The current research

This study addresses the question of whether we indeed observe the dyad emerging as the typical group size across a wide variety of social situations more than would be expected by chance. We provide preliminary empirical evidence from four survey studies (total N = 4124) that examined the size of the group in which people found themselves across social activities including dinner, movies/concert, off work chats, chats at work, projects, holidays, sports activity, and bars, as well as an experience sampling study (N = 274) of randomly selected situations people experience throughout daily life. Note that few scholars have tried to put forward situation taxonomies, and that these have not been successful in gaining strong support [22]. Therefore, we did not follow one particular situation taxonomy and instead our situations are eclectically derived from previous work. We define a group as “two or more persons who are interacting with one another in such a manner that each person influences and is influenced by each other person” [23] (p. 11). As such, when two friends go out to watch a football game, they are a group of two, within a larger aggregate of fans. In Study 1, we focused on Dutch participants’ past social activities. Study 2 replicated the results of the first study in a sample of US Americans, while Study 3 focused on US participants’ preferred group size for imagined future activities. In Study 4, same-sex social activities were assessed to examine sex differences in a US sample. The final experience-sampling study involved a real-time data capture approach using Dutch participants’ smartphones (Study 5).

The dyad is frequently the most common group size

Full information on data screening can be found in the Supporting Information (SI). We asked people to report the size of their social groups for a wide variety of activities. Group size frequencies are displayed in S1 Table and Figs 1 and S1–S3 in the main text. In all activities sampled, except for restaurant dinners, the dyad is the most common group size. This general pattern holds for Dutch participants in Study 1, American participants in Study 2, American participants reporting on preferred group size in Study 3, and for American participants reporting on activities involving same-sex others in Study 4. However, there are three cases where the dyad is the mode by only a small margin. These concern off-time chats in Study 1 (24.4% dyads, 24.3% groups of four), movies in Study 3 (34.6% for dyads and 30.8% for groups of four) and projects in Study 3 (29.6% for dyads, 25.9% for triads, and 25.5% for groups of four).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Frequency distributions of group size in eight daily activities sampled in Study 4.

https://doi.org/10.1371/journal.pone.0244188.g001

The typical group size for dinner parties was less clear. Future dinner parties of size 4 were strongly preferred in Study 3. Yet when people actually go out for dinner, they are typically with 2, 3, or 4 persons (Studies 1, 2, and 4). Similarly, participants in Study 3 preferred to go to the movies with two or four persons (34.6% vs. 30.8%), yet for actual movie visits, participants in Study 2 overwhelmingly chose to go as a dyad (51.7%) instead of as a group of four (15.5%). To a lesser extent this also holds true for chats off work and projects, as indicated in S2 and S3 Figs.

Moreover, as a general trend and perhaps not surprisingly, as group size increases the frequency of participants reporting to have been in such a group decreases. Except for sports activity, group sizes containing more than 10 individuals are rarely reported.

The over-representation of dyads compared to other group sizes

The most commonly used count models are Poisson and negative binomial. If smaller groups are easier to coordinate, and coordination costs increase exponentially with group size, the observed data would best fit a Poisson distribution (shifted so that the minimum value is 2 rather than zero). In contrast, we found that it does not, having better overall model fit with zero-inflated Poisson distribution. This improvement in model fit using a zero-inflated rather than a standard Poisson distribution indicates that there were more dyads in the observed data than would be expected simply on the basis of ease of coordination. We tested the over-representation of dyads using zero-inflated Poisson and zero-inflated Negative Binomial regression. As noted above, we first reparametrized the outcome as the count of additional group members beyond the dyad. That is, by subtracting 2 from every observed group size (dyads would then obtain the value “0”), a count measure of group size greater than two resulted. If we observe an excess of dyads, the zero-inflated part of the model thus reflects dyad inflation. We first tested the fit of the Poisson model (with Participant sex as a predictor). We then fit the two zero-inflated models (i.e., zero-inflated Poisson and Negative Binomial, both of which accommodate overdispersion) and used a Vuong test to determine if they yielded better fit than the Poisson model. (A Vuong test is a test of the zero-inflated model against the noninflated model.) In almost every case they did (S2 Table, third column), indicating dyad-inflation.

The dyad-inflated models similarly support the notion that women are somewhat more dyadic—out of a total of 25 tests for all activities for four studies (5, 6, 6, and 8 activities in Studies 1 through 4, respectively), 13 Poisson regression tests show significant sex differences, and 12 of these effects survive the move to dyad-inflated models with at least p < .10 (see also S4 Fig). In no case do men show significantly more dyads or smaller group sizes relative to women. However, this also means that in about half of the activities, there were no significant sex differences and the differences we did find were generally small with some exceptions, such as same-sex sports, in which women were noticeably more dyadic than men (see S5–S8 Figs, for an overview of the sex differences by activity).

Age and group size

Pearson correlations were performed to assess whether group size was associated with age. The correlations are reported in S3 Table. The significant associations were typically between small (.10) to medium (.30) in size. There was a positive correlation between age and dinner size for Dutch participants in Study 1, r = .11, p = .002. As age increases, so does the size of the dinner party. Yet, this result did not replicate with American participants in Studies 2–4. There was a negative correlation between age and group size at the movies for participants in Study 3 (r = -.08, p = .010) and men in Study 4 (r = -.15, p = .004). As age increases, group size at the movies decreases. Note, however, that this relationship was not found for participants in Studies 1–2, and for women in Study 4. For Dutch participants specifically, there was a positive correlation between age and conversation sizes (off work chats, r = .20, p < .001; chats at work, r = .18, p < .001). As age increases, conversation sizes become larger. There was also a positive correlation between age and project group sizes for participants in Study 1 (r = .21, p < .001), Study 2 (r = .07, p = .030), and Study 4 women (r = .11, p = .012) but not for participants in Study 3 and Study 4 men. Moreover, there was a negative correlation between age and holiday group size for participants in Study 2 (r = -.10, p = .003) but not for participants in Studies 3–4. Finally, Study 4 data indicated a negative relationship between age and sports group size in women (r = -.13, p = .009), but not men (r = -.02, p = .682); and a negative relationship between age and group size while visiting bars for men (r = -.14, p = .005), but not women (r = -.07, p = .109).

Relationship status and group size

In order to assess the association between group size and relationship status (single versus in a relationship), Pearson correlations were performed and reported in S4 Table. For dinner parties there was a consistent positive relationship between group size and relationship status. Participants in a relationship reported larger dinner sizes than participants that were single. This holds true for Study 1 (r = .13, p < .001), Study 2 (r = .11, p < .001), Study 3 (r = .08, p = .013), Study 4 women (r = .09, p = .037) and Study 4 men (r = .16, p = .002). These effects can be considered small (that is, around .10). There were no significant correlations between group size and relationship status for movies, projects, and bars. For conversation sizes the pattern was less clear with only significant results for off work chats in Study 1 (r = .10, p = .003); chats at work in Study 2 (r = .09, p = .005) and Study 4 men (r = .14, p = .006). The positive sign indicates that participants in a relationship had larger conversation sizes. Finally, there was a significant negative correlation between group size and relationship status for Study 2 participants’ holidays (r = -.08, p = .013), indicating that participants in a relationship had smaller holiday groups; and a significant positive correlation for Study 4 men in the activity of sports (r = .10, p = .047), that is, men in a relationship engaged in sports activity with larger group sizes. The significant correlations for conversation sizes and holidays should be interpreted with caution as they are not consistently found across the studies reported here.

Real-time experience-sampling data indicate that dyads predominate in daily life

Over the course of one week, participants (N = 274) were asked seven times a day to report the last situation they had experienced with another person since the last survey (i.e., situations were not sampled more than once). The experience sampling data yielded 10,933 total responses, 7,248 (66.29%) of which involved social situations. Forty-eight of these did not specify the number of other individuals present; therefore, for this analysis N = 7200 responses. The dyad was the most common group size (52.6%), followed by triads (18%), groups of four (9.1%), five (4.3%), six (2.5%), seven (1.6%), eight (0.9%), nine (1.0%), ten (0.5%), eleven (0.6%), and ‘more than 11’ (8.9%).

We again tested whether women encountered a greater proportion of dyadic situations than men. To do so, we fitted a logistic mixed effects model with random intercepts for participants and with gender as a level-2 predictor of the number of interaction partners in social interactions (7172 situations from 274 participants). Here, the outcome variable indicated whether the situation was dyadic (coded 1) or involving any other non-zero number of interaction partners (coded 0). We computed odds ratios for the fixed effects estimates. Analyses were conducted in R software version 3.4.4 using the tidyverse and lme4 packages. The hypothesis was not substantiated by the data. Women did not experience a greater proportion of dyadic (rather than non-dyadic) interactions than men, OR = 1.056, d = .03.

The detection of a sex difference in dyadic situations may be obscured by the overrepresentation of romantic couples for this unit size. Situations with romantic partners were indeed more dyadic than other situations, yet conducting the same analysis while excluding the situations with a romantic partner still found no significant sex difference (see S1 File).

The core configurations model

Caporael [5,6] proposes that face-to-face groups are hierarchically structured in four core configurations. These represent kinds of interdependent interactions between people. The interdependency is determined by the type of task to be performed, the situated environment, and the physical attributes of the participants (i.e., “body morphology”). Group size is an important feature of the model but the core configuration sizes have little meaning without invoking the associated activities to be performed. The notion of being “core” configurations reflects the idea that these group configurations can be repeatedly observed in hunter-gatherer groups (and hence presumably across human evolutionary history), across an individual’s lifespan, and in daily life. These configurations occur for a specific purpose and the successful accomplishment thereof accounts for their continued recurrence. Moreover, it is proposed that cognitive processes have evolved and developed in the context of the core configurations and thereby also cause their recurrence over time.

The first core configuration is the dyad, an interaction between two entities (e.g., two humans, one human and an animal, or human-AI interaction). Tasks addressed by the dyad are, for example, internal fertilization in the context of sex, providing infant nutrition while the mother is breastfeeding the infant, or a child’s interaction with a peer or adult. One proposed function of the dyad is microcoordination (e.g., during facial imitation, gait adjustment between two people, or interactional synchrony during courtship initiation).

Second, the work group, family group, or team has a modal group size of 5 individuals and a range of 3–7 people. It refers to interactions in small face-to-face groups that have a common task orientation. Examples of small group tasks are foraging, hunting, and gathering. A modern world example would be the completion of an assignment by ad-hoc groups of students working together for brief periods in class (e.g., devising and assembling a basic technological device). The workgroup affords the function of distributed cognition across group members. This refers both to the sharing of cognitive resources (e.g., perception, knowledge, cues, focus of attention, inference, classification) in the pursuit of a shared representation of the task or problem and to a division of cognitive effort where there are specialties in the group over time allowing role-based trust.

Third, the deme, band, microband, or conceptual deme has a mode of 30 individuals with a group size range of 25–50 people. Note that Marilynn Brewer (a close collaborator of Caporael) proposed a group size range of 50–200 people for the deme/community in her 2015 keynote at the ICSD conference in Hong Kong, China [24]. The deme is similar in size to the extended family and modern-day classrooms. Common tasks of the deme are the movement from one location to another in hunter-gatherers, providing college students skill-based education (physical or cognitive), and the integration and redistribution of resources retrieved from smaller workgroups (e.g., meat from the hunt or research results from the lab). This configuration has the function of constructing a shared reality or worldview, a common bond identity and common knowledge. It also allows cooperative alliances to emerge which may lead to the breakaway of group members to form their own group in case of conflict or when the environment’s carrying capacity is reached.

Finally, the macrodeme or (seasonal) macroband has a modal group size of 300 individuals with a range of 100–500 individuals. Brewer [24] has instead provided a range of “300–1000 and beyond” people. The task of the macrodeme is the (seasonal, annual) gathering of people (bands of hunter-gatherers, scientists, businessmen) in the pursuit of exchanging resources, people, or information about more distant places or groups. The between-group mobility of people may involve mates in the case of hunter-gatherers or staff in the case of business or science. The function of the macrodeme has been proposed to be the stabilization and standardization of language. Stabilization indicating words referring to the same thing and standardization indicating that the members often use and understand the same words.

Whereas the dyad can rely on synchronization and reciprocity to obtain goals [25,26] and whereas workgroups can rely on mutual performance monitoring [27] and self-regulation [28] to achieve their objectives, members of the deme have to resort to other mechanisms to sustain cooperation and coordination. Several mechanisms have been proposed in the literature. These include informal sanctions [29]; indirect reciprocity [30]; and descriptive norms [31]. However, whereas in demes people are all individually known, this is not the case in macrodemes. As such the likelihood of knowing someone’s reputation may be insufficiently high for indirect reciprocity to serve as an effective mechanism to foster and maintain cooperation and coordination in very large groups [32]. In addition, it has been argued that there are mental constraints preventing people to keep track of the reputation of a large number of people at the same time [13,33]. Hence, other mechanisms such as proscriptive norms, institutions, and formal sanctions [34,35] may be necessary for people to work effectively together toward the accomplishment of their goals in macrodeme configurations.

Four reasons for the prevalence of the dyad

More than larger sizes, dyadic interactions enable benefits from direct reciprocity [26]. In repeated dyadic interactions, the threat of non-reciprocation supports cooperative strategies such as tit-for-tat [26,36], but this mechanism breaks down in larger groups where freeriding is possible [37]. In a dyad, such strategies are efficient, as one’s choices are noticeably affecting one’s own and other’s outcomes. The partner may readily perceive that the other prefers stable cooperation over mutual selfishness, where exploitation would meet swift retaliation. However, even the small step from dyads to triads causes complexities in the workings (and effectiveness) of reciprocity, as one’s retaliatory action can no longer be delivered to the desired target only, the third person is equally affected [38]. When group size increases further, one is increasingly less able to induce cooperation in others through strategic signalling. One individual’s actions are both less likely to be perceived, and less likely to affect the others’ outcomes [37]. Thus, being part of a dyad (compared to larger sizes) allows for more control over the social situation toward the accomplishment of mutually beneficial outcomes.

Second, the detection of emotions and mental states via nonverbal cues is most likely to occur in dyads. Yet as group size increases, it becomes increasingly challenging to attend n-1 communication channels in a group comprising n members [39]. Moreover, as group size increases, the number of interpersonal linkages along which coordination may be required increases sharply [39]. Indeed, coordination with multiple individuals is computationally complex and therefore individuals should prefer interaction partners with whom coordination is easier [40]. Coordination is more efficient with familiar others as learning about others’ preferences, intentions, and traits allows for improved behavioural anticipation [41]. As mentioned above, dyads are expected to be in a better position to attribute mental states (e.g., intentions, emotions, beliefs, desires, knowledge), to be clearer communicators, and hence to be behaviourally more predictable than larger groups. Thus, if people prefer those with predictable behaviours as interaction partners, such as familiar others [41], people may also have a preference for dyadic activities given their relatively predictable form and hence lower cognitive effort, compared with larger group sizes.

Third, group living provides various benefits. These include cooperation in the pursuit of difficult tasks, protection against danger, directing and receiving altruistic acts toward and from kin, the availability of allies, and high concentrations of mates [42]. However, group life also comes with liabilities, mostly in the form of conflict and competition. Examples include competition over material resources, high-status positions, or romantic partners. As such, individual competitors may want to exclude others who pose a threat to their interests (e.g., through derogating one’s competence and appearance or through spurious accusations [42]). Being excluded in an ancestral environment yielded dire prospects for survival and reproduction. Therefore, it is not surprising that (the threat of) social exclusion gives rise to anxiety [43] and social pain [44] and that people may be afraid to deviate from the group. Moreover, it has been argued that humans are highly sensitive to actual and threatened rejection and may possess an ostracism-detection system biased toward overdetection [45,46]. Whereas in larger groups social exclusion is a possibility, sometimes a threat, in a dyad one cannot exclude the other person without bearing the cost of becoming alone oneself. Seeing others together when one is alone may trigger affective distress, being a reminder of (the threat of) social exclusion [43] and as such is uncomfortable. Thus, even if one is generally embedded in a larger group and as such reaping its benefits, the dyad specifically may provide a relatively comfortable unit for social interaction in which self-monitoring and self-censorship can be somewhat relaxed, allowing for more authentic behaviour (i.e., in line with one’s own preferences and idiosyncrasies).

The final argument that may help explain the prevalence of the dyad in daily life concerns reproduction and infant-rearing activities, which take place mostly in pairs [5,6]. As mentioned above, this includes interactional synchrony during the initiation of courtship, which is a process confined to the mating couple [47]. For instance, it is difficult to conceive how three people would be able to nearly continuously look into each other’s eyes. Moreover, most people choose to engage in sexual activity with one other person during a sexual encounter. Next, the provision of infant nutrition through breastfeeding typically involves the mother and the infant in a breastfeeding dyad. Finally, alloparents (e.g., the grandmother [48], great-aunt or an older sibling [49]) can assist the parents in infant-rearing by watching over or feeding the child, allowing the parents to allocate their time in the pursuit of other activities. We argue that even with extensive alloparenting, the dyad may still be the most functional unit for interaction between caregiver and young child (e.g., allowing for facial imitation [5,6] and more efficient feeding).

Yet how do these theoretical reasons connect to the activities that were probed in the present studies? Our activities may have had low risk of exploitation, low costs in terms of commitment, and involved the exchange of resources likely of little consequence. However, the four proposed mechanisms we suggest are all attuned to the two-person interaction in daily life. This applies to reciprocity where the interactants engage in exchange and turn-taking. For instance, paying for the drinks at the movies, one may expect the other to pay next time. Reciprocity can happen “on the spot” (e.g., mutual self-disclosure), but more frequently unfolds in a step-by-step manner. As social transactions recur and gradually expand in significance, reciprocity results in fortified interdependent social bonds [10] (in the words of Allen et al. [50]: strong pair-wise ties). Besides trust, these processes are often accompanied by a series of emotions that may serve as internal pressures to maintain interpersonal cooperation: feelings of indebtedness, personal obligation, appreciation, and gratitude after receiving a favour [51]; feelings of guilt when failing to reciprocate [52]; anger when receiving substandard exchange [53], and forgiveness to pardon having been short-changed [54]. If the above line of reason is correct, it suggests the interesting hypothesis that various feelings and emotions such as indebtedness and gratitude are experienced most commonly and most strongly toward one other person instead of toward groups.

Second, in terms of coordination, in a dyadic interaction one enjoys relatively noise-free information. Compared to larger group sizes, dyadic interactions offer less room for hiding (e.g., question evasion, self-concealment [55]) and ambiguity, and this facilitates coordination. For instance, when a group of friends are planning a holiday trip, it may require significantly more effort to agree on a departure date and destination, not to mention all the small decisions to be made once there, compared with a dyad planning a similar trip together. Third, in a range of situations (e.g., holidays, projects, bars), being in a dyad may also entail being tied to another person, incentivizing both persons to be accommodating and cooperative to prevent social exclusion and having to fend for oneself in a sometimes daunting world. In a bar or club, for instance, it may feel uncomfortable to have no one to fall back on if needed. Finally, dyadic interaction also allows for relatively unconstrained flirting behaviour compared to situations where third-parties are present [56], reflecting the role of reproduction in the prevalence of the dyad in daily life. During a dyadic conversation, for example, a subtle courtship attempt (e.g., a wink or prolonged eye contact) can more freely be sent without potential interception costs from third parties. These mechanisms help to explain the primacy of the dyad because each of them supports cooperation in smaller groups–by which all individuals involved benefit. This is not to imply that cooperation does not occur in larger groups. We suggest that it is more challenging, however, and it involves qualitatively different mechanisms, such as indirect reciprocity [30] or third-party punishment [57]. In this empirical paper we echo the importance and prevalence of the (cooperative) dyad in society that has recently been demonstrated in mathematical and simulation analyses by Allen et al. [50].

Sex differences

For our mixed-sex diary data (Study 5) we did not find significant sex differences. There were some significant differences for mixed-sex group activities sampled in Studies 1–3, but these were generally small. However, for same-sex groups in Study 4, the sex differences were somewhat more pronounced, where women are more dyadic than men. This applies for instance to sports (36.6% vs. 21.1%) or going to the movies (55.7% vs. 42.2%) and to a lesser extent going on holiday (49.9% vs. 43.1%) or out for dinner (37.8% vs. 32%). This latter finding is consistent with research on same sex friendship [58]. Using a sample of 112,000 Facebook profile pictures the authors found women to favour dyadic relationships, while men preferred larger male groups. Moreover, it has been demonstrated that men and women process social information differently in line with these differences in social structures, where women focus more on individuals while men focus more on groups [59,60]. Although the exact mechanism is unknown, various complementary explanations have been put forward to account for these sex differences (e.g., patrilocality [61]; by-product of pair-bonding [62]; maternal caregiving and empathic potential [63,64]).

Limitations

Studies 1–4 have three limitations in common that should be taken into account. The situations that were sampled may not have been exhaustive or fully representative of daily life. Although we asked for the last time someone engaged in the activity, for some people it may have been weeks ago. Participants may also selectively remember an instance they particularly enjoyed. These concerns were addressed by Study 5, which used a real time approach in which the reference period is the present, or the last hours, and in which situations are sampled randomly. A limitation of Study 5, however, is that we did not code the situations that people reported. This is a fruitful avenue for future research given that situation taxonomies are currently lacking consensus and may be advanced by diary studies.

The current research focused on social activities and hence we have not probed individuals conducting activities alone [14,15]. Therefore, we do not know whether dyads would also be most frequent compared to individuals for various different tasks. In our studies, most participants engaged in activities with a few other people. However, there are certainly activities that require larger groups (e.g., barn-raisings). Future research could investigate whether even there, dyads may form the most effective subcomponents (“you hold the spike while I swing the hammer”). Indeed, given the aforementioned arguments (reciprocity, coordination, social exclusion, reproduction), these activities may also predominantly yield the dyad as the most common subunit in a larger collective. Related to this, because we focused on direct interaction, and used a definition of a group as “two or more persons who are interacting with one another in such a manner that each person influences and is influenced by each other person” [23], our results do not address the fact that groups can have psychological significance beyond direct interaction [65] and measuring groups differently may yield different results.

Future research

One issue remaining for future research is to provide empirical evidence that the dyads that are so prevalent in various domains are also stable over time. Moreover, future research would do well to investigate the prevalence of the dyad in collectivistic daily life, to determine whether dyad inflation is generated by cultural factors as a necessary requirement for the effect to occur or whether an evolutionarily ingrained predisposition is sufficient in and of itself. If the former holds, then the effects should be observed only in individualistic countries, whereas if the latter holds, as argued by Caporael’s core configuration model [5,6], then the effects should be observable across individualistic and collectivistic cultures, and more generally, around the world.

Study 1

All studies reported were approved by our faculty’s ethical review board (Vaste Commissie Wetenschap en Ethiek; VCWE). Participants indicated approval with our informed consent form which was displayed on a computer screen. In order to obtain a representative sample of the Dutch population we collected data through Flycatcher (www.flycatcher.eu), an online research company in the Netherlands. The Flycatcher panel consists of 16,000 participants who frequently participate in research. Participants receive points for every questionnaire that they complete. The accumulated points can subsequently be exchanged into gift vouchers.

Study 2 and 3

In Study 2 and 3 we aimed to replicate Study 1 while changing the population from Dutch to American participants and changing the data collection platform through which the responses were collected, from the Flycatcher Panel to an online crowdsourcing service: Amazon’s Mechanical Turk (https://www.mturk.com). In addition to the five activities sampled in Study 1, we added holidays as a sixth activity. In Study 2 respondents were asked to recall past activities in which they took part, while in Study 3 participants were asked to imagine themselves in a hypothetical future activity. The studies were run at the same time; participants were assigned to either Study 2 or to Study 3 after accepting our HIT (Human Intelligence Task) on M-Turk.

Study 4

In the previous studies we did not identify strong sex differences between men and women in the group sizes that were reported. It can be argued that no sex differences are expected in these situations as many or all participants may have reported on mixed-sex situations. Therefore, in Study 4 we aimed to focus on sex differences between men and women in the group sizes they report in various activities by asking explicitly about same-sex daily activities and by including two additional activities in which sex differences may be more likely: sports activity and visiting bars and clubs.

Study 5

In Studies 1–4, we asked people to report on a wide variety of common situations. However, these categories may not have been exhaustive, and may not fully represent daily life. We therefore used experience sampling to obtain reports on random social situations. Moreover, in Studies 1, 2, and 4 we used retrospective reports to investigate group size. Although we asked for the last time someone engaged in a particular behaviour (e.g., going out for dinner), for some participants the last event may have been weeks ago. Finally, participants may remember a dinner that they especially enjoyed rather than the last dinner per se. To address these memory-related concerns we decided to use a real-time data capture approach in which the typical reference period is brief (i.e., the last hours, or even the present [66]).