Social structure emerges via the interaction between local ecology and individual behaviour


  • Colby J. Tanner,

    Corresponding author
    1. Department of Zoology, Trinity Centre for Biodiversity Research, School of Natural Sciences, Trinity College Dublin, Dublin, Ireland
    2. Department of Ecology and Evolution, Université de Lausanne, Lausanne, Switzerland
    Search for more papers by this author
  • Andrew L. Jackson

    1. Department of Zoology, Trinity Centre for Biodiversity Research, School of Natural Sciences, Trinity College Dublin, Dublin, Ireland
    Search for more papers by this author

Correspondence author. E-mail:


1. The formation of groups is a fundamental aspect of social organization, but there are still many questions regarding how social structure emerges from individuals making non-random associations.

2. Although food distribution and individual phenotypic traits are known to separately influence social organization, this is the first study, to our knowledge, experimentally linking them to demonstrate the importance of their interaction in the emergence of social structure.

3. Using an experimental design in which food distribution was either clumped or dispersed, in combination with individuals that varied in exploratory behaviour, our results show that social structure can be induced in the otherwise non-social European shore crab (Carcinus maenas).

4. Regardless of food distribution, individuals with relatively high exploratory behaviour played an important role in connecting otherwise poorly connected individuals. In comparison, low exploratory individuals aggregated into cohesive, stable subgroups (moving together even when not foraging), but only in tanks where resources were clumped. No such non-foraging subgroups formed in environments where food was evenly dispersed.

5. Body size did not accurately explain an individual’s role within the network for either type of food distribution.

6. Because of their synchronized movements and potential to gain social information, groups of low exploratory crabs were more effective than singletons at finding food.

7. Because social structure affects selection, and social structure is shown to be sensitive to the interaction between ecological and behavioural differences among individuals, local selective pressures are likely to reflect this interaction.


Social organization is among the most fundamental features in animal societies, characterizing ecological as well as social selective pressures of a species (Ross & Keller 1995). Social organization can lead to extremely complex societies (de Waal & Tyack 2003). One of the most basic structural components of any society is the formation of groups. Following Wilson (1975), we define group as a set of conspecific individuals that remain together and interact with each other more than with other conspecifics. Although the process by which individuals decide to aggregate has interested scientists for over a century (Galton 1871), there are still many unanswered questions regarding how this decision is made (Krause & Ruxton 2002).

Social structure (i.e. the formation of groups via connections among individuals) can have important effects on selective pressure. Grouping can provide benefits such as protection from predators, help in finding mates, thermoregulatory assistance, increased movement efficiency and food finding benefits (Krause & Ruxton 2002). For group living to be adaptive, however, the advantages of being in a group must generally outweigh the costs (Alexander 1974).

Because social structure is of such widespread importance, understanding its development, or how individuals organize themselves into groups, is an essential ecological component of every grouping species. Both ecological (Bronikowski & Altmann 1996; Pusey & Packer 1997; Johnson et al. 2002) and individual (Conradt & Roper 2005; Croft, James & Krause 2008) factors are known to affect social structure. Group formation is often associated with spatial features, such as resource distribution, which can affect social structure by altering the costs and benefits of sharing a territory (Pusey & Packer 1997; Johnson et al. 2002). Groups can form when individuals direct their attention towards a site containing active individuals (i.e. area copying), such as the location of a resource (Giraldeau 1997; Budden & Wright 2005; Leu, Kappeler & Bull 2011). Finally, differences among individuals for traits such as morphology (Krause et al. 2005) and behaviour (Viscido, Parrish & Grünbaum 2004) can affect group formation, composition and structure.

Although ecological conditions and individual differences are known to affect social organization for many species, how these two types of factors interact to shape social structure has not yet been explored. Here, we investigate how the interaction between food distribution (clumped or dispersed) and phenotypic differences among individuals (morphological and behavioural) affects the social structure of the European shore crab (Carcinus maenas). We use social network analysis, which provides a way of measuring how interactions among individuals give rise to the properties of a network or population (Krause, Croft & James 2007). It is also a useful tool for determining how individuals’ traits are associated with their positions in a network, which can explain their roles within the society (Krause, Croft & James 2007; Clifton, Turkheimer & Oltmanns 2009; Schürch, Rothenburger & Heg 2010).

