“Born to run? Quantifying the balance of prior bias and new information in prey escape decisions”
Nicholas M. Sutton and James P. O’Dwyer (Sep 2018)
Optimality modeling, risk assessment, and Bayesian inference of prior experience predict prey escape decisions
Past experience and new information explain decision-making in white-tailed deer
Wildlife, and specifically deer, are a common sight at state parks. But why do some deer seem to run away at the first sight of a human, while others will happily eat from the palm of your hand? And how can you tell how the deer at any given park are likely to respond to humans? In this manuscript, Nick Sutton, PhD student at the University of Illinois at Urbana-Champaign (UIUC), and James O’Dwyer, Assistant Professor at UIUC, find that this specific question provides insight into a deeper debate relating to the balance of old and new information in animal decision-making. Sutton and O’Dwyer collected data on the escape behavior of deer under human approach, and combined this with a model evaluating the costs and benefits to deer of remaining around humans versus fleeing from them. The resulting predictions depend on a balance of previous experience of humans for a given deer, and new information gathered during an encounter.
Using this approach, Sutton and O’Dwyer inferred the effects of past experiences on deer behavior at the population level, and evaluated multiple sources of information that deer are using when making the decision to flee, such as how fast humans are moving or how far away they are. By inferring the prior experiences of deer and comparing models based on different types of information, Sutton and O’Dwyer found that they can accurately predict the distance at which deer decide to flee, and that how close humans can get will vary between parks. This new modeling approach provides general methods for predicting the optimal moment to make a high-stakes decision, such as deer deciding when to flee from humans. The approach can also apply to other taxa, even humans, whenever decisions are based on a combination of prior experience, costs/benefits and new information.
Animal behaviors can often be challenging to model and predict, though optimality theory has improved our ability to do so. While many qualitative predictions of behavior exist, accurate quantitative models, tested by empirical data, are often lacking. This is likely due to variation in biases across individuals and variation in the way new information is gathered and used. We propose a modeling framework based on a novel interpretation of Bayes’ theorem to integrate optimization of energetic constraints with both prior biases and specific sources of new information gathered by individuals. We present methods for inferring distributions of prior biases within populations rather than assuming known priors, as is common in Bayesian approaches to modelling behavior, and for evaluating the goodness of fit of overall model descriptions. We apply this framework to predict optimal escape during predator-prey encounters, based on prior biases and variation in what information prey use. Using this approach we collected and analyzed data characterizing white-tailed deer (Odocoileus virginianus) escape behavior in response to human approaches. We found that distance to predator alone was not sufficient to predict deer flight response, and show that the inclusion of additional information is necessary. Additionally, we compared differences in the inferred distributions of prior biases across different populations and discuss the possible role of human activity in influencing these distributions.