A statistical analysis
This approach was designed to independently support or refute the first approach through statistical analysis. The visual analysis had established the geographical units that were of greatest interest.
The statistical approach used a number of features that were different to those that were most conspicuous to the eye. It was not impressed by technique or beauty and had few of the inbuilt prejudices of the personal observer. The Correspondence Analysis looked at the same images, but it did so in a different way. There was a bias as a characteristic of this particular algorithm but trends were disclosed from this additional perspective.
I used a Correspondence Analysis algorithm produced by Professor Richard Wright.2
Correspondence Analysis is a method of reducing a large number of objects having a large number of variables to two XY graphs. One is composed from the object scores, the other from the variable scores.
As Baxter (1994:100) says:
A feature of CA [Correspondence Analysis] that has attracted favourable comment is the possibility of simultaneously representing both the rows and the columns of a data matrix as points on the same plot. This allows structure in both rows and columns, as well as their interrelationship, to be examined.
It was fundamental to this analysis that I was comparing images of !people. I was not comparing people. There could be no assumption that any image at any time represented the full array of the visible constituents that went to make up a person. The situation where an image displays a person with no head should not be read as an indication that the head had been removed or weathered away. The first assumption must be that the artist chose not to depict the head or to depict the head in a manner that was foreign to us. The image, if portrayed as active, was possibly the depiction of a person who was alive in the mind-model of the artist at the time of making the figure. We can assume that the person did have a head but it is fallacious reasoning to assume that the figure also had a head.
999999 is the convention used by Professor Wright to express an attribute that is indeterminable. You will find no 999999s in my data entry table for this analysis. This concept does not apply to rock art images.
Correspondence Analysis is designed to use presence/absence data. An attribute is either there or it is not there.
The attributes
The full list of attributes used is a subset of the attributes that could be used to classify or describe all of the individual images.
The attributes were devised in order to be appropriate for the procedures of numerical taxonomy (Sokal 1963). This system follows polythetic rather than monothetic classification procedures (Sneath 1962:291). The requirements of a polythetic class, cluster or genre is that each item within it shares a number of properties that distinguish it from members of other classes.
This is eminently suitable for rock art analysis and assists in avoiding the possible circularity inherent in one person assessing general likenesses and picking attributes.
Clark (1968:36) displays the monothetic/polythetic distinction with the following diagram:
Figure 12.
A model for archaeological entities. A model suggesting that archaeological entities are structured as polythetic group populations and may not be treated as monothetic group populations.
In the following descriptions the use of any particular term should be read as a device intended to particularise an attribute or state. The terms should be regarded as labels rather than definitions.
Directional character
States
'Front Facing': The image is recorded as front facing if there is no indication of any part of its being deflected to the left or the right. In a full infill image it could equally well be back facing. I have chosen to ignore this possibility.
'Right Facing': The image is in full profile and is facing towards the observer's right.
'Left Facing': The image is in full profile and is facing towards the observer's left.
'Other Right': The image is facing in any direction between front facing and right facing.
'Other Left': The image is facing in any direction between front facing and left facing.
'Other Facing': The image has attributes that indicate two or more directional facings. Not to be confused with twisted perspective or contorted where the attitude is considered impossible to exist in the normal world.
Image voice character
State
'Active': The image is not at rest. This state covers the gamut of movements from a raised arm to a flat-out run. Standing, sitting or lying are not factors in this state.
Positional character
States
'Standing': The image is portrayed in a vertical or near vertical position. Motion is not a factor in this attribute.
'Sitting': The image appears to be of a !person bearing its weight on its bottom.
'Lying': The image is in a horizontal or near horizontal position.
'Splits': The legs form approximately 180 degrees with each other.
'Other': It is unclear whether the image should be assigned to one of the previous three states.
Image character
States
'Infill': The image is not an empty outline. It can be a full infill, a stippled infill or a shaded infill. This may include a pattern that would also be entered as a decoration.
'Outline': The image is depicted by an outline. It is often difficult to determine in a photograph if shading infill is the natural colour of the rock or has been depicted in a different colour to the surrounding surface.
'Stick Figure': The image is composed of single lines. A line is a mark best described as being longer than it is wide, with parallel sides. An engraved line can be quite wide but it cannot be described as having infill.
'Enhanced Stick': The image is mostly thin. The lines (sticks) that make it up vary in thickness so are not parallel sided. Unlike stick figures some 'shape' is depicted but less than would be expected in a photographic likeness.
'Free Form': A !human figure that has developed in a symbolic direction. This is one of the rare cases when it is necessary to view a group of images in order to describe one. The image has been transmogrified, a process that requires a series for recognition.
An example:
Figure 13: (after Joussaume 1981: figure 5, 8, 15)
'Contorted': An image that could not represent reality due to the positions of its bodily parts.
An example:
Figure 14: (after Chaloupka 1984: figure 21)
'Swollen': The image of the body is much thicker than is common in rock art. It may be dead, it may be pregnant or grossly obese. This is not an element that was subjected to hypothesis.
'Shaped': The image has a shape that while presumably human, does not appear to relate to reality. The shaping, rather than the portrayal of action or position appeared to be the aim.
'Realistic Shape': The image approaches the normal proportions of a human for the majority of the depiction—length to thickness.
'Rectangular Shape': The image has a rectangular trunk.
'Triangular Shape': Some part of the image is shown as a triangle.
'S-Form': The trunk of the body has a pronounced S shape. Common in earlier Indian images.
'No Head': The image has no head. It may or may not have had a head when originally made. This was not an aspect that was considered.
