New publication on network dynamics

Networks are now used to describe all sorts of systems – social worlds, protein interactions, food webs, and so on. For example, here you see a litter of marmot pups, where a network could be used to describe relationships between each individual animal. These networks are useful because they allow us to identify interesting structures and relationships, as well as test hypotheses about how groups function. For example, are individual marmots with larger body sizes more likely to have a higher number of dominant social relationships than smaller ones?

Networks like the one shown above (with individuals as dots, and relationships as lines) are useful but limited. Building a network implies that the system is stable – equivalent to assuming that relationships don’t change. This is problematic for many systems that do change over time. An individual marmot that is dominant as a juvenile may lose its dominance as an adult. And two marmots that appear to be able to interact in one network (that is, there is a path between their dots in the visual representation) may actually not be able to. Consider individuals A, B, and C. A interacts with B at 10AM; B interacts with C at 11AM. If we lump all these interactions together into one big network, we infer that A, B, and C are connected. But if the question is about the spread of some disease, then we should infer that A can spread disease to C but C cannot spread disease to A. We miss this with most networks.

These examples are only a few of the many issues that arise when networks become dynamic instead of static. I recently published a paper (with Tina Wey, Anna Dornhaus, Dick James, and Andy Sih) that surveys these issues and highlights multiple methods to resolve them and to better ask dynamic questions. You can read it in Methods in Ecology and Evolution here. A PDF is also available from my website. Please check it out!

It is one thing to point out issues, but another to provide practical tools. We also developed a free software package for R called timeordered which implements many of the algorithms and data analysis tools needed to ask questions about network dynamics. Above, you are seeing an object called a ‘time-ordered network’ that captures all the scientist’s information about the timing of interactions and the relationships between them. The network you see above can be thought of as a simplification of this more complex structure. Try the software out – it is useful for many kinds of data. We think this dynamic perspective will open up a whole new realm of questions to multiple areas of research.

Ecological networks, pollination networks (here shown using a Tetragonisca bee), disease networks… these are all dynamic systems. We hope this paper will bring us closer to dynamic understandings of them!

3 Comments Add yours

  1. Nick Dowdy says:

    Pretty cool! Are there issues with time or difficulty in getting time-ordered network data that could slow the adoption of this method? Or is it fairly common for model systems in network studies to easily allow for that? How do you handle missing data in a time-ordered network study? That seems like a significant problem! Keep up the good work!

  2. bblonder says:

    Hi Nick! Thanks for reading this. You’re asking all the hard questions that we try to address in the latter part of the paper. Time-ordered network data is very hard to get because each interaction’s timing needs to be recorded. Some systems are more amenable to this than others; generally the more automated data collection can be, the easier it is to get this data. For example one could mine facebook to easily get all the records of messages between individuals, but recording dominance interactions between whales would be very hard indeed. Missing data is a big issue also because many network statistics can depend critically on single links (which, if missed during sampling, drastically change inferences). There are really no good solutions here (the unknown unknowns) but for some systems where we have other constraints like known topology or individual dynamics we can make some guesses. Read the table in the paper for the guts of it.

    How are you doing? I ran into Nicole (from Hawaii nee Arizona) at ESA this summer and she says hello to you.

    1. Nick Dowdy says:

      Great! I will be sure to check out the paper. Things are going well. Just finished field season #2 and am hanging out in the Chiricahuas taking a Lepidoptera class with Bruce Walsh and many others. Hope things are going well for both you and Nicole! Enjoying all your great pictures from your travels!

Leave a Reply