The spatial focus

An important reason for the primordial role allocated to space in archaeological research is no doubt the nature of archaeological data. Even when we are not sure how to date an object or building, we usually know where it was found - at least if we are not dealing with looted material. Conolly and Lake (2006, 234) in their handbook on Geographical Information Systems in Archaeology, point out that ‘the bulk of archaeological data is ultimately point based’. The very practice of excavating or collecting material from the surface results in ‘physical spatial location’ being a primary quality allocated to our finds, at least since nineteenth-century archaeologists started paying attention to the location of their finds. With the rise of positivism and proces-sual approaches in the archaeological discipline in the 1960s and 1970s, the interest in space all but intensified and it was not until much more recently that social interaction and the active manipulation of culture came to be scrutinised (Hodder 1982, 2004; Jones 1997; Preucel and Meskell 2004b; Knappett 2005, 2011, 2013). Building on a long tradition that stretches back to Heidegger (1927) and the school of phenomenological philosophy, space is now considered to be a social construct that shapes and is reshaped by social interaction (Lefebre 1991; Soja 1989, 1996; Ingold 1993; Tilley 1994; Blake 2004; Preucel and Meskell 2004a).

From the very beginning, the use of the word ‘networks’ was tightly intertwined with spatial metaphors. The very origins of the word have been traced both in French and English to the description of tightly-knit fabrics before the twelfth century (Addison 1721, 390; Parrochia 1993, 5, 2005). It is thought that the ancient Greeks did not have a concept of ‘network’ or ‘connectivity’ as spatial metaphor, but the use of ¿k twv ¿mjXXayjisvwv, derived from ‘something intertwined or closely connected’ to describe an intertwined series of tracks in Xenophon’s essay On hunting (6.14) seemsto point to the contrary. The Roman reticulum (net) or opus reticulatum (a building style involving regularly placed bricks as to shape a regular pattern of intersections between the bricks) indicate that the metaphor existed certainly in the Roman world. In early modern times, the concept of réseau or net-work gradually came to be seen as a suitable metaphor for describing various new ideas of complex intertwined entities, going from transportation systems such as rivers, roads and railroads to natural phenomena such as crystals and blood vessels (Parrocchia 2005). As such, the close connection between spatial proximity, or at least a physical link between entities became intrinsically linked with the idea of ‘networks’.

Similarly, the mathematical description of networks was born from a spatial conceptualisation. In the earlier part of the eighteenth century, one of the most important mathematicians of the time, Swiss-born Leonard Euler, tackled a trivial pursuit of the bourgeoisie in a small Prussian town, Kôningsberg, now Kaliningrad in Russia (Newman et al. 2006, 1-2). The Kôningsbergians would attempt to make their walks through town, crossing all of the seven bridges that the town counted, only once. Euler constructed a visual representation of the question and proved that such was not possible. With his game, he laid the basis of what we now recognise as the visual representation of networks or, scientifically speaking, graph theory (Biggs et al. 1979).

The invention of the sociogram, the characteristic way to depict a specifically social network as dots connected by lines, was, however, not spatial in nature at all. In fact, there seems to have been no connection between the mathematical developments and the invention of graphs to depict social networks. Its invention is usually attributed to the American psychologist Jacopo Moreno in a work from 1934 (Moreno 1934). Moreno himself, however, describes the development of the sociogram as a process that started as early as 1923, when he developed a method to chart spontaneous interaction, ‘Stegreifnoten’ or ‘spontaneous interaction diagrams’ (Moreno 1923, 1953, 140). Throughout the years, he developed his charts further to encompass more forms of human interaction (Moreno 1953, 140). Moreno eventually applied the methods to study how mass hysterics spread in a reform school in New York - the work that is now, incorrectly, considered the start of the visual representation of networks (Freeman 2004). An important theoretical foundation of networks as a unit of analysis is clearly formulated in Moreno’s work. Rather than seeing individuals independently from each other, he focused on the dynamics as a group and the specific place of individuals within this group. This idea was close to the foundations of the important twentieth-century sociologist Simmel’s theory, which is considered another important theoretical foundation of social network theory (Freeman 2004; Marin and Wellmann 2011). Moreno, in line with contemporary sociologists, claimed that individuals’ behaviour could be explained by the place the subject occupies within his/her social network and society as a whole. Important to observe, however, is that spatial

Archaeological networks and interaction 5 proximity was for Moreno and his associates still the first step in drawing a subject’s social circle (Moreno 1953, 98).

Despite the common attribution of Moreno as inventor of the sociogram, he was not the first to develop a graphic application in order to study human relations. The visual representation of ‘networks’ was used in late nineteenth-century Russia already, where the secret police (the Ochrana) would use visual representations to document groups of anarchists and other adepts of philosophies considered a danger to the establishment -very much like terrorist cells or other groups are mapped today (Laporte 1935,39-40; Ruud and Stepanov 1999,69-70). Suspects would be depicted in circles and the lines between them represented their connections. The larger the circle and the lines, the more frequent the interaction and the more important the node to the network. In a way, even medieval depictions of family trees can be seen as a precursor, as well as early anthropologists’ studies on kinship ties (Freeman 2004). These depictions and studies are not formally recognised as formative in network theory, but they indicate that visualisations of networks were more commonly known than is usually recognised. Another interesting element is that among these early popular networks, space was not considered to be an important factor.

