Well shoot, I guess it is publication week! The special issue of the Journal of Statistical Software about the statnet package for advanced network statistics has finally been released. It is great to see the years of hard work from everyone on the statnet team finally in print! I did some work for the project to hook up the package with SoNIA software to create movies from output of statistical models of dynamic networks of disease transmission. One of the articles in the issue documents the process and includes some example movies:
Skye Bender-deMoll, Martina Morris, James Moody (2008) “Prototype Packages for Managing and Animating Longitudinal Network Data: dynamicnetwork and rSoNIA” Journal of Statistical Software. Vol. 24, Issue 7.
One of the exciting things for me about the project was exploring ways to display time, connectivity, and transmission simultaneously. An example of one of the movies demonstrating the effects of varying levels of “concurrent partnerships” on the paths of transmission of simulated “infection”: (70mb quicktime movie). In the last few frames of the movie, the perspective shifts to show a timeline image of the infection “trees” in occurring in the simulated network.
In the tree image (created by James Moody) time advances vertically down the page, so the seed nodes for each infection appear at the top and the “depth” of the tree indicates the time step of the infection. The color of each edge indicates the concurrency status of the corresponding relationship when the transmission occurred.
One of the other points the work drove home was how important it is to think about the time-scale being used to aggregate relationships when creating networks from any kind of relationship data. Consider these three movies of the same simulated disease transmission network:
1 “day” time window.
On the daily level, the simulated sex partnership network appears to be mostly dyads with very little connectivity. Only when looked at over a longer time period is it possible to see the connective structure that permits disease to spread. At the same time, the longer the time period used, the harder it is to see the order of the links that determine the paths by which an infection can actually spread—some parts of the network appear linked but the links actually occurred in the wrong order for a path to complete. This is why I think network dynamics (and Jim and Martina’s work) are important. Note: this data shows the “infected” component of simulations of much larger networks, see the paper for details, or watch this talk were Martina explains the work and its implications.