Neighbor effects in marmosets: social contagion of agonism and affiliation in captive Callithrix jacchus

Researchers have demonstrated the neighbor effect for affiliative and agonistic neighbor vocalizations in captive chimpanzees. We extend the investigation of the neighbor effect to New World monkeys, Callithrix jacchus. We collected data on vocalizations and behaviors of 31 focal individuals and concurrent neighbor vocalization within three behavioral categories: intragroup and intergroup aggression and intragroup affiliation. We investigated whether there was an influence of neighbor vocalizations on focal behavior within the same behavioral category. For data analysis we used approximate randomization of paired‐sample t‐tests. We found that marmosets performed intergroup aggressive behavior (bristle, anogenital present for neighbor loud shrill only) for significantly longer, and emitted significantly more intergroup agonistic vocalizations (twitter, loud shrill), at a high frequency of intergroup agonistic neighbor vocalizations (twitter, loud shrill) than at low. The marmosets were also significantly more likely to engage in bristle behavior immediately after hearing a neighbor intergroup aggressive call (twitter, loud shrill) than directly beforehand. High neighbor intragroup agonistic calls (chatter) were associated with significantly longer spent in related behavior (composite of: attack, chase, steal food). Affiliative behaviors (share food, grooming invite) were engaged in by marmosets for significantly longer at higher frequencies of affiliative neighbor chirp calls than at low. Marmosets were also significantly more likely to perform food sharing and active affiliative contact immediately after rather than before hearing a neighbor chirp call. Our findings suggest that neighbor vocalizations influence marmoset behavior through social contagion and indicate that the neighbor effect for affiliation and aggression generalizes to the marmoset. Am. J. Primatol. 72:549–558, 2010. © 2010 Wiley‐Liss, Inc.


INTRODUCTION
Researchers investigating social influences in nonhuman primates have recently tended to focus on the most cognitively complex social processes such as imitation. However, despite being relatively simple cognitively, processes such as social contagion are likely to be functional in their own right, and may also support and influence cognitively complex social processes. For example, social facilitation may enhance the advantages of group living by increasing behavioral synchrony [Clayton, 1978;Coussi-Korbel & Fragaszy, 1995] and may indirectly influence social learning [Hoppitt & Laland, 2008]. It has been suggested that social contagion may be a precursor to the human capacity: empathy [e.g. Anderson & Gallup, 1999]. Consequently, it is an important enterprise to establish how such simple processes operate in species of nonhuman primate.
Several studies have investigated whether social contagion could be demonstrated in chimpanzees through the influence of neighboring group vocalizations on individual social behavior, termed the ''neighbor effect'' [Baker & Aureli, 1996;Videan et al., 2005]. The term ''neighbor effect'' does not refer simply to the mere presence of conspecifics [Videan et al., 2005] as do some definitions of social facilitation [e.g. Zajonc, 1965]. Neither does the term refer to the contagion of identical behaviors only, a necessary condition in many definitions of social facilitation [e.g. Clayton, 1978]. In relation to the neighbor effect, Videan et al. [2005] use the relatively broad, although somewhat anthropomorphic, definition of social contagion given by Levy and Nail [1993]. We have defined social contagion slightly differently, with nonhuman primates explicitly in mind. Our definition for the purposes of this paper is, ''the spread of affect or behavior from one individual (A) to another individual or other individuals (B)'' [adapted from Levy & Nail, 1993]. We here use the term ''neighbor effect'' to refer specifically to social contagion resulting from the influence of the vocalizations emitted by conspecifics on the behavior of nearby individuals. Baker and Aureli [1996] investigated the neighbor effect resulting from aggressive neighbor calls. They found that higher levels of chimpanzee agonistic vocalizations and noises in neighboring groups were associated with an increase in intragroup agonistic behavior and vocalizations in focal groups. Videan et al. [2005] replicated these findings and also extended the evidence supporting the neighbor effect to affiliative behavior. They found that the time spent by focal individuals in allogrooming behavior and in giving grooming vocalizations increased significantly at higher levels of neighbor grooming vocalizations [Videan et al., 2005].
This study aims to explore the generality of the neighbor effect. Researchers directly investigating the neighbor effect in nonhuman primates have, to date, confined their focus to apes, specifically chimpanzees. We here investigate the neighbor effect, for aggression and affiliation, in a species of New World monkey: common marmosets (Callithrix jacchus).
The common marmoset is an ideal species for the investigation of the neighbor effect. Marmosets live in large social groups in the wild and, being cooperative rearers, they are highly socially oriented [e.g. Stevenson & Rylands, 1988]. Vocalizations play a very important role in both intergroup and intragroup communication in wild marmosets especially as the dense vegetation in which they live makes visual communication difficult [e.g. Epple, 1968]. Marmosets are known to retain a large proportion of their natural vocal repertoire in captivity and to show a high rate of calling [Pook, 1976].
Anecdotal reports indicate that the behavior of neighboring groups influences the behavior of captive marmosets. For example, Pook [1976] described twitter vocalizations, an intergroup agonistic call, as contagious, with calls in one social group being followed by calls in nearby groups.
However, there remains a lack of comprehensive, quantitative information on the effects of spontaneous neighboring conspecific calls on marmosets, i.e. the neighbor effect. To date, investigations into the contingent effect of neighbor calls on nearby common marmosets have involved some degree of experimental manipulation involving isolation of individuals and/or playback of calls [e.g. published: Chen et al., 2009;Yamaguchi et al., 2009;unpublished theses: Jones, 1993;Pook, 1976]. Thus, these studies have taken place under relatively artificial conditions and social behaviors were not recorded. Several more naturalistic observational studies have been carried out, but these have not been systematic investigations of the type we now report. For example Pook [1976] noted the responses of marmosets to surrounding calls but recorded only those calls that elicited some sort of response; he did not record the total frequency of each call type investigated. Investigating the neighbor effect in marmosets has relevance to the study of the behavior, vocalizations and captive welfare of marmosets, in addition to the potential implications for social cognition in general.
We predicted that neighbor vocalization of a particular behavioral category emitted by marmosets would influence nearby conspecifics to initiate or increase performance of behaviors within the same behavioral class. We examined social contagion within three behavioral categories: intergroup aggression; intragroup aggression; and affiliation. We investigate the effect of neighbor vocalizations on the behavior of focal individuals, within the same behavioral category, in two ways [following Baker & Aureli, 1996]. First, we assessed the overall effect of neighbor vocalizations by comparing the behavior of focal individuals between observation periods with high and low levels of neighbor calls. Second, we investigated the immediate effect by comparing the proportion of behaviors carried out by individuals just before neighbor vocalizations with the proportion performed directly after. This, latter, sequential analysis allows for firmer conclusions regarding causality.

Subjects
The study animals were housed within four colony rooms at the Medical Research Council (MRC) Human Reproductive Sciences Unit, Edinburgh, Scotland. In total, the study population numbered 121 individuals: 61 breeding adults; 15 nonbreeding adults; 30 juveniles and 15 infants. The focal individuals in this study were 32 individuals, initially; eight in each of the four colony rooms (four breeding males and four breeding females, each housed in either a breeding pair or a family group). One male focal individual was later removed from the study because excessive aggression from his partner necessitated separation, leaving 31 focals. Over the study period there were also some changes to the nonfocal (''neighbor'') population in the study rooms: one adult and one infant died, two infants were born, and three individuals (including the separated pair) were removed from the study rooms.
Each colony room (4.5 m wide by 6.5 m long) contained two rows of four cages along the two longest facing walls (each cage measured: 1.1 m deep by 1.5 m wide by 2.3 m high). At the outset of the study, the mean total number of individuals was 32 per room divided among a mean number of 9 social groups, with all 8 cages per room occupied except 2 cages in one room. Individuals were housed in pairs, family groups or same sex groups. Individuals were in auditory (and olfactory) contact with the other individuals in the same colony room and in visual contact with those individuals housed in the cages directly opposite to them.
Home cages were furnished with a variety of enrichment items including logs for gouging and rubber matting platforms to facilitate allogrooming. The marmosets were fed daily on a diet consisting of a mixture of fresh fruit and vegetables supplemented every other day with: commercially available pellet diet soaked in Ribena s , dried fruit and whole peanuts in their shells; or alternately with a mixture of yoghurt and baby rice with supplements. A scatter feed was added weekly to the sawdust on the cage floor. Water and commercially available pellet diet was available ad libitum.
The marmosets were habituated to the presence of the observer (Watson). The marmosets had also been habituated to the combined presence of the audio equipment and the observer over 11 days, of practice data collection, before the study. The observer did not interact with the marmosets before or during the study and dressed in a uniform different to that of laboratory staff. Observational data collection was carried out during March and April 2008. The study was approved by the University of Stirling Psychology Department Ethical Committee and complied with UK legal requirements.

