Page 139 - Kennemerland VOC ship, 1664 - Published Reports

NAUTICALARCHAEOLOGY, 9.1
Weight ( kg 1
Figure 12. The weight distribution of the lead ingots, measured at intervals of
0-3kg.
classes on which they are found, varies widely. Sixteen of the 37 types are found on only one class on ingot (in six cases because they only occur once), while at the other extreme one type (El) is found on 11 of the 13 identified. The former observation suggests a possible relationship between marks and shape classes, although not necessarily a strong one. The best way of summarizing all the information contained in this evidence is to undertake a cluster analysis of the assemblage to search for interrelationships within it. A collection of 119 objects sharing a range of 37 discrete features is actually a perfect data-set on which to undertake such an exercise.
In practice, ten ingots could not be included in this study since no stamps were visible on their upper surfaces, probably as a result of corrosion and abrasion. I t was also decided t o ignore the frequency of occurrence of the different stamps on each ingot, since it was felt that this too had been considerably affected by post-depositional damage. The analysis was thus conducted on the presence or absence of each particular type of stamp on 109 ingots, using the CLUSTAN package on the Cambridge University 3701165 computer. A number of differing clustering routines were used, namely single link, furthest neghbour, average link, and Ward's method (Ward, 1963). As usual in such exercises, average link clustering and Ward's method produced the most satisfactory results, with only minor variations between them; since
the latter is preferred mathematically (Doran & Hodson, 1975: 177) and gave marginally closer agreement with shape classes, discussion will proceed on the basis of the results produced by it. In order to refine further their quality, a relocation routine was applied at the level of 16 -12 clusters, where the most significant discrimination was achieved. The dendrogram produced is presented in Fig. 13, including an indication of those ingots which experienced relocation.
The most striking feature of this den- drogram is the clear identification of nine groups, as indicated by the brackets to the left. In terms of shape classes these groups represent, reading from the top, classes 9 and 6,8,5A,3, 6, 1A and lB, 2A and 2B, and 5B. At the very least this seems t o be useful confirmation of the basic veracity of the classes identified, but it goes further than that. It confirms, for example, our feeling that classes 1A and lB, and 2A and 2B, were related pairs of shapes, while calling in question our pairing of classes 5A and 5B. It also suggests that class 6 might contain two discrete subdivisions, and that one of these is closely related to class 9, something not otherwise suspected. And it arranges the classes in an order of similarity in a way no detailed analysis of shape could ever have done.
Apart from these clearly defined groups, there are 28 ingots misplaced or left floating between groups. Of these, 8 come from classes 7 or 10, possessing only two members res-
20