Growing Collaborations
Literature Background
Change in Social Networks : Scientific Coauthorship and Email Communication Networks
The study of social network evolution has so far focused on networks of coauthors of scientific papers and email communication networks . Scientific coauthorship networks are formed by linking researchers who have coauthored published research into a broad network that can either cover a wide range of scientific disciplines ( Barabasi et al , 2002 ), be limited to a specific scientific field ( Tomassini and Luthi , 2007 ), or be limited to a certain geographic area ( Perc , 2010 ). Kossinets and Watts ( 2006 ) build a network by linking individuals who exchange emails on a university email system , creating an email communication network . Both the coauthorship and email communication networks are social networks since they link people together through inter-personal interactions .
Barabasi et al ( 2002 ), Tomassini and Luthi ( 2007 ), and Perc ( 2010 ) all seek to describe in detail how the structures of coauthorship networks change . Barabasi et al ( 2002 ) study the role of preferential attachment , where agents prefer to add new links with already well-connected agents . Tomassini and Luthi ( 2007 ) also use preferential attachment while adding in four whole-network measures : mean degree centrality , clustering coefficient , average path length , and the network ’ s degree distribution . Perc ( 2010 ) uses preferential attachment , small world networks , the clustering coefficient , and the network ’ s diameter . In all three studies , the researchers examine how their measures change over time and draw conclusions from the pattern of change of each measure without attempting to explain why or how the observed patterns emerged . This kind of analysis is known as dynamic network analysis ( Carley , 2003 ; Abbasi & Kapucu , 2015 ). Kossinets and Watts ( 2006 ) use the same approach to show how network-level measures stabilize over time in the email communication network , reaching what Kossinets and Watts label as an equilibrium , while individual-level measures remain unstable . All of these studies saw the measurements of their network statistics change over time , but none of these studies examined why . They only examined what the pattern of change was , not what mechanisms generated the observed patterns of change . This leaves an important gap between in our knowledge of network evolution , between how agents choose new partners and how specific network structures emerge from those agents ’ choices .
Agent-Level Link Selection
Network Link Accretion : Homophily , Heterophily , Transitivity , and Preferential Attachment
Whether networks develop organically ( Axelrod & Cohen , 2000 ) or through the direction of policy and planning ( Koliba , Reynolds , Zia , & Scheinert , 2015 ), they follow a set of rules for link selection that is now well recognized among network scholars ( Wasserman & Faust , 1994 ). Network agents will sometimes seek partners who are similar to themselves , leading to existing groups drawing themselves closer together through increased bonding social capital ( Burt , 1992 ). Network analysis refers to this behavior as homophily . Conversely , agents will sometimes seek partners who are different from themselves as those agents gain leverage by linking unconnected agents in the network ( Granovetter , 1973 & 1983 ), where that lack of a link is known as a “ structural hole ” ( Burt , 1992 ).
5