CASOS Working PAPER

"Linking Ego-Networks using Cross-Ties" (PDF file)
Author(s): Ju-Sung Lee

Abstract
Ego-networks, or ego-centric networks, have traditionally been studied in isolation. Information on where and how a single ego-network fits into the greater social structure is often unavailable, whether due to the absence of unique identifiers, such as full names or social security numbers (e.g. in HIV networks or networks of drug users), or a sampling strategy designed to obtain networks of disparate individuals (e.g. the 1985 GSS Social Network module). Computational approaches have allowed network researchers to probabilistically reconstruct complete networks (i.e. sociometric data) from discrete ego-networks, allowing the analyses to, then, include aggregate level network measures. However, the current technology employs attribute information of egos and alters, and not the links between the alters of a single ego-network. My research contributes to this method by introducing an algorithm that connects ego-networks using both information sources: attributes and alter-to-alter ties, including tie strengths. Furthermore, the gains from the inclusion of alter-to-alter ties are assessed as well as the expected error in the reformed network as a function of the size, density, network type (i.e. random vs. empirical), and the distributions of attributes and alter-to-alter ties. Results show that the error, hence the accuracy of the completed network, varies non-linearly and significantly with all of these parameters.