Complex behaviors such as foraging, social, and antipredator decisions are notoriously difficult to predict. The most common approach to making behavioral predictions tractable is via optimality theory and modeling. What food items to consume and when to flee from predators are important problems that carry significant fitness consequences, and under optimality theory individuals should seek to optimize these behaviors with respect to fitness. While useful in forming qualitative predictions of behavior, observed animal behavior typically deviates from quantitative predictions, with high intraspecific behavioral variation commonly cited as an explanation for these deviations. We perceive two gaps in the ability of existing optimality models to make precise, quantitative predictions in the presence of high intraspecific behavioral variation. First, individuals may bias decision-making due to personality effects or the effects of prior experience, such as habituation or sensitization. Second, individuals may be limited in their capacity to acquire and utilize information, leading to decisions based on a subset of the total information available. Both individual biases and variation in information use can lead to deviations from predicted optimal behaviors. Optimality models therefore need to account for these effects in order to produce accurate quantitative predictions of behavior.
To address these gaps in existing models, we propose a new modeling framework for generating quantitative predictions of behavior in the presence of significant intraspecific behavioral variation. Our approach provides methods for identifying the types of information most likely to be utilized by animals, as well as methods for inferring the effects of prior personality or learning based biases within populations. Using these methods we analyzed white-tailed deer (Odocoileus virginianus) escape behavior in response to variable human approaches. Our modeling framework successfully predicted deer escape behavior, accounting for variation in human approach and also capturing the effects of non-flight behavior. Through rigorous goodness of fit testing and model comparison methods we show how variation in information use can affect predictions of behavior, while our population level inference of prior biases successfully captures the effects of the individual behavioral variation inherent in any wildlife population. The success of our framework represents the first instance of accurate, quantitative predictions of escape behavior via optimal escape modeling. Our approach lays the groundwork for addressing further questions, such as how individual level predator-prey interactions might scale up to population level dynamics, or how prey risk assessment may coevolve with changing predator approach strategies.