1254. sredin seminar: C. Nordlund: Valued blockmodeling: a deviational approach

In essence, blockmodeling consists of partitioning the actors into sets of equivalence (of some kind), subsequently fitting the patterns of relations within and between such sets with ideal blocks that represent the building blocks of the network’s functional anatomy. Blockmodeling was originally designed for binary, rather than valued, networks. Solutions for valued networks have been suggested, however non is appropriate for clustering networks in which actors greatly differ in their relational capacity (the strengths of ties they send/ receive), if we do not wish the clustering to be based on their capacity. The presented deviational approach is appropriate exactly for such problems, that is for clustering networks in which actors greatly differ in their relational capacity. Presenting the specifics and the idea behind the approach, the seminar will exemplify the approach using journal citation data, trade flow data and the classroom note-sharing data of Hlebec.