- Орангутаны научились использовать орудия с выгодой для себя [2019-03-22]
- Мозг собак отличил настоящие слова от тарабарщины [2019-03-22]
- Какаду Гоффина выклевали палочки из картонки и использовали их в быту [2019-03-22]
- Орангутанов уличили в создании крючков [2019-03-22]
- Любовь самок гуппи к ярким самцам объяснили генами и освещением [2019-03-22]
- Орангутаны рассказывают друг другу о прошлом [2019-03-22]
- Блеск и нищета этологии [2018-05-03]
- Почему хозяева похожи на своих собак? [2018-03-15]
- Интервью с Анатолием Протопоповым [2019-02-13]
- Цирки: развлечение vs мучение [2018-12-15]
• 2. Study 1: age, sex, and clique size
• 3. Study 2: age, sex, and ToM skills as an influence on clique and group size
• 3.1.1. Group size observations
• 3.1.2. The imposing memory task
• Appendix A. Lucy's birthday cake
• Appendix B. Rescaled weighted arithmetic mean
1. Introduction
A good deal of comparative data indicate that group or network size in social primates is set by constraints on the ability of individuals to sustain adequate levels of interaction (Dunbar, 1992, Henzi et al., 1997, Kudo & Dunbar, 2001). In humans, the maintenance of relatively large networks is held to be possible because of the increased efficiency of language as a transmitter of appropriate social information (Aiello & Dunbar, 1993, Dunbar, 1993). Nevertheless, there is an upper limit to primary network size that corresponds to the average size of conversational cliques (Dunbar, 1993), making the investigation of limits on clique formation appropriate to the analysis of population structure and its relation to cognitive ability.
In an analysis of freely forming adult conversational cliques, Dunbar, Duncan, & Nettle (1995) showed that clique size (the number of individuals actively participating in a single conversation) increased with increasing group size (the total number of individuals present in an interacting group) until it reached asymptote at around four individuals, after which further increases in group size had little effect. Following Cohen, 1971, Dunbar et al., 1995 concluded that this asymptote was due essentially to the mechanical constraints of human speech production (sound attenuation over distance, rising ambient noise) acting, perhaps, in conjunction with increasing difficulties in discerning the visual cues necessary for fluent conversational exchange (Argyle, Lallje, & Cook, 1968). Nevertheless, given that there is a strong interspecific relationship between mean group size and relative neocortex size among the primates (Barton & Dunbar, 1997), and that overall group size can be set by primary social units or cliques (Henzi et al., 1997, Henzi et al., 1997), it may well be that cognitive constraint is involved, to a greater or lesser degree, in the capping of clique size reported for humans. The rising demands—for any one participant—of monitoring the input of others, formulating appropriate responses, while also managing conversational pragmatics (Steiner, 1972), may disincline individuals to join larger cliques or may make these unstable over time.
We have begun to address the issue of cognitive constraint and clique formation by examining the relationship between conversational clique and group size in children of different ages. Children provide us with a neat “model system” that allows us to test between these alternative explanations. If mechanical factors alone were operating, we would expect to find no age effect, either within our sample set or between it and the adult data presented by Dunbar et al. (1995). While it might be argued that differences in the acoustic properties of children's voices would lead to smaller clique sizes, this is unlikely. Dunbar et al. (1995) found no difference between the clique sizes of adult males and females, despite obvious acoustic differences between male and female voices. Whatever constraints are imposed by human speech mechanisms, therefore, they are unlikely to be sensitive to features such as pitch and volume, even though these may nevertheless affect participation in conversation (Dunbar et al., 1995).