More than one archaeologist has expressed astonishment about the fact that archaeologists did not really pick up on the concept of social networks or the possibility to visualise interactions between entities (Knappett 2011, 2013; Brughmans 2013). Previous research has pointed out that the network pioneers were processual archaeologists who turned to the fields of geography and formal methods to study physical space (Knappett 2011, 2013, Brughmans 2013). A paper by the historical geographer, Forrest Pitts (1965) is usually cited as the initial impetus for graph theoretical approaches in archaeology. Pitts studied the Russian river system to trace the centrality of Moscow and explain the city’s rise within the fluvial system. Pitts’ approach was adopted, with mixed success, to study questions such as the road system in Roman Britain and the Roman road and city network in Pannonia (Dicks 1972; Hutchinson 1972, Burghardt 1979). Other archaeologists started using graph theory to develop an early form of space syntax, despite the anthropologist Edmund Leach (1978) questioning space syntax as a heuristic tool, thus anticipating the still ongoing archaeological debate. Another geographical application explored by archaeologists interested in networks was proximal point analysis (PPA). John Terell (1977) adopted this particular perspective to reconstruct hypothetical journeys that would be able to explain biological and cultural similarities on an archipelago in Oceania. PPA is still used in archaeological applications of network analysis today and is recognised as an example of macro-level network analysis (knappett 2011).

Tom Brughmans has identified a study by Cynthia Irwin-Williams (1977) as one of the earliest non-strictly geographical archaeological applications of network analysis (Brughmans 2013). Irwin-Williams used network analysisto analyse prehistoric trade, but just like her more geographically-minded colleagues, she took sites as unit of analysis. Her innovative example focusing on exchange rather than distance, however, did not incite others to adopt similar approaches. Brughmans seeks an explanation for this in the absence of readily available computer power (Brughmans 2013, 634). Carl Knappett explains the lack of interest in social networks by archaeologists in terms of the location of interaction at the level of sites and regions, and thus ‘zones of interaction’ rather than links at the level of agents (Knappett 2011,18).

Another scholar who picked up network concepts in the early days was Colin Renfrew. In his landmark study on the emergence of the prehistoric Cyclades (1972), he developed a concept of an ‘interaction model’, not entirely unlike our multiscalar concept today. He used visualisations involving dots and lines to represent settlement hierarchies which were then linked to the interaction model. In later work, Renfrew and Cooke (1979) unsuccessfully tried to model interaction mathematically. Others also explored the concept of interaction matrix in more depth (Alden 1979). Alden studies political units in the Valley of Mexico in the Toltec period. He starts with a rather sophisticated concept of interaction with several hypothetical levels, from face-to-face to institutional, drawing on Renfrew’s initial interaction model (Renfrew 1972). However, for the mathematical definition of the model, Alden draws on a geographically defined gravity model, in which size is considered to be equivalent to complexity. Despite thinking about institutions and decision-making processes, interactions are unequivocally placed at the level of sites. How scales of interaction were integrated or affected each other is given little consideration. The interaction thus is situated exclusively on a geographical scale.

Some years later, Renfrew started to use graphs to represent interacting polities (1986). By now, Renfrew has come to place much more emphasis on the nature of the interactions being studied. It is less the exchange of commodities that interests Renfrew, as it is the flow of information (Renfrew 1986, 8). Renfrew (1986, 18) refers explicitly to recent developments in social network analysis, specifically the early work of Boissevan and Mitchell (1973), discussed further below, but he does not proceed with challenging the profoundly geographical basis of network analysis in archaeology. Indeed, he points out how others (in the same volume) rather prefer to locate social networks in a defined geographical space.

Social interaction became more of an interest to archaeologists in the later twentieth century, but when visualising interaction with graphs, it proved hard to escape the spatial constraints of archaeological datasets. Todd Whitelaw and Catherine Morgan used graphs to represent interaction in the Iron Age Peloponnese, as attested through formal similarities in painted pottery (Morgan and Whitelaw 1991). They developed statistics to assess the degree of similarity in pottery assemblages and plotted the results as a graph projected on a map. The larger nodes and ties represent closer similarity and, thus, interaction. Morgan and Whitelaw cite a number of books on mathematical approaches in archaeology as their source of inspiration

Archaeological networks and interaction 7 (Morgan and Whitelaw 1991, 96) but, interestingly, they do not refer to Renfrew or network analysis explicitly, despite their use of graphs. However, their approach announces the contemporary similarity matrices and communities of interaction approach.

The ‘new wave’ of network analysis in archaeology thus draws on a long geographical tradition of studying physical space (Knappett 2011, 2013; Brughmans 2013; Mills 2017), despite several attempts to break away from the spatial constraints. The spatial determination of archaeological network analysis therefore seems inevitable; but it would be wrong to blatantly equate all current approaches of network analysis with static spatial analysis. In fact, there is a plethora of network approaches in archaeology, and the way in which space figures in the analysis ranges from absolute to a factor considered when other types of interaction and exchange are discussed. The idea that network analysis is necessarily spatial and ignoring agency or the social construction of reality is, thus, wrong. Soren Sinbaek (2007a, 2007b) explores in a series of groundbreaking papers how Viking Age Scandinavia was a highly connected region in which connectivity gave rise to the formation of towns. Anna Collar (2013) studies the role of networks for the transmission of religious ideas in the Roman Empire and Emma Blake focuses on the rise of regional identities in societies, based on the circulation of a selected amount of elite consumption goods (Blake 2013, 2014). Based on similarities of selected attributes between sites that were at a maximum distance of 50km of one another, she reconstructs changing network dynamics in Late Bronze Age Italy. Barbara Mills and collaborators compiled a massive pottery database from the prehispanic Southwest of the United States, which they query for similarities to detect patterns of interaction and migration (Mills et al. 2013, 2015). Other archaeological applications of network analysis are indeed more explicitly spatial and several recent contributions have set out to systematise current archaeological network studies - mostly within this spatial framework (Brughmans 2010, 2013; Knappett 2013; Ostborn and Gerding 2014; Leidwanger et al. 2014; Collar et al. 2015; Mills 2017).