Social Network Analysis

Jeroen Bruggeman (2008) Social Networks: An Introduction. New York: Routledge.

Humans are the most social of all animals, learning from and being interdependent with many others, thereby forming relationships that span our complex world-society. Understanding humans and humanity therefore requires comprehending social relations, which may be hard to grasp systematically. Fortunately, the variety and turmoil of social relations can be mapped out as clear-cut networks. With the aid of network theory, characteristics of social life can be pointed out and explained that nobody imagined before, such as small worlds, highly skewed distributions of social contacts, and the structure of social inequality and cohesion. Furthermore, phenomena that are not yet well understood can be grasped more clearly, such as organizations and the growth and diffusion of knowledge. Many social network models are also applicable to, or inspired by, other fields, such as economics, biology, political science, statistical physics, and organization science.

This book introduces social networks to a general audience, from novices in all kinds of fields to experts wanting to catch up, and from academics on the one hand to practitioners in consultancy, management, policy, and social work on the other hand. Sophisticated models are lucidly explained and comprehensible without math (which is put in boxes, footnotes, or references), and are illustrated with network diagrams and examples ranging from anthropology to organizational sociology. A free to use software tool - R's igraph package - is explained in a new manual so readers themselves can depict and analyze networks of interest to them.

Table of contents

   1. Introduction

   2. Representation and conceptualization

   3. Small worlds

   4. Searching and fat tails

   5. Communities: detection, conflict, cohesion, and culture

   6. Social inequality: prestige, power, brokerage and roles

   7. Organizations as networks

   8. Methods: data and software. A new computer manual replaces the software part of Ch.8, and there is a new introduction to R.

Updates after the manuscript was completed (early 2008).

   1. The software has been further developed to the point that the treatment of it in the book has become obsolete, and is replaced by a new manual.

   2. Community detection has become also possible for networks with both positive and negative ties, which is implemented in igraph. 

   3.  Data from earlier studies can be found at the Siena site, as well as at Matthew O. Jackson, SocioPatterns, Wikipedia, Stanford, Enron's email traffic, Mark Newman, Fowler's legislative cosponsorship, Epinions with trust and distrust, KONECT, reality mining, international trade, and DANS. Some journals such as PLOS and Am. Econ. Rev. require that authors publish their data, for example in Dryad. Network data can be combined with node data, e.g. country data for international trade.

   4. As strategic information for brokers appears to be limited to direct contacts (Burt 2010), for betweenness centrality you should leave out paths longer than 2 by using the command betweenness.estimate(graph, cutoff=2)

   5. For power of individuals, Bonacich' centrality measure is advocated, which turns out to correspond neatly to centrality in information spreading (Kristina Lerman, Rumi Gosh et al). It has a couple of practical drawbacks, though: paths can be longer than actual influence, local cycles are amplified beyond their importance, matrix inversion can fail, and only nodes within the same network can be compared (in contrast to degree and betweenness). Therefore the book recommends to use Page Rank instead, which, however, may have less validity. A solution to these practical problems is to simplify Bonacich' measure to distance two. 

   6. It has become clear that for ties to contribute to social cohesion they have to be bi-directional and treated as binary, possibly after rethinking the data and setting thresholds for bi-directionality and for presence versus absence; then the ties can be represented as binary edges, just like in the book.

   7. Ego network size. People in the USA do not know on average about 750 people, as was thought in 2006, but about 600 (McCormick, Salganik and Zheng 2010).

   8. A network component is a set of nodes where each node can be reached from any other through a path (a concatenation of lines), as said in the book. However, a component is also maximal in the sense that every node that can be reached belongs to the component in question. For example, organizations are components for the authority relation, at least for a period of the day (as members can work for other organizations at other times.)