Network-level consequences of outgroup threats in banded mongooses: Grooming and aggression between the sexes

1. Animal groups are heterogeneous assemblages of individuals with differing fitness interests, which may lead to internal conflict over investment in group territo rial defence. Differences between individuals may lead to different behavioural responses to intergroup conflict, particularly between the sexes. These potential impacts have been little studied. 2. We used social network analysis to investigate the impact of simulated intergroup conflicts on social relationships in groups of wild banded mongooses Mungos mungo , in which intergroup fights are more costly for males than females. We pre dicted that social cohesion (specifically male-to-male and female-to-male

individuals (Jones, Aplin, Devost, & Morand-Ferron, 2017;Krause, Croft, & James, 2007;Krause, Lusseau, & James, 2009;Kurvers, Krause, Croft, Wilson, & Wolf, 2014;Rozins et al., 2018;Shizuka & Johnson, 2020). Social network analysis could be a useful tool to test whether social cohesion or individual social relationships change after intergroup conflicts, and whether this is affected by individual traits such as age or sex. Previous work has used social network analysis to investigate the impact of internal disturbance on social relationships (Formica, Wood, Cook, & Brodie, 2016;Maldonado-Chaparro, Alarcón-Nieto, Klarevas-Irby, & Farine, 2018;Shizuka & Johnson, 2020;Wilson et al., 2015); however, the impact of an external intergroup conflict event (which does not necessarily disturb networks by removing individuals or shuffling groups) on internal group dynamics has not been studied. However, social network traits have been correlated to participation in group defence in female white-faced capuchins Cebus capucinus (Crofoot, Rubenstein, Maiya, & Berger-Wolf, 2011), but the impact of individual intergroup conflict events on animal social networks has not yet been investigated.
In this study, we quantify how individuals differ in their response to intergroup conflicts, and how these individual differences scale up to influence group behaviour. Specifically, we test how banded mongoose Mungos mungo individuals differ in their social response to simulated intergroup encounters. As banded mongoose groups are heterogeneous, being made up of multiple males and females of different ages, we anticipate that there are differences in how males and females respond to intergroup conflict that lead to changes to grooming and aggressive social relationships, which may not be clear when measuring these behaviours at the group level.
Banded mongooses live in stable multi-male, multi-female groups of between 10 and 30 individuals. Dispersal is relatively rare for both sexes: around 85% of males and females are born and die in the same group (Cant, Nichols, Thompson, & Vitikainen, 2016). Males sometimes disperse in groups voluntarily, but most dispersal occurs as a result of mass evictions, where groups of females (or, in mixed-sex evictions, groups of males and females) are attacked and forced out of their group by members of their own sex . Groups of mongooses are highly territorial, defending their territories from other groups during frequent, and sometimes lethal, intergroup conflicts (Nichols, Cant, & Sanderson, 2015;. Mongooses respond to sighting a rival group by standing alert and calling to other members of their group, they often congregate and stand looking for the rival mongooses (Cant et al., 2016). Small groups often flee from larger groups, with larger groups sometimes giving chase. If groups are more evenly matched in size, then the individuals may bunch together and approach in 'battle lines' (Cant et al., 2002(Cant et al., , 2016. Fighting is often highly aggressive involving biting and scratching, and sometimes individuals are held down and attacked by multiple rival mongooses. On the basis of previous studies, we predict that intergroup conflict will be associated with increased social cohesion after conflict, represented by within-group affiliative behaviour (Bruintjes et al., 2015;Radford, 2008aRadford, , 2008bRadford, , 2011Radford & Du Plessis, 2004;Schaffner & French, 1997). However, the concept of social cohesion is rarely defined explicitly, and could in theory be manifested as a reduction in within-group conflict, or reduced aggression (Reeve & Hölldobler, 2007;Thompson et al., 2020). We note, however, that previous studies have recorded no change in within-group aggression (Bruintjes et al., 2015;Morris-Drake et al., 2019;Nunn & Deaner, 2004), or an increased rate of within-group aggression following intergroup conflict (Bruintjes et al., 2015;Cooper et al., 2004;Polizzi di Sorrentino et al., 2012;Schaffner & French, 1997), rather than any decrease. We also predict that there will be differences in responses to intergroup conflicts between males and females because males experience higher mortality costs from intergroup encounters (F.J. Thompson, unpubl. data), whereas females can benefit from extra-group paternity (Nichols et al., 2015). If rival groups present no threat to females, we would not expect females to respond behaviourally in a manner that increases group social cohesion.
We make the following predictions: 1. Social cohesion will increase in response to intergroup conflict.
Specifically, we predict that grooming eigenvector centrality (a proxy for cohesion) will increase after simulated encounters with rival mongooses.
2. Male-to-male grooming and female-to-male grooming will increase after simulated encounters, as a reward for male participation or as a form of group cohesion.
3. Aggression will on average decrease after simulated encounters (following theory Reeve & Hölldobler, 2007). We predict that this decline will be particularly marked for aggression between males and females because males may seek to discourage dispersal by females after exposure to stimuli from rival males.

