5 A graphical site-typological analysis of the Michelsberg sites
For graphic analysis of multivariate data we can also use pie charts and segment diagrams, beside the Tukey stars mentioned before. Pie charts have the drawback that the value of a variable is not always in the same position in the circle; furthermore, they can only be used for percentages. The Tukey stars and segment diagrams do not have these drawbacks. In a Tukey star the value is expressed by the length of the line from the centre. A connecting line between the far ends expresses the multivariate data in the shape of a star. The sequence of the variables is important, however, the variables have become dependent visually because of the connecting line. A segment diagram, where the value is expressed in the radius of the segment, does not have this drawback.
![[Pie, Tukey and segment diagram]](images/segment.gif)
![[Michelsberg sites]](images/mkschkl.gif)
Fig. 10 The Michelsberg sites in diagrams
Order the sites yourself (Javascript)
The segment diagrams of the progressive class values provide, we feel, a good picture of the artefact composition of the Michelsberg sites. It combines into one type of graph much of the information also expressed in other data and diagrams, such as the absolute size of the sites and the range of artefact types. The segment diagrams of the sites may be printed side by side, but also on the distribution map of the core region of Venray, allowing a direct impression of the site types and their spatial context.
The highly intuitive character of such a graphic analysis raises the question whether there is a more statistical technique to effect a reduction of the information, with a clearly graphical base. The technique to fit in best seems to be Multi Dimensional Scaling (MDS) (Doran & Hodson 1975). It aims to represent graphically all differences between the sites, as expressed in a dissimilarity coefficient (we have chosen the Phi value), as best as possible. The lower the dimensions of the resulting picture (3D-space, 2D-plane, 1D-line), the harder it is for MDS to do justice to all differences. When the strain in the graph is not too large, we prefer a relatively easy to interpret 2-D figure. The 23 Michelsberg sites have in this way been arranged in a MDS graph and represented as segment diagrams.
more details about:
Résumé of the archaeological goal
Multivariate statistical methods
Displaying similarity matrices
Methodical and archaeological conclusions about the Michelsberg sites
![[Multi Dimensional Scaling: Michelsberg]](images/mkmds.gif)
Fig. 12 Multi Dimensional Scaling diagram for the Michelsberg phase. The sites are depicted as segment diagrams and the 'balloons' indicate differing artefact types
MDS appears to be a good method of arranging sites typologically. Just like many other multivariate techniques, however, the assumptions and choices, such as the dissimilarity coefficient or dimension of the picture, greatly influence the results. We therefore only use MDS in addition to the tables and segment diagrams.
Archaeologically, three site types may be distinguished in the meagre data of the Michelsberg phase. On the basis of the combination of the graphic results we have decided to classify as follows: a group of small sites with commonly a single macrolithic artefact, a group of sites with a wider range of artefact types and much production debris, and a group of sites with arrowheads only.