If, however, the cognitive demand of conversation constrains participation, we would expect to find a developmental trend in the observed size of conversational cliques and groups. It is well established that there is general improvement in cognitive competence across the age range represented by our sample (3–13 years) (Anderson, 1992, Fry & Hale, 2000). In particular, three cognitive variables—processing speed, working memory, and fluid intelligence—develop together and, while this growth is also nonlinear overall, show a consistent improvement throughout the period of childhood and adolescence (Fry & Hale, 2000). For example, the ability to store items in working memory while also processing information online—something one could argue is central to the successful management of conversation in larger groups—improves steadily over the first 18 years (Siegel, 1994). Fry and Hale (2000) also suggest that the age-related variance in fluid intelligence that is not explained by changes in processing speed or working memory may involve changes in the frontal cortex during childhood and early adolescence (see, e.g., Goldman-Rakic, 1987). In addition, and related to this, there is also the well-established developmental trend in the emergence of sociocognitive and metarepresentational skills, often labeled as “Theory of Mind” (ToM) (Wellman, Cross, & Watson, 2001), that could influence children's abilities to manage conversational groups. It has been shown for adults that those individuals with larger social networks show better perspective-taking or ToM skills (Stiller & Dunbar, in press). While the direction of causality between these variables is not clear, the link between them is undeniable. Such findings are pertinent since they suggest that the same link between ToM abilities and social skills may apply developmentally. Constraints on both domain-general processing abilities and domain-specific sociocognitive skills both imply that younger children should be less capable of manipulating the information necessary to sustain larger conversational cliques.
Here, we present data from two studies, one undertaken in South Africa and one in the UK. In the first study, we used observational data to test for a developmental trend in children's clique size. Given evidence of a trend consistent with the notion of cognitive constraint, we used the second study to investigate whether the developmental trend in sociocognitive skills maps onto children's clique sizes, and to test the hypothesis that children with above-average scores for their cohort were able to manage and participate in larger conversational cliques.
2. Study 1: age, sex, and clique size
2.1. Methods
We collected data on three different age cohorts from four different schools in KwaZulu-Natal, South Africa. The youngest cohort came from a single preprimary school. In the other three schools, the second cohort (Junior Primary) had a playtime that differed from that of the third cohort (Senior Primary), which thus determined the age allocation on which the study was based. Class records provided data for the range and mean age of cohorts (Table 1). All cohorts contained children of both sexes from a variety of ethnic backgrounds.
We obtained playground data from a single observation period for each of the cohorts at each school. From a vantage point at the edge of each playground, we used instantaneous scan sampling (Altmann, 1974) to record data on children's conversational groups during each observational period. Conversational groups were defined as those containing children whose main activity was either speaking or listening to a speaker, and who were not engaged in any other on-going playground activity (e.g., playing games). Definitions of group and clique size followed those of Dunbar et al. (1995) to ensure comparability with adult data. Group size was defined as the total number of children present in a conversational group (i.e., the total number of children in close proximity to at least one speaking child, and who were closer to each other than to any other child in the vicinity; this effectively demarcated conversational groups both from other kinds of grouping and other conversational groups). Clique size was defined as a sub-division of a group consisting of the number of children actively participating in a particular conversation, either speaking or visually attending to the speaker. Gender and language group (either English- or isiZulu-speaking) of each participating child was also recorded. Recording language group allowed us to control for any possible cultural differences. We obtained data on a total of 390 cliques.
Statistical analyses were performed on relevant subsets of the database, with the effects of age, sex, and language assessed using the generalized linear model (GLM) procedure in SPSS. GLM is robust with respect to deviations from normality provided the data are symmetrical.
2.2. Results
As mentioned above, we defined clique size as the number of children actively participating in a particular conversation, either speaking or attending to the speaker. The mean clique sizes observed in our sample were 2.0 for Cohort 1, 2.16 for Cohort 2, and 2.66 for Cohort 3. Fig. 1 gives the distribution of clique sizes for each age cohort.
;)
|
|
Fig. 1. Distribution of clique sizes for each age cohort (Study 1). |
Two things are apparent: the distribution changed with age and the maximum clique size increased from 2 to 5. As clique size is potentially affected by other factors, such as gender, we tested for age effects with gender and cohort entered as factors and group size as a covariate. To do so, we removed from the sample all mixed-sex cliques. A main effect was found for cohort [F(2,345)=22.8, p<.001], but not for gender [F(1,345)=0.013, NS], nor for the interaction of gender and cohort [F(2,345)=1.09, NS]. Group size covaried significantly with clique size [F(1,345)=153.2, p<.001]. Regression analysis revealed that the best fit between clique and group size was provided by a quadratic model [R2=.47; F(2,345)=153.5, p<.001]. The same relationship was obtained with the full data set. The relationship between mean clique size and group size is illustrated in Fig. 2, to enable comparison with the data presented in Dunbar et al (1995). The shape of the relationship for children is identical to that of adults over the range of group sizes covered by our sample, with clique size reaching an asymptote at a group size of approximately four individuals.
