Do cross-scale correlations confound analysis of nest site selection for chestnut-backed chickadees?

Abstract

Multiscale models of nest site selection often ignore cross-scale correlations (correlations between predictor variables at different scales). We reexamined nest site selection of Chestnut-backed Chickadees (Poecile rufescens) at two scales within territories to isolate: (1) variation associated purely with variables measured at the patch (0.031 ha) and tree scales, and (2) variation shared by variables measured at the patch and tree scales. We used conditional (fixed-effect) logistic regression to build a patch, tree, and full model, and subtracted pure and shared components of variation from the deviance explained by the full model. Tree scale and patch scale variables accounted for 85% and 9% of the total explained variation, respectively. Only 6% of explained variation in nest site locations was due to cross-scale correlations. We suggest that multiscale habitat selection studies incorporate a diagnostic tool like variance decomposition to avoid spurious results caused by lack of independence of habitat relationships. © The Cooper Ornithological Society 2008.