Comments on predictability and overdifferencing in vigilance sequences

摘要

The degree of predictability in behaviour sequences is a central ethological question in many studies. For example, individuals, while routinely scanning for predators or locations of conspecifics, may be at a disadvantage if their scans can be predicted. Two empirical issues arise in regard to the predictability of these vigilance sequences, here defined as time series consisting of durations of sequential inter-scan intervals (ISIs). First, to what degree are ISI durations predictable from observations of past ISI durations? Second, does knowledge of the time spent in the current interval (time since the previous scan) provide information useful for predicting how much longer the current interval will last? Desportes et al. (1989) refer to unpredictability of ISI dur-ations based on past ISI durations as sequential randomness, and unpredictability of the remaining length of an ISI based on knowledge of the time spent in the current interval as instantaneous randomness. The first issue relates to time series (autocorrelation) properties of ISI durations, the second to the probability distribution of ISI durations. These two issues are essentially distinct, both conceptually and statistically. In this paper we discuss the first issue in reference to a recent article of Roberts (1994).