Some critical aspects of our method are explained here. Feel free to comment on any part using the ‘reply’ function below.
Evaluation of previous archaeological applications of network analysis
Interdisciplinary bibliographical research
Archaeological case-studies
Software
2-mode networks
m-slices
k-cores
Closeness centrality
Betweenness centrality
Domain
Evaluation of previous archaeological applications of network analysis
For now, the full diversity of network analysis has not yet found its way into Roman archaeology. To explain this, the network principles used in archaeological applications will be traced back to their origins, and their original meaning will be compared with how they were used to answer archaeological questions. This will provide a clearer image of which network analysis ideas were adopted and how they can help enhance Roman societal analysis.
Top
Interdisciplinary bibliographical research
Relational thinking in terms of networks has been discussed in academic literature since the 1960’s (Harary 1969) and new network models are still being developed (in recent years especially in physics, Albert & Barabási 2002). A vast body of literature exists on the application of network analysis in other disciplines, which provides excellent examples of the diversity with which network analysis can be applied. These sources will be consulted to explore how network ideas evolved over the past decades and how they can be critically adopted to fit a Roman archaeology framework. In addition, the use of quantitative tools in these applications will be studied, focussing on the algorithms quantitative tools are based on, their implied assumptions and their possible archaeological application. Some tools with good potential for understanding social interaction in the Roman world will be tested in the archaeological case-studies.
Top
Archaeological case-studies
The potential use for Roman archaeology of network principles, methods, reasoning, visualisation and quantification explored in the first phases of the project, will be tested in two Roman case-studies. These are focused on the following questions: how can networks of social interaction be identified and defined? When can a networks approach be justified? How are archaeological networks of social interaction to be interpreted? The two-case studies purposefully originate from different research environments and cover different data types, with the aim of being confronted with a diversity of social relationships as they are reflected in material culture. Both case-studies embody a different aspect of social interaction that can be examined archaeologically.
Case study 1
The FWO funded ICRATES project (G.0788.09) of Prof. Poblome (KUL) assembled exhaustive information on table wares from the Roman East. Given the broad chronological scope of the ICRATES project (2nd c. BC – 7th c. AD), the dataset provides the opportunity to examine evolving networks of table ware consumption, in an important and wide research area. By creating a network where table ware forms are represented as points which are related to each other through their simultaneous presence on sites, table ware distribution and consumption patterns can be examined. In contrast to most previous studies of Roman ceramic distribution patterns (e.g. Fulford 1989), this approach is less concerned with the spatial logic in these patterns. Instead, it aims at understanding the social logic of archaeologically attested relationships from a structural rather than a spatial point of view. Moreover, a non-geographical approach will allow for evaluating the archaeological use of non-geographical network analytical software packages such as Pajek and UCINET. By examining the archaeologically attested relationships in table ware consumption directly, we will be able to explore how particular social interactions are reflected in ceramics, archaeologists’ most abundantly available material source from the past. Indeed, in a previous stage the ICRATES project explained the attested distribution patterns of table ware in the Roman East mostly against the background of wider political and economical developments in these regions of the empire, leaving the potential role of the consumer of these wares mostly undiscussed (Bes 2007; Bes & Poblome 2008). The project wishes to contribute to explaining how complex patterns in a data-rich environment can help reveal social patterning in consumption. The applicant built up familiarity with the ICRATES data during his MSc thesis at UoS.
Case study 2
In the framework of the British AHRC funded “Urban Connectivity in Roman Southern Spain” project, Prof. Keay and Dr. Earl (UoS) collected a large dataset on relevant sites and their interconnections (through roads or visibility). A geographical transport network of sites connected by roads will shed light on how the Roman road network facilitated communication between communities, and how goods, people and information were disseminated over this network. This geographical approach will allow for the archaeological use of ArcGIS network analyst to be explored, the most popular tool for studying geographical networks.
Top
Software
Although a large number of programs for network visualisation and analysis are available these days, we decided to use Pajek for this project for a number of practical and methodological reasons. First of all, Pajek is free to download and use for non-commercial purposes. The program offers an integrated package, combining visualisation services with a vast array of analytical functions. Although Pajek supports a limited number of import formats, compatibility with other popular network analysis programs is ensured (e.g. UCINET), and Pajek’s own file formats are sufficiently simple to allow swift conversion from standard data storage formats (e.g. spreadsheets and databases). Lastly and most crucially, the documentation provided by the authors of Pajek is exhaustive and accessible, allowing social network analysis novices to grasp the basics through online tutorials and an introductory textbook (Nooy et.al. 2005), while more advanced users will find all functionality explained in detail in the online manual (Batagelj & Mrvar 2009).
