Application of a bayesian hierarchical regression model to explain multiple trait distributions of lotic insects across environmental gradients

Reclassifying a community from taxonomic designations into functional trait categories provides a mechanistic link between environmental selective forces and community composition.  However, redefining a community according to traits generates multivariate response variables (multiple trait states) that are inter-correlated and thus pose statistical challenges.  We applied a Bayesian hierarchical regression model, which can account for the conditional relationships between multivariate variables, to explain the abundance of aquatic insects with specific trait combinations using environmental variables for 197 streams in Colorado, Oregon and Washington. In the model, aquatic insects were classified according to their membership in one of nine possible combinations of the traits voltinism (3 states) and thermal preference (3 states). We were able to explain the abundance of individuals within each trait combination in terms of several environmental variables at the reach (e.g., substrate size), valley (e.g. stream power) and watershed (e.g., winter temperature, landcover) scales.  We found that trait distributions were meaningfully correlated with environmental variables associated with habitat degradation (e.g. percent forest, sand and fine stream substrates).  Bayesian models can clarify the complex relationships between correlated response variables and correlated environmental variables and thus provide more appropriate estimates of trait-environment relations than standard regression approaches.