Our first objective is to test whether individual shore crabs form groups that persist beyond foraging. The evolution of sociality in some species is thought to coincide with ecological factors such as clumped food distribution (Jarvis et al. 1994). We therefore would expect any groups of C. maenas to form in tanks where food is clumped. Our second objective is to test the hypothesis that phenotypic traits influence group formation. Because groups often contain individuals of similar phenotypes (Krause & Ruxton 2002), we expect to find shore crab groups, if they form, to contain individuals of similar morphologies and/or behaviours. Aside from forming groups, there are many other ways that individuals can affect social structure according to their social roles (Flack et al. 2006). Our third objective is to test whether ecological factors interact with phenotypic traits during the emergence of social structure. In the presence of an interaction between ecology and phenotype, we would predict that behavioural traits will correspond to different social roles according to ecological conditions.

Materials and methods

Study organism and laboratory conditions

We collected male shore crabs c. 2 km north of Dublin, Ireland, and immediately transported them to the laboratory where we labelled each crab (1 × 1 cm2 label glued to the central dorsal carapace region) and measured its carapace width. We then put individuals into isolated containers of sea water in an environmental chamber (14 °C, 12-h light:dark) for 7 days prior to beginning the experiment (Sneddon & Swaddle 1999).

Experiments and observations

To begin the experiment, crabs were placed into tanks (nine crabs per 0·4-m2 tank, c. 22 crabs m−2) filled 25% with aerated filtered sea water and rocky substrate. Individuals were assigned to a tank based on body size, so that each tank had a range of body sizes (45–100 mm across all tanks), but no tank had a gap in body sizes between consecutively ranked individuals >4 mm. Natural shore crab population densities vary over large areas (e.g. 3–16 individuals m−2; Young et al. 1999) and can reach densities as high as c. 100 crabs m−2 (Beukema 1991). Crabs were therefore in relatively high densities, but well below the maximum of their natural habitat.

On Day 1 of the experiment, for the first 30 mins immediately after placing crabs in their tanks, digital video cameras were used to record the amount of time each crab spent exploring the novel environment. These data were used to rank each crab’s relative exploratory behaviour. This procedure was repeated on Day 15 to determine how consistent relative exploratory behaviour was among individuals throughout the experiment. On every second day after Day 1, we placed 27 g of eel meat into each tank for 30 min and video recorded the tanks. Crabs in treatment tanks (clumped food distribution; n = 10 tanks) received a single piece of eel placed near the central region (but never in the same exact location) of the tank, while crabs in control tanks (dispersed food distribution; n = 6 tanks) received nine separate 3-g pieces scattered throughout the tank. On non-foraging days, we video recorded the tanks for 30 min to analyse population structure when food was not present. Starting times were varied for both feeding and recording on non-feeding days randomly within a 5-h window each day to avoid crabs developing ‘circadian-like’ activities (e.g. Boujard & Leatherland 1993). The experiment lasted 19 days.

On Day 20, one individual from four of the treatment tanks was removed to determine how stable the existing structure was to perturbations in the social environment (Flack et al. 2006). In two of the tanks, we removed a high exploratory individual, and in the other two tanks, we removed a low exploratory individual. We continued the feeding/recording protocol for 4 days and then reversed the feeding method (switching to the dispersed food protocol) to determine how stable the existing structure was to resource distribution for an additional 7 days.

Description of network metrics

One essential concept when analysing network structure is centrality. There are many ways of measuring centrality, but all describe various aspects of an individual’s role in network cohesion (Freeman 1977; Borgatti & Everett 2006). Here, four different network measures were used to quantify centrality or subgroup cohesion. Degree centrality (weighted, non-directional) describes the total amount that an individual interacts with others in the population, and is involved in communication and synchronization processes (Borgatti & Everett 2006). Also important for communication and synchronization is eigenvector centrality (weighted, non-directional) (Bonacich 1972; Borgatti & Everett 2006), which extends degree centrality by measuring the relative centralities of an individual’s associates (i.e. a measure of how well connected an individual’s associates are). Betweenness centrality (weighted, non-directional) describes an individual’s role in connecting otherwise poorly connected individuals and is important in controlling the spread of information and/or disease within a population (Freeman 1977). Finally, clique census (binary, non-directional) is a measure of aggregation that reflects an individual’s capacity to form stable bonds with other individuals (Butts 2010).

Social network statistical analysis

Using video recordings on four non-foraging Days (3, 7, 13, and 19), association degree, eigenvector and betweenness centralities (weighted individual properties), as well as clique size (un-weighted individual property), were quantified using the SNA package in R (Butts 2010). Edge weights for the first three metrics represent the amount of time that crabs spent in contact with each other (physical contact with walking legs or claws, or within one body length from the anterior portion of the carapace). Because of spatial constraints and our ability to record 100% of the interactions among individuals within each population, we filtered network data prior to clique size analysis to include connections (defined the same as for the other three metrics) between individuals that interacted ≥1% of the total observation time (Croft, James & Krause 2008). Using averaged values for each tank, clumped and dispersed tank metrics were compared using general linear mixed-effects models with treatment as a fixed effect, time as a linear covariate and tank identity as a random factor to correct for repeated measures inherent in longitudinal studies (Bates & Maechler 2009).

To examine the relationship between body size or exploratory behaviour and centrality metrics within the population, we performed permutation tests because of the non-independence of individuals’ positions within a tank (Croft, James & Krause 2008). We first found the correlation coefficients among each individual’s four network metrics and both their ranked body size and exploratory behaviour generated from the observed data. We then compared these observed correlation coefficients with 104 coefficients derived from randomized networks in which edges (or connections) between individuals within a tank were permuted (Croft, James & Krause 2008). New P-values for each comparison represent the proportion of correlation coefficients from randomized networks that were greater in magnitude (depending on the direction of the correlation) than the observed correlation coefficient. We repeated this for the clumped and dispersed tanks on Days 3 and 19 (initial and final non-foraging observation days). This same procedure was used to determine whether there was a significant correlation between each individual’s exploratory behaviour or body size and rank of arriving to food (on foraging Days 2 and 18).

To determine the stability of individuals’ network positions, as well as group composition within networks, we performed Hemelrijk (1990)Rr modification of the Mantel test comparing the similarity of association matrices for each tank on Days 7 and 19 (after individuals had time to experience the treatment and at the end of the experiment). This permutation test derives a correlation statistic between non-independent similarity matrices after ranking edge values within rows and permuting the edges of one matrix, constraining permutations to retain each individual’s overall level of association. We constructed 104 matrices with these constrained edge permutations to derive independent P-values for each tank and used binomial tests for treatment and control tanks separately to determine whether the resulting P-values differed from 0·05 (where a significant value indicates a significant overall correlation among tanks for a type of food distribution, suggesting individuals’ positions and connections between networks on the 2 days were similar).

Individual spatial structure

To determine whether crabs established territories or showed a preference for a particular site in a tank, each tank was visually divided (in the video playback) into eight equally spaced square cells (all nine individuals could conceivably be recorded in the same cell). At the beginning of non-feeding recordings on Days 3, 5, 17 and 19, each individual’s location within the tank was noted (cell number 1–8). For each tank, we used a permutation test (104 permutations per tank) comparing individuals’ locations between Days 3 and 5 at the beginning of the experiment, as well as between Days 17 and 19 at the end. We then used separate binomial tests (= 0·05) for the control and treatment tanks to determine whether the derived P values were significant within each group, correcting for multiple comparisons on each tank with a sequential Bonferroni adjustment (α = 0·05, k = 2).

At the beginning of the observation period on Days 3, 5, 17 and 19, for tanks with clumped food, we recorded whether each crab was located at the same location (cell number 1–8) as the food site from the previous day. This could not be done for control tanks because food was scattered throughout those tanks. To compare the observed probability of finding an individual at the previous day’s food site to that which would be expected by chance (eight possible locations; probability of success = 1/8), separate binomial tests were used for each of the 10 tanks and a sequential Bonferroni correction if required (α = 0·05, k = 10).


General network structure

At the beginning of the experiment, crabs in clumped and dispersed food tanks did not differ in their average level of connectedness (degree centrality: = 0·60, = 0·56) or group size (clique size: = −0·12, = 0·91). After 19 days, however, crabs in clumped food tanks developed more associations (degree centrality: = 10·5, < 0·001) and formed larger groups (clique size: = 4·28, < 0·001) than did crabs in dispersed food tanks (Fig. 1).

Figure 1.

 Minimum, 25%, median, 75% and maximum values of mean degree centrality (a) and clique size (b) per tank for clumped and dispersed food tanks throughout the experiment.

In contrast, food distribution did not affect the average connection strength of individuals’ associates (eigenvector centrality: = 1·86, = 0·07) or the average ability of individuals to link other individuals (betweenness: = 0·22, = 0·83).

Individual characteristics and network metrics

Individuals’ relative exploratory behaviours were significantly correlated between Days 1 and 15 (Spearman’s rank correlation; clumped food tanks: r = 0·87, < 0·001; dispersed food tanks: r = 0·83, < 0·001). In dispersed food tanks on Day 3 (Fig. 2), an individual’s exploratory behaviour was positively correlated with how well it was connected to other individuals (degree centrality: < 0·001) and the relative strength of its associates’ connections (eigenvector centrality: < 0·001) and how effective it was at connecting other individuals (betweenness: < 0·001) and its group size (clique census: < 0·001). The same positive relationships with exploratory behaviour existed on Day 3 for crabs in clumped food tanks (degree centrality: < 0·001, eigenvector centrality: < 0·001, betweenness: = 0·01, clique size: < 0·001).

Figure 2.

 Correlations between individual initial exploratory behaviour and weighted centrality measures (degree, eigenvector and betweenness) or binary membership in network substructure (clique size) at the beginning (Day 3) and end (Day 19) of the experiment. See results in text for significance levels.

In dispersed food tanks on Day 19, the relationships between degree centrality, eigenvector centrality, betweenness and clique size all remained positively correlated (< 0·001 for each metric) with an individual’s exploratory behaviour (Fig. 2). In contrast, for clumped food tanks on Day 19, high exploratory behaviour remained positively correlated only with how effective an individual was at connecting other individuals (betweenness: < 0·001). In clumped food tanks on Day 19, both high and low exploratory individuals were found in similar group sizes (clique census: = 0·37) and low exploratory individuals were better connected than high exploratory ones (degree centrality: < 0·001, eigenvector centrality: = 0·05).

Individual body size showed a much weaker correlation with individual network measures. On Day 3, body size was not significantly correlated ( 0·10) with any network metrics for either food distribution type. On Day 19, body size remained insignificant ( 0·09) in the dispersed food tanks for all metrics, as well as for eigenvector and clique size in clumped food tanks ( 0·33). In clumped food tanks, body size was significantly negatively correlated with degree (= 0·01) and positively correlated with betweenness (= 0·02).

Individual characteristics and foraging

Initially (Day 2), crabs with high exploratory behaviour found food faster in both the dispersed (r = 0·60, < 0·001) and clumped food tanks (r = 0·48, < 0·001). This relationship persisted at the end of the experiment (Day 18) for crabs in dispersed food tanks (r = 0·64, < 0·001), but not in the clumped food tanks, where low and high exploratory individuals were equally good at quickly finding food (r = 0·09, = 0·40).

Similar to network metrics, body size was poorly correlated with food finding ability. On Days 2 and 18, for both the dispersed and clumped food tanks, the connection between body size and food finding rank was not significant (P ≥ 0·10).

Stability of individual positions within a network

For clumped food tanks, there was a strong correlation (binomial test on 10 Rr Mantel results: mean < 0·001) between individuals’ positions in the network on Days 7 and 19, suggesting the connections being formed on Day 7 persisted throughout the experiment. In contrast, there was no significant correlation between Days 7 and 19 for individuals in dispersed food tanks (binomial test on six Rr Mantel results: mean = 0·08).

Network graphs for two treatment tanks show how non-foraging population structure emerged (Fig. 3). A comparison of the upper (1% threshold) and lower (5% threshold) rows of Figs 3a,b reveals how individuals with high relative exploratory behaviour (low numbers in the figure) make many transient connections among population members, while individuals with low relative exploratory behaviour (high numbers in the figure) make fewer but longer-lasting connections with other low exploratory individuals. Once formed, network structure was robust to changes in the social environment, as removing a high (Fig. 3a) or low (Fig. 3b) exploratory individual on Day 20 did not affect connections among the remaining individuals. Network structure remained sensitive to food distribution however, as subgroups disbanded after food distribution was switched from clumped to dispersed on Day 24.

Figure 3.

 Two non-foraging social networks from clumped food tanks throughout the experiment. Individuals are ranked according to initial exploratory behaviour (from highest = 1 to lowest = 9). Days 3, 13 and 19 include all crabs. A single crab was removed on Day 20 [#1 – high exploratory removed from (a); #6 – low exploratory removed from (b)]. Food distribution switched to dispersed on Day 24. Top row includes connections among individuals after 1% threshold filter to visualize central role of high exploratory individuals in connecting individuals. Bottom row includes connections after 5% threshold filter of the same networks to visualize cohesive subgroups of low exploratory individuals.

Individual spatial structure

We found no evidence that individual crabs established territories or showed a spatial preference when not foraging. Individuals’ positions between Days 3 and 5 showed no significant correlation in either dispersed food tanks (binomial test on six permutation tests: mean = 0·32) or clumped food tanks (binomial test on 10 permutation tests: mean = 0·42). Neither were there significant position correlations between Days 17 and 19 (dispersed tanks: mean = 0·54, clumped tanks: mean = 0·51). Furthermore, crabs in clumped food tanks were not found at the previous days’ food site more often than by chance (10 binomial tests:  0·68).


Our experiment reveals that the interaction between ecological and behavioural factors has a pronounced effect on how social structure develops. Based on the interaction between food distribution and individual exploratory behaviour differences, we provide a robust mechanism to explain the emergence of social structure. Differences in exploratory behaviour were correlated with differences in individuals’ emergent social roles within a group when not foraging. High exploratory individuals made different types of connections with other members than did low exploratory individuals, resulting in differences between their respective effects on network structure. Regardless of food distribution, crabs with high exploratory behaviour were vital in linking otherwise poorly connected members of the population (high betweenness). These individuals in societies are often referred to as ‘gatekeepers’ or ‘powerbrokers’ because their strategic position within the network means that they regulate information transfer and mediate interactions between other individuals (Scott 2000). In comparison, low exploratory crabs became central to the network by forming cohesive subgroups (high-degree centrality, eigenvector centrality and clique census), but only in clumped food tanks. No subgroups formed in dispersed food tanks, where individuals remained isolated both when foraging and not foraging.

Furthermore, once low exploratory crabs in clumped food tanks formed subgroups, membership within each group was stable as long as food distribution remained clumped. There are several explanations for this stability. Low exploratory crabs could have been segregated spatially, by forming territories, copying other individuals or tracking resource location (Bronikowski & Altmann 1996; Giraldeau 1997; Johnson et al. 2002; Budden & Wright 2005). Our results, however, refute each of these explanations. Individuals’ current spatial locations were independent of both their location on the previous day and where food was previously located. Therefore, crabs did not form spatial territories, and they did not track resource placement. Alternatively, group stability could depend on the costs and benefits of being in a group (Alexander 1974; Ranta, Rita & Lindström 1993). Because resource distribution influences the effectiveness of competing foraging strategies (Milinski & Parker 1991), and social structure affects information flow among individuals (Naug 2008), foraging as a group can be advantageous when food is difficult to find (Krause & Ruxton 2002), such as when it is clumped (Milinski & Parker 1991). Cohesive subgroups such as those formed here provide group members with the potential, either actively or passively, to increase the level of influence they have on each other, share information and synchronize their activities (Freeman 1977; Conradt & Roper 2005; Borgatti & Everett 2006). Throughout the experiment, high exploratory individuals most quickly found dispersed food. The same was true for clumped food at the beginning of the experiment, when low exploratory individuals foraged separately. But when in groups, low exploratory individuals were able to find clumped food just as quickly as high exploratory individuals, which is consistent with an increase in communication and synchronization (either actively or passively) made possible by the formation of cohesive subgroups of interacting individuals. Likewise, Laidre (2010) found that, despite their lack of selection for recruitment, aggregating hermit crabs inadvertently provide social information to conspecifics regarding resource location.

By monitoring one another and synchronizing their activities, low exploratory crabs benefited from social information when a group member found food. Sharing social information has been linked theoretically to cultural as well as biological evolution via simple mechanisms such as imprinting, imitation, teaching and learning (Danchin et al. 2004). Furthermore, social interactions are mediated by interacting phenotypes, and the covariance among interacting phenotypes can affect the strength of social selective pressure on a particular trait (West-Eberhard 1983; Wolf 2003; Higgins, Jones & Wayne 2005). Non-random associations, such as the formation of groups, can have a strong effect on this covariance. Accordingly, factors that lead to group formation have the potential to play an important role in the evolution of competitive interactions (Wolf, Brodie & Moore 1999; Bowles, Choi & Hopfensitz 2003). Our results here show that the interaction between ecological and behavioural factors affects the emergence of groups. Therefore, this interaction itself can be an important evolutionary selective force.

The social networks that emerged among crabs in clumped food tanks indicate the importance of resource distribution in the formation of social structure. Indeed, similar patterns are found in the observational studies of crab species in which ecological context plays a role in social organization. For example, in grapsid crabs (Pachygrapsus spp.), individuals compete for access to spatially clustered rock crevices, and complex social structure has evolved in which individuals form coordinated groups based on phenotypic traits to defend home ranges (Abele, Campanella & Salmon 1986). In the closely related congener P. crassipes, however, individuals do not defend clustered resources and no such complex social organization has evolved. In conclusion, our study reveals that interactions between factors such as ecological and behavioural differences provide a rich avenue for future studies into the emergence of social structure, and we suggest that similar interactions will be found among all species that exhibit various levels of social behaviour.


We thank Peter Stafford and Gul Deniz Salali for facilitating laboratory experiments. Fred Adler, Richard James, Laurent Keller and two reviewers provided helpful comments on earlier versions of the manuscript. C.T. was funded by an Irish Research Council for Science, Engineering and Technology postdoctoral fellowship, as well as by a joint IRCSET/Marie Curie postdoctoral fellowship.