Searching for structure within multivariate archaeological data

In the interests of objectivity and clarity, archaeologists are turning more to a quantitative representation of their data, and tables, histograms and graphs are now commonplace in the more serious archaeological publications. However, presentation of the data is merely a first stage: objectivity is no doubt honoured, but unless some further treatment of the data is possible, clarity may be missed. This is especially true where the quantified data refer to more than two interacting aspects of the same material, such as the features of a set of pots or the trace-elements in a bronze axe. A direct two-dimensional plot, or even a whole series of these, cannot accommodate this kind of multivariate information. Fortunately, the high-speed computer has made feasible the analysis of multi-dimensional data in ways that were completely impracticable a dozen years ago. It is uncertain at the moment which procedures will ultimately prove to be of most use to archaeologists but, in very general terms, we are looking for methods that will simplify complex data and will reveal structure latent in it. In order to test some likely procedures of this kind, I have taken three contrasting sets of multivariate archaeological data and have analysed them by available numerical methods. The aim at the moment is to search for useful techniques rather than useful results; however, some of the results themselves already seem of interest. It is hoped to publish a full account of these experiments shortly as a monograph. This paper is a brief progress report, originally prepared for a meeting of the Biometric Society (British Region) on Statistical Methods in Archaeology, held on 7 December 1967. The aim of the paper was to show to statisticians the kind of multivariate data with which the archaeologist is faced, and to suggest to archaeologists procedures that are beoming available to help them. The three sets of data chosen for experimental analysis are: