For children in peer groups it is important to gain resources, such as material objects or alliances with and support from others (Hawley and Bower, 2016). Children's natural differences in their ability to prevail in resource competition result in social dominance and in their position in the group's dominance hierarchy (Hawley, 1999). Dominance has been defined as an individual's ability to gain access to and/or maintain control over resources (Hawley, 2002) and can be influenced by certain person characteristics. Hawley (2002) found that older children were rated as more dominant by their preschool teachers and were more successful at controlling a valuable resource in a dyadic play situation with a peer. Results regarding an influence of gender on dominance, however, are inconclusive. When observing children's interactions among familiar peers, Charlesworth and La Freniere (1983) found that boys exhibited higher dominance by maintaining access to a valuable toy significantly longer than girls. However, other studies suggest that both genders can achieve high dominance, with females often using more subtle strategies to do so (cf. Pellegrini et al., 2007). That dominance plays a significant role in preschooler's peer interactions is supported by the fact that, as early as 3 years of age, children infer others' dominance via social cues such as age or decision power and expect dominant individuals to be more competent in games and have better access to resources (Charafeddine et al., 2015).

Experimental evidence points to a negative association between children's dominance and their prosocial behavior when sharing resources. Preschoolers that prevail in resource conflicts with peers have been found to donate fewer valuable items to an anonymous recipient (Guinote et al., 2015) and to acquire more resources than their partners in cooperative tasks (Grueneisen and Tomasello, 2017). Similarly, preschoolers with higher teacher-rated dominance had a tendency to withhold valuable items from a familiar peer in a resource allocation experiment (Horn et al., 2024b). In natural social groups, however, children sometimes use seemingly prosocial actions to acquire and maintain resources, such as sharing a non-preferred toy with a peer in order to gain access to a preferred toy (Hawley, 1999). In a longitudinal study conducted with children across a whole year of preschool, dominance was positively associated with aggression and coercion at the beginning of the study period, whereas dominant children were particularly prosocial toward the end of the school year (Roseth et al., 2010). When investigating whether the recipient's dominance affected children's prosocial choices, Charafeddine et al. (2016) found that 3-4-year-olds preferentially donated valuable items to a dominant puppet in a forced-choice sharing task, while 5-year-olds did not show a clear preference. Moreover, more than 90% of the tested 8-year-olds gave a bigger piece of chocolate to the subordinate, suggesting that the influence of moral norms and fairness principles that may underlie sharing gets stronger with age (cf. Schmidt and Sommerville, 2011). These findings suggest the association between dominance and prosociality might not be as straightforward as experimental studies suggest and that socio-ecological factors such as the stability of hierarchical structures in a social group might influence whether and how strongly dominance affects prosocial behavior in preschoolers. Such associations can only be tested in the microsystem of the peer group, where children interact regularly and over an extended period of time. Further, as is shown in Spinrad and Eisenberg's (2023) heuristic model of prosociality, dominance might not be the only modulating factor of prosocial behavior in natural peer groups.

Popularity can be defined as a measure of peer acceptance or likeability, with popular individuals having a higher status and a more central position within the group (McDonald and Asher, 2018). Neither age nor gender were found to have a clear influence on popularity in preschool children (Walden et al., 1999). However, observations of peer interactions suggest that popularity and dominance are linked, with higher dominance often going along with more popularity in children (Hawley, 2002; Vaughn and Waters, 1981). Dominant individuals receive more attention from their peers (La Freniere and Charlesworth, 1983), pointing toward their central position in the group. In addition, gender seems to interact with popularity and dominance. Sebanc et al. (2003) found that in same-sex interactions around a valuable resource, more dominant boys were more popular among their peers than less dominant boys, whereas more dominant girls were less popular.