| Study site
Data were collected for this study from wild banded mongooses on the Mweya Peninsula in Queen Elizabeth National Park, Uganda was performed on banded mongooses that are part of a long-term study population, detailed descriptions of which can be found in Cant (2000), Cant et al. (2016) and Rood (1975).
All mongooses in the study population are individually marked using unique hair-shave patterns, and are habituated to close observation from 2 to 4 m. One to two mongooses in each group are fitted with a radio collar weighing 26-30 g (Sirtrack Ltd.) with a 20-cm whip antenna (Biotrack Ltd.) to allow the groups to be located. Five focal groups (which are habituated to being followed) were used in this study.

| Experimental timeline
We adopted a controlled experimental approach in which we compared social networks before and after a simulated intergroup intrusion because it is difficult to predict when and where intergroup encounters will occur for any given study group. Trials took place over 5 days (see schematic in Figure 1). On the first and second days, we recorded baseline social interaction data that were used to build pre-conflict social networks. On the third day, we carried out simulated intrusions or control presentations. On the fourth and fifth days, we recorded social interaction data again, to build post-conflict social networks, representing social responses to intergroup conflict. Hormonal changes are likely to lead to short-term behavioural changes via allo-grooming, preening and other affiliative behaviours (Crockford, Deschner, & Wittig, 2018;Dunbar, 2010;Madden & Clutton-Brock, 2011) but they may also impact behaviour into the longer term. Hormone levels decline, but can persist into the days and weeks after an event, as seen in banded mongoose glucocorticoid levels between breeding events (Sanderson et al., 2014), additionally changes could persist into the longer term through feedback loops of social behaviour (or other adaptive feedback loops; Sih et al., 2015), perhaps forming or breaking reciprocal relationships that last into the longer term after hormonal (or other physiological) impacts of intergroup encounters have dissipated. However, this is an area of limited research, and the mechanisms of potential behavioural change into the longer term are not yet known.

| Presentations
We carried out simulated intrusion presentations and control presentations on each of five focal groups. These presentations were designed to simulate an intergroup conflict with escalating cues, following a natural progression from sensing rivals indirectly, to direct contact. In natural encounters, mongooses will typically encounter indirect stimuli first, such as faeces or scent marks, which alert them to the potential presence of rival groups in the area, prior to any direct encounters. Since we did not know which of the many stimuli, or combination of stimuli, are most salient to simulate a natural encounter, we presented three major classes of sensory stimuli-olfactory, auditory and visual-in a single day, mirroring the sequence of exposures that characterize natural encounters. In total, we carried out 22 control presentations and 22 simulated intrusion presentations. Presentations to each focal group were separated by at least 2 weeks to prevent habituation of the mongooses to the stimuli being presented.