Selection of Behaviors and Vocalizations
Table I displays the behavioral definitions used for all recorded behaviors and the behavioral category to which they belong. The definitions used for the visual and spectrographic identification of the four coded marmoset calls, sample spectrograms and a list of the call names used in previous studies equivalent to the call types coded in this study are available at the Stirling Online Research Repository at http:// hdl.handle.net/1893/1883 or alternatively contact C.F.I. Watson. We selected behaviors and vocalizations based on two criteria. They were selected first as indicators of the four behavioral categories, according to proposed functions and observed contexts given in the relevant literature. Second they were selected for reasons of practicality involving the level of accuracy of recording that was possible. During the live coding the observer had to be able to localize vocalizations easily to a particular individual by sight so we chose only those calls made with the mouth open or partially open. Additionally, during the video coding, all open mouth calls had to be distinguished easily, reliably and consistently from similar, nonopen mouth calls, on the basis of examining the spectrogram visually and listening to the call (e.g. chatter is distinct from the spectrographically related calls: cough and ek). Phee calls that were over 1.3 sec in length were coded and labeled as loud shrill calls to ensure that only open-mouthed phee calls were coded.

Live Coding
A continuous focal sampling method was used [Altmann, 1974;Martin & Bateson, 2007] to record the behavioral states and vocalizations of each of the 32 focal individuals, over a 5 min observation period, on each of 15 study days. A total of 465, 5 min observations were recorded, although 2 were subsequently lost (totaling 38.6 hr in all and 1.25 hr per focal individual). To allow the focal calls to be identified and differentiated from neighbor calls during subsequent audio coding, the precise event time of each focal individual vocalization was recorded, along with the probable call type. Pilot studies indicated that it was possible to localize all selected call types to a particular focal individual. Observation of vocalizations is further facilitated by the fact that calls are predominantly made during periods of low activity or when the individual is stationary [Pook, 1976].
During each observation period the observer sat about 140 cm from the cage housing the focal individual. To minimize disturbance to the subjects, the observer sat quietly for 5 min following the initial setup of the audio recording equipment in each room, and for 2 min after the movement of equipment between cages. Data were collected using a Psion Workabout (a handheld computer) running real-time event recording software (Observer 5; Noldus Information Technology, Wageningen, The Netherlands).
We counterbalanced the order in which the eight focal individuals within each room were observed and also the time of day that observations were carried out in each room. Observations were made in four sessions, between 08.30 and 16.30, with each room being observed on an equal number of occasions in each of the four sessions. Counterbalancing was considered to be especially important as the level of particular behavior is known to be subject to diurnal variation [e.g. vocalizations: Jones, 1993]. No observations were carried out in the 30 min following daily feeding at 12.30.

Audio Recording
During each 5 min observation session we made a simultaneous 5 min audio recording to collect the vocalizations made by the focal individual and all other individuals within the colony room. We used a tripod-mounted Zoom H4 digital recorder (sampling rate of 96 kHz and 24 bits with the gain level set to 127 dB) and an AKG-c 1000 directional microphone, mounted on a microphone stand. For each observation the sensitive end of the microphone was placed 20 cm from the center of the front of the cage housing the focal individual. The use of one directional microphone was considered sufficient to collect the vocalizations of all individuals within the colony room following a pilot study comparing the recording output from a directional microphone with that from a centrally placed omnidirectional microphone.

Audio and Spectrogram Coding
The identity of the different call types (loud shrill; twitter; chatter; chirp) was established during the pilot study by visual and audio examination of spectrograms of pilot recordings in relation to the verbal descriptions (and in some cases spectrogram images) given by various researchers [Epple, 1968;Goldman, 2000;Jones, 1993;Pook, 1976;Stevenson & Rylands, 1988;Winter, 1978]. The four different call types were considered to be sufficiently distinct to allow reliable and accurate identification by visual and auditory examination alone.
The 5 min audio track for each observation period was converted into a video, consisting of a spectrogram (a plot of frequency against time) and the original audio track, using a custom-designed program. Each video was played and coded within the Observer 5 PC observation module, allowing us to identify and code focal and neighbor vocalizations (any individual audible within the room other than the focal individual) by a method of continuous recording [Martin & Bateson, 2007] and to superimpose these codes onto the existing observational data collected during live coding. Each video was synchronized precisely with the time of the existing live observation using the beep emitted by the Psion Workabout at the end of each observation as a point of reference. When accurate measurements of frequency and time were required for call and call bout identification during the audio coding, we viewed the audio track as a spectrogram using the freely available software program ''Sonic Visualiser 1.0'' (http://www. sonicvisualiser.org/; licensed under the GNU general public license). Details of additional coding issues and how we dealt with them are available online at http://hdl.handle.net/1893/1883, as above.

