So I have now been visiting with the Bascompte Lab for about a week now. Jordi has been very helpful in discussing with me my research ideas and goals. He has also started to walk me through the basics of doing network based science, going over the tools of the trade and important concepts about statistical reasoning with network properties. I’ve also been able to discuss some of the main tools for understanding networks with the lab members.
Yesterday we discussed how to measure nestedness and modularity in networks. We went over several measures of nestedness, and talked about the pros and cons of using nestedness temperature, NODF, and the measure described by Bastolla et al. 2009. Modularity the method that Jordi suggested was that of Guimerá and Amaral (2005), which uses a simulated annealing algorithm in addition to one for finding modules by Newman and Girvan (2004).
The last part of our discussion was based on null models. Basically, without null models how would we know whether that value we calculated for nestedness is meaningful or not? The problem, however, is how to know which null model to use, since there are several out there and their implementation can be important for whether a significant result is obtained. In our discussion we did not arrive at a good answer for which null model is best, but our conclusion was basically that you should use several different null models and if you arrive at significance with each then you can be confident as to your result.
It has been good learning so far…