Abstract:Animal behavior can be decomposed into a continuous sequence of discrete activity bouts over time. Analyzing the statistical structure of such behavioral sequences provides insights into the drivers of behavioral decisions in animals. Laboratory studies, predominantly in invertebrates, have suggested that behavioral sequences are characterized by multiple timescales and exhibit long-range memory, but the extent to which these results can be generalized to other taxa and to animals living in natural settings remains unclear. By analyzing accelerometer-inferred predictions of behavioral states in three species of social mammals (meerkats, white-nosed coatis, and spotted hyenas) in the wild, we discovered surprisingly consistent structuring of behavioral sequences across all behavioral states, all individuals and all study species. Behavioral bouts were characterized by decreasing hazard functions, wherein the longer a behavioral bout had progressed, the less likely it was to end within the next instant. The predictability of an animal’s future behavioral state as a function of its present state always decreased as a truncated power law for predictions made farther into the future, with very similar estimates for the power law exponent across all three species. Finally, the distributions of bout durations were also heavy-tailed (typically truncated power-law or lognormal). These patterns suggest similar structural principles in behavioral sequences across all study species. Why such shared patterns emerge remains unknown. We explore multiple plausible explanations, including environmental non-stationarity, behavioral self-reinforcement via positive feedback, or a combination of multiple relevant timescales driven by the hierarchical nature of behavior. The existence of highly consistent patterns in behavioral sequences across all three study species suggests that these phenomena could be widespread in nature, and points to the existence of fundamental properties of behavioral dynamics that could drive such convergent patterns.

Long Timescales, Individual Differences, and Scale Invariance in Animal Behavior

Long time scales, individual differences, and scale invariance in animal behavior

Fluctuating landscapes and heavy tails in animal behavior

Capturing continuous, long timescale behavioral changes in $\textit{Drosophila melanogaster}$ postural data

Behavioral sequences across multiple animal species in the wild share common structural features

Searching for long timescales without fine tuning

Individuality across environmental context in

Temporal autocorrelation: a neglected factor in the study of behavioral repeatability and plasticity

Searching for long time scales without fine tuning

A Markovian dynamics for Caenorhabditis elegans behavior across scales

Scale-free behavioral dynamics directly linked with scale-free cortical dynamics

Unraveling linguistic patterns in dog behaviour

Changing measurements or changing movements? Sampling scale and movement model identifiability across generations of biologging technology

Consistent hierarchies of single-neuron timescales in mice, macaques and humans

Signatures of hierarchical temporal processing in the mouse visual system

A Markovian dynamics for C. elegans behavior across scales

From Fine-Scale Foraging to Home Ranges: A Semivariance Approach to Identifying Movement Modes across Spatiotemporal Scales

Dominating Lengthscales of Zebrafish Collective Behaviour

The scale-invariant covariance spectrum of brain-wide activity

Neural timescales reflect behavioral demands in freely moving rhesus macaques