The Influence of Context and Player Comments on Preschoolers’ Social and Partner-Directed Communicative Behavior

ABSTRACT To successfully navigate their social worlds, children must adapt their behaviors to diverse situations and do so in a fluid fashion. The current study explored preschool-aged children’s sensitivity to a gameplay context (cooperative/competitive) and messages from another (fictional) player (team-oriented/self-oriented) while distributing gameplay resources. To understand children’s approach to social behavior within these contexts, we focused on whether these factors affected 1) the number of resources children provided to the other player and 2) children’s verbal responses to other players. Children (4 to 6 years-old, N = 118) first provided verbal responses to audio messages, then completed a resource distribution task. Children’s verbal responses were influenced by both context and the other players’ messages; however, there was a greater influence of players’ messages in a competitive context. In contrast, children’s resource distributions were influenced primarily by the context (greater sharing of resources in the cooperative context). Children with better ToM showed a greater shift in their distributive behavior across conditions, specifically, distributing more items to the other players within a cooperative context relative to a competitive context. Also, within a cooperative context, children with better EF generated more prosocial comments for the other player. Together, the findings highlight the interplay between contextual and interpersonal factors with children’s cognitive skills for prosocial behavior.

social approach in response to distinct contextual and interpersonal factors prior to a resource distribution involving game pieces. Capturing both what children do and say, our primary foci were children's distribution of game items as well as their partner-directed communication, that is, their comments to other players. Importantly, we also explored whether children's cognitive skills were associated with these outcomes, and whether the contextual and interpersonal cues (i.e., messages from the other player) moderated associations.

Distributive contexts
Within the variety of social behaviors, children's division of resources among themselves and others (i.e., sharing, resource allocation) represents a widely studied (Jackson & Tisak, 2001) and everyday activity (Nancekivell, Van de Vondervoot, & Friedman, 2013). Although their distributions are initially indiscriminate, as they get older, children become more selective with whom they share resources (Warneken & Tomasello, 2009). With greater social-cognitive understanding and experience, children increasingly use various factors to assist in this differentiation.
Resource distributions can be particularly costly and effortful endeavors (Malti et al., 2016) relative to other prosocial behaviors, like helping or comforting. Beyond the resource costs of parting with highly desirable items or those already in one's possession (i.e., fewer items for oneself), there may also be social costs associated with inappropriately generous distributive behavior. Against the backdrop of competition, for example, sharing with an opponent may suggest that social norms and expectations are not understood. Three to 5-year-old children are sensitive to and communicate about social norms, including within games (e.g., Dahl & Kim, 2014;Gockeritz et al., 2014), and will protest violations of such norms (Hardecker, Schmidt, Roden, & Tomasello, 2016;. For example, children object when others breach the jointly agreed upon rules that facilitate competition, even at the cost of winning a desired prize . Thus, competent distributive behavior relies on a nuanced appreciation of social climate. There have been consistent findings showing that children's understanding and application of fairness principles (as per general distributive behavior and comments about the distributions) evolve throughout childhood. Mainly, sharing and resource distribution behaviors increase in frequency with age, and come to reflect increasingly complex principles of equality (i.e., all parties receive equal amounts; e.g., Blake & McAuliffe, 2011), need (i.e., those in greater need receive a greater share; Rizzo & Killen, 2016), and merit (i.e., those who work harder deserve more; e.g., Hamann, Bender, & Tomasello, 2014). Children as young as 3 years old understand, promote, and expect an equal or fair division of resources between two parties, yet they often have difficulty enacting these fairness principles before reaching school age (Smith, Blake, & Harris, 2013;Ulber, Hamann, & Tomasello, 2015).
In addition to what children do, research has examined what children communicate about distributions. Much of the existing literature examines children's verbal responses to others' distributions, such as protests and agreements regarding perceived fairness (Wörle & Paulus, 2018). For instance, 5-to 6-year-olds selectively affirm and protest social partners' behavior accordant with norms of charity, relative to 3-to 4-year-olds (Wörle & Paulus, 2018). However, social interactions also require the coordination of behavior (e.g., Hamann, Warneken, & Tomasello, 2012) and negotiations with partners (e.g., Ram & Ross, 2008). Children become increasingly sophisticated at employing social and communicative strategies to synchronize their actions with those of their social partners after the age of 3 years (Warneken, Gräfenhain, & Tomasello, 2012).
In sum, children develop increased sensitivity to social conventions surrounding resource distribution and demonstrate this through their distributive behavior and communication regarding the distribution. To more fully understand how children navigate these situations, it is important to explore the influence of social and contextual cues (and combination of cues).

Influences on children's distributive behavior
Aspects of the distributive setting may offer cues for appropriate behavior. Past work has modified the conditions under which children make allocations (e.g., suggestions that an agent would like resources; altering agent characteristics, such as friendliness, generosity, age, gender) to determine how children's distribution may vary because of such factors. In the present study, the two focal aspects were the task goals (cooperative versus competitive) and a social partner's comments.