Several studies indicate that there might be a connection between popularity and prosociality in children. When observing interactions between characters in video sequences, preschool children expect an individual with high regard from the other characters to help more in reaching a joint goal, but at the same time refrain from taking more than their relative proportion from a common resource (Stavans and Diesendruck, 2021). In line with these findings, preschoolers whose popularity was determined via high levels of visual regard from their peers had a higher tendency of choosing prosocial reward distributions in a resource sharing task with a familiar peer (Horn et al., 2024b). Moreover, 8–12-year-old children with high levels of visual regard were found to be particularly prosocial by donating resources and intervening in fights on behalf of the losing opponent (Ginsburg and Miller, 1981). Conversely, in a resource sharing experiment with 6–9-year-old children whose popularity was first measured by counting the number of interaction partners during a group observation period, children with low popularity preferred prosocial resource distributions, both when this choice benefitted another peer and when they were tested alone in a control situation (Horn et al., 2018). This highlights that the connection between popularity and children's prosocial behavior still needs to be investigated in more detail. Particularly the question whether a recipients' popularity influences the likelihood of becoming a target of prosocial actions remains under-researched.

Bronfenbrenner and Morris (2007) describe friendship as an emotional relationship between two individuals, that also shapes a developing individual's surroundings and is part of the way their surroundings are being perceived. Friendship formation occurs early in childhood and early friendships are characterized by spatial proximity and frequent social interactions (Bagwell and Bukowski, 2018). Friendships seem to mediate children's prosocial behavior. Children expect more prosocial acts such as sharing or helping between individuals who are friends than those who are not (Afshordi, 2019; Liberman and Shaw, 2017). Anonymized sharing studies, in which no peer is present but valuable items are said to be delivered to a specific other child, suggest that preschoolers are more likely to share with friends than peers that are not their friends. Paulus and Moore (2014) found that 3–5-year old preschoolers expected sharing to occur more often toward a friend than a disliked peer in an individual story-telling task and also shared more with a toy figure used as a stand-in for a friend than with a toy representing a disliked peer. In a setting, where an experimenter asked 3- and 5-year-olds to share stickers with a friend from their kindergarten group, a hypothetical child who would join the kindergarten group the next day or a hypothetical stranger, participants of both age groups gave more stickers to their friends than to any other category, regardless of whether they expected their prosocial act to be reciprocated or not (Lenz and Paulus, 2021). However, Berndt (1981) found that while children claimed they would share more with friends than acquaintances when being asked, actual sharing behavior did not differ between friends and non-friends when the children were paired with familiar peers in direct interactions. Several other studies in which preschoolers were tested in dyadic direct interactions with familiar peers similarly failed to find preferential sharing with friends (e.g., Horn et al., 2024b; Messer et al., 2017). Helping, on the other hand, seems to be influenced by friendship, even when looking at direct interactions between peers. In a forced-choice task conducted in triads of familiar peers, 3-year-old children preferentially provided help to the peer they had previously identified as their friend in a sociometric interview than toward the other, neutral peer from their kindergarten group (Engelmann et al., 2019). The same children also provided more help toward a friend in a second task, where they were paired either with a friend or a neutral peer from their group (i.e., children cleaned up a larger amount of paper shreds for their friends; Engelmann et al., 2019). These findings imply that some relational factors might affect preschoolers' sharing and helping behavior differently, indicating that it would be valuable to test and compare these two types of prosocial behavior and investigate how they are used among familiar preschool peers.

Existing research suggests that young children's dominance, popularity, and friendships influence their prosocial behavior. However, most of this evidence comes from laboratory studies testing children individually or in dyads. Only few studies have investigated how children use prosocial behavior in groups of familiar peers (e.g., Charlesworth and La Freniere, 1983), where interaction partners have pre-existing relationships that guide not only children's decisions about whether to help and share or not, but also which group members to direct these prosocial actions to. Moreover, while most studies and theoretical models placed a strong focus on actor characteristics predicting prosocial behavior, few studies have taken into account whether the recipient's characteristics might also modulate an acting child's prosocial behavior (e.g., Charafeddine et al., 2016; Paulus and Moore, 2014). Therefore, it remains unclear how prosocial decisions are made in preschoolers' natural environment—their peer group microsystem (Bronfenbrenner and Morris, 2007)—where dominance, popularity, and friendship are all important aspects of their everyday social life.

