Workshops

We are pleased to offer eight three-hour training workshops over two days. There are two rough "tracks." The first provides training in the basics of Social Network Analysis (SNA), and ways in which one can visualize and find patterns in network data. All workshops in this track take place in Perkins 217. The second provides training in developing and testing theories of networked behavior, and ways to perform empirical inference in the presence of network interdependence. All workshops in this track take place in Rubinstein 153 (in the library, not the policy school!). Below can be found a schedule and expanded descriptions of each workshop.

Workshop Descriptions

Track 1

Title: Introduction to Social Network Analysis

Trainer: Michael T. Heaney (University of Glasgow and University of Michigan)

This workshop introduces the questions, concepts, theories, methods, and important findings of social network analysis for scholars who have no prior training in this field. It begins by discussing the logic of relational thinking and motivations for the study of networks, including the identification of appropriate research questions and fertile areas for research. The collection of network data using various methodologies is considered, such as surveys, interviews, experiments, archival analysis, and ethnography. It discusses the dilemmas of boundary specification, the trade-offs of ego-network and whole-network approaches, and the challenges of asking network questions in surveys. It presents definitions of key objects (e.g., links, nodes, modes, components, paths, triads), one-mode versus multi-modal analysis, and network statistics (e.g., degree, degree distribution, indegree, outdegree, density, centrality, centralization). It surveys the major theories in network analysis (e.g., balance, embeddedness, resource dependence, brokerage, status, homophily, multiplexity, structural holes, preferential attachment, and small world theory), as well as recent findings from studies of political behavior, legislatures, parties, law, advocacy, trade, and international conflict and cooperation.

Materials: Slides

Title: Introduction to Network Analysis in R

Trainer: Bruce Desmarais (Penn State University)

This workshop will introduce participants to network analysis using the R statistical programming language. We will cover the basics of R programming for network analysis, as well as the fundamentals of network analysis. Topics will include network analytic terminology, network data collection and storage, network-level measures of global structure, actor/node-level measures of centrality, community detection, basic network simulation methods, and hypothesis testing with network structure. Example code and data will be provided online for the workshop.

Materials: Link to folder

Title: Network visualization with R

Trainer: Katherine (Katya) Ognyanova (Rutgers University, @ognyanova)

This workshop will cover network visualization using the R language for statistical computing (cran.r-project.org) and RStudio (rstudio.com). Participants should have some prior knowledge of R and network concepts. The workshop will provide a step-by-step guide describing through a series of examples the path from raw data to graph visualization in the igraph and Statnet frameworks. The advanced portion of the workshop will touch on dynamic visualization for longitudinal networks and combining networks with geographic maps. We will also discuss ways of converting graphs in R to interactive JavaScript/3D-based visualizations for the Web.

Materials: Online slides, Data and Code

Title: Community Detection in Networks

Trainer: Peter Mucha (University of North Carolina)

Community detection identifies the organization of a network in terms of patterns of connection. As one possible outcome, the detected communities may be clusters of nodes that are better connected within the communities than to the rest of the network. A wide variety of methods for community detection have been proposed, with a number of available software packages. In the past decade, these efforts have included increased interest in community detection for multilayer networks. In this workshop, we will summarize some of the main methods of community detection. We will then model the use of some of the community detection software available across MATLAB, Python and R on political networks examples including Congressional data and international trade.

Materials: Link to files

Track 2

Title: Theorizing Networked Behavior

Trainer: David A Siegel (Duke University)

Social Network Analysis often focuses on how networks matter (i.e., the empirical effect of interdependence on individual and group behavior), leaving imperfectly specified why networks matter (i.e., the causal mechanisms underlying the effect of interdependence). In this workshop, we will explore different answers for why networks matter and discuss how we might develop theories that can elaborate and distinguish among these answers. Our goals in doing so are to develop novel hypotheses and to better understand the role of networks in behavior. We will begin by specifying different causal mechanisms that have arisen in the literature and some evidence for each. We will then discuss two different, complementary approaches for deriving theoretical insight via formalization: game theory and simulation modeling. For each, we'll provide some examples of how they have been productively used, introduce the tools needed to apply the approach, and discuss how one can translate one's verbal theory into a formal one from which one can derive hypotheses. No prior experience with either game theory or computational modeling is required.

Materials: Slides

Title: Empirically Studying Networked Behavior

Trainer: Jennifer Larson (Vanderbilt University)

This workshop focuses on the move from theory to empirics in the study of networks. Designing an empirical networks study entails complex choices with big consequences. Because observations in a network are interconnected, well-established data collection practices from non-network settings can lead researchers astray, and some design choices that seem sensible at first introduce serious problems that cannot be overcome once the data are collected. This workshop will cover approaches to collecting network data that shed the most light on a researcher’s theory of networked behavior. We will consider issues such as how to select the nodes and links to study, how to navigate tradeoffs between network precision and affordability, how to avoid accidentally masking network effects, how to deal with noise in the data, and how to assess confidence in observed patterns.

Materials: Slides

Title: Exponential Family Random Graph Modeling (ERGMs) Using statnet

Trainer: Lorien Jasny (University of Exeter)

This workshop will provide an introductory tutorial on using exponential-family random graph models (ERGMs) for statistical modeling of social networks, using a hands-on approach to fitting these models to data. The ERGM framework allows for the specification, estimation, and simulation of models that incorporate the complex dependencies within networks, and provides a general and flexible means of representing them. The session will demonstrate ERG modeling using the statnet software in R. Topics covered within this session include: an overview of the ERGM framework; defining and fitting models to empirical data; interpretation of model coefficients; goodness-of-fit and model adequacy checking; simulation of networks using ERG models; degeneracy assessment and avoidance; and (time permitting) modeling and simulation of complete networks from egocentrically sampled data. Familiarity with basic descriptive network concepts and statistical methods for network analysis within the R/statnet platform is recommended. Attendees are expected to have had some prior exposure to R, but extensive experience is not assumed.

Materials: Slides, R code, Edge List

Title: Latent space network models

Trainer: Bailey Fosdick

Regression models are often used to quantify the association between a network and observed covariates on the nodes and edges, however network dependencies are complicated and observed covariates rarely adequately describe the observed connection patterns, e.g. clustering and homophily. Latent space models are a powerful extension of network regression models that represent residual variation in a network not captured by the covariates by mapping nodes to locations in an underlying social space. Nodes are positioned in the latent space such that nodes nearby one another have a high probability of being linked in the network. In this workshop, we will introduce latent space network models and discuss their specification, estimation and interpretation for various data sets.

Materials: Slides