Simulated intrusion presentations
Faeces, urine and scent marks on plastic from a neighbouring rival group (considered to be the largest threat to the focal group) were presented to the focal group on the morning of the presentation day (07:43-10:27 hr). Faeces, urine and scent marks from the rival group were collected early in the morning, as the group emerged from the den or from the first group marking site of the day. Plastic sheets were presented to encourage urination and scent marking.
These samples were collected from multiple individuals in the group, both males and females from different age classes, and a standardized volume of faeces was used in each trial (100 × 137 mm ziplock bag). Samples were collected within 30 min, transferred as quickly as possible to the presentation site, and presented within 2 hr of collection, but generally much faster and as such were not stored on ice. The presentation site was placed in the foraging path of the focal group, to ensure that the mongooses encountered the stimulus. The samples were arranged in a semi-circle on open ground, with faeces placed around the sheets of plastic (spaced over 70-100 cm) as mongooses often use open patches for territorial marking (adapted from (Müller & Manser, 2007).
After 3 min of exploring the scent marks, or slightly before if the animals began to move away from the area, a playback of war cries (that had been recorded earlier from the same rival group that the scent marks were collected from) occurred. Playbacks were conducted using a portable USB speaker (iHome IHM60) hidden in vegetation. These war cries were recorded in advance using an H1 Zoom recorder attached to a Sennheiser directional microphone, and were emitted in response to rival mongooses presented in traps.
F I G U R E 1 A schematic diagram of the timeline of each trial, showing the process in both control (top row) and intrusion (bottom row) trials These recordings were made when individuals from the rival group were vocalizing at rival mongooses that were presented in traps (see Supporting Information for an example). The recordings were taken from 2 to 3 m away from the mongooses, and recorded calls from multiple individuals as the group were calling together. The recordings were cut into 30 s sections in which vocalizing was occurring, and the amplitude of each clip was standardized using the normalize function in Audacity 2.1.2 to −1 dB (http://audac ityte am.org). Each 30 s playback clip was used only once to prevent habituation of the mongooses to particular recordings.
On the afternoon of the same day (16:35-18:18 hr), four adult male individuals from the rival group were trapped and presented to the focal group. The traps were washed with soap and water to reduce scents from any previous trapping events before the males were captured. Trapped animals were transferred from the rival group to the focal group covered with a black cloth to minimize stress. The traps were placed in the foraging path of the focal group to ensure they encountered the traps, and the cloth was removed. After 5 min, the rival males were removed (and the traps re-covered with the cloth) then returned to their own group, to minimize stress levels.
Mongooses appear to react to these simulated intergroup encounters in a similar manner to their reaction to natural encounters, despite their artificial nature. Banded mongooses assess their rivals by approaching, either in battle lines in natural encounters, or by approaching caged animals in simulated encounters. Once they have engaged, they then split their time between group scent marking slightly away from the rivals and engaging with the rival group (either direct fighting, or snapping at and moving around cages)-this happens in both simulated and natural encounters.

Control presentations
The same procedures were carried out for control presentations.
However, faeces and marks were collected from the focal group, and re-presented to them. War cries were replaced with close calls (a non-threatening communication call between group members) from the focal group as the recordings used for the playback. The close calls were recorded from the focal group during normal foraging behaviour when there were no threats from rival groups or other sources. These recordings were cut and standardized in the same way as the war cry recordings. Four adult males were presented in traps, as before, but these were males from the focal group, which were trapped and removed for half an hour (to a safe, shaded location) before they were presented to the rest of the focal group.

