Testing hypotheses in ecoimmunology using mixed models: disentangling hierarchical correlations.

Considerable research in ecoimmunology focuses on investigating variation in immune responses and linking this variation to physiological trade-offs, ecological traits, and environmental conditions. Variation in immune responses exists within and among individuals, among populations, and among taxonomic groupings. Understanding how variation and covariation are distributed and how they differ across these levels is necessary for drawing appropriate ecological and evolutionary inferences. Moreover, variation at the among-individual level directly connects to underlying quantitative genetic parameters. In order to fully understand immune responses in evolutionary and ecological contexts and to reveal phylogenetic constraints on evolution, statistical approaches must allow (co)variance to be partitioned among levels of individual, population, and phylogenetic organization (e.g., population, species, genera, and various higher taxa). Herein, we describe how multi-response mixed-effects models can be used to partition variation in immune responses among different hierarchical levels, specifically within-individuals, among-individuals, and among-species. We use simulated data to demonstrate that mixed models allow for proper partitioning of (co)variances. Importantly, these simulations also demonstrate that conventional statistical tools grossly misestimate relevant parameters, which urges caution in relating ecoimmunological hypotheses to existing empirical research. We conclude by discussing the advantages and caveats of a mixed-effects modeling approach.