Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Kuril Island Social Network Analysis Methodology Note (DRAFT) S. Colby Phillips and Erik Gjesfjeld
Social Networks Analysis
Social network analysis (SNA) is an approach that focuses on the relationships between social entities, the patterns that are present based on those relationships, and the implications of those relationships (Wasserman and Faust 1994). The networks utilized in SNA consist of nodes (such as actors, players, agents, or organizations) and ties between the nodes (such as links based on communication, economic transactions, or kinship) (Thompson 2003). While the conventional analysis of social science data compares actors based on their attributes, network analysis compares actors based on their relationships (Hanneman and Riddle 2005).
The conceptual and analytical use of network analysis provides a means for modeling prehistoric regional systems as networks formed by connections among sites. Analytically, the technique provides a tool for measuring the characteristics of regional systems quantitatively, and in turn, for objectively comparing systems with one another. Because networks are dynamic, and the types of interactions that produced network ties may shift over time, it is necessary to incorporate a diachronic element to the analysis and compare network structures as they changed through time (Nietzel 2000). Additionally, SNA provides a way to superimpose a measure of non-Cartesian social geography represented by human relationships on top of cartographic space to make comparisons between geographic and social space (Mackie 2001; Thomas 2001).
For this example, the relationships between archaeological sites in the Kuril Islands is explored via several methods, beginning with proximity similarity measures based on the distances between archaeological sites, and then through the analysis of ties between sites using specific sources of obsidian that are found in each site as a proxy for site ties. Thirteen archaeological sites present across the southern, central, and northern parts of the Kuril archipelago were chosen for analysis based on the presence of at least five obsidian flakes. A discussion of each of the methods used and an outline of the results follows below. Using Proximity Measures to Model Perceived Similarities
Several network analysis methods are used to model perceived or hypothesized similarities between network nodes based on geography. The technique of proximal point analysis (PPA) (also known as nearest neighbor analysis) is a way to explore what patterns of contact or relationships might exist if these relationships are influenced by geographic distance, and to hypothesize “neighborhoods” within a network. A 1st, 2nd, and 3rd order PPA for each site simply uses lines (edges) to connect each site of interest (nodes) to the 1st, 2nd, and 3rd closest nearest (proximal) sites (Terrell 1976, 1986, 2009 in press; Knappett et al. 2008). For the Kuril example, a distance matrix was created which lists the straight-line geographic distances in kilometers among the thirteen archaeological sites included in this study (Table 1). Based on these distances, for each site a line was drawn to the 1st, 2nd, and 3rd nearest sites; the lines were superimposed on a map of the Kuril Islands to show these linkages (Figure 1).
Another way to display the proximity of network nodes on a network map is to use multidimensional scaling (MDS). MDS represents proximities in two-dimensional space to show that nodes that are more proximate to each other based on the input data are closer together in network space. This may allow for further study of subgroups that are shown to be more proximal to each other (Wasserman and Faust 1994). The goodness of fit measure for MDS representations are based on stress measures, normalizations of the sum of squared differences between the MDS representations and the input data. High stress indicates a poor fit caused by MDS distortion of the input data, low stress indicates a good fit and little MDS distortion (DeJordy et al. 2007). Using the same one-mode symmetric matrix of site distances shown in Table 1, the matrix was transformed into a network map using the Ucinet 6 software program (Figure 2). The MDS network map shows a very close approximation to the location and proximity of sites shown in Figure 1 with little map distortion as measured by the low stress value of 0.005.
Graphic layout algorithms (GLAs) can also be used to model proximity data as networks, and are commonly used by social network analysis tools. Most GLAs are designed to work with binary data, and proximity data must be dichotomized in order to visualize relationships of interest (DeJordy et al. 2007). Figure 3 displays the site distance matrix data from Table 1 in an un-dichotomized network. Figure 4 on
the other hand shows the connections between sites when the distance between sites is 300 km. This network representation (which was rotated around and re-sized to better approximate the geographic layout of the Kuril archipelago) provides a way to examine relationships at specific distances of interest.