| Social data collection
Social interaction data were collected during observations of the focal group for 1 hr in the morning (between 07:00 and 12:00 hr) and for 1 hr in the afternoon (between 16:00 and 19:30 hr) on each day when mongooses had moved away from the den and begun foraging. Banded mongooses spent time foraging, moving and resting during these periods, but changed activity regularly, and there was no systematic difference between observations. Throughout the observation, every affiliative and aggressive interaction between individuals was recorded. This was possible in this system as banded mongooses forage at ground level in cohesive groups in relatively open habitat so all individuals could be seen by either one, both or all of the observers at any one time. A minimum of two observers were present during each observation.
All affiliative interactions, that is, grooming and 'nubbing' (mutual genital sniffing) were recorded by noting the identity of the individuals involved and the direction of the interaction (see Table 1 for detailed descriptions). As most affiliative interactions recorded were grooming interactions, we refer to all affiliative interactions and networks made from these interactions as grooming interactions or grooming networks. All aggressive interactions, including food competition and dominance aggression were also recorded in the same way (see Table 1 for detailed descriptions). Interactions that were observed but where individual identity could not be confirmed were not analysed as part of the networks. Social interaction data from 2 days before the presentation day (total 4 hr of observations) were pooled to create a pre-conflict grooming and a pre-conflict aggressive social network. Social interaction data

Behaviour Description
Affiliative interaction (grooming, or 'nubbing') Grooming-one mongoose grooms another mongoose using their mouth, manipulating the fur with the teeth, the head moves in a distinctive backward and forward motion. One bout of grooming was defined as active grooming between the same pair of individuals with short breaks of no longer than 30 s of rest. If 30 s elapsed and the same pair began grooming again this was considered to be a second interaction. Grooming between multiple individuals switching from one partner to the other was recorded as one interaction per actor-recipient pair. Returning to a previous partner was not recorded as a separate interaction, unless 30 s of rest (no grooming of any partner) occurred Nubbing-wo mongooses perform 'nubbing' behaviour-a mutual genital sniff with raised tails which may also include marking each other and vocalizing Aggressive interaction One mongoose is aggressive to another mongoose. This can include lunging, biting, growling or snarling vocalizations, or physical displacement of another individual. Aggressive interactions happen over food resources, during mateguarding and as part of dominance interactions. One aggressive interaction was defined as aggression between the same pair of individuals with short breaks of no longer than 30 s between aggressive behaviours (e.g. lunging, vocalizing) from 2 days after the presentation day (total 4 hr of observations) were pooled to create a post-conflict grooming and a post-conflict aggressive social network.

| Social network creation and analysis
The pre-and post-conflict social networks for both grooming and aggression were created from the edge lists (lists of observed social interactions, with the identity of each actor and recipient) collected during observations, using the igraph package in r (Csardi & Nepusz, 2006). The networks were both directional (included the direction of the interaction) and weighted (i.e. they included the strength of the interaction between individuals-in this case, the total number of interactions observed between that pair of individuals during the observation session). In total, 10,641 grooming interactions and 7,435 aggressive interactions were observed over a total of 348.8 hr of observation across 44 trials and five groups.
On average, 30.23 ± 1.92 (range = 0-142) grooming interactions, and 21.12 ± 1.44 (range 0-108) aggressive interactions were observed per hour, and each individual was involved in, on average, 15.6 grooming and 10 aggressive interactions per pre-or post-conflict sampling period. To test the robustness of these networks, we ran Mantel tests in three pilot control trials pre-and post-conflict grooming and aggression matrices, which were significantly correlated (Mantel tests: all p < 0.05), suggesting that the observed social interactions were stable over the study period. We also ran bootstrapping methodology (adapted from Lusseau, Whitehead, & Gero, 2008). Mean deviation of each mongooses eigenvector centrality from 1,000 bootstrapped samples of the raw data was 0.062 ± 0.20 in the grooming networks and 0.055 ± 0.21 in the conflict networks. Modal deviation was 0 in both grooming and conflict networks. These methods are based on those available (Lusseau et al., 2008;Whitehead, 2008), but so far there are no specific methods for addressing robustness of networks when there is more than one network, and for interaction rather than association data. These should be developed more in the future.
The social networks included individual attributes for each node (in this case, an individual mongoose), including individual identity, group identity, age and sex. The networks also included an edge (the link between two nodes-here based on social interactions) attribute, which denoted the identity of each edge in terms of the sex of the two individuals it connected, for example, male-male for an interaction from a male towards another male, or female-male for an interaction from a female directed to a male. All network metrics analysed were chosen a priori based on the predictions and hypotheses outlined in the introduction, and again below.

