Sex- and age-based social structures have been well-documented in animals with visible aggregations. However, very little is known about the social structures of snakes. This is most likely because snakes are often considered non-social animals and are particularly difficult to observe in the wild. Here, we show that wild Butler’s Gartersnakes have an age and sex assorted social structure similar to more commonly studied social animals. To demonstrate this, we use data from a 12-year capture-mark-recapture study to identify social interactions using social network analyses. We find that the social structures of Butler’s Gartersnakes comprise sex- and age-assorted intra-species communities with older females often central and age segregation partially due to patterns of study site use. In addition, we find that females tended to increase in sociability as they aged while the opposite occurred in males. We also present evidence that social interaction may provide fitness benefits, where snakes that were part of a social network were more likely to have improved body condition. We demonstrate that conventional capture data can reveal valuable information on social structures in cryptic species. This is particularly valuable as research has consistently demonstrated that understanding social structure is important for conservation efforts. Additionally, research on the social patterns of animals without obvious social groups provides valuable insight into the evolution of group living.

We examined the influence of sex, age, and body condition on social patterns in Butler’s gartersnakes. We did so with data from a long-term capture-mark-recapture project monitoring a population of snakes under threat from a road construction project. Coverboards were primarily used to locate the snakes across a ~2 ha area with 3 major sampling sites. The sex, weight, and snout-vent length (SVL) of the snakes were recorded at the time of capture as well as the sampling site, the exact location, and the time. Body condition was defined by the scaled mass index which is derived from weight and SVL. Weight was used as a proxy for age. Social networks were constructed from the capture time and location data, and we modeled the relationship between demographic factors, weighted degree, and community betweenness. To derive community betweenness, we subdivided our networks into communities of individuals who tended to associate with each other and calculated the betweenness of the individuals within these communities. We additionally examined sex and age homophily as well as the order of arrival of dyads that shared a capture location. Below, we outline how we used these data to construct social networks and how we then analyzed the resulting data set.

The social networks were constructed based on the temporal and physical proximity of the snakes at the time of capture. In the resulting networks, the edges represent the probability of an association occurring. As such, a 1 indicates that two snakes were found at the same place and the same time. A value less than 1 indicates that the snakes were found further apart (temporally and/or physically). We constructed networks at different scales of proximity. We identify these as the precise and three broad-scale networks. The precise network is derived only from direct associations. The broad-scale networks expand the possible temporal and physical proximity allowances to include possible associations within 14, 10, or 5 days.

The analysis was done with R v4.2.1. We tested for non-random associations using null models that controlled for temporal and physical space use. To test for intra-species communities, we used a Louvain partition and tested the robustness of the resulting communities compared to communities generated from random graphs using the robin package. To test for relationships between phenotypes and network measures, we used mixed-effect hurdle models with node-label permutations.  We tested the inherent type I error rates of these models by simulating the same analyses but with randomly generated values for our independent variables. We derived homophily values from the networks using the igraph package and tested them against values derived from permutated graphs. To examine the order of arrival in shared space, we designated snakes as leaders and followers based on their order of capture at the same location - with leaders captured first. We then used a multi-membership model constructed with the brms package to examine the relationship between sex and age on the order of arrival. More details on these analyses are available in the manuscript and/or by request.