Generalized joint attribute modeling (GJAM) provides a common framework for synthesis of ecological attribute and abundance data, both for estimating responses and for prediction. GJAM was motivated by the challenges of modeling distribution and abundance of multiple species, so-called joint species distribution models (JSDMs), where species and other attributes are recorded on different scales. Because species are not independent, they must be modeled together.  But how does one combine species recorded on different scales?  Data may be continuous, discrete, censored, composition, nominal, and ordinal—such combinations of observations are not described by standard distributions.  Equally challenging is the fact that most of the values in species-abundance data sets are typically zero.  Finally, application of non-linear link functions as used in GLMs make it hard to interpret estimates from a model, because they apply to a log or logit scale.  I introduce the basic ideas, followed by examples showing how GJAM is used integrate biodiversity data from networks.

The first example comes from NEON, including pitfall traps for ground beetles, 1-m^2 plots for plants, and traps for small mammals. Each has a different scale and level of effort, including numbers of traps, length of time trapping, and plot area.  The second example includes microbiome data on fungal pathogens and host status.  This high-dimensional example combines composition microbiome data and categorical host status data.

Participants should have a working knowledge of R and arrive with Rstudio loaded on a laptop.  Files will be available on Github in advance of the workshop.