On the whole, these proximity measure methods of modeling similarities between Kuril Islands archaeological sites provide basic network representations which might form the basis for developing hypotheses about distance-related ties between sites. Incorporating archaeological data provides the next step in examining site relationships at a different level. Obsidian Source Data as a Proxy for Network Ties
Similarities in archaeological assemblages can be used as a proxy measure to infer ties between sites that can be modeled using social network analysis tools (Terrell in press). For the Kuril Island example, data from the source provenance analysis of obsidian artifacts recovered from Kuril archaeological sites was used to explore potential connections between sites (Phillips in press; Phillips and Speakman 2009). Table 2 is a matrix that shows the number of obsidian flakes from each of 11 different obsidian sources present in each of the 13 archaeological sites that are part of this study. This matrix was then converted into a binary presence-absence correlation matrix that represents which sites share at least on obsidian source with any other site (Table 3). The Table 3 matrix was then represented as a network using the NetDraw GLA tool to graphically show the ties between archaeological sites (Figure 5). As might be expected from observing the presence-absence matrix, this network structure is not very informative in terms of exploring network relationships – many site assemblages share at least one of the same obsidian sources, and therefore there are many site ties represented in the network.
One more informative way to explore the network is to compare the sites in terms of all of the different obsidian sources that are present in their artifact assemblages as a way to discover patterned groupings with as few assumptions as possible (Shennan 1997). Table 4 is a matrix of the presence or absence of each source in each site (a binary conversion of Table 2). This presence-absence matrix was converted into a matrix of Jaccard similarity coefficients, a statistic used to compare the similarity and differences in a sample set (Table 5). The Jaccard similarity coefficient ignores negative matches (sites are not grouped together because of the absence of specific obsidian sources), and coefficient values close to 1 indicate a strong similarity while values close to 0 indicate a strong difference. This similarity matrix is represented graphically as a network in Figure 6 based on similarity values > 0.40, which now shows which sites are more similar based on the presence of obsidian sources in their assemblages and demonstrates that the specific sources that each of the sites shares influence the network ties that are created. From this graphic we might infer social relationships (trade, exchange, familial, ceremonial, etc.) under the assumption obsidian source assemblages are material traces of social similarities (and by definition ties) that existed among the inhabitants of Kuril Island sites. References DeJordy,R., S.P. Borgatti, C. Roussin, and D.S. Halgin (2007). Visualizing Proximity Data. Field Methods 19(3):239-268. Hanneman, R.A. and M. Riddle (2005). Introduction to social network methods. Riverside, CA: University of California, Riverside (published in digital form at http://faculty.ucr.edu/~hanneman/ ). Knappett, C., T. Evans, and R. Rivers (2008). Modelling maritime interaction in the Aegean Bronze Age. Antiquity 82:1009-1024. Mackie, Q. (2001). Settlement Archaeology in a Fjordland Archipelago. BAR International Series 926, Oxford: Hadrian Books. Neitzel, J.E. (2000). What is a regional system? Issues of scale and interaction in the prehistoric Southwest. In The Archaeology of Regional Interaction, M. Hegmon (ed): 25-40. Proceedings of the 1996 Southwest Symposium. Boulder, Colorado University Press of Colorado. Phillips, S.C. Bridging the Gap Between Two Obsidian Source Areas in Northeast Asia: LA-ICP-MS Analysis of Obsidian Artifacts from the Kurile Island of the Russian Far East. In Crossing the Straits:
Prehistoric Obsidian Source Exploitation in the Pacific Rim, edited by M. Glascock, Y.V. Kuzmin, and R.J. Speakman. Smithsonian Institution Press: Washington, D.C. In press. Phillips, S.C. and R.J. Speakman (2008). Initial source evaluation of archaeological obsidian from the Kuril Islands of the Russian Far East using portable XRF. Journal of Archaeological Science 36(6): 1256-1263. Shennan, S. (1997). Quantifying Archaeology, 2nd Edition. Iowa City: University of Iowa Press. Terrell, J. (1976). Island biogeography and man in Melanesia. Archaeology and Physical Anthropology in Oceania 11(1):1-17. Terrell, J. (1986). Prehistory in the Pacific Islands. Cambridge: Cambridge University Press. Terrell, J. Language and Material Culture on the Sepik Coast of Papua New Guinea: Using Social Network Analysis to Simulate, Graph, Identify, and Analyze Social and Cultural Boundaries Between Communities. Journal of Island and Coastal Archaeology, in press. Thomas, J. (2001). Archaeologies of Place and Landscape. In Archaeological Theory Today, I. Hodder (ed.): 165-186. Cambridge: Polity Press. Thompson, G.F. (2003). Between Hierarchies and Markets: The Logic and Limits of Network Forms of Organization. New York: Oxford University Press. Wasserman, S. and K. Faust (1994). Social Network Analysis. Cambridge: Cambridge University Press.
Table 1: Matrix of distances (in kilometers) between Kuril Island archaeological sites.
Rikorda Sernovodsk Peschanaya 2 Berezovka Kubyushevskaya
Ainu Creek
Peschanaya Bay Vodopodnaya
Broutona Bay Drobnyye Savushkina Baikova Bolshoi
Rikorda 0 29 36 148 219 378 525 640 651 878 1113 1119 1123
Sernovodsk 29 0 8 119 193 355 503 620 626 847 1091 1092 1096
Peschanaya 2 36 8 0 122 191 351 501 614 624 845 1089 1091 1093
Berezovka 148 119 122 0 70 228 379 491 503 727 971 972 977
Kubyushevskaya 219 193 191 70 0 164 312 425 434 655 901 903 905
Ainu Creek 378 355 351 228 164 0 154 265 276 504 755 756 759
Peschanaya Bay 525 503 501 379 312 154 0 112 122 351 604 604 607
Vodopodnaya 640 620 614 491 425 265 112 0 10 243 498 498 502
Broutona Bay 651 626 624 503 434 276 122 10 0 234 489 490 492
Drobnyye 878 847 845 727 655 504 351 243 234 0 255 255 258
Savushkina 1113 1091 1089 971 901 755 604 498 489 255 0 7 7
Baikova 1119 1092 1091 972 903 756 604 498 490 255 7 0 3
Bolshoi 1123 1096 1093 977 905 759 607 502 492 258 7 3 0
Figure 1: Map showing PPA with lines drawn between each Kuril Island archaeological site and its 1st, 2nd, and 3rd nearest neighbor.
Rikorda Sernovodsk Peschanaya 2
BerezovkaKubyushevskaya
Ainu CreekPeschanaya BayVodopodnayaBroutona Bay
Drobnyye
SavushkinaBaikovaBolshoi
Figure 2: MDS network map representation of Kuril Island site distances from Table 1.
Figure 3: GLA network representation of Kuril Island site distances from Table 1.
Figure 4: GLA network representation of Kuril Island site distances from Table 1 when network ties are based on distances < 300 km.
Table 2: Frequency matrix of the presence of obsidian flakes from 11 obsidian sources in 13 Kuril Island archaeological sites.
Rikorda Sernovodsk Peschanaya
2 Berezovka Kubyushevskaya Ainu Creek
Peschanaya Bay Vodopodnaya
Broutona Bay Drobnyye Savushkina Baikova Bolshoi
Shirataki‐A
12 0 6 4 4 145 0 3 0 0 0 0 0
Shirataki‐B
30 2 11 8 9 71 0 0 0 2 0 0 0
Oketo‐1 69 20 16 15 7 353 0 3 0 2 0 1 0
Oketo‐2 3 4 3 0 0 13 0 0 0 0 1 1 0
Kam‐01 1 0 0 0 0 1 7 10 1 39 35 59 0
Kam‐02 0 0 0 0 0 2 1 34 2 84 6 16 5
Kam‐04 0 0 0 0 0 0 0 20 0 4 55 2 1
Kam‐05 16 0 1 0 0 1 0 10 0 6 0 0 0
Kam‐07 0 0 0 0 0 0 2 25 7 1 0 0 0
Group‐A 0 0 0 4 0 0 0 0 0 0 0 6 0
Group‐B 4 1 0 0 0 1 6 4 0 0 1 2 0
Total 135 27 37 31 20 587 16 109 10 138 98 87 6
Table 3: Correlation matrix of sites that share at least one obsidian source.