;)
|
|
Fig. 2. Relationship between mean group size and mean clique size for the total sample of children in Study 1 (black diamonds). The two comparative data sets for adults (open squares, open circles) are taken from Dunbar et al. (1995) and cover only the range of group sizes for which there are data for children. |
Fig. 3 gives the distribution of group sizes (the total number of children present in an interacting group) for each age cohort. Mean group sizes were 2.5 for Cohort 1, 3.0 for Cohort 2, and 3.5 for Cohort 3. We assessed the effect of gender and cohort on group size by entering them as factors in a two-way analysis of variance (ANOVA). There was a significant main effect for cohort [F(2,346)=18.7, p<.001], but not for gender [F(1,346)=0.006, NS]. There were no significant interaction effects. Tukey's test (p<.05) revealed that each of the three cohorts differed significantly from the other two.
;)
|
|
Fig. 3. Distribution of group sizes for each age cohort (Study 1). |
Limitations on sample size precluded the inclusion of language as a factor in the above models. To test the effects of language, we extracted from Cohorts 2 and 3 all cliques composed entirely of iSiZulu or English speakers. This was entered with cohort as a factor in an ANOVA model with clique size as the dependent variable. There was no main effect for language [F(1,144)=1.3, NS], although age remained significant [F(1,144)=23.2, p<.001]. There was no interaction between language group and age [F(1,144)=0.4, NS].
2.3. Discussion
As predicted, there was a significant developmental trend in children's grouping behavior, with younger children maintaining smaller clique and group sizes than older children, supporting our notion that cognitive constraints may limit clique size. In addition, although mean clique size significantly increased with age, even that of the oldest age cohort fell below adult values and suggests that older children, while more competent, were nevertheless constrained in their ability to sustain large cliques compared to adults. However, we also found that there was no observable difference in the shape of the relationship between clique size and group size across the range of group sizes shared by adults and children. This latter result would seem to rule out any significant role for experience or growing social competence in setting clique size and, given the absence of gender effects, is supportive of the argument that some general constraints of speech production and coordination limit the growth of conversational cliques.
This apparent contradiction is resolved, however, if we accept that cognitive constraints do not affect the maintenance of single, small conversational cliques but, rather, the ability to achieve this while also monitoring other cliques or individuals within the larger group. The primary difference between our sample set and the adults observed by Dunbar et al (1995) lies in group size. Whereas adults are able to form groups of up to 14 individuals, the largest social group recorded for children in the 9–13 age range was six, with the youngest cohort only managing a maximum of four. Lower average clique size values for children, then, are likely to be due to lower average group sizes, which reflect the developmental constraints suggested by our other analyses.
By and large, the conversational clique and the group are coincident for young children. Even for the 11-year-old cohort, coherent groups contain at most one clique and a single bystander who is not, in the short-term, participating. Given that two cliques are likely to be pursuing different conversations, the task of monitoring these, as suggested by experiments on listening span (Siegel, 1994), may simply be beyond young children. In short, it may be that the cognitive constraint acts on group size, rather than clique size; children's lower sociocognitive competence means that the number of individual interactions with which a child can cope is absolutely lower than that of adults, resulting in smaller group sizes and by necessity smaller clique sizes on average.
Given these results, we therefore conducted a further study to investigate the relationship between a measure of children's sociocognitive capacity (specifically, the number of levels of intentionality capable of being understood) and group size. Our intention here was (1) to establish that the increase in group size with age mirrors the increase in sociocognitive capacities over this age range and (2) to test whether variation in sociocognitive abilities among children has any effect on group size over and above this developmental trend. Specifically, whether children with above-average skills for their age were found in larger groups.
3. Study 2: age, sex, and ToM skills as an influence on clique and group size
3.1. Methods
3.1.1. Group size observations
Data were collected in two primary schools and one high school in the northwest of England. In all cases, observations were made from the edge of the playground, and group and clique size was defined in a manner identical to Study 1. Instantaneous focal sampling was used to collect data on group size for individual children. A total of 120 individuals (60 male, 60 female) were selected at random and observed for a total of 15 min each during school playtimes. Table 2 gives details of the sample.
A sample period of 1 min was used, with the interacting group size and clique size recorded on each sample point. In the primary schools, all observations were made during playtime between 10:30 and 10:45 a.m. or during the lunch break between 12:30 and 1:15 p.m. In the high school, observations were made between 11:00 and 11:20 a.m., 12:20 and 3:00 p.m., or 4:10 and 4:25 p.m. during break periods between lessons.
3.1.2. The imposing memory task
Once observations on group and clique size had been completed for all subjects, they were subsequently approached for their participation in an “imposing memory task” (IMT) (Stiller & Dunbar, in press) designed to test their level of sociocognitive skill. This task was performed either in the classroom or in a quiet area of the playground where the child could complete the test alone, free from distraction. The subject's age was determined either by asking them to write it on the top of their answer sheet or by asking an adult supervisor who knew all the pupils and their ages. As the IMT was carried out on the same day as the group size observations, children could remain anonymous at all times. All participants were allocated a three-digit code that gave their age, sex, and enabled their answer sheet to be matched to the group data.
The IMT consisted of a simple illustrated story (“Lucy's birthday cake”; see Appendix A) concerning a social situation involving three people. Following the story, subjects were presented with eight questions in which they were required to select the correct option from two alternatives. The order in which the questions appeared and the order of true and false alternatives were counterbalanced across subjects. Four of the questions were concerned with memory for the actions of the story's characters and made no demands on the children's ability to mentalize. The remaining four questions were designed to test the subjects mentalizing ability by asking them to recall information about the characters' beliefs and desires. The four questions relating to the characters' actions required only first-order intentionality (i.e., the subject was required to have a belief about the facts of the story and nothing else). The remaining four questions varied in complexity. The simplest required an understanding of the mental state of a character. In the case of higher-order intentionality questions, these mental states concerned the mental states of others. The highest of these was a fifth-order question. Beyond this level of intentionality, most adults fail IMT questions (Kinderman et al., 1998). The story and the accompanying list of questions are given in Appendix A. In the case of young children (6 years old and below), the questions were read out to them by the observer. However, all subjects, regardless of age, were read the story only once.
To calculate subjects' test scores, the number of correct answers was calculated for both the intentionality questions and the memory questions. Only data on intentionality questions are presented here. For these questions, a weighted means procedure was used to convert children's scores into a level of intentionality (see Appendix B). However, unlike Stiller and Dunbar (in press), where aggregate scores over several stories were calculated for each subject, the use of only one story and one question at each level means that subjects' scores potentially are more prone to bias due to guessing (i.e., if a child answers, say, Level 2 and 3 questions incorrectly, but gets the Level 5 question right by chance, then the score will be inflated artificially). Consequently, we also present analyses using a “cutoff” score, whereby we scored each subject according to the level of the intentionality at which they first failed to answer correctly (i.e., if a child answered a Level 3 question incorrectly, but went on to answer Levels 4 and 5 correctly, their score would be nevertheless be only 2, which is the last level at which they answered correctly). This measure is also subject to the same bias (i.e., subjects could answer incorrectly by chance), but, as it tends to underestimate children's level of intentionality, it is more conservative than the weighted mean score with respect to the hypotheses under investigation.
The sample was divided into three age categories corresponding to those used in Study 1 in order to ensure comparability. However, direct statistical comparison between Studies 1 and 2 is not possible due to the differences in data collection procedures.
3.2. Results
The mean clique sizes for each cohort in Study 2 were 2.05 for Cohort 1, 2.99 for Cohort 2, and 3.49 for Cohort 3. Fig. 4 gives the distribution of clique sizes for each age cohort. A two-way ANOVA with cohort and gender entered as factors found a significant main effect of cohort [F(2,114)=32.5, p<.001], but not gender [F(1,114)=0.47, NS], and there was no interaction between the two [F(2,144)=1.20, NS]. Tukey's test (p<.05) revealed that all three cohorts were significantly different from each other.
