Abstract
Social network analysis has gained prominence in recent years by offering a powerful means to represent social phenomena. While much of this explosion has been with respect to large-scale digital networks and big data, just how well such sources can help answer long-standing sociological questions remains to be seen. For some relational questions, traditional surveys of population-based samples offer the most practical approach. Brea Perry, Bernice Pescosolido, and Stephen Borgatti show us how in Egocentric Network Analysis: Foundations, Methods, and Models, part of Cambridge’s Structural Analysis in the Social Sciences series.
As backdrop, social network research has historically proceeded along two tracks. One focuses on “complete” networks, defined as a set of actors and some type of relation(s) between them (e.g., friendships among students or advice between coworkers). Such data constitute a census, making them expensive and difficult to collect, particularly longitudinally. Hence, this approach is infeasible when the population of interest is large (e.g., the population of adults in the United States). An alternative means is to generate a probability sample and survey individuals about their relationships and the attributes of those people (termed “alters”). The product is “egocentric” or “local” network data on the set of relationships that constitutes each respondent’s immediate social context. Such data allow for population-level inferences regarding distributions of local network configurations (e.g., network size, relationship strength, cohesion, role overlap, demographic composition, and access to resources) and their associations with individual outcomes.
Egocentric networks have a storied history in sociological research, having been the focus of prominent studies by Edward Laumann, Mark Granovetter, Frank Fischer, and Ronald S. Burt (among others). However, it is only recently that a handful of books has appeared that focus on the methodological aspects of collecting and analyzing such data. Egocentric Network Analysis is among the most comprehensive. Perry and Pescosolido have extensive experience working with egocentric data to address questions related to health, and Borgatti is an esteemed social network methodologist. As such, the book offers a balance of methodological rigor, with numerous examples drawn from the health field, that is still intuitive to a wide readership.
The book begins with a summary of several theoretical traditions (i.e., social support, social capital, social influence), which offers needed background to the kinds of explanations egocentric network data can provide. This is followed by a discussion of how the egocentric approach differs from a complete network design. These first two chapters will be valuable for anyone who has an inkling that networks matter for their particular outcome and is trying to determine what the relevant network is and how to measure it.
The bulk of the book is devoted to issues surrounding data collection and analysis. Chapters Four and Five are logically divided into measuring relationships with alters through “name generator” questions and gathering information about alters through “name interpreter” questions. The authors enumerate various ways of eliciting the alters in one’s network, a step whose importance cannot be overemphasized given that representations of network structure are wholly dependent on the relation measured. They discuss the tradeoffs of different strategies of eliciting networks and provide several practical suggestions. I appreciated the discussion of what types of information survey respondents can reasonably provide about their alters. Accuracy is a key concern with proxy reports, and it is vital to understand the limits of this method.
Chapters Six through Nine are a guide to analyzing egocentric network data. Chapter Six, on visualization, offers several examples and advice for visualizing a single egocentric network using the NetDraw software. However, the strength of the egocentric approach is in gathering data on dozens, if not hundreds or thousands, of personal networks. The authors present several sophisticated means to summarize and visualize multiple networks (e.g., through blockmodels or multi-dimensional scaling) as well as change in networks across time.
I greatly appreciated Chapter Seven (written with Ann McCranie), which is an in-depth presentation of ways to summarize egocentric networks. The authors don’t just provide formulas; they offer theoretical explanations for what various facets of network structure represent and how they might matter for outcomes of interest. For instance, we learn the measures necessary to evaluate how social learning and normative pressure may influence women’s contraception decisions. Chapters Eight and Nine discuss how linear models can be used to examine egocentric networks and their change over time. The authors review methods to test both respondent-level and relational-level outcomes (e.g., social support from each alter). In the latter case, multilevel models are used to overcome the inherent interdependence between respondents’ relationships. The authors also provide a good discussion of issues that arise when analyzing network data, such as non-normal distributions and sources of error. Many examples rely on publicly available data, and Stata code is provided.
Chapter Ten offers ways to bridge sociometric and egocentric approaches by using methods designed for one type to analyze the other. Although the hybrid approach is notable for its creativity, the authors fall short in offering a compelling case for the advantages of such hybrid approaches—that is, what is to be gained from analyzing egocentric data using sociometric methods? This contrasts with much of the book, which is replete with practical examples.
Just as important as learning a method or statistical test is learning the appropriate time to use it. The discussion of the tradeoffs between the egocentric and sociometric approaches was rather brief, but it made several good points. That said, the assertion that egocentric data is preferable for evaluating questions of social influence is sure to raise eyebrows and would be disputed by the many scholars who study diffusion with complete network data. The authors also assert that respondent burden is greater for complete versus egocentric network studies. Yet many complete network studies use a nomination method that is similar to the egocentric approach. In so doing, respondent burden is far lower than in egocentric studies because respondents are only reporting on themselves and their relationships, not also providing proxy reports on their alters’ attributes.
Relatedly, the book lacks a discussion of causality and the thorny problem of selection. Much of the book focuses on network characteristics as predictive of some outcome. However, typical “outcomes,” such as mental health, health-related behavior, or attitudes, are often endogenous and can affect selection into relationships—something I’ve spent my career demonstrating. How can researchers use egocentric data and methods to make inferences regarding causal direction (to the extent possible with observational data)? A discussion of the complexities that make this a challenging issue would help readers to situate these methods.
Perhaps the authors need to make an overly strong case for the egocentric approach in order to offset the attention given to complete networks. As they note, any egocentric analysis can be done with complete network data, but not vice versa. The issue is practicality. The book rightfully highlights the advantages of the egocentric approach, the times when it is the most feasible avenue, and considerations that should go into gathering and analyzing such data. Despite my quibbles, this is a valuable book that offers a worthwhile guide to conceptualizing and conducting egocentric research.