| Prediction 1: Social cohesion will increase following simulated intergroup encounters
Linear mixed models were used to investigate the change in grooming eigenvector centrality of individuals following simulated intergroup encounter, or control, presentations. The response variable for the model was the change in the grooming eigenvector centrality, that is, the difference between the pre-conflict and post-conflict grooming eigenvector centrality of each individual present in the networks. Eigenvector centrality is a measure of a node's connectedness, including indirect connections, that is, the nodes that the focal node is connected to. High eigenvector centrality indicates a node which is connected to other nodes which are also highly connected in the network (Wasserman & Faust, 1994 intrusions (Bruintjes et al., 2015;Radford, 2008aRadford, , 2008bRadford, , 2011Radford & Du Plessis, 2004;Schaffner & French, 1997). In contrast, a reduction of an individual's grooming eigenvector centrality suggests that the network is less connected, and less grooming is being directed at or given by the focal individual and its direct connections-this might indicate less grooming in general, or less grooming among certain parts of the network, for example older females. These changes will give us an insight into how cohesive and well connected the network is following intrusions.

| Prediction 2: Grooming directed towards males will increase
Linear mixed models were used to investigate the change in groom-

| Prediction 3: Between-sex aggression strength will decrease following encounters
Linear mixed models were used to investigate the change in aggres-

| Null models and network permutations
Variables calculated from social networks are not independent, so the observed model coefficients were compared to the coefficients from models of randomly shuffled network permutations.
As sampling was even within each time period we built null models using node label permutations. Node label permutations in this case meant that the node labels of each observed network (e.g. pre-experimental observations of group 1 in trial 1) were shuffled, to separate individual identity from the age or sex of an individual and test for the effect these factors had on their social relationships.
We then applied our models to each of these permuted networks to generate a distribution of potential coefficient values given the nonindependence of our data [following the methods of Croft, James, and Krause (2008), Croft, Madden, Franks, andJames (2011), Farine andWhitehead (2015)]. Model coefficients stabilized at 5,000 permutations, tested using the method from Bejder, Fletcher, and Brager (1998). We therefore ran 5,000 permutations to generate a distribution of random network coefficients. Observed model co- grooming and aggression strength: three post-hoc tests, α = 0.008).
All analyses were run in r 3.6.1 (R Development Core Team, 2019), and all models were run using the lmer function in the lme4 package (Bates & Maechler, 2009).

Additional permutations which randomized the treatment label
were also performed. These permutations meant that the treatment type label (control or intrusion) was randomized within group, within paired experiment, within period (before or after), and within individual. Change in network metric was then recalculated after randomization so that treatment type was separated from the network metric value. Models were then re-run on subsets of the data for each sex, or edge sex, but included only Treatment type and, as a random factor, individual identity. Age could not be included in these models as individuals changed their age (measured as a continuous variable) between paired control and intrusion experiments, and labels were shuffled between these paired experiments, leading to mismatches in age when recalculating change in network metric.

| Prediction 1: Social cohesion will increase following encounters
There was a significant interaction between treatment type and sex (estimate = −0.099, p = 0.025, Table 2 control data, female-male estimate = 0.075, p = 0.02; intrusion data, female-male = −0.032, p = 0.81, Table A1]. There was no relationship between change in grooming eigenvector centrality and either age, or an interaction between treatment type and age (Table 2). This suggests that there is no evidence for the first prediction that social cohesion (represented by eigenvector centrality) will increase following encounters.