;)
|
|
Fig. 4. Distribution of clique sizes for each age cohort (Study 2). |
Mean group sizes were 2.6 for Cohort 1, 4.2 for Cohort 2, and 4.2 for Cohort 3. Fig. 5 gives the distribution of group sizes for each age cohort. Again, ANOVA revealed a main effect of cohort [F(2,114)=21.4, p<.001], but not gender [F(1,114)=0.29, NS], and no interaction between the two [F(2,144)=2.89, NS]. Tukey's test (p<.05) revealed a significant difference between Cohort 1 and the other two cohorts, but no significant difference between Cohorts 2 and 3.
;)
|
|
Fig. 5. Distribution of group sizes (Study 2). |
As for Study 1, the distribution of both these measures changed with age and maximum clique/group size increased across age cohorts. As before, the relationship between clique size and group size was best explained by a quadratic relationship [Fig. 6: R2=0.80, F(2,117)=240.30, p<.0001], although this explained only slightly more of the variance than a linear model [R2=0.74, F(1,118)=335.87, p<.0001].
;)
|
|
Fig. 6. Regression of clique size on group size (Study 2). The quadratic fit (±95% confidence interval) is shown. |
As expected, the mean weighted IMT (MWIMT) scores increased across cohorts, from 2.61 for Cohort 1 to 3.09 for Cohort 2 to 3.91 for Cohort 3. A two-way ANOVA revealed a significant main effect of age cohort on MWIMT score [F(2,114)=12.28, p<.001], but no effect of gender [F(1,114)=0.516, NS], nor was there any significant interaction between the two [F(2,114)=0.088, NS]. Tukey's test (p<.05) revealed that Cohort 3 differed significantly from the other two, but that there was no significant difference between the two younger age cohorts. Inspection of the distribution of scores across age cohorts revealed a wide variability within the overall developmental trend; the maximum score of 5 was achieved even in the youngest cohort, while some of the oldest children only achieved the minimum score of 1 (Fig. 7).
;)
|
|
Fig. 7. Distribution of MWIMT scores for each age cohort (Study 2). |
Values calculated using cutoff scores gave mean values of 2.15 for Cohort 1, 2.25 for Cohort 2, and 3.15 for Cohort 3, suggesting that there was some overestimation of intentionality level reached by Cohort 2 subjects using MWIMT scores. However, a 2-way ANOVA gave the same results as those obtained using MWIMT scores [age cohort: F(2,114)=9.527, p<.01; gender: F(2,114)=0.775, NS; Age Cohort × Gender: F(2,114)=0.018, NS]. Tukey's test (p<.05) again revealed that Cohort 3 differed significantly from the other two, while there was no significant difference between the two younger cohorts.
Finally, to investigate whether sociocognitive skills had any effect on group size beyond the developmental trend identified, we calculated the standardized residuals from regressions of both MWIMT and cutoff scores against subjects' true age. We then entered each of these into separate regressions against group size to test for a positive relationship between group size and age-controlled ToM score [i.e., to test whether children with relatively high (or low) ToM scores for their age were able to maintain significantly larger (or smaller) groups]. There was no significant relationship found for either MWIMT [R2=−.008, F(1,118)=0.006, NS] or cutoff scores [R2=−0.006, F(1,118)=0.263, NS].
3.3. Discussion
The results for Study 2 mirror those obtained for Study 1: clique size and group size increased significantly with age, while at the same time the relationship between mean clique size and mean group size was equivalent to that obtained for adults. However, the maximum size of cliques and, in particular, groups fell below adult values, and there appeared to be a much stronger cap on group size in this sample of children, as mean group size did not differ significantly between the two oldest age cohorts. While children were therefore capable of forming conversational cliques of adult size, they did so in the context of much smaller group sizes.
In addition, we found that IMT and cutoff scores increased with age, as expected, but that there was no effect of IMT/cutoff scores on group size once age effects were taken into account: children with relatively high ToM scores for their age were not found in significantly larger groups. Even those children in the oldest age cohort who achieved the highest possible scores on the IMT (n=32; mean age, 10.75) were capable of a maximum group size of only 6.5 (mean, 4.1). These data support our hypothesis that there exist cognitive constraints on group size, but implies that the nature of the constraint does not lie with children's ability to take the perspective of other individuals.
These findings suggest three possibilities: (1) the cognitive constraint that prevents large groups is not fully measured by the IMT, (2) something other than cognitive constraint limits children's ability to form large groups, or (3) group size is too crude a measure for assessing children's sociocognitive skills in a group context.
We believe that all three of these possibilities have some currency. First, the IMT deals solely with what can be termed cold cognition (Tager-Flusberg & Sullivan, 2000). The ability to reason about others' mental states within the context of an experiment represents a different skill to that of monitoring the behavior of others in an ongoing, dynamic social situation. The latter calls on what have been termed socioperceptual skills or hot cognition (Stone, 2000, Tager-Flusberg & Sullivan, 2000). Measuring children's ability to do the former may reveal very little about the latter, as the results from studies of children with Asperger's syndrome amply demonstrate; while these children can pass standard ToM tasks by calling on acquired, language-based skills, they nevertheless fail to perform well in real social situations, since the rigid structure and predictability of the experimental laboratory setting are absent. As Happé (2000) has noted, we must be careful to distinguish between the cognitive mechanisms that are necessary for ToM and the social understanding that springs from ToM; the ability to function well in a social setting requires more than an ability to mentalize in a purely cognitive fashion.
The second possibility is that something other than sociocognitive skills limits children's group-forming abilities. This itself is composed of two possibilities: first, that children are limited not by their own cognitive skills but by those of the children with whom they interact. No matter how sociocognitively competent a particular child is, if the other children potentially available to compose a group are more limited in this respect, then groups will tend to fracture as those with lower abilities fail to cope with the social monitoring demands made on them and drift away from the group. This could be tested by comparing levels of metarepresentational skill to group sizes across different study populations; higher levels of metarepresentational ability should, on average, be associated with larger group sizes.
Second, it may be that the successful monitoring of groups, as opposed to cliques, is largely dependent on experiential, rather than purely maturational, factors. Understanding how cohesive groups can be maintained through the shifting interactions of individuals within them, and the behavior patterns that individuals in groups are likely to display, may require considerable experience of group interactions. Thus, even when children have acquired the requisite sociocognitive skill that enables them to maintain adult-like clique sizes, this “raw” ability cannot be translated into larger group sizes since they lack experience of larger groups (which is, in part, brought about by the developmental limitations on clique size at younger ages). Of course, these explanations are not mutually exclusive and it is, in fact, highly likely that both are operating simultaneously.
Focusing attention on contingent and experiential factors in this way leads to the third and final possibility which is that group size may fail to measure the relevant aspects of children's social competence. If group sizes are, at least in part, determined by the sociocognitive skills of the least competent children, then assessing a child's ToM scores against the average group size within a single population is an inappropriate test since group size will tend to reflect the average ability of all the children in the study population, and not the individual child's ability to maintain a group. Highly skilled children cannot maintain larger group sizes, even if they are attempting to do so, since this is not something that they can achieve alone. The differences between highly skilled and less skilled children may therefore manifest as behavioral differences within groups. Highly skilled children may be more dominant, for example, and dictate the group's activity or direct the conversation more effectively.
By the same token, if experiential factors are also important, then measures of group size alone are, again, uninformative, and it is the interactions within groups that are important. More highly skilled children may display behaviors that allow them to acquire necessary experience more quickly and effectively, such as monitoring a higher proportion of group members and doing so more frequently. This suggests that if we are to make progress in understanding how children's sociocognitive skills map onto their ability to interact socially with others, then we need to focus on within-group dynamics, individual differences in children's exposure to group situations and the processes by which children's groups undergo fission and fusion.
