Summary: Across very different ecosystems and species, animals from meerkats to coatis and spotted hyenas show a common pattern in how they switch between activities. Using accelerometer tracking and machine learning classification, researchers found that the longer an animal remains in a behavior—resting, foraging, or moving—the less likely it is to switch out of that state in the next moment. This conserved pattern is described as a decreasing hazard function.
The research also identifies a consistent “predictivity decay” curve: as the time horizon stretches, current behavior becomes steadily less useful for predicting future actions. Together, these findings point to shared structural principles in mammalian behavioral sequences.
Key Facts:
- Shared Algorithm: Meerkats, white-nosed coatis, and spotted hyenas all display a decreasing likelihood of switching behaviors over time.
- Predictivity Decay: Predictability of future behavior decreases with time in a consistent, truncated power-law form across species.
- Cross-Species Insight: Results imply an underlying architecture to decision-making and behavioral dynamics in social mammals.
Source: Max Planck Institute
Animal behavior is remarkably diverse, yet these new results suggest a hidden order beneath the variety.
A collaborative study led by researchers at the Max Planck Institute of Animal Behavior analyzed high-resolution movement data from three social mammal species living in very different habitats: meerkats in the Kalahari Desert, white-nosed coatis in Panama’s rainforest, and spotted hyenas on the Kenyan savanna. Despite differences in size, ecology, and lifestyle, the animals’ daily activity patterns exhibited strikingly similar statistical features.
Animals were fitted with accelerometers—the same compact sensors used in consumer devices to record motion—generating continuous posture and movement traces over days or weeks. Machine learning models then classified those traces into discrete behavioral states such as lying, foraging, walking, and resting. From these labeled sequences, researchers examined how long behaviors lasted, how the probability of switching changed over time, and how well current states predicted future actions.
Uncovering underlying patterns
Across behaviors, individuals, and species, a clear principle emerged: behavioral bouts followed decreasing hazard functions. In plain terms, the longer an animal remains engaged in a given behavior, the lower the instantaneous probability that it will change behavior in the next moment. This “lock-in” effect contradicts the intuitive expectation that animals would become more likely to switch behaviors the longer they remain in one state.
Alongside decreasing hazard functions, researchers quantified predictivity decay—the decline in forecasting accuracy as predictions extend farther into the future. Predictability consistently dropped as a truncated power-law function for all three species, and the estimated power-law exponents were very similar across animals. Bout-duration distributions were heavy-tailed, indicating that while many bouts are short, longer episodes occur more often than would be expected under simple exponential models.
These regularities suggest that behavioral dynamics are not purely random or entirely species-specific; instead, they may reflect common constraints or strategies shaping how animals sequence their activities over multiple timescales.
Why these patterns?
The study explores two broad, non-exclusive explanations for these shared patterns. One possibility is behavioral self-reinforcement or positive feedback: remaining in a state can increase its immediate value, making continued engagement more likely. For example, resting may be rewarded by thermoregulation or social safety, and foraging may be reinforced by recent success.
A second explanation invokes multi-timescale decision-making. Rather than a single internal timer triggering switches, animals may integrate many overlapping signals—hunger, fatigue, risk cues, social dynamics—each operating on different timescales. The combined influence of these nested processes could produce long-tailed bout durations, decreasing hazard rates, and the observed predictivity decay.
Future work can test whether these structural features appear across broader taxonomic groups, in non-social species, across life stages, or under varying ecological pressures. Researchers also aim to determine whether such behavioral architectures confer adaptive advantages, for example by optimizing energy use, attention, or group coordination.
As co-author Meg Crofoot, Director of the Department for the Ecology of Animal Societies, notes: this study points to hidden organizational principles in real animal behavior that may be conserved across diverse branches of life.
About this evolutionary neuroscience and behavior research news
Author: Carla Avolio
Source: Max Planck Institute
Contact: Carla Avolio – Max Planck Institute
Image: The image is credited to Neuroscience News
Original Research: Open access.
“Behavioral sequences across multiple animal species in the wild share common structural features” by Pranav Minasandra et al., published in PNAS.
Abstract
Behavioral sequences across multiple animal species in the wild share common structural features
Animal activity can be represented as sequences of discrete behavioral bouts. Statistical analysis of these sequences reveals drivers of behavioral decisions and the temporal structure of behavior. Previous laboratory work—mainly in invertebrates—has indicated multiple timescales and long-range memory in behavior, but whether such patterns generalize to wild vertebrates was unclear.
Analyzing accelerometer-inferred behavioral states in three social mammal species (meerkats, white-nosed coatis, and spotted hyenas), the study found consistent structural features across states, individuals, and species. Behavioral bouts exhibited decreasing hazard functions: the longer a bout had lasted, the less likely it was to end in the next instant. Predictability of future states declined with time according to a truncated power-law, with similar exponent estimates across species. Bout-duration distributions were heavy-tailed.
The origins of these shared structural principles remain unresolved. Plausible explanations include environmental variability, behavioral self-reinforcement, and hierarchical, multi-timescale decision processes. The high consistency of these patterns across the studied species suggests these phenomena may be widespread in nature and points toward foundational properties of behavioral dynamics that warrant further investigation.