Lori Rosenkopf, Management, The Wharton School, and Anindya Ghosh, Indian School of Business
Abstract: While much strategic management research examines network evolution through a micro-level focus on the dyads composing networks, far fewer studies have attempted to analyze network structure and evolution at the more macro level of aggregate networks. When they do, studies typically focus on one of a variety of tie types (such as alliances, board interlocks, patent collaborations or technical committee participation) and explore network characteristics such as size, degree, clustering, and path length. While these network statistics appear, at first blush, to be directly comparable across studies, it is important to note that researchers’ particular design choices regarding network boundaries may have important consequences for the observed structural characteristics as well as their evolution. For example, studies with fixed sample sizes (Fortune 500 firms) may find network evolution constrained in a way that increasing sample sizes (inventors in a particular technology) may not. Likewise, studies limited to firms in a particular industry may observe more constraints to structural change than studies that encompass a wider variety of firms and burgeoning technologies.
Our purpose in this paper is to compare and contrast extant network-level studies to assess commonalities and differences in network structure and its evolution as a result of the design choices made by researchers. We develop of typology of networks which accounts for the roles of organizations, individuals, and additional affiliative groupings which connect these actors. Taking care to specify each of these attributes clarifies when different tie types are likely to yield similar or different network structures and/or patterns of network evolution. Such findings are important for more accurate specification of network structures which informs studies of the antecedents and consequences of networks as well as providing more accurate network models for simulation studies.
We have generated networks from alliance data, patent data, board data, and standards body data, which will enable explicit empirical comparisons. We expect this paper to find a home in a journal like SMJ, and we hope this effort will “reset” some of the overgeneralized findings currently held as exemplars in the network literature.