Withn= 51 for our data set we must minimize the number of variables in multiple linear regression models to prevent the possibility of over-fitting.
To do this, we first examine univariate models and select most useful predictors from these models for examination in multi-variate models.
To further reduce the number of predictors, and to aid inference regarding the most
important independent variables, we use Bayesian Model Averaging (BMA).
This approach is increasingly being used in ecological species distribution modeling (e.g.,Wintle et al., 2003; Thomson et al., 2007), and provides a method to account for
model uncertainty by calculating (approximate) posterior probabilities for each possible model that could be constructed from a suite of independent variables (Hoeting et al., 1999).