Finding the smoothest path to success: Model complexity and the consideration of nonlinear patterns in nest-survival data

摘要

Quantifying patterns of nest survival is a first step toward understanding why birds decide when and where to breed. Most studies of nest survival have relied on generalized linear models (GLM) to explore these patterns. However, GLMs require assumptions about the models' structure that might preclude finding nonlinear patterns in survival data. Generalized additive models (GAM) provide a flexible alternative to GLMs for estimating linear and nonlinear patterns in data. Here we present a comparison of GLMs and GAMs for explaining variation in nest-survival data. We used two different model-selection criteria, the Bayes (BIC) and Akaike (AIC) information criteria, to select among simple and complex models. Our study was focused on the analysis of Red-winged Blackbird (Agelaius phoeniceus) nests in the Rainwater Basin wetlands of south-central Nebraska. Under BIC, our quadratic model of nest age had the most support, and the model predicted a concave pattern of daily nest survival. We found more model-selection uncertainty under AIC and found support for additive models with ordinal effects of both day and age. These models predicted much more temporal variation than did the linear models. Following our analysis, we discuss some of the advantages and disadvantages of GAMs. Despite the possible limitations of GAMs, our results suggest that they provide an efficient and flexible way to demonstrate nonlinear patterns in nest-survival data.