Re-Analysis of Breeding Bird Density in Eastern Pennsylvania Woodlots

Abstract

Previously, we developed statistical models for the densities of 36 breeding bird species occurring in 176 woodlots in eastern Pennsylvania, USA. Here, we present a re-analysis of the same data after eliminating many highly correlated variables (|r| > 0.5), decreasing our number of candidate models from thousands to 12, and correcting a statistical error that we had made previously. Our distance-to-edge variable was a measure of depth in the woodlot that depended on distances to 4 woodlot edges. We could not use our area variable because of its very high correlation with our distance-to-edge variable. Our isolation variable was not highly correlated with either our distance-to-edge or area variable. Two other key variables related to buildings at the woodlot edges. As previously, our models were nonlinear, and we used bootstrapped data sets to help ascertain the best model for each species. We determined that 3 species were forest-interior species, and 2 of these also responded negatively to increasing isolation. We found 3 edge species, plus 2 modified edge species that occurred only at suburban or urban edges; both of the latter are known to nest on buildings. Our results agreed well but were conservative compared to an extensive literature. However, they did not agree well with our previous analysis. In particular, we were not able to confirm our previous conclusion that buildings near the edges of woodlots were important to many bird species in eastern Pennsylvania, USA.