"This case study uses 3D scans of Central European Bronze Age swords (~1400–800BC) to recreate community networks of knowledge. 3D scans of 111 bronze swords were analyzed, from which measurements including blade profile, hilt profile, and decorative shape data were collected. The data were analyzed using a variety of statistical methods. Cluster analysis was used to create links between the nodes of networks that were modelled. A community detection algorithm was run on the networks to examine potential communities of bronze smiths based on theorized manufacturing decisions. These analyses suggest there were four distinct areas within which craft workers were sharing knowledge."
Construction
"Links of the network were established using similar clusters as an indication of a network link. Two types of networks were created; one network used individual swords as the nodes and shared clusters as the links, and the second network used the blade or hilt clusters as the node and the inverted minimum spanning tree (MST) value of the distance between the means of the clusters. For the clusters, the distance between the group means were calculated using ANOVA (Analysis of Variance) or MANOVA (Multivariate Analysis of Variance) values as appropriate. From these, the differences in means were analyzed using a tukey adjustment. This type of adjustment is used to help eliminate Type 1 error across unequal groups and is appropriate for unequal groups of unknown size (Dallal, 2012). A Type 1 error is a false rejection of the null hypothesis. These numbers were used to create a minimum spanning tree. Since the MST interprets higher numbers as objects that are weakly connected, and network analysis interprets higher numbers as a stronger bond, those measurements had to be flipped. The matrices were inverted using the following formula: New variable = largest value + smallest value – original value. This formula results in a change where the largest number becomes the smallest number and each value in between inverts according to that scale. A weight of 1 was assigned every time two swords shared a cluster. Thus, every sword included in the cluster containing all of blade profile 1 shares a link with a weight of 1. Likewise, every sword included in the first cluster of hilts would contain a separate weight of 1. Separate networks were created for each decoration using clusters based on shape data for blade profiles, hilt profiles, concentric circles, dashes, parallel curves, and parallel straight lines; in this case, each link was given a weight of 1. Secondary matrices included a combination of the original matrices where a weight of 1 represented either a shared blade or hilt cluster and a weight of 2 represents both a shared hilt and shared blade cluster. Finally, combined matrices were created using blade and hilt data, all decorative data, and a combination of blade and hilt plus decorative data where the link weight ranged from 1–8 based on the number of clusters shared. These data are available in the supplemental DRUM files (Golubiewski-Davis, 2016)."
Kemencezei, Tibor. 1991. Die Schwerter in Ungarn II. Vol. Praehistorische Bronzefunde IV Band 9. M nchen: C.H. Beck'sche Verlagsbuchhandlung.
Von Quillfeldt, Ingeborg 1995. Die Vollgriffschwerter in S ddeutschland. Vol. Praehistorische Bronzefunde IV Band 11. Stuttgart: Franz Steiner Verlag.
Harding, Anthony. 1995. Die Schwerter im ehemaligen Jugoslawien. Vol. Praehistorische Bronzefunde IV Band 14. Stuttgart: Franz Steiner Verlag.
Waestemann, H. 2004. Die Schwerter in Ostdeutschland. Vol. Praehistorische Bronzefunde IV Band 15. Stuttgart: Franz Steiner Verlag.
Friedrich Laux. 2009. Die Schwerter in Niedersachsen. Vol. Praehistorische Bronzefunde IV Band 17. Stuttgart: Franz Steiner Verlag