Sunbelt XXV Workshops

A beginners tutorial on the concepts, methods and data analysis techniques of social network analysis. The course begins with a general introduction to the distinct goals and perspectives of network analysis, followed by a practical  discussion of network data, covering issues of collection, validity, visualization, and mathematical/computer representation. We then take up the methods of detection and  description of structural properties such as centrality, cohesion,  subgroups, cores, roles, etc. Finally, we consider how to frame and test network hypotheses. An important element of this workshop is that all participants are given a demonstration version of UCINET 6 for Windows and the Netmap visualization software, which we use to provide hands-on experience analyzing real data using the techniques covered in the workshop. In order to participate fully in the workshop, participants should bring laptop computers so that they can run the analyses on their machines at the same time as they are being demonstrated by the instructors.

Longitudinal social network data are understood in this workshop as two or more repeated observations of a directed graph on a given node set (usually between 30 and 100 nodes, sometimes up to a few hundreds). In other words, this workshop is about statistical modeling of the dynamics of complete networks. The workshop teaches the statistical method to analyze such data, as described in Sociological Methodology - 2001, p. 361-395, and implemented in the SIENA program. The statistical model used for the network evolution allows various network effects (reciprocity, transitivity, cycles, popularity, etc.), effects of individual covariates (covariates connected to the sender, the receiver, or the similarity between sender and receiver), and of dyadic covariates. One interpretation of this model is an actor-oriented model where the nodes are actors whose choices determine the network evolution. Further information about this method, including references and a JAVA demo, can be found at website http://stat.gamma.rug.nl/snijders/siena.html. The statistical analysis is based on Monte Carlo simulations of the network evolution model and therefore is a bit time-consuming. The computer program SIENA is included in the package StOCNET which runs under Windows. The workshop will demonstrate the basics of using StOCNET and SIENA. Attention will be paid to the underlying statistical methodology, to examples, and to the use of the software. The morning session will focus on the intuitive understanding of the model and operation of the software. The afternoon will continue this, and also present some further mathematical background. Special attention will be paid to a recent development: models for the simultaneous dynamics of networks and behavior. Participants are requested to check website http://stat.gamma.rug.nl/snijders/siena.html in the week before the workshop to download the workshop materials.

The workshop consists of three parts. In the first part we will give an introduction to the use of Pajek based on our textbook on social network analysis 'Exploratory Social Network Analysis with Pajek'. In the second part we will explain how to use multi-relational networks (new in Pajek 1.02, November 2004) and present some efficient approaches (valued cores, triangular and short cycle connectivity, citation weights, pattern search, generalized blockmodeling, islands) to analysis and visualization of real-life large network (genealogies, collaboration networks, citation networks, Internet networks, dictionary networks, 2-mode networks). We will also discuss the 'fine-tuning' of Pajek's layouts (pictures) and combining Pajek with statistical program R.
In the last part participants will have an opportunity to discuss about the use of Pajek (questions, suggestions, analysis of specific data...). Jurgen Pfeffer, from FAS.research, Vienna will present his program Text2Pajek that converts excel/text file datasets into Pajek format.
To actively follow the workshop bring your laptop with you. Program Pajek is available at http://vlado.fmf.uni-lj.si/pub/networks/pajek/.

This is a non-technical introduction to social network analysis. It describes the development for social network analysis, some key concepts, and some key substantive methods and findings. It is aimed at newcomers to the field, and those who have only seen social network analysis as a method.

MultiNet is an interactive computer program for the analysis and display of discrete and continuous network data. It simultaneously examines characteristics of links and nodes. The program is menu-driven, it has context-sensitive, interactive, on-line help, and always presents a color graphic representation of the data or the results of analysis as well as a textual report. The program does univariate descriptive statistics, crosstabulation, analysis of variance, regression, correlation, p*, and eigen analysis. It has powerful and flexible data manipulation capabilities. It performs continuous and discrete transformations, such as ordination, quantiles, recategorization; linear, log, power, and z transforms. New variables can be created by transforming or combining existing ones in any manner describable by algebraic equations. The program also provides file viewing and editing.
Part 1. Managing complex data

MultiNet is a program designed for exploring many types of relationships in complex network data. We discuss the univariate and multivariate methods currently available for exploring both attribute (node) and network (link) variables. These include discrete and continuous data recoding and bivariate and trivariate methods applied to node and link variables by themselves, as well as within networks. These methods will be demonstrated on real network data.

Part 2. Spectral Analysis

MultiNet does four types of eigen decomposition for spectral analysis of networks with up to 5,000 nodes with interactive graphical display of results in 1, 2, or 3 dimensions. We will demonstrate the analytic procedure; explain the various options available for interactive display of results; and show how the results from this procedure are integrated with the rest of the program and how both coordinates in eigen space and partitions can be used as variables in any other subsequent analysis.

Part 3. Hybrid methods

We describe hybrid methods which allow creating node variables from networks, such as  eigenvectors, partitions, and various centrality measures. We also describe methods for creating link variables from node attributes, and groupings of link variables. These methods will be demonstrated on real network data.

Part 4. p* in MultiNet

We describe the implementation of p* in MultiNet, and discuss various aspects of p* fitting with special types of data: large; symmetric; bipartite; multiple network. Since the current version can handle up to 5,000 nodes and 256 parameters, managing the displays and reports can be quite complex. We demonstrate how this implementation may be applied to some moderately large datasets.

Part 5. MultiNet in action

We apply topics covered in the proceeding parts to analyse moderately large, complex datasets from medicine. Topics applied include eigenspaces, hybrid data creation and recoding, bipartite p* fitting, and network crosstabs.