Seriation Based on Agglomerative Clustering: An Example Using Ceramics Imported to Sulawesi, Indonesia

Abstract

Similarity coefficients between artifact samples can be used for constructing larger groupings or for seriating the artifact samples. Indeed, these two approaches work well together, because the construction of groupings assists the seriation of the samples within and across groups. Sequentially grouping the samples into a single total sample, through hierarchical analysis of their coefficients, enables the use of these coefficients to seriate the artifact samples in the reverse order of their grouping. If the main point of interest is the composition of the groupings (for instance, as a summary of attribute similarities between artifact samples), seriation is still valuable in providing the hierarchical structure with an overall orientation that would otherwise be lacking. Alternatively, if the main point of interest is the seriated order (for instance, based on patterns of co-occurrences between the artifact samples), grouping analysis provides a structure to the seriation that would otherwise be lacking. Moreover, because we are dealing with samples, these can be subjected to stepwise agglomeration (rather than hierarchical clustering as commonly understood) during the grouping process, and then subjected to reverse-order sequential partitioning during the seriation process. The advantage of using agglomerated samples is that the coefficients between them can be calculated directly rather than derived indirectly from the constituent sample coefficients. How this approach can be applied to the seriation of artifact types is illustrated through a revisited analysis of high-fired ceramics imported to Sulawesi, Indonesia. Show more