Contextual cues: cooperative versus competitive task goals
A major distinction in social contexts is whether they involve competition or cooperation. Competition and cooperation are defined by the characteristics of a situation (i.e., the "rules"; Richard et al., 2002). Inherent to competitive tasks is the goal of outperforming opponents. Such tasks are interpreted as zero-sum, wherein only one party can succeed (e.g., Johnson, Maruyama, Johnson, Nelson, & Skon, 1981). Cooperative tasks, on the other hand, entail working with others to reach a commonly held goal. Given that competitive and cooperative tasks have different overarching aims, there are different expectations for behavior; what is considered an appropriate advancement strategy when competing (e.g., taking objects from others) may be less appropriate in cooperative settings.
Sensitivity to task goals emerges early and strengthens with age, with prior research showing that children understand the difference between cooperation and competition by the age of 3 years . At this age, children can work together with peers in cooperative contexts and show fewer self-serving behaviors (Huyder, Nilsen, & Bacso, 2017). By 5 years old, children hold a greater appreciation for competitive contexts and understand that competitors work alone to reach their own goals . Correspondingly, Nilsen and Valcke (2018) found that context goals impacted both preschoolers' (aged 4-6 years) and school-aged children's (aged 7-9 years) overall sharing: items required to win a game were more readily shared by participants with those described as collaborators versus competitors. However, these findings emerged in the absence of social partner input or feedback-an unlikely scenario in a natural setting such as a playground or classroom. It may be that children's context-dependent behavior is further shaped by the nature of a social partner's utterances (e.g., a friendly gesture), a premise examined in the current work.

Interpersonal cues: social partner
Even in the absence of cooperative or competitive contextual cues, children become increasingly selective with their resource distributions, attending more to partner characteristics as they get older. Specifically, partner-related factors such as familiarity and reciprocity provide important cues that facilitate or limit children's distributive behavior (Martin & Olson, 2015). Given that reciprocity could lessen the perceived costs of sharing (Sebastián-Enesco & Warneken, 2015), it is no surprise that children become increasingly aware of a social partner's distributive behaviors over time: Two-year-old children share regardless of whether a partner returns the favor, but 3-year-olds are less willing if a partner never reciprocates (Martin & Olson, 2015). With age, children also become more likely to distribute more items to known others, such as friends, compared to strangers (Yu, Zhu, & Leslie, 2016). However, a commonly used resource distribution paradigm involves children distributing a finite amount of "windfall" resources or prizes to an unfamiliar, absent peer (e.g., Warneken, Lohse, Mélis, & Tomasello, 2011). Without information about a peer's sharing tendencies or track record, the key information children use to make distributive decisions is absent.
What a partner says can provide a window into their social intentions. Moreover, given that verbal communication may precede action in naturalistic distributive contexts, it is reasonable to believe that it may serve as an important clue about a social partner's future behavior, particularly in the absence of behavioral cues. As noted above, existing research has mainly focused on children's interpretations of or verbal reactions to others' distributions (e.g., responding to an already completed transaction; Wörle & Paulus, 2018), rather than how children modify their own distributive or verbal behaviors in accordance with utterances from others. While post-distribution comments can provide us with important information about children's perceptions of resource distributions, the partner-directed communication prior to a resource distribution are also informative.
First, verbal utterances from a social partner prior to a distributive interaction can provide children with critical information about their partner that they can use to adjust their distributive approach. In other words, a partner's comments (e.g., "I want all the stickers") may impact a child's subsequent resource distribution behavior (e.g., shares fewer stickers with partner). Second, partner-directed communication from a child (e.g., "That's not very nice") in response to their partner's comments (e.g., "I want all the stickers") can provide important information about children's perceptions of a partner, providing a backdrop against which their subsequent distributions can be interpreted. In this vein, consideration of the partner-directed communication that occurs prior to the resource distribution allows us to tease apart the influence of a partner's messages from that of the context or individual differences in children's distributive tendencies.
Given that communication has been shown to facilitate the synchronization of children's actions with those of their social partners , there is reason to believe that both children's distributive behavior and partner-directed communication would be impacted by a partner's comments. Past work has examined the reciprocal nature of children's behavior within competitive and cooperative contexts (e.g., Huyder & Nilsen, 2012). During interactive puzzle tasks, Huyder and Nilsen (2012) found that dyads of 5-to 8-year-old children showed strong reciprocity in cooperative behaviors (e.g., giving each other puzzle pieces) and verbal statements (e.g., "We're going to win!"), but not competitive behaviors and verbal statements, regardless of context. In other words, children's cooperative behaviors were related to their partners' cooperative behaviors, which suggests that children modify their prosocial behavior based on the behavior of their partner in a reciprocal fashion. However, the observational design employed by Huyder and Nilsen did not isolate the impact of verbal cues from behavioral cues, therefore limiting the conclusions about the unique influence of what a partner says.

