Discrimination of British Lower and Middle Palaeolithic handaxe groups using canonical variates

Abstract

The British handaxe material discussed in this paper has been assiduously collected by Derek Roe and a gazetteer of the British Lower and Middle Palaeolithic sites has been published by the Council for British Archaeology as their research report number 8 (Roe i968a). Further publications cover the metrical approach that Roe has adopted in dealing with the handaxes, a preliminary report (I964), and a more comprehensive coverage later (x968b). In the preliminary paper Roe argues cogently for the primary study of 38 sites that 'all provide a substantial number of implements of which it can be stated with fair or clear certainty that they all belong together, that is, that they occur together within very close limits indeed of time and space' (I964: 249 and I968b: 2). It is these 38 groups of implements (listed in table i) that have been analysed. Archaeological details of the sites and a distribution map are given by Roe (i968b: 3, fig. i) and are not repeated here. From table i it will be seen that the 38 groups give a grand total of some 4,800 axes, yet despite Roe's initial hope that all sites would provide a substantial number of implements site 4 Round Green, site 2I Holybourne and site 30 Swanscombe Upper, have only 14, 19 and i8 examples respectively. None the less, taken as a whole, the 8 measures that have been determined for each of nearly 5,000 axes provides a highly satisfactory set of data for analysis. Such a large sample is all too rare in archaeology. While Roe explains in full (I964 and i968b) how he derived his measures, the vast body of raw data is not published but has been punched on paper tape suitable for computer input. It is this body of data which has been used in the canonical analysis. Before examining the handaxe data in any more detail, it will be as well to have clearly in mind the questions that we are hoping to answer by its further study. The problem is not the usual one in archaeology of trying to divide artefacts into like groups or types, for the axes are already grouped according to their find-site. Instead we wish to see if the 38 assemblages can be satisfactorily distinguished purely by the 8 measured variables. It is to solve this multivariate problem of discriminating between assemblages that canonical analysis has been developed (Bartlett 1965, with references). A canonical variate is a combination of the simple measures in certain proportions, known technically as a linear combination. That is to say, it is a sum of terms of the form: