The emergent properties of a dolphin social network

• Complex networks can keep vertices connected through short chains of intermediate vertices, enabling rapid information transfer.Source: §1 · p. 1
• Existing network models distinguish homogeneous Poisson connectivity from heterogeneous scale-free power-law connectivity at large k.Source: §1 · p. 1
• Animal societies rely on information transfer, but their organisational principles remain poorly understood.Source: §1 · p. 1
• The paper investigates whether a small bottlenose dolphin society exhibits emergent network properties comparable to complex human and technological networks.Source: §1 · pp. 1–2

• Dolphins were identified by dorsal-fin markings, and repeated school observations from 1995–2001 recorded which adults were seen together.Source: §2 · p. 2
• Pairwise association strength was measured with the half-weight index, comparing observed co-occurrence with expected co-occurrence when either dolphin was encountered.Source: §2 · p. 2
• Only individuals surviving the first 12 months were analysed, ensuring sufficient information to estimate association preferences.Source: §2 · p. 2
• Association significance was tested using 20,000 within-group permutations while preserving group size, fjord, and individual sighting counts.Source: §2 · p. 2
• Pairs were classified as preferred companions when observed HWI exceeded more than 95% of permuted expected values.Source: §2 · p. 2
• The observed network was compared with random networks having the same numbers of vertices and links using diameter and clustering coefficient.Source: §2 · p. 2

• The dolphin network contained 64 individuals linked by 159 preferred companionships, with average connectivity k = 4.97.Source: §3 · p. 2
• Its degree distribution was not Poisson (G_adj^2 = 26.48, df = 12, p = 0.009), and the tail for k ≥ 7 resembled a scale-free distribution with γ_dolphin = 3.45 ± 0.1.Source: §3 · pp. 2–3
• The dolphin network had substantially higher clustering than matched random networks (C_dolphin = 0.303; C_random = 0.081).Source: §3 · p. 3
• Random removal of 20% of individuals increased network diameter by only 0.4, indicating resilience to random attacks.Source: §3 · p. 3
• Removing individuals with the most associates increased the shortest path by 1.6 after 20% removal, but the network did not fragment.Source: §3 · p. 3
• Even after removing more than 30% of individuals randomly or selectively, a large cluster still contained most individuals alongside isolated singletons.Source: §3 · p. 3

Random network

• The study illustrates the dolphin network alongside a random network constructed with similar characteristics; the dolphin network is inhomogeneous, whereas the random network is homogeneous with Poisson-distributed vertex link counts.Source: Figure 2 · p. 3, Random network · p. 3
• The dolphin network was inhomogeneous with a few highly connected vertices, whereas the random network was homogeneous with Poisson-distributed link counts.Source: Random network · p. 3

Individual roles

• Hubs were mainly adult females, except for one adult male, and seemed to be older individuals identified by scars and larger size.These individuals appeared to help maintain short information paths between population members.Source: §4 · p. 4

Network resilience

• Removing highly connected individuals altered network diameter, but the increase was trivial compared with other scale-free networks, where removing only 2% of highly linked nodes can more than double diameter.The dolphin network did not fragment under targeted attacks, unlike typical scale-free networks.Source: §4 · p. 4

Mechanisms

• The network’s high clustering may provide redundant connections, leaving several short paths between individuals and increasing resilience to deaths.This interpretation links redundancy to the network’s ability to remain cohesive after removals.Source: §4 · p. 4

Connectivity structure

• The dolphin network’s resilience to individual removal may be related to its link-number distribution, which differed from typical scale-free networks and flattened for k<7.Source: §4 · p. 4
• Individuals with intermediate numbers of associates (4–7) seemed important for maintaining redundant links between different sections of the network.The network also contained association centres, indicating that individuals did not all play equivalent structural roles.Source: §4 · pp. 3–4

Network resilience

• Even after removing more than 30% of individuals, the largest cluster still contained most network members under both random and targeted attacks.The fragmentation pattern was similar for both attack types, contrasting with fragmentation reported for other small-world networks under targeted attacks.Source: Figure 4 · p. 4, §4 · p. 4

• The largest cluster retains most individuals after both random and targeted attacks, unlike other small-world networks under targeted attacks.Source: §1 · p. 4
• The benefits of this emergent resilience could also be relevant to networks vulnerable to targeted attacks, including the World Wide Web and Internet.Source: §1 · p. 5
• Even after removing more than a third of the population, the network can maintain a cohesive society.Source: §1 · p. 5
• Its scaling properties permit the evolving network to assimilate new vertices without disrupting cohesion.Source: §1 · p. 5
• Observed scale-free properties in this very small network suggest that the self-organizing phenomena may reflect general laws of evolving networks.Source: §1 · p. 5
• The network’s resilience to selective and random attacks allows cohesion to persist despite catastrophic losses.Source: §1 · p. 5