Statistical Analysis
Focal behavior was summarized as the percentage of time spent in each behavior by each individual in every 5 min observation period. Percent time was used as a measure across all behaviors for consistency and was considered a suitable measure even for behaviors of relatively short duration, such as attack, as duration and frequency of such behaviors are well correlated [e.g. Badihi, 2006]. Focal vocalization was summarized as the total number of each call type emitted by the focal individual in each observation. Neighbor vocalization was summarized as the total number of each call type produced by all individuals in the room except for the focal individual, including other members of the focal group.
To investigate the effect of the overall level of neighbor vocalization the data sets were split in one of two ways. First, the call type that occurred in greater than 75% of observations (twitter) we split the data set into quartiles, according to the number of neighbor vocalizations occurring within each observation period, and compared the focal behavior in the upper quartile observations with the lower. Second, the data set was split into those observations with zero neighbor vocalizations of a particular type (low neighbor vocalization) and those with more than one such vocalization (high neighbor vocalization). We had intended to split the entire data set using the first method but the second method was necessary for call types (loud shrill; chatter; chirp) with more than 25% of observations containing zero neighbor vocalizations because there was no way of identifying a lower quartile. Hereafter the terms ''high neighbor vocalization'' and ''low neighbor vocalization'' [after Baker & Aureli, 1996;Videan et al., 2005] are here used to refer to the conditions created by both methods of splitting the data set.
The mean level of behavior for each focal individual was calculated from a variable number of observations depending on how many observations for each individual were in the high and in the low condition for each particular neighbor vocalization. Therefore, due to the intended within-subjects analysis, some individuals were excluded by default from the analysis of certain neighbor vocalizations, due to an absence of data from one or other of the conditions. We used approximate randomization tests (without replacement) [e.g. Adams & Anthony, 1996] to determine the mean result of related samples t-tests, carried out on 200,000 iterations [the recommended minimum number of iterations being 5000: Adams & Anthony, 1996], to compare focal behavior and vocalization between observations with high and low neighbor vocalization within the same behavioral category. We used a program, written in MATLAB, to shuffle the data without replacement exclusively for each focal individual in order to conserve the maximum amount of information in the analysis (i.e. the direction of difference, between the mean values for behavior in the upper and lower neighbor vocalization conditions, respectively, was randomly determined for each focal individual). We considered this to be the most appropriate and statistically powerful method available given that our data do not meet parametric assumptions and are drawn from a relatively small sample size.
To investigate the immediate effect of neighbor vocalizations in a particular behavioral category on focal individual behaviors within the same behavioral category, the mean proportion of focal behaviors occurring in the 15 sec interval before each type of neighbor vocalization was compared with the proportion occurring in the 15 sec interval after, following the method of Baker and Aureli [1996] but using a shorter time interval. Mean proportions for each focal individual, for each combination of neighbor vocalization and focal behavior, were calculated using the ''sequence lag analysis'' function in the ''Observer 5'' program (Noldus Information Technology, Wageningen, The Netherlands). We considered an interval of 15 sec to be appropriate because marmosets change behavioral state relatively quickly. Mean proportions for the intervals preceding and following neighbor vocalizations were compared using approximate randomization tests of paired sample t-tests (as above).
All the statistical tests were one tailed as clear directional predictions were made (except for the two-tailed post hoc tests on feeding and on foraging) and the a level was set at 0.05.  focal individual at high and low neighbor vocalization within the same behavioral categories.

