Categorical Data Analysis of Single Sociometric Relations

Methods based on fitting loglinear models are adapted for the analysis of sociometric relationships among a group of actors, represented as a directed graph.By arranging directed graph data in a four-dimensional cross-classified table, the dyadic relationships between individual actors can be fully studied with a variety of models.These models are based on Holland and Leinhardt's probability density for directed graphs, p 1 , but extend their.approach to model data from single sociometric generators which include variables measuring nodal attributes.We show how both and these new models can be fit using standard iterative proportional fitting algorithms.A network of organizations from a midwestern community is used to illustrate these new ideas.