Guided by a developmental-ecological stance, the aim of the current study was to assess preschoolers' prosocial behavior, as well as their dominance, popularity, and friendships in groups of familiar peers and test how actor and recipient characteristics influence sharing and helping behavior. By testing participants in their respective preschools, in their socio-ecological niches with familiar peers, and in naturalistic settings, our research was conducted in an ecologically valid environment (cf. Bronfenbrenner, 1974; Read and Szokolszky, 2018). We first assessed peer friendship with age-appropriate sociometric interviews. In group game settings, we tested dominance in a contest and a scramble competition situation, as well as two types of prosocial behavior (i.e., sharing and helping). We chose games because games are a familiar part of preschoolers' everyday lives and we could at the same time carefully design these games to assess dominant and prosocial behavior. As external measures for dominance, popularity, and prosociality, we handed out questionnaires to preschool teachers. At each point during our age-appropriate experimental games, children could freely choose not only whether they wanted to engage in prosocial behavior or not, but also who would (or would not) become the target of their prosocial actions.

In the first step, we examined the predictors of children's dominance in peer groups. We hypothesized that dominance would increase with age (Hawley, 2002). We further hypothesized that there would be an interaction effect of popularity and gender, with more popular girls being less dominant and more popular boys being more dominant (Sebanc et al., 2003). For our main research question, we examined whether dominance and popularity of the actor and the recipient, as well as friendship between the children predicted sharing and helping. We further controlled for effects of actors' and recipients' gender and age. Considering actor characteristics, we expected higher dominance to decrease children's sharing and helping behavior (Guinote et al., 2015; Grueneisen and Tomasello, 2017). We further predicted that higher popularity would be associated with more sharing and helping (Ginsburg and Miller, 1981; Horn et al., 2024b) and that the children would display more prosocial actions toward friends than toward children who were not their friends in the group setting (Engelmann et al., 2019; Lenz and Paulus, 2021). We assumed that there would be an effect of gender, with girls being more prosocial, as well as an effect of age, with older children sharing and helping more often (for a meta-analysis, see Fabes and Eisenberg, 1998). Regarding recipient characteristics, there was very little literature from which we could derive our hypotheses. Due to popular children's central position in the social network, we expected higher recipient popularity to favor becoming the target of sharing and helping (McDonald and Asher, 2018). Similarly, we predicted higher dominance to favor becoming the target of prosocial actions among preschoolers (Charafeddine et al., 2016). We could not formulate clear hypotheses regarding the influence of recipient age or gender on becoming the target of prosocial actions, but chose to explore these factors in our analyses.

The study consisted of three parts: (1) a sociometric assessment of popularity and friendship within the peer group, (2) three experimental sessions with age-appropriate games that were designed to elicit dominant and/or prosocial behavior in a group context, and (3) a teacher questionnaire, designed to provide external validity of children's popularity, friendships, dominance, and prosociality. The sociometric assessment was always conducted one day before the first experimental session. The experimental sessions were conducted 4–14 days apart (Mdn = 7). Each experimental session lasted ~30 min and consisted of three games. In the first game, designed as a contest competition, children could show dominance by gaining control over a limited resource (i.e., one novel toy that could only be used by one child at a given time). They could also show prosociality by passing the toy to another peer (i.e., “sharing”). In the second game, we measured dominance in a scramble competition. Children had to queue to obtain stickers of varying desirability and could show dominance by taking positions at the front of the queue. The third game was a freeze-unfreeze game, where children could show prosociality by “unfreezing” a peer, thereby allowing that peer to resume playing the game (i.e., “helping”). Teachers were asked to name a maximum of five best friends per child and rated each child's perceived prosociality and dominance.