Agonistic Intergroup
Neighbor twitter and loud shrill vocalizations occurred in 94.4 and 53.1% of observations, respectively. The amount of time spent by focal individuals in intergroup agonistic behaviors was generally significantly higher in observations with a high level of neighbor twitter vocalizations (bristle: N 5 30, P 5 0.004) and with a high level of neighbor loud shrill vocalizations (anogenital present: N 5 31, P 5 0.026; bristle: N 5 31, P 5 0.009) than in observations with low levels of the respective neighbor vocalizations (Fig. 1A). However, although marmosets spent longer in the intergroup agonistic behavior, anogenital present, during high neighbor twitter than during low, the difference was nonsignificant (N 5 30, P 5 0.062). Intergroup agonistic vocalizations were all produced significantly more often by marmosets in observations with high levels of neighbor twitter vocalizations (twitter: N 5 30, Po0.001; loud shrill: N 5 30, Po0.001) and neighbor loud shrill vocalizations (twitter: N 5 31, Po0.001; loud shrill: N 5 31, Po0.001) than in observations with low levels of these neighbor calls (Fig. 1B).

Agonistic Intragroup
Neighbor chatter vocalizations occurred in 39.7% of observations. The intragroup agonistic behaviors were collapsed into a single category as they occurred too infrequently to allow statistical analysis of the separate variables. Marmosets spent significantly more time in intragroup agonistic behaviors (composite of: steal food; chase; and attack) during observations with high neighbor chatter vocalization than in observations with low neighbor chatter vocalization (N 5 27, P 5 0.043; Fig. 2). However, although the mean rate of focal chatter was higher during high neighbor chatter vocalization than during low the difference was not significant (N 5 27, P 5 0.188; Fig. 2).

Affiliative
Neighbor chirp calls occurred in 64.6% of observations. During observations with high neighbor chirp vocalization marmosets spent significantly more time sharing food (N 5 31, P 5 0.012) and in grooming invite behavior (N 5 31, P 5 0.024) and emitted significantly more chirp calls (N 5 31, P 5 0.031) than in observations with low neighbor chirp vocalization (Fig. 3). Marmosets spent more time allogrooming and in active affiliative contact  during high neighbor chirp vocalization observations than during low, but the differences did not reach significance (allogrooming: N 5 31, P 5 0.369; affiliative contact: N 5 31, P 5 0.149; Fig. 3). Conversely, marmosets spent less time in social play during high neighbor chirp observations than during low.
The data were analyzed, post hoc, to determine whether marmosets spent more time feeding and more time foraging during observations with high neighbor chirp vocalization than with low, and a significant difference was found for feeding (N 5 31, two tailed, P 5 0.003) but not for foraging behavior (N 5 31, two tailed, P 5 0.853). Table III displays the results of the statistical tests comparing the proportion of 15 sec intervals before and after neighbor vocalizations during which focal behaviors and vocalizations in similar behavioral categories occur. For the four comparisons for which the difference does not lie in the predicted direction the P values are not reported; however, in all these four cases the two means are very similar to one another.

Agonistic Intergroup
As predicted, marmosets displayed a significantly higher mean probability of performing the agonistic intergroup behavior, bristle, in the 15 sec following a neighbor agonistic intergroup call (twitter; loud shrill) than in the 15 sec before such a neighbor call (N 5 31, P 5 0.019; N 5 31, P 5 0.010; respectively). Also in the predicted direction, but not significant, were the following results: marmosets were more likely to perform anogenital present, or vocalize a twitter, following a neighbor twitter call; and were more likely to emit a twitter, or loud shrill, following a neighbor loud shrill (see Table III). Two comparisons were not in the predicted direction: marmosets had a lower mean probability: of performing anogenital present directly before a neighbor loud shrill call than after; and of producing a loud shrill call before a neighbor twitter call than after.

Agonistic Intragroup
The mean probability of marmosets engaging in intergroup agonistic behavior (composite of chase;  Am. J. Primatol. attack; steal food) was higher following neighbor intragroup chatter vocalization than preceding, but this difference was not significant (N 5 31; P 5 0.343). The other comparison for this category was not in the predicted direction: there was a slightly lower mean probability of marmosets emitting intragroup agonistic calls (chatter) after neighbor chatter calls than before.

Affiliative
Consistent with our predictions, there was a significantly higher mean probability of marmosets engaging in the affiliative behaviors of food sharing (N 5 31; P 5 0.008), and active affiliative contact (N 5 31; P 5 0.031), after the neighbor affiliative call chirp than before. Also in the predicted direction, but not significant, marmosets were more likely to engage in allogrooming, and in grooming invite behavior, after neighbor chirp calls than before (see Table III). There was no difference between the mean probability of focal individuals engaging in social play before and after neighbor chirp vocalization. For one of the comparisons, the means were not in the predicted direction: marmosets produced more chirp calls before a neighbor chirp call than after. Regarding the post hoc tests: although marmosets had a higher mean probability of feeding and foraging directly following a neighbor chirp call than directly before the difference was not significant (feeding: N 5 31, P 5 0.836; foraging: N 5 31, P 5 0.344; both two tailed).

