Latent layers in social networks and their implications for comparative analyses

Animal social systems are remarkably diverse, ranging from solitary individuals to well-connected cooperative groups. Understanding the drivers of this variation is a key question in behavioral ecology and has been the focus of numerous studies linking social structure to ecological, demographic, and life history patterns within groups, population, and species. Equipped with this information, researchers are now turning to investigations that are comparative in nature. However, comparing social networks remains a considerable logistical and analytical challenge. Here, we present the latent layers framework, which outlines how observed social networks are linked to the 2 underlying latent networks that are of interest for most research questions: the realised social network (the actual pattern of social interactions), and the social preference network driving these interactions. This conceptual framework provides a clear and unified approach to understand when and why differences in network properties and sampling protocols can introduce discrepancies between observed and latent networks, potentially biasing or confounding statistical inference. We then use this conceptual framework to outline some of the central challenges to comparing animal social networks, describe why and how they create challenges for comparative analyses, and suggest potential directions for solutions. The latent layers framework can help researchers to identify networks they can (or cannot) compare. In doing so, this framework facilitates advances in comparative social network studies with the potential to generate new and important insights into the ecological and evolutionary drivers of variation in social structure across the animal kingdom.