In total, 108 3–6-year-old children from 14 preschool classrooms from five different childcare facilities in Vienna participated in the study (53 females, 55 males; age: M ± SD = 60.61 ± 11.47 months, min = 38, max = 80). There was no significant age difference between females and males [Welch's t-test: t (105.72) = 1.74, p = 0.084; females: M ± SD = 62.75 ± 10.68 months; males: M ± SD = 59 ± 11.68 months]. The children were tested in 11 groups of familiar peers, consisting of 6 to 13 children each (Mdn = 11). All children aged 3–6 years of the 14 preschool classrooms were invited to participate in the study. Final group size and composition of the tested groups depended on parental consent for participating in the study. The majority of groups was composed of children that attended the same preschool classroom with three exceptions: due to a low number of participants from six classrooms, we combined children from two classrooms that knew each other well and regularly interacted with each other into three suitable peer groups, respectively (i.e., groups C, I, and J). Due to individual absences, not all children took part in all sessions. One hundred children participated in the sociometric assessment and at least one of the game sessions. Two additional children did not participate in the sociometric assessment, but participated in at least one game session. Of these, 63 children participated in all three game sessions. Six additional children participated only in the sociometric assessment.

To assess popularity and friendship, children participated in two rounds of sociometric assessment [adapted from Birch and Billman (1986)]. We chose a sociometric approach, because it provided an easy but reliable and age-appropriate measure for peer relationships in children. Each child was tested individually in a separate room at their childcare facility. To ensure peer recognition, we first showed the participant portrait photos of each other participating child from their group and asked them to name the peer depicted in the photo. In the first round, the participants were asked to assign each photo individually to one of three categories indicated by colored faces (green smiling face = “I like to play with this child,” red frowning face = “I do not like to play with this child,” yellow neutral face = “I do not know/care, whether I like to play with this child”). In the second round, the experimenter shuffled the peers' photos and handed them to the participant again. The participant was then asked to assign all photos corresponding to a specific category to one of three piles next to the three respective faces, until each photo was assigned to a category. During all instructions, the word “friend” was omitted to avoid socially desired answers (e.g., the participants stating that all peers are their friends). Overall, children showed substantial agreement in their ratings over the two rounds of the sociometric assessment (weighted κ = 0.69). In the various groups, agreement ranged from moderate to perfect (M ± SD = 0.67 ± 0.15, min = 0.41, max = 0.96).

To assess popularity, we used a standard procedure (e.g., Birch and Billman, 1986; Horn et al., 2024b; Ladd and Coleman, 1997) where we calculated each participant's acceptance and rejection scores by first counting the number of green and red nominations, respectively, that they had received from all other group members during the sociometric assessment. We then standardized the score by group size by dividing it by the number of all group members who rated the participant (i.e., group size −1). The minimum reachable score was 0 (i.e., no nominations by other children in the respective category) and the maximum was 1 (i.e., all nominations were green or red, respectively). Acceptance scores ranged from 0.05 to 1 (M ± SD = 0.47 ± 0.20) and rejection scores ranged from 0 to 0.85 (M ± SD = 0.24 ± 0.18). Both scales showed good to excellent internal consistency between first and second round (acceptance score: Cronbach's α = 0.90; rejection score: α = 0.88). Acceptance and rejection scores were moderately negatively correlated (r = –0.64, p < 0.001). Due to slightly higher internal consistency and broader range, we chose the acceptance score as our main variable for popularity. Females received significantly higher popularity scores than males (t (99.98) = 2.16, p = 0.041; females: M ± SD = 0.51 ± 0.19, males: M ± SD = 0.43 ± 0.20).

As friend nominations (N = 379) we counted those instances where a participant had allocated a peer to the green face in both rounds. All other nominations (N = 623) were summarized in a “non-friend” category. Children reported having between 0 and 12 friends (Mdn = 3). There was no difference in the number of friends reported by females and males (t(103.78) = 0.23, p = 0.819).

