Applying ERGMs in analysis of large ego-network data - accuracy and culnerability for alter selection and alter sample size

Abstract

In Social Network Analysis, one of the ways of gathering data of large networks is the use of surveys where individuals are asked about their connections. This type of data-collection does not always lead to a fully recorded, complete network. However, research questions can be about general properties of this complete network. These questions may be answered by using an extension of the already often used Exponential Random Graph Models (ERGM), suitable to analyze sampled ego-centered network data. In this research, the accuracy of the ego-centered ERGM, as well as its vulnerability for certain biases that egos may have are tested using a large scale complete socio-centric data set from an online social network with approximately 10, 4 million users. Biases that are included are biases that follow from egos nominating alters non-randomly and the maximum number of alters an ego can have. A model about gender homophily is used to discover the accuracy and vulnerability. The findings suggest a low coverage and a high bias in the estimations of the ego-centered ERGMs, and little differences when changing the maximum number of alters per ego occur. However, the biases in alter selection seem to barely influence the results.