Testing determinants of the annual individual fitness: An overall mean mixture model for de‐lifing data

Abstract

Abstract The de‐lifing method (Coulson et al., 2006, Proceedings of the Royal Society of London B: Biological Sciences, 273, 547), though very promising for studying ecological and evolutionary changes, has been scarcely used to identify factors influential on fitness. Through simulations representative of a variety of iteroparous species, we establish that a two‐component normal mixture usually provides a much better representation of de‐lifing data than the single normal distribution assumed in classical linear models. To test determinants of the annual individual fitness, we propose the overall mean mixture model (O3M), a mixture model parameterised in terms of the overall mean of the mixture distribution. We examine the gain in performance and accuracy when using the O3M over classical linear models and bootstrap methods on simulated finite normal mixture distributions for different regression shapes and variance structure, and apply it to a real dataset to study how the annual individual fitness varies with age in alpine marmots. The O3M improves considerably the precision of the estimates and hence the power of the analysis. We discuss the adaptation of the O3M model to more complex distributions and advise on its use.