| Prediction 2: Grooming directed towards males will increase following encounters
There was a significant interaction between treatment type and  Table A4].
The change in male-to-male and male-to-female grooming strength was also confirmed to be decreasing in intrusion trials compared to TA B L E 2 Model parameter estimates from the grooming eigenvector centrality model, and p values from network permutations (p values are calculated as a proportion of randomized model coefficients that are larger/smaller than the observed model coefficient, α = 0.025 as these were two-tailed tests). Model was fitted with individual identity nested in group identity as a random intercept (LMM, N = 857 observations (274 = female, 583 = male) across 100 individuals in five groups and 44 trials). The reference category for treatment type was control and for sex was female, the intercept therefore represents the estimate for females in control trials. Significant terms are given in bold   Table   A3]. In contrast, in intrusion trials, both female-to-male grooming decreased compared to female-to-female grooming [post-hoc tests (α = 0.004): FF-FM estimate = −1.279, p = 0.00, Table A3]. This suggests that female-to-female grooming relationships are not affected by intergroup conflict, but other grooming relationships weaken after intrusion, but not control, trials.
Older individuals reduced their grooming more (negative change in grooming strength) than younger individuals after intrusion trials (

| Prediction 3: Between-sex aggression strength will decrease
Male-to-female aggression decreased significantly more in intrusion trials compared to control trials [estimate = −0.442, p = 0.00, Table 4; post-hoc test (α = 0.004): MF data control-intrusion estimate = −0.468, p = 0.00, Table A7; Figure 5]. This decrease in male-to-female aggression TA B L E 3 Model parameter estimates from the grooming strength model, and p values from network permutations (p values are calculated as a proportion of randomized model coefficients that are larger/smaller than the observed model coefficient, α = 0.025 as these were two-tailed tests). Model was fitted with individual identity nested in group identity as a random intercept (LMM, N = 1,714 observations (FF = 274, MM = 583, MF = 583, FM = 274) across 100 individuals from five groups in 44 trials). The reference category for treatment type was control and for edge sex was female-to-female, the intercept therefore represents the estimate for female-to-female grooming strength in control trials. Significant terms are given in bold

F I G U R E 3
The change in mongoose individual grooming strength from before presentations to after presentations for males and females in intrusion (orange triangles) and control (blue circles) trials. This figure shows that female-female grooming is not affected by intergroup conflict, but male-male, male-female and female-male grooming decreases after exposure to simulated conflicts. Points shown are means from the raw data and error bars are standard errors on these means. N = 1,714 observations (FF = 274, MM = 583, MF = 583, FM = 274) across 100 individuals from five groups in 44 trials. **p < 0.01, ***p < 0.001

F I G U R E 4
The change in mongoose individual grooming strength from before presentations to after presentations in intrusion (orange triangles) and control (blue circles) trials. This figure shows that in intrusion trials older individuals reduce their grooming more than younger individuals. Points shown are raw data binned into categories (statistical analysis used a continuous measure) and lines are predictions from the raw data. N = 1,714 observations across 100 individuals from five groups in 44 trials in intrusion trials was significantly different from zero, suggesting a decrease in real terms [post-hoc test (α = 0.008): MF data, intrusion estimate = −0.308, p = 0.00, Table A8]. The change in male-to-female aggression was also confirmed to be decreasing in intrusion trials compared to control trials by treatment permutations (estimate = −0.411, p = 0.01). There was no significant interaction between treatment type and male-to-male aggression (estimate = 0.289, p = 0.03) or female-tomale aggression (estimate = −0.163, p = 0.82), suggesting that changes in these relationships do not differ between trial types (Table 3; Figure 5).
The effect of both control and intrusion presentations on maleto-female aggression differed depending on the age of the actor (estimate = 0.073, p = 0.002; Table 3). Specifically, male-to-female TA B L E 4 Model parameter estimates from the aggression strength model, and p values from network permutations (p values are calculated as a proportion of randomized model coefficients that are larger/smaller than the observed model coefficient, α = 0.025 as these were two-tailed tests). Model was fitted with individual identity nested in group identity as a random intercept   Table A9]. There was therefore mixed evidence for the third prediction, that between-sex aggression strength would decrease, as male-to-female aggression decreased, but female-to-male aggression was not affected by simulated intergroup conflict ( Figure 6).

