Relationships we have with our friends, family, or colleagues influence our personal decisions, as well as decisions we make together with others. As in human beings, despotism and egalitarian societies seem to also exist in animals. While studies have shown that social networks constrain many phenomena from amoebae to primates, we still do not know how consensus emerges from the properties of social networks in many biological systems. We created artificial social networks that represent the continuum from centralized to decentralized organization and used an agent-based model to make predictions about the patterns of consensus and collective movements we observed according to the social network. These theoretical results showed that different social networks and especially contrasted ones – star network vs. equal network - led to totally different patterns. Our model showed that, by moving from a centralized network to a decentralized one, the central individual seemed to lose its leadership in the collective movement's decisions. We, therefore, showed a link between the type of social network and the resulting consensus. By comparing our theoretical data with data on five groups of primates, we confirmed that this relationship between social network and consensus also appears to exist in animal societies.
Citation: Sueur C, Deneubourg J-L, Petit O (2012) From Social Network (Centralized vs. Decentralized) to Collective Decision-Making (Unshared vs. Shared Consensus). PLoS ONE 7(2): e32566. https://doi.org/10.1371/journal.pone.0032566
Editor: Timothy Ravasi, King Abdullah University of Science and Technology, Saudi Arabia
Received: September 23, 2011; Accepted: January 28, 2012; Published: February 29, 2012
Copyright: © 2012 Sueur et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was supported by Wallonia Brussels International, the Belgian National Fund for Scientific Research, the Franco-American Commission, the Alsace Region and the Fyssen Foundation. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Every day humans make decisions. Any decision made by an individual is influenced by the relationships he or she has with people in different circumstances. For example, the head of a firm, or the parents in a family, may exert greater influence in the decision-making process than other contributors do. For a presidential election, people generally decide individually which candidate they will vote for, even if friends and family may influence their decision to a certain extent. Broadly speaking, two systems of decision-making are often described: on the one hand, one individual, or one sub-group, decides for the rest of the group (unshared consensus, [1]); on the other hand, each individual can make independent decisions and take an equal part in the vote (shared consensus, [1]). This study aimed to understand how the social network – the structure of social relationships between the members of a group - may affect the influence of these individuals on collective decision-making and thus lead to unshared or shared consensus [1]–[4].
Consensus decision-making in animal groups has already been described by several authors (for a review, see [1]). Many animal species live in groups and have to reach consensus in order to maintain cohesion [5]. One of the most tractable ways of understanding how group members attain consensus is to study collective movements [6]–[9]. In this context, consensus decision-making for group movements has been described as a continuum – from an unshared consensus to an equally shared consensus [1], [10].
The influence of ecological constraints has often been used to explain the type of consensus observed [11], [12]. Studies have reported that specific individuals lead groups with the aim of gaining better personal access to food (Papio ursinus [9]; Equus burchellii [13]; Pan troglodytes [14]). In other studies, individuals who know where to find the best food resources can become the leaders [15]. These two general cases – leading according to needs or according to knowledge - can be qualified as unshared or partially shared consensus. On the other hand, shared consensus allows information to be pooled and may lead to more appropriate decisions for all group members [7], [16].
Nevertheless, these previous studies did not explore the influence that social network could have on the decision-making process. This type of direct link between the properties of social networks and the kind of consensus has been suggested [1], [10], [17], but has never been empirically tested. Moreover, as stated in [18], “models of collective motion typically do not consider social network structure”, despite the fact that an increasing number of studies illustrate how group social networks are both complex and crucial for understanding the synchronisation of group activity. Some authors have modelled the effect of social network on collective motion in human crowds [19] and in fish shoals [20], but their results were not compared to empirical data and the social networks used in their model were not representative of networks observed in animal groups. Indeed, they used simple networks, whereas Erdos-Renyi random networks or scale-free networks were found in several animal species (see [17], [21]–[24]). These social networks described in animals may constrain many social phenomena such as information or disease transmission, cooperation and group fission in species ranging from amoebae to primates [25]–[30]. The strength of the social relationships of group members is not only based on ecological constraints but also on species-specific and group-specific internal factors [31]. In the genus Macaca [32] or Cebus [33], different social styles [34] have been described, ranging from despotic to egalitarian societies [35]. Vehrencamp [36] first described these societies as follows: “Variation in the balance point between the forces of cooperation and competition is common from society to society. In egalitarian societies, benefits are divided roughly equally or in proportion to the risk or effort taken. In despotic societies, on the other hand, benefits accrue disproportionately to a few individuals in the group at the expense of others. Societies can thus be ranked along a continuum in terms of the degree to which fitnesses of individuals within social groups are biased”. This variation among social networks can be observed through behavioural patterns that co-vary [32]. In rhesus and Japanese macaques (Macaca mulatta and M. fuscata), for example, most conflicts are unidirectional, high-intensity aggression is common, and few conflicts are reconciled. The dominant male appears to be very central, managing conflicts and receiving the most grooming or other affiliative interactions [21]. In Sulawesi macaques, most conflicts are bidirectional, aggression is generally of low intensity, conciliatory tendencies are frequent and grooming is distributed between all individuals rather than centralized on the dominant male [31]. There is a recursive feedback loop between the social network and individual behaviour. Sueur and Petit [10] suggested a similar link between social networks, - especially centrality - and consensus in their studies on collective movements in macaques. An equally shared consensus was found in the egalitarian Tonkean macaques (M. tonkeana), whereas the more despotic rhesus macaques used a partially shared consensus when deciding to move. In the same way, it has previously been reported that species with strict hierarchies appear to have unshared, or partially shared, consensus (Canis lupus [37]; Helogale parvula [38]; Equus caballus [39]; Gorilla gorilla berengei [40]).