Agonistic
As predicted, we found that the levels of focal intergroup agonistic behaviors investigated (anogenital present, bristle, twitter and loud shrill) were all significantly higher at high levels of the neighbor intergroup vocalizations (twitter and loud shrill), than at low (with the exception of focal anogenital present during neighbor twitter). This finding is consistent with a neighbor effect for intergroup aggression. The close temporal association found between the production of neighbor twitter and loud shrill calls and bristle behavior in nearby marmosets provides support for a causal link. We speculate that contagion of intergroup agonistic behavior may improve co-ordination of group aggression toward conspecific groups encountered in the wild [e.g. Clayton, 1978].
Regarding intragroup aggression we found, as expected, that the composite of marmoset intragroup aggressive behavior was significantly higher at higher levels of the intragroup neighbor agonistic vocalization (chatter) than at lower levels. This is consistent with the social contagion of intragroup aggression. Lack of evidence for a close temporal association between the neighbor call and focal behavior indicates that we should treat the inference of a causal link with some caution.
Previously, researchers have found evidence for an effect of agonistic intergroup neighbor calls and noises on aggressive intragroup behavior in chimpanzees [Baker & Aureli, 1996;Videan et al., 2005]. Our results are broadly consistent with these findings in that they also support the social contagion of aggression. Furthermore we have demonstrated the neighbor effect of agonistic behavior for marmosets that is specific to each subclass: intergroup and intragroup, respectively.

Affiliative
Consistent with our prediction, we found that high overall levels of neighbor chirp calls were associated with significantly longer time spent by nearby marmosets in affiliative behaviors: grooming invite and share food and with more affiliative vocalizing by focal marmosets (chirp call). Focal individuals were also significantly more likely to engage in food sharing and in active affiliative contact immediately after a neighbor chirp vocalization rather than just before. The neighbor effect for active affiliative contact appears to be largely confined to the 15 sec interval postneighbor call. Our results support the existence of an affiliative neighbor effect.
We suggest that the chirp call may function as an invitation toward conspecifics to engage in affiliative contact. This would provide an explanation for the increase in grooming invite behavior in focal individuals through the contagion of behavior inviting affiliative contact and for the immediate increase in active affiliative contact following neighbor chirp. However, this explanation does not fully account for the greater time spent by marmosets in food sharing at relatively higher levels of neighbor chirp calls.
We categorized the chirp call as an affiliative call based on previous literature [Epple, 1968;Stevenson & Rylands, 1988]. However, there appears to be no consensus on the exact function or context of this call. As well as being described as an affiliative vocalization, the chirp call has also been frequently linked to feeding [Epple, 1968;Goldman, 2000], especially on highly favored foods. Indeed, Vitale et al. [2003] refer to a particular C. jacchus vocalization as a solely food-associated call. They state that they have heard marmosets making this call only while handling, eating or viewing favored foods. However, given the low resolution of their spectrogram, it is difficult to confirm that the vocalization they use is the same call we identify as chirp. A post hoc analysis of our data supports the association of chirping with feeding, although not with foraging. However, although an overall high rate of neighbor chirp calls was found to be associated with a significantly longer time spent in feeding by marmosets, there was no evidence for an immediate effect. Kitzmann and Caine [2009] investigated the effect of chirp playback in C. geoffroyi. Spectrograms provided in the study closely resemble those of C. jacchus chirp calls in this study. They noted an overall increase in the level of feeding behavior following chirp playback, consistent with the overall neighbor effect found in this study and, contrary to our findings, in foraging behavior. They did not investigate social behaviors or the immediate effect of chirp playback on feeding related behaviors. In the current study the link between spontaneous chirp calls and feeding may indirectly explain the apparent association of neighbor chirp with focal food sharing behavior but it does not explain the association with other affiliative behaviors unrelated to feeding. Pook [1976] stated, more specifically, that chirping was mainly associated with the excitement of feeding following the presentation of food. Another possible explanation, then, is that the chirp call is a more general expression of anticipatory excitement in expectation of a desirable and pleasurable event. The apparent contagion of chirp calls may indeed reflect the contagion of this positive arousal.
Regardless of the precise function of the chirp call, it does appear to be associated with nonthreatening events at least some of which are affiliative in nature. Thus, our findings provide supporting evidence for affiliative social contagion in accord with the findings of the previous study for chimpanzees [Videan et al., 2005]. We have also presented here new evidence bearing on the contextual use and function of the marmoset chirp vocalization.
It is unclear as to exactly how affiliative social contagion through the chirp call relates to wild marmoset behavior. Bezerra and Souto [2008] do not include an equivalent to the chirp call in their repertoire of wild C. jacchus vocalizations. It remains to be established whether the chirp call is confined to captive marmosets or whether the chirp was not heard in the wild group due to the context in which calls were collected. Certainly, the chirp call seems to be an important call for the captive marmosets in this study because they use it frequently. The chirp call clearly merits further research attention.