| D ISCUSS I ON
Banded mongooses adjusted their grooming and aggressive interactions between group members after simulated intergroup encounters. Following an intergroup encounter, and contrary to our predictions, we found that grooming decreased in male-to-male, male-to-female and female-to-male interactions. We also found that male-to-female aggression was reduced (following our predictions) but female-to-male aggression did not change. Additionally, we found that older individuals reduced their grooming more after intrusion trials than younger individuals, and male-to-female aggression increased in older males compared to younger ones. These results highlight both sex and age differences in the responses of banded mongooses to intergroup encounters.

| Prediction 1: Social cohesion will increase following encounters
Contrary to our simple prediction, we did not find an overall increase in grooming eigenvector centrality in intrusion trials; however, we did find that female eigenvector centrality increased compared to control trials, but that male eigenvector centrality did not change. However, the increase in female eigenvector centrality in intrusion trials was not significantly different from zero, suggesting that simulated intrusions did not result in a significant change in social cohesion among females. This suggests that, despite differences between control and intrusion trials in female eigenvector centrality, there is no evidence for increased social cohesion after exposure to simulated intergroup conflict. Social cohesion may be masked by this measure if there is increased spatial cohesion of individuals, but these individuals are more vigilant or otherwise not interacting with each other. This could be investigated in the future if individuals can be followed and spatial structure of the group and interactions measured simultaneously.

| Prediction 2: Grooming directed towards males will increase following encounters
In contrast to our prediction, we found that male-to-male, maleto-female and female-to-male grooming declined after intrusion presentations. Unlike in primates, there seems to be no 'reward' given to males from females in the form of grooming for their participation (Arseneau-Robar et al., 2016;Cooper et al., 2004). Male investment in grooming relationships may not be as important after conflicts, but equally males may be investing more time than females in other behaviours, like searching or patrolling, or alarm calling and scent marking, and not engaging in grooming. Previous studies in dwarf mongooses have found that vigilance behaviour increases after exposure to a simulated intergroup conflict (Morris-Drake et al., 2019); however, we found no evidence of increased alarm calling or scent marking at a group level (Preston, 2019). It is possible that males contribute more to alarm calling or scent marking that is not clear when measuring group-level behaviour but unfortunately, these behaviours were only recorded at a group and not individual level in this study. It seems plausible though, that as males face a greater risk from rival groups than females, they may direct more time and energy into combating these external threats through such behaviours, rather than investing in internal relationships. Females may then reciprocate grooming less, as males are not grooming them, leading to a by-product reduction in female-male grooming.
Individual differences in behaviour may also be attributed to individual contributions to intergroup conflicts, for example in this case males contributing more than females. In this study, it was not possible to separate out contributions of individuals, even by sex and age class, to intergroup conflicts, as almost all individuals interacted with presented stimuli and detailed individual behaviour could not be followed accurately during presentations. Future work aims to address this using technology to disentangle contributions and link these to behavioural change at an individual level. A meta-analysis of grooming relationships and intergroup conflict in primates found that increased female grooming was linked to high levels of intergroup conflict, but male grooming was not (Majolo, de Bortoli Vizioli, & Lehmann, 2016). This meta-analysis suggests that this sex difference in affiliative behaviour linked to intergroup conflict might be widespread, although here we find the opposite result, with males grooming less. Grooming after a conflict may present itself as a trade-off, in which males reduce investment in internal relationships and increase defensive behaviours. An example of a similar trade-off has been observed in meerkats Suricata suricatta (Mares et al., 2012). Males chased intruders more than females as they suffer a greater threat from the intruders, and reduced pup care when intruders were present (Mares et al., 2012). Our results also provide evidence that males are affected more by intergroup encounters than females, as all grooming strength changes involving males were significantly negative, and female-to-female grooming was not af-

