The Analysis of Multilevel Networks in Organizations: Models and Empirical Tests

Abstract

The presence of hierarchically interdependent levels of action is a defining feature of formal organizations. As a consequence multilevel statistical models are being increasingly proposed as uniquely appropriate for understanding networks within organizations. Conventional multilevel models, however, suffer from two main limitations that reduce their analytical value in the study of intraorganizational networks. The first is inherent in the assumption that the sub-units used to classify organizational members (like, for example, subsidiaries, divisions, functions, and teams) are not independent – but linked through multiple relations of hierarchical subordination. The second limitation derives from the fact that conventional multilevel models assume that common membership is the only source of interdependence among members. This assumption is inconsistent with results produced by decades of systematic empirical research showing that social networks among individuals in organizations represent differentiated sources of dependence both within as well as across the boundaries defined around organizational units. The objective of this paper is to show how Multilevel Exponential Random Graph Models (MERGMs) may be specified and estimated to alleviate these limitations. MERGMs are a class of new statistical models that may be adopted to analyze the complex multilevel dependences that networks create within and between organizational units. We illustrate the empirical value of our methodological proposal using data on relations of hierarchical subordination and informal communication between top managers in an industrial group containing five subsidiary companies. We discuss the implications of our results in the context of the current theories of organizations as connected multilevel systems.