4. General discussion
Overall, our data support the hypothesis that children are constrained cognitively in their ability to support adult-like conversational cliques, although the relationship was more complex than we initially envisaged. While there was a significant developmental trend in the size of cliques in which children were found, it was also true that average clique sizes mirrored those of adults. Maximum clique and group sizes, however, fell below those of adults, pointing to the maintenance of large coherent groups as the limiting factor on children. The fact that there was no effect of mentalizing ability on group size, beyond the expected developmental trend, suggests that the task of maintaining groups requires more than just the ability to take another's perspective.
Aiello and Dunbar (1993) have argued that the evolutionary/strategic significance of language as a shaper of human societies lies in its ability to transmit efficiently relevant social information (“gossip”) about others, thereby reducing the likelihood of falling prey to “free-riders.” Such transmission typically occurs during casual conversation, where the efficiency of information transmission is set by the size of the conversational unit. Dunbar (1993) has shown that social network size in humans is a product of conversational clique size. What our data on children suggest is that the primary determinant of this relationship is not constraints on speech production within cliques, but the cognitive demand of sustaining coherent groups.
By definition, a coherent social group is one where, over the course of its existence, all individuals interact with one another. As the conversational unit grows, the amount of information required for continued participation (both in terms of monitoring the flow of the conversational topic itself, as well as monitoring the response of all other participants to particular utterances) rises rapidly to the limits of efficient processing. If the group then increases in size it should divide into conversational cliques. The advantages of maximizing access to useful information means that participants should still attempt to monitor other conversations and to switch cliques appropriately. Such eavesdropping is a common feature of human conversational interactions. This ability to switch attention rapidly back and forth between different conversations may hold the key to explaining the difference between the group sizes of adults and children; children will be more constrained in this ability, due to lower processing speeds and more limited working memory capacity.
While eavesdropping is possible the group as a whole remains coherent. Obviously, as the number of participants increases, mechanical aspects of speech production, such as the attenuation of sound, will constrain the efficacy of eavesdropping, and the group will then fragment. Consequently, conversational clique sizes are limited by both the cognitive capacities of the participants and the mechanical aspects of speech production. For adults, these mechanical constraints become relevant since, for the most part, adults cope better with the cognitive demands of large groups. For children, however, because the cognitive demands of larger groups are excessive, these mechanical constraints rarely, if ever, come into play since groups never reach a size where the ability to eavesdrop is limited.
The results of our studies ultimately raise more questions than they answer with regard to understanding the cognitive limits on conversational groups sizes. In order to address these questions, developmental analyses of the dynamics of conversational groups are needed. A more experimental approach designed to explore the differences between children in terms of their ability to monitor others within groups, and to participate in more than one conversation at a time, in relation to ToM, executive function, and individual experience, will allow a more precise test of the hypotheses presented here, as well as provide valuable information on how children apply their cognitive skills during interactions with their peers.
Acknowledgments
We thank the participating schools for permission to conduct the research and Christine Liddell, Neil Humphrey, and Tom Liggett for reading earlier drafts.
Appendix A. Lucy's birthday cake
Katy came home from the shop with a cake that she had bought for Lucy's birthday. Katy put the cake on the kitchen table and went to play in the garden with Lucy. While Katy was out, Tom went into the kitchen and accidentally knocked the cake off the table. The cake was totally ruined so Tom decided not to tell anyone what he had done and he put the cake in the bin. But just then, Katy came back into the kitchen and saw Tom putting the cake into the bin. Tom told her that Lucy had knocked the cake onto the floor and he was tidying it up for her. But Katy knew that Tom was lying because Lucy had been in the garden with her.
Questions:
1.
2.
3.
4.
5.
6.
7.
8.
Appendix B. Rescaled weighted arithmetic mean
Equation for weighted mean (Szulc, 1965)
In this study, the weight is the level of intentionality corresponding to any given question. The sum of the weights is 15 and the sum of wixi is 18. This means that the highest score obtainable with the weighted mean is 1.2. To obtain a result representative of a value along the scale of weights, the initial equation can be rescaled as follows:
By multiplying the sum of WiXi by a scaling value v=0.347, we obtain a number that when divided by the sum of the weights yields an answer that represents the level of intentionality at which the participant falls on a scale of 0–5. The benefit of using this method is that it takes into account that a participant might fail a low-order question, yet by chance alone succeed at a higher level (Stiller and Dunbar, submitted for publication).