'Stick Head': The head of the image is represented by a single straight line—a very distinctive feature.
'Twisted': The image is portrayed in twisted perspective. Usually obviously contrived and could not occur naturally.
'In Proportion': The lengths of legs, arms, body and head are in natural or near natural proportion.
'Bow Legs': The legs have a pronounced bow, mostly depicted in front facing images.
'Pointed Feet': The feet have pronounced points where the toes would be.
'Pronounced Bottom': The bottom protrudes. This is a realistic aspect of the physiology of the current San.
'Bird Head': The image has a bird head or mask.
'Animal Head': The image has an animal head or mask.
'Eye': An eye or eyes are depicted.
Image accoutrement character
States
'Large Disc': A large solid circular object is displayed held or attached to the body. It is possible that it is spherical. Some have decorated edges but none have internal decoration. No distinction has been made here between the decorated and undecorated forms.
'Head-dress': The head in the image carries a decoration—very common. abundant hair could account for this in some cases.
'Beehive Head-dress': A specific type of head-dress common to many Australian images.
'Apron Small': The image has an apron or lap-lap. It can be at the front or rear or both but does not go all around the body.
'Apron Large': Similar to the previous but is much bigger. Frequently longer than the legs.
'Skirt': Covers legs or bottom and appears to go around the body. It can be a part of a more complete garment and of any length.
'Boots': The image appears to be wearing high boots with a roll at the calf—restricted to Spain in this collection.
'Waist Band': A band around the waist that appears to be of circular section.
'Tri-tassel': An apparently decorative or ceremonial item that consists of three cords with decorative ends. Usually worn at the waist.
'Tri-sash': Similar to the above but with infill between the cords.
'Nose Bone': A stick like object that appears to pass through the dividing membrane of the nose.
'Wing': An image decoration that appears to be the wing of a swimming bird.
'Feather': Appears to be a feather.
'Fan': An item of decoration that has a fan-like appearance.
'Mirror': An object that appears to be a mirror.
'Hat': A hat-like object.
'Object': An unidentifiable item.
'Bag': A bag.
'Ring': A continuous line forming a circle, oval or compressed in one dimension to an even greater degree.
'String': A piece of string or twine.
'Rope': The image appears to be utilising rope.
'Basket': A basket.
'Vessel': An apparently solid object that by appearance does not fit in the basket or bag class.
'Musical Instrument': An object that appears to be intended for making noise.
'Food': An object that appears to be a consumable at some time.
'Animal': Usually an association.
'Leaf': An item of decoration that appears to be a leaf. May often be erroneously classified as a feather and vice versa.
'Fish': Usually an association, may also be entered as a food attribute.
'Trophy Head': A severed head is being carried by the image. The name is not intended to denote the head is a spoil of war, it could as well be a part of funeral rites or a ceremony. The word 'trophy' is used to differentiate it from the images own head.
'Baby': A baby is depicted on the image's back, papoose style.
Weapon character
States
'Spear Barbed': A spear having one or more barbs.
'Spear Multipoint': A spear having more than one point.
'Spear Pointed': A spear that has one point and no barbs and appears to be too long to be described as a stick.
'Spear Thrower': An item that could have been used as a spear thrower. In some cases it may have been a killing stick or it may have been a multipurpose weapon. I have chosen to classify all such objects as spear throwers.
'Boomerang': A boomerang. No distinction has been made between curved and angled boomerangs.
'Axe': An implement that appears to be an axe, either hafted or unhafted.
'Sword': Where a sword or its natural companion a scabbard are depicted I have indicated their existence as 'sword'.
'Knife': A knife.
'Shield': A shield.
'Bow': Where a bow or its natural companions, arrows or quiver are depicted in an image I have indicated their existence as 'bow'.
'Stick': Appears to be a stick and is shorter than a spear.
'Hook': Any item having a hooked shape. Its purpose has not been considered.
'Club': Thicker than a stick. A purely subjective evaluation.
Other characters
States
'Decoration': An item or items that are associated with or attached to an image that appears to serve no survival function.
'Sexed': An image is sexed if it displays sexual anatomy. Male and female are not differentiated.
Results
The statistical result was impressive.
The Bradshaws of the Kimberley were shown to be unlike overseas images.
Figure 15 is a plot of the first two columns of the object scores of a Correspondence Analysis of 2230 images with 79 variables. The mean centroid of each geographical unit is located at the centre of its name.
The position of each geographical unit on Figure 15, when related to the others, displayed the degree of similarity of the images collected in each unit to those collected in each other unit. One can see the isolation of the Bradshaws in the Kimberley from all other units. One can also see that their closest affinity lay with the Arnhem Land images. This supported the visual analysis.
Figure 16 shows all of the images in the analysis.
It was revealing to inspect the graphs of the images in the geographical units.
Images from the geographical units of Algeria, Arnhem Land and the Kimberley are shown in Figures 17 to 19. The spread of the images from each area was prominent. Possibly it was caused by the difference in the era of the images resulting in the formation of different genres at different times. Variations in the landscapes within a geographical area could have affected the images. There may have been geological factors that divided geographical units of which I am unaware. The intrusion of another culture could have contributed to a diversity of attributes. But above all, from each area, there was a clustering of the images that was significant. It was a factor that contributed to the integrity of the analysis as a whole.
Figure 17: Graph by Correspondence Analysis of all 206 images in Algerian Geographical Unit.
Figure 18: Graph by Correspondence Analysis of all 315 images in Arnhem Land Geographical Unit.
Figure 19: Graph by Correspondence Analysis of all 199 images in Kimberley Geographical Unit.