Analyzing Animal Escape Data with Circular Nonparametric Multimodal Regression

Analyzing the escape direction of animals subject to covariates is a problem that requires statistical techniques beyond classical regression methods. Apart from the periodicity of the angle of direction, which demands the use of circular statistics, animal escape data usually call for the exploration of the preferred orientations rather than the expected orientation. In this paper we propose the use of a nonparametric method to estimate the conditional local modes of the escape directions of animals from a regression perspective. We present the estimation algorithms and study the asymptotic properties of the estimators as well as its finite sample performance through some simulation experiments. Our proposal is used to model the escape behavior of a group of larval zebrafish escaping from a robot predator. More broadly, the approach presented in this paper can be applied to many existing problems related to animal behavior or other fields.