Beyond urbanization metrics: Using graphical causal models to investigate mechanisms in urban ecology and evolution

As the fastest-growing form of land use, urbanization drives profound environmental change that reshapes biodiversity. Studies often characterize biodiversity patterns along urban-rural gradients with statistical models that include one or more generic indices of urbanization. Such models may be useful for prediction, but they do not permit explicit tests of causal hypotheses by which urbanization mediates ecological and evolutionary processes. Here, we show how a graphical causal modeling framework with directed acyclic graphs (DAGs) can be used to design clear conceptual models and inform appropriate statistical analysis to better evaluate mechanistic hypotheses about the effects of urbanization on biodiversity. We first introduce the basic structure of DAGs and illustrate their value with simulated datasets. We then apply the framework to a case study on coat color variation in eastern gray squirrels (Sciurus carolinensis) along an urbanization gradient in Syracuse, New York, USA. We show how statistical models ungrounded in causal assumptions are difficult to interpret and can lead to misleading conclusions about mechanisms in urban ecology and evolution. In contrast, DAGs - by making causal assumptions transparent - help researchers identify appropriate control variables for statistical models to estimate the effects of interest. When applied to our case study, statistical models informed by a DAG revealed a surprising finding: although squirrel melanism was more prevalent in urban than rural populations, the prevalence of melanism was constrained by components of environmental change common to cities, namely roads, forest loss, and predator activity, contrary to expectations. Managing biodiversity in an increasingly urbanized world will require a mechanistic understanding of how urbanization impacts biodiversity patterns; graphical causal models such as DAGs can provide a powerful approach to do so.