A new stochastic path-length tree methodology for constructing communication networks

Abstract

Network analysis has become a popular method for identifying the communication structure in a system where positional and relational aspects are important. In this paper, a maximum likelihood based methodology is presented that allows for the analysis of binary sociometric data. This methodology provides a network representation via estimated path-length or additive trees that indicate the distance between all pairs of members. The methodology is distinguished from traditional hierarchical clustering based procedures by its direct consideration of the asymmetry in a typical communication process, the simultaneous representation of structural characteristics (e.g., clique membership, clique cohesiveness), and the identification of the specialized communication roles of each member (e.g., opinion leader, liaison). A penalty function algorithm is developed and its performance is investigated via a Monte Carlo analysis with synthetic data. An application examining information flows among managers is presented. Finally, directions for future research are suggested.