Collective motion diminishes, but variation between groups emerges, through time in fish shoals

Abstract

Despite extensive interest in the dynamic interactions between individuals that drive collective motion in animal groups, the dynamics of collective motion over longer time frames are understudied. Using three-spined sticklebacks, Gasterosteus aculeatus, randomly assigned to 12 shoals of eight fish, we tested how six key traits of collective motion changed over shorter (within trials) and longer (between days) timescales under controlled laboratory conditions. Over both timescales, groups became less social with reduced cohesion, polarization, group speed and information transfer. There was consistent inter-group variation (i.e. collective personality variation) for all collective motion parameters, but groups also differed in how their collective motion changed over days in their cohesion, polarization, group speed and information transfer. This magnified differences between groups, suggesting that over time the 'typical' collective motion cannot be easily characterized. Future studies are needed to understand whether such between-group differences in changes over time are adaptive and represent improvements in group performance or are suboptimal but represent a compromise between individuals in their preferences for the characteristics of collective behaviour.

Keywords: collective personality; plasticity; reaction norm; repeatability.

Figure 1.

Variation between the groups in collective motion over time; (a,d) show the predicted reaction norms of the 12 groups across days from the best-supported models to explain polarization, centroid speed, convex hull area and maximum cross-correlation in speed, and (e) shows the predicted reaction norms from intervals during a trial from the best-supported model to explain the rate of transitions in state. The predicted reaction norms for each group are coloured by the slope (highest (yellow) to lowest (dark purple)) within each panel and estimated using the predict function in R package lme4 [60]. For predictions, the interval (in a–d) and day (in e) are held at their mean value. The circles are the raw values (i.e. per interval per trial, n = 416). For the convex hull area, values are square-root transformed to meet parametric assumptions. In (a–d), the lines are truncated to the day where the experiment ended for a particular group; (f–i) show the trends in between-group variance over days of the experiment. Between-group variance is the model-estimated variance in the random intercept of group identity when the variable is centred at day two, four, six, eight, 10 and 12, respectively, shown as filled squares with upper and lower 95% bootstrap confidence intervals shown as vertical lines for each estimate; (j) shows the trend in between-group variance in the rate of transitions in state within trials for the model-estimated variance in the random intercept of group identity when the variable interval is centred at interval one to five, respectively. The trend in the model-estimated variance in group identity reveals how inter-group differences in collective motion change over time.