The first experimental game was designed as a contest competition over a single valuable resource (cf. Pellegrini, 2008), which is a widely used method to assess dominance in social dominance research (Charlesworth and La Freniere, 1983; Grueneisen and Tomasello, 2017). In our game, children got a limited amount of time to play with a novel toy that could only be used by one participant at a time (3 min for groups with <10 participants, 5 min for groups with ≥10 participants). Children sat down in a circle and the experimenter explained that they would get to play with a toy for a certain amount of time. The experimenter explicitly stated that she or he would not interfere during that time, meaning that children had to settle toy possession themselves. When all children agreed to having understood the instructions, the experimenter placed the toy in the middle of the circle and turned around an hourglass indicating the duration of the game. The experimenter informed the children when half the time had passed. At the end of the game, the experimenter collected the toy and put it out of sight. A different toy was used in each of the three experimental sessions. All instructions were given without using the word “share,” to avoid socially desired sharing behavior.

We recorded the total duration of resource possession of each child (i.e., the total duration that the toy was in a specific child's hands during the game). For those children that participated in all three game sessions, resource possession was correlated across all sessions (Spearman's rank correlation: n = 63; sessions 1 and 2: ρ = 0.45, p < 0.001; sessions 1 and 3: ρ = 0.28, ρ = 0.029; sessions 2 and 3: r = 0.54, p < 0.001), indicating good internal consistency of this variable. We also recorded the sequence in which the children obtained possession of the toy as the contest rank (i.e., rank 1 = had the toy first, rank 2 = had the toy second, etc.). If a child took possession of the toy more than once, this respective rank was skipped and the next-higher rank was assigned to the child that got possession afterwards. We then standardized the contest rank by group size by dividing it by the number of participants in the respective session. Children who never took possession of the toy did not receive a contest rank score in this session and were omitted from analyses including this variable (session 1: n = 5; session 2: n = 19; session 3: n = 14). Contest rank was not significantly correlated across sessions (n = 40; sessions 1 and 2: ρ = 0.25, p = 0.124; sessions 1 and 3: ρ = 0.08, p = 0.623; sessions 2 and 3: ρ = 0.08, p = 0.614) and was therefore dropped from further analyses due to insufficient internal consistency. As a variable for prosociality, we counted all instances of sharing (i.e., passing the toy to another child). Sharing was correlated across all sessions (n = 63; sessions 1 and 2: ρ = 0.48, p < 0.001; sessions 1 and 3: ρ = 0.47, p < 0.001; sessions 2 and 3: ρ = 0.51, p < 0.001), indicating good internal consistency of this variable.

The second experimental game was designed as a scramble competition (cf. Pellegrini, 2008), where all children received stickers of varying desirability, but by having to queue, some children got access to the stickers earlier than others (adapted from the ticket paradigm; Pellegrini et al., 2007). While seated in the circle, the experimenter read a picture book in which animal characters had to queue to receive stickers. The first animal in the queue could choose the “coolest” sticker, while the last in the queue had to take a “boring” sticker. Each child was asked two comprehension questions (i.e., “Who can choose a sticker first?”, “Who is the last one to choose a sticker?”). Thereafter, the experimenter told the children that they would have 1 min to form a queue to receive stickers themselves and showed them a box with stickers of varying desirability. The experimenter explicitly stated that she or he would not interfere during that time, meaning that children had to settle their positions in the queue themselves. Upon the experimenter's signal, the children got up and had 1 min to form a queue. When the queue was final, children could approach the experimenter one-by-one and each choose one sticker from the box.

We used the position in the queue to determine each child's scramble rank as a variable for dominance, with lower numbers indicating a better position in the cue and therefore higher dominance. We then standardized the scramble rank by group size by dividing it by the number of participants in the respective session. If a child failed to answer one or both of the comprehension questions correctly, their position in the queue was considered when assigning other children's scramble rank, but they did not receive a scramble rank score in this session and were omitted from analyses including this variable (session 1: n = 15; session 2: n = 6; session 3: n = 3). For those children that received a scramble rank in all three game sessions, scramble rank was correlated between sessions 2 and 3 (Spearman's rank correlation: n = 48; ρ = 0.51, p < 0.001), but not between sessions 1 and 2 (ρ = 0.22, p = 0.141) or sessions 1 and 3 (ρ = 0.04, p = 0.796), indicating limited internal consistency of this variable.

