5.5 Multivariate statistical methods

Yet an intuitive description of a graphical site typology may offer difficult choices. What should be done with a site like 52B-223, which displays a reasonable range, but is small in size and has little production debris. Should it join the intermediate sites on the basis of the range, or the small sites on the basis of the limited size? These dilemmas are part and parcel of an intuitive approach. We simply have to make a decision based on all our (personal) archaeological knowledge and experience. Such an approach is equally valid as allowing a mathematical criterion to govern the decision-making process. This is supported even more when we include in our considerations various factors such as the collection of the data, knowledge of the research area and the period, all of which can hardly be expressed in the 'cold' numbers of the data files.

When a site contains a large number of different artefact types, it is often no longer possible to display this graphically and we don't get a picture of the pattern in the data. For the Michelsberg sites we have for this purpose chosen a specific order of the segments: the artefact types with similar function or interpretation are side by side. First are the three types of arrowhead, followed by pointed blades, macroliths and other tools. The next group (axe, pottery and grinding stones) represent 'domestic' activities and finally there are hammer stones, cores and production debris. This facilitates a visual interpretation, even when the number of artefact categories is large.

However, both these drawbacks have led us to search for a more statistical technique to allow a reduction of the amount of information, which yet has a clearly graphic basis. We have chosen a technique called Multi Dimensional Scaling (MDS), which aims at displaying the differences in artefact composition between all sites in a 2- or 3-dimensional diagram (Doran & Hodson 1975; Orton 1980; Shennan 1988). The result is a simple graphic display, expressing as best as possible all differences among the sites. This technique therefore meets most of our requirements.