How species of plants and animals interact with one another is an important question of interest to community ecologists. While this question can be investigated by specifically collecting data on species interacting (for example pollination or predation) or measuring species abundances over time, it is also possible to extract information on species interactions from routinely collected co-occurrence data. Existing methods, like null models and hierarchical models, define species interactions in terms of correlations. Species are thought to interact if they are correlated, possibly after accounting for known covariates. There are however several reasons species might be correlated. These include a joint response to missing environmental variables, or a common interaction with other species in the community.
In order to distinguish between species that interact directly, and those that are correlated due to shared interactions with other species, we need to investigate conditional dependence relationships. These can be studied with graphical models, which have not been widely employed in ecology.
Results/Conclusions
We use graphical models to find conditional dependence relationships from co-occurrence data, which we interpret as species interactions. This method allows us to obtain additional information about how species interact, which is complementary to information we obtain from currently used visualization methods like non-metric multidimensional scaling and latent variable models. We will demonstrate this method by investigating species interactions of predatory spiders.