The role of social and cognitive skills in children's distributive and partnerdirected communicative behavior
Children's distributive behavior and partner-directed communication within resource allocation situations also depend on their cognitive skills, such as executive functioning (EF) and Theory of Mind (ToM). That is, the ability to think about and reason with social information in a more complex and nuanced manner facilitates children's ability to learn, understand, and strategically apply behaviors (Ding, Wellman, Wang, Fu, & Lee, 2015). These skills, given their importance in children's social awareness and behavioral regulation (Best, Miller, & Jones, 2009;Longobardi, Spataro, & Rossi-Arnaud, 2019;Weimer et al., 2021) may be particularly pertinent when adapting behavior to a variety of contexts and social partners. However, in keeping with the Beauchamp and Anderson's (2010) Social-Cognitive Integration of Abilities model, while ToM and EF independently show associations with children's social behavior (Imuta, Henry, Slaughter, Selcuk, & Ruffman, 2016), contextual factors impact these relations (e.g., Huyder et al., 2017). Thus, to fully understand the role these factors play, research needs to account for, or in the present work, strategically manipulate, contextual factors.
Within the context of distributing resources, children with better inhibitory control at 30 months of age shared more stickers with others when they were 5 years old, potentially allowing them the ability to inhibit the tendency to keep limited items for themselves (Paulus et al., 2015). In addition, preschool (but not school-age) children with better EF skills showed greater willingness to provide future social partners with resources (Nilsen & Valcke, 2018). A gap in this literature, however, is the degree to which EF skills are required for resource distributions with particular players and/or in certain contexts (e.g., when in more socially challenging situations, such as working with peers who demonstrate less collaborative behavior). Moreover, while EF shows associations children's ability to use communication strategies in social interactions (i.e., pragmatic skills; e.g., Blain-Brière, Bouchard, & Bigras, 2014), the degree to which the context may affect associations is unknown.
Given the theoretical importance that EF plays in children's ability to navigate differing social contexts (Best et al., 2009), we expect that well-developed EF skills will contribute to more contextually appropriate behaviors, including their partner-directed communication, particularly increased prosociality within cooperative contexts. Within a competitive setting, we expect that the EF abilities required to self-regulate (e.g., inhibit prepotent responses) are relied upon less, as this context may align more with self-interested motivations.

Theory of Mind (ToM)
Theory of Mind (ToM) is a socio-cognitive ability that allows for the attribution of mental states to oneself and others, and facilitates the understanding that others have mental states that may differ from one's own (Premack & Woodruff, 1978). ToM has an interesting relation to resource distribution. On one hand, better ToM facilitates more sharing (Takagishi, Kameshima, Schug, Koizumi, & Yamagishi, 2010;Wu & Su, 2014). Yet, ToM also enables children to recognize divergent views of social partners, wherein children with stronger ToM skills share fewer resources during a dictator game (Cowell, Samek, List & Decety, 2015) and engage in more "poaching moves" (i.e., taking more resources from a competitor) during resource distribution (Priewasser, Roessler, & Perner, 2013). Thus, better ToM skills can be associated with different social behavior depending on the task goals.
Using a within-subjects design, the present study addresses these diverging results by exploring how children adjust their behavior across contexts and how their ToM contributes to their ability to do so. Specifically, we might expect that better ToM converges with cooperative cues, enabling a greater appreciation of both the cooperative task goals and a team-focused partner and leading to more generous item distributions. However, there may be opposing forces at play when the cues are competitive in nature. Although children with more sophisticated ToM skills may have a greater appreciation of their competitor's divergent goals, they may simultaneously better understand that they are bound by the cooperative structure of competitive games . Thus, an association may not emerge within competitive contexts.
While EF and ToM are considered distinct, it is important to note their associations. The "emergence" account of ToM development posits that EF is a precursor to ToM (Moses, 2001) in that to take on another person's perspective, one must be able to mute their own perspective, shift their thinking, and hold both perspectives in mind. The development of EF enables children to attend to and understand mental states, with early EF predicted ToM understanding throughout the preschool (Carlson, Mandell, & Williams, 2004) and schoolage years (Lecce & Bianco, 2018).

Present investigation
Prior research suggests that context (task goals and social partner) and individual differences in ToM and EF are independently associated with preschoolers' sharing behavior (Nilsen & Valcke, 2018;Wu & Su, 2014). The current study builds on past work by presenting task goals and player-message cues together, and explores which cues (or combination of cues) are most salient in guiding both what children do (distributions) and say (partner-directed communication). Further, we explore how cognitive skills relate to children's distributive behavior and verbal partner-directed communication. A withinsubjects design and mixed linear modeling allow us to investigate how children shift their behavior across task goals and players, and whether associations with cognitive skills emerge within particular social contexts.
To meet these goals, we created a resource distribution scenario wherein 4-to 6-year-old children determined who, between themselves and a series of (fictional) peers, should receive items that were integral to the execution of a series of later games. This age range was chosen due to increasing sensitivity to cues within resource distribution scenarios (e.g., Schmidt et al., 2016). So as to limit children's prior exposure to specific games as a confound, our resource distribution task was not intended to imitate a specific game situation. Thus, the task provided children with a unique experience that allowed us to quantify the degree to which children enact social behaviors (i.e., resource distribution, partner-directed communication) within a specific set of circumstances.
More specifically, the distributive context was manipulated in two ways: first, children were informed of the task goals, such that they understood they would either be working with (i.e., cooperating) or playing against (i.e., competing) others; second, prior to distributing gameplay items, children heard audio messages from other (virtual) players, which either conveyed a willingness to collaborate or not (i.e., team-oriented vs. self-oriented messages). We selected this early time point within the resource distribution process to share the audio messages to better understand the directionality of the relationships between the partner's verbal statements and children's subsequent resource distribution and partner-directed communication (as measured by their verbal responses). By examining both these variables, we could assess the causal, and perhaps differential, impact of the comments from social partners on different aspects of social behavior. As well, our task allowed us to explore how children's own individual characteristics are associated with their responses to the different contexts.
We anticipated that children would shift their behavior across contexts such that greater distributions would be made toward teammates (cooperative context) versus competitors (competitive context) and that greater resources would be provided to players who offered team-oriented messages. However, as the meaning of speech is dependent on the context in which it is uttered (Mishler, 1979), we anticipated these two factors may interact. For instance, a self-focused statement may be interpreted as particularly negative in a competitive context, leading to greater resource hoarding. Regarding children's verbal responses toward partners, we anticipated similar main effects wherein children would generate more prosocial statements toward teammates and players who sent team-oriented statements, but that greater reciprocity might be found for statements in certain contexts (e.g., team-oriented messages in a cooperative context). Finally, it was anticipated that better cognitive skills would enable children to demonstrate greater shifts in distributions/verbal responses across contexts, for example, appropriately providing more resources and generating more prosocial statements within cooperative (versus competitive) contexts. As past work revealed gender differences in both distributive (Dunham, Baron, & Carey, 2011;Rizzo, Elenbaas, & Vanderbilt, 2018;Zaleskiewicz & Helka, 2011) and communicative choices (Mewhort-Buist, Nilsen, & Bowman-Smith, 2020), participant gender was also examined.

Participants
Participants were 137 children (aged 4-6 years) recruited from local kindergarten and grade one classrooms in a mid-sized Canadian city. Parents who provided consent also completed a demographic survey, and were offered the opportunity to choose whether their child would subsequently participate in a school setting or in the laboratory. Children's verbal assent was obtained at the initial testing session. One-hundred and thirty-one children participated at school, while 6 children participated in the laboratory.
Participant data were excluded for children under 48 months of age (n = 1) or older than 84 months (n = 2). Data were also excluded for participants who experienced examiner error (n = 4), did not provide assent to participate (i.e., their parent-reported demographic data was excluded; n = 3), or had incomplete data for the resource distribution task (i.e., rescinded assent; n = 5). The distribution task was considered "complete" when there were data for at least one of two trials per condition (as described in more detail below). Finally, children whose parents reported neurodevelopmental concerns (i.e., Autism Spectrum Disorder, Fetal Alcohol Spectrum Disorders) were also removed from analyses (n = 4). Thus, 19 participants were removed, leaving N = 118 participants ranging in age from 50.50 to 83.20 months (M age = 66.59 months; SD = 7.78; Mdn age = 67.85 months; 55% female).
Of the parents who provided information regarding their child's ethnic background (n = 103), 13% identified their children as Canadian (with no other information); 31% indicated their child was Caucasian/Canadian, White, or from a European background; 27% indicated they were from one or more Middle Eastern, Asian, or South Asian backgrounds; 6% indicated they were from Central or South America; and 11% reported having multiple backgrounds. Seventy-nine percent of parents indicated that their child had spoken English since birth, although 51% of families reported speaking other languages in the home. Sixtyseven percent of mothers and 59% of fathers reported having completed at least an undergraduate, professional, or graduate degree.

Materials and procedure
This project was approved by the Research Ethics Committee at the University of Waterloo. During a 45-to 60-minute session, participants completed tasks in a standardized order: 1) resource distribution, 2) inhibitory control, 3) working memory, 4) cognitive flexibility, and 5) ToM tasks.

Resource distribution task
The design. The resource distribution task was a 2 (task goals: cooperative versus competitive) x 2 (player message: self-oriented versus team-oriented) repeated measures design. Conditions were counterbalanced across participants. The dependent measures were children's resource distributions (i.e., number of items distributed to their social player) and verbal responses (i.e., messages sent back to the player).
Task introduction. Seated side-by-side at a table, the experimenter described a series of 8 "games" (i.e., trials; two per condition) that the children presumably would later be playing on a tablet (i.e., Google Pixel C Tablet). Participants were told that they would play these games with four different children of the same age and gender (one at a time; see Figure 1b) via the tablet once everyone was logged in (i.e., two games with each player). To ensure that conditions were consistent, the other players were fictional in nature. Players were depicted on-screen as silhouettes during the individual instructions for each game and were described as being from different schools to ensure that children did not extrapolate preexisting relationships.
While "waiting" for the other players to log in, the objectives of all games were explained to participating children. They were tasked with "setting up" games by preemptively distributing required gameplay items. This distribution process, in anticipation of the game, was where the dependent measures for our purposes came from. Although specific gameplay items differed, the overarching theme for each anticipated game was explained as interacting with another player and manipulating 25 items as quickly as possible (e.g., put leaves on a tree, build a tall tower, assemble a scrambled puzzle). Children were explicitly told the criteria for winning each anticipated game (e.g., put the most leaves on the tree, build the tallest tower). Although the speed of the anticipated game was emphasized, a specific timeframe was not disclosed (i.e., "Work as fast as you can . . . there probably won't be enough time to [complete the specific activity]"). This was done to allow for the consideration of context-specific strategies during the distribution (e.g., dividing items equally with a teammate to maximize efforts within the allotted time). Children were granted as much time as needed to distribute the items.
When the anticipated game had cooperative task goals, the experimenter introduced the other (virtual) player as a "teammate" with whom they would be working to complete the task in order to win prizes for both team members. When task goals were competitive, the experimenter told children that they would be working independently against the other player for a single prize.
Participants also received an audio message from each virtual player via the tablet (see Figure 1c) after hearing instructions for that specific condition, but prior to distributing items. Audio messages were prerecorded voices of 5-and 6-year-old actors (4 boys, 4 girls). Eight messages, categorized into two types: 1) self-oriented, and 2) team-oriented, were matched in content, except for the pronoun (e.g., "I hope I do really well"/"I hope we both do really well"). Participants were then asked to reflect on what they thought their partner would do, send a message, and then distribute items.
Within the pre-game distribution, we measured children's distributive behavior and verbal responses (described further below) to determine whether our manipulated variables Figure 1. Example of the sequence of screens shown to children for a game within the task (i.e., tree game). First the game is presented, and children complete a demonstration (a). Next the player silhouette (b), then the message notification (after which, the participant sends a verbal response back) (c), and finally the resource distribution screen (d).
(task goal -cooperative versus competitive; partner message -self-versus team-oriented) influenced their behavior. (See Supplemental file for full instructions).

Partner-directed communication (i.e., verbal responses).
Children were invited to generate a verbal response for the player before distributing resources. Responses were transcribed verbatim and later coded. Participants could generate one response (without word limits) for each received message (i.e., two responses per condition). Individual means were calculated for each participant who provided at least one response, based on the number of responses they provided. For those responses, mean response lengths ranged from 2.40 to 15.75 words (M = 6.39, SD = 2.60).
Participants' responses were coded into 12 categories, which were subsumed under three major groupings. Previous literature has coded children's verbal utterances within collaborative settings using categories based on valence (e.g., negative, neutral, positive; Stanton-Chapman & Snell, 2011) and relevance (e.g., task-related vs. task-unrelated; Castellaro & Roselli, 2015). The current study adopted a more detailed, hybrid coding scheme, wherein "Prosocial" responses were positively-valenced messages coded as Pro-Team, Pro-Other, Friendly Overture, and Agreement. "Antagonistic" responses included negatively-valenced messages coded as Pro-Self, Anti-Other, and Disagreement. These two response categories were our measures of partner-directed communicative behavior. 1 We also recorded "Neutral" responses, which were non-valenced, and included responses not directly related to gameplay (i.e., Random; e.g., "I like unicorns"), Acknowledgments (e.g., "Thanks"), and two very rarely used statements: those that were negative toward oneself (i.e., Anti-Self) or one's team (i.e., Anti-Team). Instances where children did not respond were also noted. Given that Neutral responses provide less insight into children's communicative intentions, we focused our analyses on Prosocial and Antagonistic verbal responses.
Categories were mutually exclusive, however, when responses involved multiple components, each component was coded separately. The primary coder was a research assistant who was unaware of the research hypotheses and conditions. A second researcher, who was also unaware of the specific conditions, coded the responses of all children (100%) to ensure reliability. Reliability between coders ranged from good to excellent (see Appendix for descriptions of the codes and reliability analyses).

Resource distribution.
Finally, participants allocated the game materials among themselves and the other player by sliding items across the tablet screen into one of two boxes (see Figure 1d). Reflecting the key measure of their resource distribution, the number of items they moved to the other player's box on each game set up (/25) was recorded.

EF and ToM tasks
Inhibitory control. To assess children's ability to inhibit prepotent responses, a computerized version of the Red Dog/Blue Dog task (a Stroop-like task modified from Beveridge, Jarrold, & Pettit, 2002) was administered. Participants were shown two dogs named "Red" and "Blue," whose fur colors were incongruent with their names (i.e., "Red" had blue fur). Children called out each dog's name as it appeared on screen for 3 seconds, with 1 second between trials. Total scores ranged from 0-28, with higher scores indicating stronger inhibitory control.
Working memory. Children's verbal working memory was assessed using the Digit Span subtest from the Wechsler Intelligence Scale for Children-Fourth Edition (WISC-IV; Wechsler, 2003). Participants repeated digits in the same sequence as they were heard, then a second series of digits in backwards order. Both tasks (i.e., forwards, backwards) consisted of 8 items with 2 trials each and were discontinued when children gave incorrect responses for both trials of the same item. Scores were summed to create a total score, ranging from 0 to 32.
Cognitive flexibility. The Object Classification Task for Children (OCTC), developed for children aged 3-to 7-years (Smidts, Jacobs, & Anderson, 2004), examined children's ability to shift mind-set and think flexibly. Following at least one practice trial, children completed test trials consisting of categorizing 6 objects into 2 groups, with each group sharing a common feature (i.e., color, size, and function). If children were unable to do so, the experimenter sorted the objects and asked children to identify similarities and/or asked children to sort into specific groupings. Children received a total score ranging from 0 to 12 points.
Composite EF score. As per previous research suggesting that EF abilities within the preschool period are represented by a single construct (e.g., Brocki & Bohlin, 2004), as well as critiques of using a latent variable (Camerota, Willoughby, & Blair, 2020), a composite score was created by calculating the mean of the standardized scores (z-scores) of all the EF tasks. 2 The composite reflected the three measured components of EF, which past work has found to be undifferentiated within this age range (Hughes, Ensor, Wilson, & Graham, 2009;Wiebe, Espy, & Charak, 2008;Wiebe et al., 2011). Within this sample, Digit Span significantly correlated with both Red/Blue, r(94) = .38, p < .001, and OCT r(95) = .23, p = .022, but Red/Blue and OCT were not significantly correlated r(92) = .07, p = .51. The consistency of the composite did not improve greatly when any measures were removed, thus the three components were retained. (2004), which is considered a reliable measure of children's understanding of mental states (Wellman, Fang, & Peterson, 2011). Children's understanding that two people may have diverse desires (DD), diverse beliefs (DB), and access to knowledge (KA), as well as their understanding that a person may have a belief that differs from reality (i.e., contents false belief; CFB) and can display emotions that differ from their internal state (i.e., hidden emotion; HE) were assessed. As the conditions of the resource distribution task involved varying contextual and verbal cues regarding children's social partner's (potentially divergent) motivations, we chose a ToM task that assessed children's ability to understand that individuals can hold different mental states. Sub-tasks included control questions, which had to be answered correctly to receive full credit for the subsequent target questions. Target questions were worth one point each, with the DD, DB, and KA contained one target question, while the CFB and HE tasks contained two questions, which were both factored into the score (with this approach departing from the traditional scoring of Wellman and Liu (2004) with the rationale of obtaining a more sensitive measure of children's performance in the more complex sub-tasks). Target item scores were summed across the ToM sub-tasks, resulting in total ToM scores ranging from 0 to 7.

Data analytic plan
To account for the dependency of observations (i.e., multiple trials per participant), we used mixed linear modeling (MLM). The models were fit using lmer package in R. We defined a random intercept for each of models, with all observations nested within each participant. Model fit was estimated using Marginal R 2 (i.e., fixed effects only). The data that support the findings of this study are available from the corresponding author, EN, upon reasonable request.

Preliminary analyses
Preliminary analyses revealed no significant difference in children's distributive behavior with respect to order (competitive vs. cooperative first; p > .14); therefore, order was not included in further analyses. The frequencies for each verbal response category (across conditions) were as follows: children generated a total of 536 prosocial responses (Mean = 4.54, SD = 3.01), 294 antagonistic responses (Mean = 2.49, SD = 2.42) and 328 neutral responses (Mean = 2.78, SD = 2.75).
Children's mean performance on the EF tasks, which were amalgamated into a composite score for each participant, was as follows:

Main analyses
Nine models were tested, three per dependent variable (i.e., resource distribution, prosocial responses, antagonistic responses), using MLM. Each dependent variable was modeled as a function of age (in months), gender, task goal, message, ToM, and EF. Prior to analyses, age, ToM, and EF were centered at the grand mean. The first exploratory model for each set of analyses included only main effects, while the second included all possible two-way interactions. To obtain the most parsimonious model, non-significant predictors and interactions were trimmed, retaining the two primary variables of theoretical interest (i.e., task goals and message). The main effects and trimmed models were used for interpretation. Two sets of coding schemes were used for categorical predictors (i.e., gender, task goal, message); dummy coding for main effects models and effects (or sum) coding for models with interaction terms. Thus, gender was coded such that girls were "0" (or "-1") and boys as "1." Similarly, task goal was coded such that cooperative context was "0" (or "-1") and competitive context as "1." Message was coded such that team-oriented messages was "0" (or −1) and self-oriented messages as "1."

Resource distribution
We first examined the average number of items (/25) distributed to other players. The intraclass correlation coefficient (ICC) was calculated using a baseline (or null) model with a random intercept (Hox, 2002). The ICC value of .40 suggests that 40% of the variance can be explained by using repeated measures of the same participant. Marginal R 2 for the main effects model was .09, while the Marginal R 2 for the trimmed (i.e., main effects + significant interactions only) model was .11.
As hypothesized, the main effects model demonstrated that there was a significant main effect of task goal on resource distribution, ß = −.37, t(747.48) = −7.14, p < .001, 95% CI for main effect of task goal [−0.47-−0.27], such that there were more resources distributed to the other player in the cooperative context than the competitive context. Contrary to our prediction, we did not find evidence that message type impacted children's resource distribution (p = .43). With respect to cognitive skills, EF had a significant association with children's resource distribution, p = .04, wherein children with better EF offered more items to others relative to those with weaker skills. This finding remained in the final trimmed model, ß = .18, t(104.09) = 2.78, p = .01, 95% CI [0.05-0.31]. No further predictors emerged as significant (ps > .26) in the main effects model (see Table 1 for estimates).
The inclusion of key interaction terms in the final trimmed model (see Table 1 for estimates) revealed that the within-person relation between task goal and resource distribution was moderated by ToM, ß = −.11, t(745.52) = −4.22, p < .001, 95% CI [−0.16-0.06]. Simple effects were examined post-hoc at one standard deviation above and below the mean of the continuous predictor, ToM (see Figure 2). Children with both high (+1SD) and low (−1SD) ToM shared more items overall within cooperative contexts relative to competitive contexts (high: p < .001; low: p = .03). However, the extent of this difference was approximately 3.76 times greater for those with high ToM than low ToM, suggesting that those with better ToM show greater degree of adaption across context. Looking more closely within each context, one unit increase of ToM was related to significantly more generous distributions within the cooperative context, b = 0.72, SE = 0.29, t = 2.48, p = .01. Conversely, one unit increase in ToM did not relate to distributive behavior in the competitive condition (p = .58). Taken together, these results suggest that ToM skills facilitate the allocation of more items when it is called for by the context. There was also an interaction between task goal and gender on resource distribution, ß = −.13, t(745.50) = −5.03, p < .001, 95% CI [−0.18 --0.08]. A post-hoc pairwise comparison, averaged over the levels of message, was completed to better understand the nature of this interaction. As shown in Figure 3, boys and girls both adjusted their distributive behavior across contexts, with both genders offering significantly more items to collaborators (girls: M = 11.12, SE = 0.46; boys: M = 11.99, SE = 0.54;) than competitors (girls: M = 10.35, SE = 0.46, t(745) = 2.24, p = .025; boys: M = 8.56, SE = 0.54, t(745) = 8.59, p < .001). However, the degree to which boys adjusted their distributive behavior across contexts was approximately 4.5 times larger than girls, suggesting that boys are more greatly impacted by context. In this vein, boys and girls did not differ within the cooperative context (p = .