References
Aiello & Dunbar, 1993 1.. Neocortex size, group size and the evolution of language. Current Anthropology. 1993;34:184–193. CrossRef
Altmann, 1974 2.. Observational study of behaviour: Sampling methods. Behaviour. 1974;49:227–265. MEDLINE
Anderson, 1992 3.. Intelligence and development. Oxford: Blackwell; 1992;.
Argyle et al., 1968 4.. The effects of visibility on interaction in a dyad. Human Relations. 1968;2:3–17.
Barton & Dunbar, 1997 5.. Evolution of the social brain. In: Whiten A, Byrne RW editor. Machiavellian intelligence II. Cambridge: Cambridge University Press; 1997;p. 240–263.
Cohen, 1971 6.. Casual groups of monkeys and men. Cambridge, MA: Harvard University Press; 1971;.
Dunbar, 1992 7.. Time: A hidden constraint on the behavioural ecology of baboons. Behavioural Ecology and Sociobiology. 1992;31:35–49.
Dunbar, 1993 8.. The co-evolution of neocortical size, group size and language in humans. Behavioral and Brain Sciences. 1993;16:681–735.
Dunbar et al., 1995 9.. Size and structure of freely-forming conversational groups. Human Nature. 1995;6:67–78.
Fry & Hale, 2000 10.. Relationships among processing speed, working memory, and fluid intelligence in children. Biological Psychology. 2000;54:1–34. MEDLINE | CrossRef
Goldman-Rakic, 1987 11.. Development of cortical circuitry and cognitive function. Child Development. 1987;62:1–22. CrossRef
Happé, 2000 12.. Parts and wholes, meanings and minds: Central coherence and its relation to theory of mind. In: Baron-Cohen S, Tager-Flusberg H, Cohen DJ editor. Understanding other minds: Perspectives from developmental cognitive neuroscience. Oxford: Oxford University Press; 2000;p. 203–221.
Henzi et al., 1997 13.. Fission and troop size in a mountain baboon population. Animal Behaviour. 1997;53:525–535. CrossRef
Henzi et al., 1997 14.. Cohort size and the allocation of social effort by female mountain baboons. Animal Behaviour. 1997;54:1235–1243. CrossRef
Kinderman et al., 1998 15.. Theory-of-mind deficits and causal attributions. British Journal of Psychology. 1998;89:191–209.
Kudo & Dunbar, 2001 16.. Neocortex size and social network size in primates. Animal Behaviour. 2001;62:711–722. CrossRef
Siegel, 1994 17.. Working memory and reading—a life-span perspective. International Journal of Behavioural Development. 1994;17:109–124.
Steiner, 1972 18.. Group processes and productivity. New York: Academic Press; 1972;.
Stiller & Dunbar, in press 19.Stiller, J., & Dunbar, R. I. M. (in press). Perspective-taking and social network size in humans. Social Networks.
Stone, 2000 20.. The role of the frontal lobes and the amygdala in theory of mind. In: Baron-Cohen S, Tager-Flusberg H, Cohen DJ editor. Understanding other minds: Perspectives from developmental cognitive neuroscience. Oxford: Oxford University Press; 2000;p. 253–273.
Szulc, 1965 21.Statistical Methods. London: Pergamon Press; 1965;.
Tager-Flusberg & Sullivan, 2000 22.. A componential theory of mind: Evidence from Williams' syndrome. Cognition. 2000;76:59–89. MEDLINE | CrossRef
Wellman et al., 2001 23.. Meta-analysis of theory of mind development: The truth about false belief. Child Development. 2001;72:655–684. MEDLINE
a Behavioural Ecology Research Group, University of Kwa-Zulu Natal, South Africa
b Department of Psychology, University of Central Lancashire, PR1 2HE Preston, UK
c Evolutionary Psychology and Behavioural Ecology Research Group, University of Liverpool, Liverpool, UK
Corresponding author. Department of Psychology, University of Central Lancashire, PR1 2HE Preston, UK.
PII: S1090-5138(06)00071-7
doi:10.1016/j.evolhumbehav.2006.07.002
© 2007 Elsevier Inc. All rights reserved.