Some experiments in the numerical analysis of archaeological data

Abstract
The value for archaeology of some available methods of numerical classification was investigated. Brooches from an Iron Age site in Switzerland (Münsingen) were chosen for analysis for two reasons: they show variation over a typically wide range of attributes, and their archaeological context, in rich associations in a horizontally stratified cemetery, provides independent evidence to check the significance of any proposed classifications. A further control on results was provided by an ‘intuitive’ classification of the brooches by professional archaeologists. The brooches were described numerically by quantitative and qualitative characters, and a similarity coefficient was computed between each pair. These coefficients were then submitted to two forms of cluster analysis (single- and average-link) and to a multi-dimensional scaling procedure. The average-link, but not the single-link, cluster analysis and the multi-dimemional scaling procedure were found to produce classifications of demonstrable archaeological significance.