Partner-directed communication
Prosocial verbal responses. The baseline model with random intercept for prosocial verbal responses generated an ICC value of .33, which suggested that 33% of the variance can be explained by the repeated measures of the same participant. Marginal R 2 for the main effects and trimmed models were both .12.
These main effects were qualified by a significant interaction between task goals and message, ß = -.07, t(332) = -2.12, p = .03, 95% CI [-0.14 --0.01] (see Figure 4a). Pairwise comparison analyses revealed that message type did not significantly impact the prosociality of children's verbal responses within the cooperative condition (p = .10). However, selforiented messages received within the competitive condition were particularly impactful, such that children generated fewer prosocial verbal responses (Mean = 0.66, SE = 0.10) compared to following team-oriented messages (Mean = 1.14, SE = 0.10), Mean Difference = 0.48, SE = 0.10, t(332) = 4.63, p < .001. This suggests that the specific impact of the competitive task goal may either be buffered or intensified by the type of message received. As shown in Figure 5, we also found that EF significantly moderated the effects of task goals on prosocial verbal responses in the trimmed model, ß = -.11, t(332) = -3.06, p = .002, 95% CI [−0.17 --0.04]. The interaction suggests that within a cooperative context, better EF skills resulted in more prosocial verbal responses  Error bars represent ± 95% CI. Note: the terms included in the trimmed models differed for prosocial and antagonistic responses (see Tables 2 and 3).
relative to those with weaker EF skills, b = 0.37, SE = 0.11, t = 3.48, p < .001. Within a competitive context, children were overall less likely to generate prosocial verbal responses, with EF not showing an association with the frequency of children's prosocial verbal responses, p = .52. Note. Significant estimates are denoted by * p < .05. ** p < .01. Figure 5. Frequency of prosocial responses as a function of EF composite and context. Bands represent ± 95% CI. Interrupted vertical Orange lines indicate EF score ± 1SD below and above the mean.
Antagonistic verbal responses. Finally, we modeled antagonistic verbal responses as a function of the abovementioned predictor variables. The ICC value for the baseline model with random intercept was .14, which suggested that 14% of the variance can be explained by the repeated measures of the same participant. Marginal R 2 for the main effects model was .17, while the Marginal R 2 for the trimmed model was .16. Like our findings for prosocial verbal responses, there were significant main effects of task goals, ß = . 61,t(322) = 7.94,p < .001,and message,ß = .54,t(322) = 6.98,p < .001,]. There were more antagonistic verbal responses being generated in the competitive context and following self-oriented player messages, relative to the cooperative context and following team-oriented messages, respectively. However, in the final trimmed (main effects + significant interactions only) model (see Table 9), message moderated the effect of task goal on the generation of antagonistic verbal responses, ß = .08, t(351) = 2.24, p = .03, 95% CI [0.01-0.16]. As can be seen in Figure 4b, children generated significantly more antagonistic verbal responses in the cooperative condition following a self-oriented message (Mean = 0.53, SE = 0.09), compared to a team-oriented message (Mean = 0.14, SE = 0.09), Mean Difference = −0.39, SE = 0.11, t(351) = −3.65, p < .001. Within the competitive context, children showed the same response pattern (Self: Mean = 1.28, SE = 0.09; Team: Mean = 0.55, SE = 0.09), albeit to a greater degree, Mean Difference = -0.73, SE = 0.11, t(351) = -6.82, p < .001. Moreover, when exploring difference across context within message type, a similar pattern was observed whereby team-oriented messages and self-oriented messages were responded to differentially according to task goal (team: p < .001; self: p < .001). Interestingly, the degree of the difference across task goals was larger (Mean Difference = -0.75) following receipt of a self-oriented message (Mean Difference = -0.42), relative to a team-oriented message. Together, the competitive task goal with a self-oriented message induced more antagonistic verbal responses.
Children's cognitive skills were not found to moderate the conditions for this response. That is, children's antagonistic verbal responses in the different conditions did not vary by their cognitive skills (see Table 3).

Discussion
Our aims were to assess how task goals (i.e., cooperative vs. competitive) and messages from another player (i.e., self-vs. team-oriented) individually and collectively impacted children's social behavior, namely resource distribution and partner-directed communication (i.e., verbal responses). As well, we sought to determine how EF and ToM were associated with behavior within the various conditions. Findings from this work provide crucial information as to how different aspects of children's social behavior (in our study, what children say and what they do) is differentially impacted by contextual cues, as well as how children's cognitive skills influence the extent to which children adapt their behavior to these changing social contexts. Thus, in line with current theories of children's social development, we find evidence for important interplays between the social context and the child's internal characteristics (e.g., Beauchamp & Anderson, 2010).