Rikorda Sernovodsk Peschanaya
2 Berezovka Kubyushevskaya Ainu Creek
Peschanaya Bay Vodopodnaya
Broutona Bay Drobnyye Savushkina Baikova Bolshoi
Rikorda 1 1 1 1 1 1 1 1 1 1 1 0
Sernovodsk 1 1 1 1 1 1 1 0 1 1 1 0
Peschanaya 2 1 1 1 1 1 0 1 0 1 1 1 0
Berezovka 1 1 1 1 1 0 1 0 1 0 1 0
Kubyushevskaya 1 1 1 1 1 0 1 0 1 0 1 0
Ainu Creek 1 1 1 1 1 1 1 1 1 1 1 1
Peschanaya Bay 1 1 0 0 0 1 1 1 1 1 1 1
Vodopodnaya 1 1 1 1 1 1 1 1 1 1 1 1
Broutona Bay 1 1 0 0 0 1 1 1 1 1 1 1
Drobnyye 1 1 1 1 1 1 1 1 1 1 1 1
Savushkina 1 1 1 0 0 1 1 1 1 1 1 1
Baikova 1 1 1 1 1 1 1 1 1 1 1 1
Bolshoi 0 0 0 0 0 1 1 1 1 1 1 1
Figure 5: GLA network representation of Table 3 correlation matrix of Kuril Island archaeological sites sharing at least one obsidian source.
Table 4: Presence-absence matrix of obsidian sources present in Kuril Island archaeological sites.
Rikorda Sernovodsk Peschanaya
2 Berezovka Kubyushevskaya Ainu Creek Peschanaya
Bay Vodopodnaya Broutona
Bay Drobnyye Savushkina Baikova Bolshoi
Shirataki‐A 1 0 1 1 1 1 0 1 0 0 0 0 0
Shirataki‐B 1 1 1 1 1 1 0 0 0 1 0 0 0
Oketo‐1 1 1 1 1 1 1 0 1 0 1 0 1 0
Oketo‐2 1 1 1 0 0 1 0 0 0 0 1 1 0
Kam‐01 1 0 0 0 0 1 1 1 1 1 1 1 0
Kam‐02 0 0 0 0 0 1 1 1 2 1 1 1 1
Kam‐04 0 0 0 0 0 0 0 1 0 1 1 1 1
Kam‐05 1 0 1 0 0 1 0 1 0 1 0 0 0
Kam‐07 0 0 0 0 0 0 1 1 1 1 0 0 0
Group‐A 0 0 0 1 0 0 0 0 0 0 0 1 0
Group‐B 1 1 0 0 0 1 1 1 0 0 1 1 0
Table 5: Jaccard similarity matrix for Kuril Island archaeological sites.
Rikorda Sernovodsk
Peschanaya 2 Berezovka Kubyushevskaya
Ainu Creek
Peschanaya Bay Vodopodnaya
Broutona Bay Drobnyye Savushkina Baikova Bolshoi
Rikorda 1.000 .571 .714 .375 .429 .875 .222 .500 .111 .400 .333 .400 .000
Sernovodsk .571 1.000 .500 .333 .400 .500 .143 .200 .000 .222 .286 .375 .000
Peschanaya 2 .714 .500 1.000 .500 .600 .625 .000 .300 .000 .333 .111 .200 .000
Berezovka .375 .333 .500 1.000 .750 .333 .000 .200 .000 .222 .000 .222 .000
Kubyushevskaya .429 .400 .600 .750 1.000 .375 .000 .222 .000 .250 .000 .111 .000
Ainu Creek .875 .500 .625 .333 .375 1.000 .333 .600 .222 .500 .444 .500 .111
Peschanaya Bay .222 .143 .000 .000 .000 .333 1.000 .500 .750 .375 .500 .375 .200
Vodopodnaya .500 .200 .300 .200 .222 .600 .500 1.000 .375 .667 .444 .500 .250
Broutona Bay .111 .000 .000 .000 .000 .222 .750 .375 1.000 .429 .333 .250 .250
Drobnyye .400 .222 .333 .222 .250 .500 .375 .667 .429 1.000 .333 .400 .286
Savushkina .333 .286 .111 .000 .000 .444 .500 .444 .333 .333 1.000 .714 .400
Baikova .400 .375 .200 .222 .111 .500 .375 .500 .250 .400 .714 1.000 .286
Bolshoi .000 .000 .000 .000 .000 .111 .200 .250 .250 .286 .400 .286 1.000
Figure 6: GLA network representation of Jaccard similarity matrix of Kuril Island archaeological sites.