Within many such societal organizations – despotic or egalitarian - it is still not known how consensuses (reaching a common decision in spite of conflicts of interest) emerge from then influence properties of the social network. These networks, despotic and egalitarian can be directly compared to centralized and decentralized networks respectively [41]. Dominant or central individuals are classically described as leaders but many factors may constrain this leadership and we do not know if social relationships really influence consensus, nor which of these relationships (aggressive vs. affiliative for instance) influences the consensus and to what extent it does so. Here, we based our study on the assumption that the extent of affiliative relationships may lead to a specific type of consensus, as the distribution of these relationships seems to drive many other phenomena. However, to test this assumption, we first need to combine an experimental approach on several groups with modelling, and then have to combine social network analysis with models for collective motion. We first created artificial social networks – representing the continuum from centralized to decentralized organization – and then developed a stochastic model to make predictions about the patterns of collective movements that would emerge from them. Who leads? And who is more successfully followed, both in terms of the number of followers and of the time needed for a follower to join the movement? We predict that the more centralized the network is, the more differences will appear between individuals, with the emergence of a leader during collective movements. And if it has a central position in the network, this leader will increase both the number of joiners and the joining speed of individuals. We then compared the relations between social network and patterns of collective movements to linear and non-linear functions in order to establish how leadership emerges from the social network. Indeed, many studies have already shown that the relation between the information transfer and the probability of performing a behaviour does not increase linearly but in a non-linear way due to an amplification process [42]–[44]. In order to validate or nullify our assumptions based on simulations, the theoretical data was therefore compared to observed data collected from the observation of collective movements in five different groups of primates living in similar semi free-ranging conditions.
This study involved the observation of animals without animal handling or invasive experiments carried out on studied subjects. We declare that our study was carried out in full accordance with the ethical guidelines of our institution with the approval of the latter (certificate number: 67-339, French Republic, Bas-Rhin County Hall, French veterinary services). Our experiments comply with European animal welfare legislation. The work being carried out during this study is in accordance with the weatherall report and all efforts were made to ensure the welfare of the animals and minimize suffering. Concerning the amelioration of animal welfare, the study groups were bred under semi free-ranging conditions at the Strasbourg University Centre of Primatology. They had complete access to about 0.35 ha (maximal length = 80 m; maximal width = 60 m) of wooded parkland as well as indoor housing within the enclosure. The indoor housing (20 m2) is made of cement and tiling. The enclosure area was made up of various slopes and uneven ground. The distribution of vegetation was also heterogeneous, with three layers (grass, trees and bushes) that were unevenly distributed throughout the enclosure. For each group, fresh fruit and vegetables were provided once a week, one hour after the end of the observation session. Thus, the behaviour of the animals was unlikely to be affected by this event. Animals were used to human presence in their enclosure.
The model is based on rules of mimetism/cohesion (Markov chain process) described in several studies on collective phenomena [42]–[44]. In this model, the probability that an individual will join the collective movement depends on the number but also the strength of relationships it has with the individuals already participating in the movement. The number of individuals, individual identities and the network of affiliative relationships of each artificial social network are included in the model. At the start of a simulation, all agents (N) were in an area called the resting area and had to move to another area, the foraging area. We implemented the intrinsic probability λi of each agent. This intrinsic probability is independent of the influence of conspecifics, and is, for example, a nutrient need. The departure probability of the initiator (first individual to depart) was the same whatever the social network and identity of this individual and was constant per time unit. The departure probability ψ01 of the initiator was: