
Sided Bowls and Hole Mouth vessels made up over
129
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
MOTIF*
3
4
5
6
7
8
10
11
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
32
33
34
35
37
40
42
43
44
45
46
47
49
50
51
52
53
54
55
57
58
59
60
61
62
63
65
66
67
68
69
70
71
72
73
THE
Arpachiyah
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
FIGURE 10
DISTRIBUTION OF DESIGN MOTIFS FOR ALL
VESSELS AT SEVEN
Banahilk
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
SITES
HALAF SITES
Tilki Tepe Girikihaciyan
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Tell Halaf
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Turlu
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Yunus
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
•"Numbers refer to the motifs in fig. 3
130
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
FIGURE 11
THE VALUES OF Si FOR SEVEN HALAF SITES
BASED ON ALL MOTIFS
Arpachiyah
Banahilk
Tilki Tepe
Girikihaciyan
Tell Halaf
Turlu
Yunus
Arpachiyah
29/53
9/52
34/57
48/60
36/59
20/56
Banahilk
.547
4/35
18/50
28/57
19/53
11/43
Tilki Tepe
173
114
7/40
8/56
6/45
2/31
Girikihaciyan
596
360
.175
35/58
26/54
18/44
Tell Halaf
800
491
.143
.603
41/56
21/58
Turlu
610
358
.138
.481
.732
19/47
Yunus
357
256
063
.409
.362
.404
FIGURE 12
THE RELATIONSHIPS AMONG SEVEN HALAF SITES INDICATED BY
STATISTICAL ANALYSIS OF PAINTED POTTERY
... TILKITEPE
•••***
GIRIKIHACIYAN
"
4
\ BANAHILKr
YUNUS
I
TELL HALAF ARPACHIYAH
Values of Sj for Jar Motifs
_^_ .76
__- .54— .57
— • .43 .40
.« - .16— .15
eighty percent of all the bowls found at Banahilk, while
at Girikihaciyan both Round Sided Bowls and Hole Mouth
vessels were almost non-existent. At Arpachiyah, these
two forms were almost equally represented (although due
to the selective retention of sherds this cannot be considered
more than an overall impression). Also the number of Flare
Rim Bowls seems to be greater at Turlu and Yunus than
at sites in the eastern edge of the Halaf range. It is clear
that important intersite differences in the painted pottery
exist that are not detectable merely by examining the
design motifs; however, only when sufficient unbiased
samples of pottery are available can this extremely interest
ingvariation be properly analyzed.
Various aspects of the sites other than the painted pottery
were next considered as possible correlates of the motif
relationships exhibited by these sites. Very little could
be found that could account for the observed patterns.
The three primary cluster sites were as large, but not
significantly larger than the rest of the sites, and, although
they might have had deeper Halafian deposits, there is not
enough information to confirm this. The three sites whose
dimensions we know best (Girikihaciyan, Banahilk, and
Arpachiyah) seem remarkably similar and could not have
had greatly different populations. The location of the sites
in either the foothills or the plains does not correlate well
with their relative similarity. Yunus is as much on the plains
as are Tell Halaf, and Arpachiyah, and Turlu is located
more in the foothills than Girikihaciyan.
The one aspect of these sites that does seem to correlate
remarkably well with the motif similarities is the geogra
phicaldistances between them. Except for Yunus, all the
sites are related to each other almost identically for both
motifs and geographical position. While other aspects of
these sites are surely related to other phenomena, the
distribution of design motifs is certainly closely correlated
with the geographic locations of the sites themselves.
It is necessary to point out that this study was concerned
only with the variable aspects of the painted pottery from
these sites. Although intersite variability is certainly
present, one's overwhelming impression is that the overall
similarity among these sites is tremendous. Eighty to ninety
percent of the sherds from any of these sites would not
appear out of place in the collections from any other site.
The point being made is that low coefficients of similarity
do not reflect atypical sites or merely some vague form of
131
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
cultural « influence ». The pottery from Tilki Tepe has
completely typical Halaf motifs, it is simply less like the
other Halaf sites considered here than they are to each
other. Given that the known maximal range of the Halaf is
over 360 miles (500 kilometers), and the differences among
the sites are extremely slight (not only for pottery motifs,
but for other artifacts as well), it is probably safe to
conclude that the Halaf is one of the most homogeneous
prehistoric cultures anywhere in the world. It should be a
primary goal of Near Eastern archeology to determine
carefully the overall similarities and differences within the
Halaf culture, and also to try to explain the extremely high
degree of similarity of the painted pottery designs.
ACKNOWLEDGEMENTS
Collection of the data utilized in this study was financed in
part by a Ford Foundation Traineeship and in part by
National Science Foundation grant GS-1986, and was made
under the auspices of the Prehistoric Project, Universities
of Istanbul and Chicago. We are grateful to the Co-Direct
orsof the Project — Halet Çambel and Robert J. Braidwood
— and to the Project staff for their help and encou
ragement. We should like to thank Benin Aksoy, an
advanced student in archeology at the University of Istan
bul,for her unfailingly cheerful and invaluable help with the
tedious task of recording the thousands of sherds from the
Turkish sites. We are also indebted to Jean Perrot for
giving us permission to examine the Turlu pottery stored
at the archeological museum in Gaziantep, and to the
museum authorities there for their courtesy and help. We
are grateful to David French, Director of the British Insti
tute of Archaeology in Ankara, and to his wife, Lisa, for
advice and aid generously given on several occasions. We
should also like to thank the museum authorities in Istanbul
who aided our examining of the Yunus pottery. In London,
David Oates, Joan Oates, and various staff members of the
Institute of Archaeology were exceedingly kind and helpful
in making available to us their study collections from
several Near Eastern sites, in particular those from Arpachiyah,
and we are very grateful to them. Back in St. Louis,
we were fortunate to have the assistance of a group of
enthusiastic undergraduates who did much of the basic data
tabulating, collating, and keypunching: Babette Black,
Theresa Dickerson, Nancy Haskins, Fred Leykam, Greer
Price, and Mitchell Rothman. Finally, we are grateful to
Mary Morrison for typing the tables, and to Catherine
Brandel and Steven Ward for making the original drawings
of the 78 pottery motifs.
Steven A. LeBlanc
Wichita State University, Kansas, USA.
Patty Jo Watson
Washington Universitv. St-Louis, Missouri, USA.
BIBLIOGRAPHY
BINFORD L.R.
1962 Archaeology as Anthropology, Am. Antiq., 28,
217-225.
BRAIDWOOD R.J., CAMBEL H., REDMAN C.L., and
WATSON P.J.
1971 Beginnings of Village-Farming Communities in
Southeastern Turkey, Proc. Nat. Acad. Sci.
USA, 68, 1236-1240.
BRAIDWOOD R.J., CAMBEL H. and WATSON P.J.
1969 Prehistoric Investigations in Southeastern
Turkey, Science, 164, 1275-1276.
BRAIDWOOD R.J. HOWE B. et al.
1960 Prehistoric Investigations in Iraqi Kurdistan,
Studies in Ancient Oriental civilization,31.Chica
go: The University Chicago Press.
DIRVANA S.
1944 Cerablus Civarinda Tunus'ta Bulunan Tel Halef
keramikléri. Turk Tarih Kurumu Belleten Alt
VIII : 403-420.
FRITZ J. and PLOG F.
1970 The Nature of Archaeological Explanation,
Am. Antiq., 35, 405-412.
ISAAC G.
1971 Whither Archaeology? Antiq., 45, 123-129.
LEBLANC S.
1971 Computerized, Conjunctive Archeology and the
Near Eastern Halaf. Ph.D. Dissertation, Depart
mentof Anthropology, Washington University,
St. Louis, Missouri, U.S.A.
MALLOWAN M.E.L.
1936 The Excavations at Tall Chagar Bazar, and an
Archaeological Survey of the Habur Region.
Second campaign, 1936. Iraq, IV, 91-177.
MALLOWAN, M.E.L. and ROSE J.C.
1935 Excavations at Tall Arpachiyah, Iraq, II, 1-178.
MELLAART J.
1965 Earliest Civilizations of the Near East. London
Thames and Hudson.
1967 The Earliest Settlements in Western Asia from
the Ninth to the End of the Fifth Millenium B.C.
Fascicule of the Cambridge Ancient History, I,
Chapter VII.
MELLINK M.
1964 Archaeology in Asia Minor, Am. J. of Arch.,
68, 148-166.
OPPENHEIM M. von
1943 Tell Halaf, I Die prdhistorische Funde, Bearbeitet
von Humert Schmidt. Berlin : Walter de
Gruyterand Co.
PERKINS A.L.
1949 The Comparative Archeology of Early Mesopot
amia,Studies in Ancient Oriental Civilization,
25. Chicago : University of Chicago Press.
132
COMPARATIVE STATISTICAL ANALYSIS OF PAINTED POTTERY
REDMAN C.L. and WATSON P.J.
1970 Systematic, Intensive Surface Collection, Am.
Antiq., 35, 279-291.
SOKAL R.R. and SNEA1H, P.H.A.
1963 Principles of Numerical Taxonomy. San Fran
cisco and London, W.H. Freeman.
WATSON P.J.
1956 New Halafian Material from Northern Iraq:
the Halaf "Period" Reconsidered. Master's
Thesis, Department of Anthropology, Univers
ityof Chicago.
n.d. Banahilk; an Halafian Site in Northern
Iraq, MS.
WATSON P.J., LEBLANC S. and REDMAN C.L.
1971 Explanation in Archeology; an Explicitly
Scientific Approach. New-York and London
:
Columbia University Press.
WATSONP.J., LEBLANC S. et al.
in prep. Girikihaciyan; an Halafian Site in
Southeastern Turkey.
WOOLLEY C.L.
1934 The Prehistoric Pottery of Carchemish, Iraq,
I, 146-162.
133