The third experimental game was a “freeze-unfreeze” game that the children played for 5 min. The aim of the game was to balance a plush toy on the head while moving to music. If the toy dropped from a child's head, the child had to “freeze” and stop moving. Another peer could pick up the toy and place it back on the child's head to “unfreeze” them, thereby allowing the child to resume playing the game. The game was designed to not produce any winners, so that children would not have to compete with one another. Two experimenters demonstrated the rules of the game to the children. All instructions were given without using the word “help,” to avoid socially desired helping behavior. When all children agreed to having understood the instructions, they were allowed to choose a plush toy. The experimenter then started the music and told the children that they could play the game until the music ended. If children were so good at balancing the toys that none dropped from their heads, the experimenter suggested “challenges,” such as walking backwards. After the music ended, the experimenter collected the plush toys and children were thanked for participating in this session and sent back to their classrooms.

We recorded all instances of a child picking up another child's plush toy. Instances of helping were correlated across all sessions (Spearman's rank correlation: n = 63; sessions 1 and 2: ρ = 0.58, p < 0.001; sessions 1 and 3: ρ = 0.64, p < 0.001; sessions 2 and 3: ρ = 0.60, p < 0.001), indicating good internal consistency of this variable.

As external measures for children's popularity and friendship, we asked teachers to name a maximum of five best friends from all participating peers for each participant. This method was chosen to ensure comparability to children's ratings while reducing time and effort for the teachers. The teachers also rated each child's perceived dominance and prosociality on a continuous, graphic rating scale ranging from “strongly disagree” to “strongly agree.” There were four items for dominance (i.e., “assertive,” “dominates classmates,” “tells others what to do,” “stands up for self”; Pellegrini et al., 2011) and four items for prosociality (i.e., “helps other children,” “willingly shares with other children,” “offers assistance and comfort to other children,” “seems to be moved by other children's distress”; Ladd and Profilet, 1996). Per preschool classroom, the questionnaire was filled out by two teachers that knew all participants well, with two exceptions: for groups C and E, we could only acquire a single teacher questionnaire.

Popularity was measured by counting how often a child had been nominated as a friend of another child. As with children's nominations, we standardized the popularity score by group size by dividing the score by the number of other group members. Teacher-rated popularity scores ranged from 0 to 0.83 (M ± SD = 0.32 ± 0.20). Overall, teachers' agreement on the popularity scores was high (Cronbach's α = 0.82). For further analyses, a mean popularity score over all raters was calculated for each child. For friend nominations, agreement between teachers was overall moderate (Cohen's κ = 0.58) and ranged from slight to perfect in the various groups (M ± SD = 0.50 ± 0.26, min = 0.12, max = 1). Due to the very low consistency of teacher's friendship nominations for some of the groups, we did not use the friendship nominations for external validity.

Ratings on the continuous graphic rating scale were measured with a ruler and standardized, so that the range of possible values was between 0 and 10. Dominance and prosociality ratings showed good to excellent internal consistency (dominance: α = 0.91; prosociality: α = 0.88). Therefore, for each teacher, a mean rating per child was calculated over the four items of dominance and prosociality, respectively. Agreement between raters was high for dominance (α = 0.84) and acceptable for prosociality (α = 0.73). Dominance ratings ranged between 0.10 and 9.89 (M ± SD = 5.14 ± 2.56) and prosociality ratings ranged between 0.10 and 9.80 (M ± SD = 5.79 ± 2.36). For further analyses, a mean dominance and prosociality score over all raters was calculated for each child.