General Discussion
We are aware that conclusions drawn from this study rest on the assumption that the connection between neighbor calling and focal behavior is causal. Whether or not a causal connection exists cannot be established on the basis of the current data. Rather than displaying dependent responses to conspecific vocalizations individuals may instead be reacting independently to similar external stimuli.
Another alternative explanation for the present results, instead of social contagion, is that focal behavior may indeed be contingent on neighbor calls but may reflect either an entirely instinctive or entirely learned response, either selected for, or shaped by, direct interaction with conspecifics. This criticism is probably most applicable to the apparent contagion of intergroup aggression than to the social contagion of intragroup aggression and affiliation.
Although we cannot entirely exclude alternative explanations it is likely that social contagion has an important influence given the strong and consistent associations between the overall level of neighbor calls and focal behavior of the same behavioral category. Further research involving experimental manipulation is needed to distinguish the precise mechanisms underlying the neighbor effect.
The neighbor effect has interesting implications for the within-group transmission of a particular conception of social culture, proposed by Sapolsky [2006]. He used the term ''social culture'' to denote a distinctive social style particular to a specific group of individuals expressed across a range of usual species-wide behaviors that are performed to an unusual extent. Possible transmission mechanisms for this form of social culture have been discussed, including observational learning [Sapolsky, 2006] and facilitation [de Waal & Johanowicz, 1993]. We here suggest that social contagion constitutes a possible transmission mechanism for social culture. We have previously suggested investigating this form of culture by attempting to initiate a social culture of increased affiliative behavior through the playback of recordings of affiliative vocalizations to conspecifics [Watson & Caldwell, 2009]. The results of this study suggest that social contagion of affiliation occurs in marmosets and so provide the first step toward determining whether longer-term changes in social culture can be facilitated through social contagion, via such experimental manipulation.
The demonstration of social contagion in marmosets has important implications for captive welfare. Although the apparent spread of mood and behavior from neighboring groups to nearby individuals is likely to occur also in the wild, in the captive situation it is likely to lead to more pronounced effects. Larger numbers of captive groups of marmosets are close together for much longer than might be expected in the wild and the marmosets do not have the option of moving apart. Affiliative social contagion among captive groups may function to increase positive welfare behavior, whereas the effect of the contagion of agonism is more equivocal. Further, increases in agonistic behaviors are largely incompatible with the performance of affiliative behaviors within the same room. It is important that the potential influence, both good and bad, of neighboring groups on the welfare of nearby marmosets is taken into account in the captive management of marmosets.
Taken together, the results of this study provide the first systematic evidence to support the existence Am. J. Primatol. of a neighbor effect in marmosets. We have extended research into social contagion and provided further support for the generality of the neighbor effect.

ACKNOWLEDGMENTS
We thank Prof. H. M. Buchanan-Smith for her support and helpful comments. A University of Stirling Department of Psychology PhD studentship provided funding for Claire F. I. Watson. The audio equipment used in this study was funded by The Universities Federation for Animal Welfare (UFAW) through a Small Project Award. The authors are grateful to Keith Morris, the manager of the MRC Human Reproductive Sciences Unit, for generously allowing us to conduct research at the facility and also to the staff for their valuable assistance. The authors also wish to thank: William M. J. Holland for kindly customizing the conversion program used for audio coding; Iain M. Harlow for writing the program enabling randomization test analysis of the data; and C. A. Howie for her statistical expertise. The experimental procedure was approved after review by the Stirling University Psychology Departmental Ethics Committee and the research was conducted in full compliance with animal care regulations and applicable European law.