Top
2-mode networks
m-slices
Similarity of vertices is partly determined by the strength of their ties: the larger the number of co-present forms found on two sites, the stronger their tie and the more interdependent they are. We will therefore classify vertices according to their line values, using the concept of m-slices: in an m-slice, vertices are connected by lines of value m or higher to at least one other vertex (concept introduced as ‘m-cores’ by Scott 1991; Nooy et.al. 2005: 109). M-slices consist of nested groups of vertices, as illustrated in the figure below, and the ‘m’ stands for the line value of the group or ‘slice’. This means that a 3-slice is part of the bigger 1-slice, while some of the vertices of a 1-slice are not part of the 3-slice.
When we apply this to the co-presence network of forms, the forms that are part of a high m-slice are those that are present on many sites. The m-slices in the co-presence network of sites, are an indication of the diversity of forms evidenced on these sites. For this project, m-slices will therefore be used to establish the wideness of a form’s distribution network, and the number of distribution networks a site is part of.
Top
k-cores
Similar to m-slices, k-cores are also nested and the ‘k’ stands for the core’s number. Unlike m-slices, however, k-cores represent groups of sites or forms with at least a certain number of relationships: a k-core consists of all vertices that are connected to at least ‘k’ other vertices within the core (Nooy et.al. 2005: 70-71).
In a co-presence network of forms, a high k-core consists of forms that are co-present with many other forms. For the co-presence network of sites, a high k-core indicates that a site has evidence of forms that are present on many other sites. We will therefore use k-cores as an indication of the similarity of the distribution networks of forms, and the wideness of the distribution networks a site is part of.
Top
Closeness centrality
As our distance networks represent the transportation of goods over trade routes between sites, we may wish to understand how easily they are distributed, and what sites are more easily reachable than others. The closeness centrality method builds on the idea that the closer a vertex is to all other vertices, the easier information, goods or people may reach it, and the higher its centrality. It is defined as “… the number of other vertices divided by the sum of all distances between the vertex and all others” (Nooy et.al. 2005: 127).
As our networks are directed we should make a distinction between input, output and ‘all’ closeness centrality. In this project’s distance networks, the output closeness centrality represents the relative ease with which a site’s pottery can be transported to all other sites. All closeness centrality, on the other hand, combines the input and output of vessels, and therefore reflects how easy a site can be reached by all other sites, and vice-versa.
Top
Betweenness centrality
A second centrality measure focuses on the idea that a vertex is more central if it is more important as an intermediary in the network. The betweenness centrality of a vertex is defined as the proportion of all shortest paths between pairs of other vertices that include this vertex (Nooy et.al. 2005: 131). If the flow of goods between sites can be severely disrupted by the removal of one site, then this site is a crucial go-between to the transmission of goods in the network. Betweenness centrality will therefore provide us with a tool to measure the influence and control individual sites exercise on the transportation of tabelwares.
Top
Degree
Contrary to centrality methods, the degree measure only takes a site and its direct neighbours into account. In a directed network, the outdegree of a vertex is the number of arcs it sends (Nooy et.al. 2005: 64). Defining the outdegree for every vertex allows us to identify all junctions in the trade routes, and distinguish between the number of coinciding trade routes.
Top
Domain
When goods are transported from site A to both sites B and C, the latter and all subsequent sites are dependent on the first for their provision of goods. The number of sites connected to site A serve as an indication for its domain of influence. The output domain of a vertex can be defined as the number of all other vertices that this vertex connects to by a path (Nooy et.al. 2005: 193). A site’s domain therefore represents the number of sites for which tablewares are evidenced that were transported through this site. Although this measure is implied by the betweenness centrality, the exact number of sites in a site’s domain is an interesting measure for comparison between sites. Moreover, it helps us understand and compare sites where routes diverge, i.e. sites with an outdegree of more than one.
Top
Home |Method | Blog | Bibliography
