Constructing social networks from RFID Data

PIT tag + RFID feeder systems are a very common way for researchers to quantify social networks in the wild. By attaching a PIT tag on wild birds, an RFID reader mounted on a feeder can then detect the identities of visiting birds. With an RFID data stream, the most common method is to use a Gaussian Mixed Model (GMM) to detect and cluster visiting groups, then convert into a social network. While such unsupervised clustering algorithms are appropriate for species like Great/ Blue Tits, who forms foraging flocks over winter, it might be less appropriate for more gregarious species, like House Sparrows.

We developed an alternative way to define social associations: using arrival time to a feeder instead, assuming flocks are grouped outside of the feeder, then arrives to the RFID feeder together. With the algorithm, we found fitness effects of individual sociality measures in House Sparrows on Lundy Island, UK.

Still unsure when and how we should choose the appropriate algorithm for defining association using RFID datastreams, we then did a comparison across 4 study systems and species, comparing 3 different association methods. We found that assocaitions are largely robust to association definition, with subtle differences that are driven by feeder design and social behaviours of the species. We encourage future researchers to consider the research question at hand before deciding which association definition to use.