In ecology, as in several other disciplines of research, the distribution of data are often clustered in different levels of organization. In such cases, the use of single level model with fixed slopes would be erroneous, especially since subjects (people or species) within hierarchies or clusters are more similar than when randomly sampled from an entire population. For example, subjects normally may belong to organizational bodies such as families, genus, tribes, landscapes, ecosystems etc. In such cases, a common oversight is to ignore this hierarchical structure of data and assume that the subjects are independent. In contrast to ordinary least squares regression models that are characterized by fixed effects, MLMs specify regression coefficients as random effects, and often include more than one random effect. MLMs are becoming increasingly common in ecology and researchers are continuously trying to extend and improve techniques for predictor importance in MLMs. Very recently, a novel technique using commonality coefficients have been developed to analyze predictor importance from the perspective of improving the fit of models to data in MLMs (Nimon et al. in prep).
Results/Conclusions:
In prey-predator ecological systems, prey individuals can employ several adaptive behaviours to reduce predation risk. In order to learn how behaviours interact in an overall strategy of risk management, and whether such strategies are fixed or contingent on changing conditions, Morris (in prep) conducted an experiment evaluating snowshoe hares’ (Lepus americanus) vigilance in artificial food patches in two habitats and moon phases. We use a subset of those data to demonstrate how a MLM can provide insight into hierarchically structured ecological data.
Analysis by a linear mixed-effects model yielded a single significant effect: the number of vigilant images recorded by remote wildlife cameras increased with the total number of images. A common interpretation of this result would be that neither habitat or moon phase is ‘important’ to snowshoe hare’s risk management. The more sensitive MLM reveals, instead, that habitat, moon phase and distance from safety contribute additional unique variance to hare behavior. We encourage ecologists to thoughtfully include the hierarchical structure of their data in order to more fully understand the complex dynamics of ecological and evolutionary systems.
The paper further discusses the advantages and limitations of this technique.