Theoretically Informed Network Analysis for Social Scientists

Social network analysis uses dyadic data (data about the connections between entities) to construct measures of larger structures, such as cohesive groups, bridging individuals, unique positions, or central and peripheral actors.  In more general terms, social network analysis is a paradigm that builds theoretical arguments about the impact of triadic or higher order (meso-level) structures on individual nodes, dyads, or macro level systems.  This paradigm is closely related to the relational turn in the social sciences, and provides us with the conceptual tools to avoid falling into traps of structuralism, to formulate arguments that center on strategic agents exploiting openings in their network structures. 

This course aims at helping the participants build and test such arguments, with the help of network data. Each day will focus on a particular argument, with a reading and hands-on work with data. The readings, software, and data will be provided by the instructor.  We will primarily use Ucinet 6, and Gephi 0.8 for analyzing and visualizing network data.  We will use Excel for basic data entry and manipulation tasks, although students are encouraged to explore Python, R, or Matlab tools for data collection and manipulation. Pointers for exploring such solutions will be offered in the beginning of the course.

The first argument that we will consider, as an introduction to working with social networks data, is about the diversity of the nodes connected.  Is diversity related to creative success?  During the first day, participants will learn the basic skills of recording and manipulating dyadic data, and the basics of visualizing network data with node attributes. We will discuss fundamental graph theory concepts, such as degree, path, and distance.  We will also introduce the basics of using graph visualizations, such as spring embedders, circular layouts, node and edge size and color settings, and filterings.

The second day will be about arguments centered on the node level, primarily arguments about the advantages to brokerage (the unique vantage point that actors enjoy when they are linked to distinct regions of the network).  We will introduce competing measures of nodal prominence (centrality, prestige, and brokerage measures), and we will visualize the diverse network reach of brokers.  We will also introduce hypothesis testing at the node level, addressing non-parametric methods appropriate for interdependent observations.

The third day will be about the causal significance of cohesion, illustrated through the case of political corruption in business networks. We will introduce methods to measure cohesion, and to identify cohesive clusters, and we will discuss ways to visualize cohesive groups.  We will discuss the advantages and disadvantages to methods that identify exclusive or overlapping clusters. We will discuss the basics of hierarchical cluster analysis, and modularity as a diagnostic.

The fourth theme will be about network multiplexity and power, introducing structural equivalence blockmodels of multiple networks.  We will discuss methods of finding and visualizing structural equivalence blocks, and we will introduce goodness of fit measures.  We will introduce the idea of working with block images, and relational multiplication tables.  We will also talk about the significance and challenges of working with multiplex data.

The fifth and final theme will be about the dynamics of the economic world system, through the networks of international and intersectoral trade. We will discuss valued networks, cutoffs, filterings. We will introduce regular equivalence blockmodels, and methods to analyze and visualize networks through time.  

Basic knowledge of statistics and probability (probability density functions, tests of statistical significance, ordinary least squares estimator). Basic knowledge of data entry and manipulation (for example using Excel sheets).