| Prediction 3: Between-sex aggression strength will decrease
Male banded mongooses reduced aggressive interactions towards the opposite sex after simulated intergroup encounters. This is consistent with the hypothesis that groups respond to an external conspecific threat by suppressing internal conflict to maintain social cohesion, as we predicted. In contrast, previous studies that measured postconflict aggression found either an increase (Cooper et al., 2004;Polizzi di Sorrentino et al., 2012;Schaffner & French, 1997), or no change (Bruintjes et al., 2015;Morris-Drake et al., 2019;Nunn & Deaner, 2004), rather than any suppression of conflict. Although one study exposing groups of cichlid fish to a neighbour group over an extended period of time (rather than a short-lived intrusions into the territory) did find a reduction in conflict between mating pairs (Hellmann & Hamilton, 2019 (Arseneau-Robar et al., 2016, which contrasts with the results of this study. Suppressing conflict between the sexes in banded mongooses may help promote social cohesion when there is risk of another encounter. Despite evidence that male-to-female aggression is suppressed after an intergroup encounter, there seems to be no change in aggression within each sex, or from females to males. No change in within-sex aggression might suggest that within-sex conflicts are not strongly affected by intergroup conflict, and are more heavily influenced by other factors, such as dominance hierarchies (e.g. Clutton-Brock et al., 2006) or reproductive conflict (Cant, Hodge, Bell, Gilchrist, & Nichols, 2010).
Alternatively, suppression of aggression to boost social cohesion may be balanced by intensified aggression within sex classes leading to no overall change in mean group aggression levels. Indeed, male-to-male aggression appears to increase slightly, and although this is not statistically significant, this is worth investigating in the future in case this trend represents re-directed aggression from out-group members to within-group males. Same-sex aggression may serve to encourage participation in future conflict, or to relieve tension from losing a conflict (Radford, Majolo, & Aureli, 2016).

| CON CLUS IONS
In conclusion, focusing on individual relationship changes using social network analysis can reveal important changes in behaviour after intergroup encounters. We found differences between males and females in their response to intergroup encounters, some of which were also affected by age.
In banded mongooses, males are more socially affected by intergroup conflicts than females, changing both their grooming and aggressive patterns.
This study suggests that suppression of between-sex competition, particularly from males to females, occurs post-conflict, and may be important for overcoming inter-sex conflict over entering into intergroup conflicts.
Measuring group-level behaviours can be important in recognizing general behavioural change after disturbance, but these measurements ignore the differences between individuals in groups. These individual differences may be more important when assessing changes in relationships, particularly, as in the case of intergroup conflict, when individuals have different costs and benefits associated with interacting with other groups. This study highlights the importance of studying both group-level behaviours and individual relationships to fully understand responses to intergroup encounters. Social network analysis can reveal changes in within-group social dynamics that are susceptible to being obscured in studies of group-level behaviour.

ACK N OWLED G EM ENTS
We are grateful to Uganda Wildlife Authority and Uganda National Council for Science and Technology for permission to carry out our re-

AUTH O R S ' CO NTR I B UTI O N S
E.F.R.P. carried out field work and data analysis, designed the study and drafted the manuscript; F.J.T. carried out pilot fieldwork, and critically revised the manuscript; S.E. assisted with data analysis and critically revised the manuscript; S.K. collected field data and critically revised the manuscript; D.P.C. co-supervised the work, assisted with data analysis and critically revised the manuscript; M.A.C. supervised the work, managed the long-term project and critically revised the manuscript. All authors gave final approval for publication and agree to be held accountable for the work performed therein.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are openly available in Figshare at https://doi.org/10.6084/m9.figsh are.12744992 (Preston et al., 2020).