All hypotheses and predictions were formulated before data collection started, but they were not pre-registered online. No additional data were collected after the analyses. Sample size estimation was conducted with G*power (Faul et al., 2007) and based on the linear models in the first part of our analysis (see details below). Sample size estimation for negative binomial mixed models was not available to the authors at the time of conceiving this study. Sample size estimation for linear models with five predictors yielded a sample size of 92 in order to detect medium effect sizes (f² = 0.15; α err prob = 0.05; 1–β err prob = 0.80). Since based on our previous work in childcare facilities we expected an attrition rate of about 15% between children that received parental consent for participation and those that were actually able to participate in the experimental sessions, we decided to recruit 108 participants. Actual attrition was lower than expected at 5.6% for the main analysis. Therefore, the two linear models were slightly over-powered compared to the sample size estimation. Sample size estimation for bivariate correlations yielded a minimum sample size of 67 in order to detect correlations of medium strength (ρ = 0.3; α err prob = 0.05; 1–β err prob = 0.80). Sample sizes were large enough to meet these criteria in all main analyses. Sample sizes were not large enough when conducting the internal consistency checks of the behavioral variables, since they were conducted with data obtained only from children that were present in all three game sessions (range of n = 40–63). Post hoc analyses revealed that power to detect significant correlations of medium strength was only 1–β err prob = 0.60–0.78 for these internal consistency checks. Statistical analyses were carried out in R version 4.2.1 (R Core Team, 2022). P-values < 0.05 were considered statistically significant and p-values < 0.1 are reported as non-significant trends. Due to the tested groups' similarity in gender distribution (χ² = 16.52, df = 10, n = 108, p = 0.086), age (Kruskal–Wallis: χ² = 12.593, df = 10, n = 108, p = 0.247), and number of friends (Kruskal–Wallis: χ² = 10, df = 10, n = 108, p = 0.441), we used the full sample for statistical analyses and did not include group as a predictor.

To assess construct and external validity, we calculated mean values per child for each variable and compared them to each other and to teachers' ratings, respectively. To assess construct validity for dominance and prosociality, respectively, we calculated Spearman's rank order correlations with duration of resource possession in game 1 and position in the queue in game 2 as well as sharing in game 1 and helping in game 3. To test external validity for dominance, we correlated teacher rated dominance with children's duration of resource possession with a Pearson correlation and children's scramble rank with a Spearman's rank order correlation, respectively. Similarly, we assessed external validity for prosociality by performing a Spearman's rank order correlation for teacher rated prosociality and children's sharing in game 1 and helping in game 3, respectively. External validity for popularity was examined by calculating a Pearson correlation for popularity determined via friendship nominations made by teachers and children themselves. Not all children participated in all parts of this study and/or obtained a score in each of the experimental games (see previous sections). We conducted each correlation with the maximum sample size possible for the respective pair of variables.

To test which factors influence children's overall dominance, we first calculated a mean of the variables duration of resource possession in game 1 and scramble rank in game 2 across all three experimental sessions, respectively. Further, we did a logarithmic transformation of the two variables for dominance, as the data did not follow a normal distribution. We then fitted two linear models with outcome variables duration of resource possession and scramble rank, respectively. Predictor variables were popularity, gender, their interaction, age, and group size.

To test how individual characteristics of the actor and recipient influenced each occurrence of prosocial behavior, we fitted two negative binomial mixed models. Outcome variables were the occurrences of sharing in game 1 and helping in game 3, respectively. Predictors were friendship between interaction partners, as well as dominance, popularity, gender, age of the actor and the recipient of the prosocial action, respectively. As predictor variable for dominance, we chose children's duration of resource possession in game 1 of the respective session, because this variable had greater internal consistency and fewer missing values than scramble rank. We controlled for group size and session by adding them as predictors. The individual was added as fixed effect.

All models were examined for goodness of fit by fitting simulated residuals against those predicted with the DHARMa package for R (Hartig, 2022). With the same package, we did a Kolmogorov–Smirnov–Test, an outlier test and a dispersion test for each model. Models were considered to have a good fit, if no significant deviations were detected in the residuals plot or any of the tests.