Children's resource distributions
The number of items that children distributed to the other players was influenced by the task goals, as predicted. All children distributed more game resources to players in the cooperative context, relative to the competitive context (albeit less than an equal distribution in both contexts). This work extends previous studies demonstrating that context impacts children's collaborative versus competitive behaviors during interactive puzzle tasks (Huyder & Nilsen, 2012) and replicates recent work showing that resource distribution is impacted by contextual task goals (Nilsen & Valcke, 2018). Interestingly, the hypothesis that the type of player message would impact children's resource distribution was not supported. It is unlikely that children were insensitive to these messages, as their verbal responses showed reciprocation of the message content (discussed further below). Rather, it seems that children decided how much to share based on factors outside of player messages. This non-significant finding is interesting given that past work has found that children do modify their sharing behavior (albeit a different paradigm) based on the behavior of a fellow player (e.g., Martin & Olson, 2015). From the current findings, it may be that certain gameplay norms regarding resources supersede the influence of messages. As well, children may have felt that it is appropriate within a gameplay context for players to speak competitively. Reflecting this notion, older preschoolers show skepticism of competitors who do not "compete" in typical ways, but rather show collaborative acts . Thus, in the current study, the other players' self-focused messages may have been interpreted as normative gameplay talk, and, as such, were less relevant to resource distribution.
Though gender was not of primary interest, the main effect of task goal was qualified by gender. Both boys and girls generally shared more items in the cooperative versus competitive context, but boys shared fewer items relative to girls, specifically in the competitive context. That is, in the condition where there was incentive to keep more items for oneself, girls were more willing to share than boys. This aligns with existing literature demonstrating that boys tend to be more selfish in their distributions (Rizzo et al., 2018), but are more willing to part with resources when stakes are low (Posid, Fazio, Cordes, 2015). As well, boys might have played more strategically due to increased experience with competitive games (Weinberger & Stein, 2008).

Children's partner-directed communication
In contrast to their distributions, children's verbal responses reflected sensitivity to both task goals and message type. Beyond offering assurance that children were indeed processing the messages from the players, the two main effects and interaction of task goals and message highlight that the degree to which children's verbal responses were influenced by the other player, as well as how this influence differed by task goal. As expected, children generated more prosocial verbal responses after receiving a team-oriented versus a selforiented message, and in the cooperative versus the competitive context. The interaction further confirmed our prediction that certain combinations of context and message were particularly stimulating. When children received self-focused messages from peers, the salience of the context was greater.
Prosocial statements communicate greater willingness to work in a partnership or maintain social relationships with others. Such a stance would be particularly conducive to success in the cooperative context, which (presumably) entails collaboration. In the competitive context, "trash talk" might be expected over politeness. Children seemed to recognize this, as their verbal responses were generally tailored to the context. However, message type played a role such that the difference in children's prosocial verbal responses across condition was more pronounced following self-focused messages. Put differently, the combination of competitive context and self-focused message from a partner was particularly detrimental to the generation of prosocial verbal responses to their partner.
Consistent with this, children's antagonistic verbal responses, that is, those that were selfpromoting, unfriendly, or disagreeable, showed a similar pattern, albeit in the reverse direction. First, children responded with more antagonistic utterances within a competitive context than a cooperative context. Given that competition entails working alone to achieve a goal, fostering a relationship with an opponent is not necessary for success and, therefore, children might be more inclined to assert themselves through disagreement or self-promotion. Second, in both contexts, children more frequently generated antagonistic verbal responses following a self-focused message from a peer versus a team-focused message, showing some reciprocity in valence. Like prosocial verbal responses, the interaction between message and context indicated that a competitive context with a self-oriented message resulted in more antagonistic verbal responses.

Relations between cognitive skills and social behavior
Children's resource distribution EF emerged as an important predictor of resource distribution: children with better EF offered more items to other players, regardless of task goal or message type. There was also a significant interaction between context and ToM on children's distributions, where those with better ToM gave more resources to potential teammates versus competitors. Whereas in the competitive condition ToM showed no associations with distributions, children with high (relative to low) ToM distributed more in the cooperative context. Thus, children with high ToM demonstrated more flexibility in their distributions across task goals than did their peers with low ToM, with this shift due to increased provision of resources to teammates (i.e., versus penalizing competitors). Likely, children with greater ToM abilities were better able to reason that their other teammates had similar goals in mind and, accordingly, understood that sharing was advantageous. In addition to facilitating children's social behavior in other realms, such as helping, cooperating, and comforting (Imuta et al., 2016), the present study highlights ToM's unique role for greater sharing of resources in anticipation of a collaborative task. These findings also extend past work in which preschoolers with better ToM showed more collaborative behaviors during a cooperative task (but not a competitive task; Takagishi et al., 2010).

Children's partner-directed communication
Children's EF skills showed associations with prosocial verbal responses primarily in a cooperative context. In this context, children with better EF provided more prosocial verbal responses to other players, including statements conveying agreement. Thus, children with greater EF skills met the demands of the situation more effectively based on the contextual cues. Such a finding underscores the importance of EF for socially appropriate behavior. That is, past work has found that those children with better EF tend to have more positive peer exchanges (Ciairano, Visu-Petra, & Settanni, 2007) and behave more cooperatively with others (Paulus et al., 2015). Here we find that their partner-directed communication is also more context-appropriate, specifically more prosocial within a cooperative context.

Limitations and future directions
This research offers preliminary insights into the factors that influence children's social behaviors; however, replication is warranted. As well, this work is not without limitations. One limitation is the artificial nature of the interaction between children and the (virtual) players (e.g., constraining the exchange to one or two sentences; using a novel game), which meant that an involved dialogue was not captured, and the gameplay objectives (e.g., quickly building a tower) may not fully generalize to what occurs on the playground. However, this situation was likely novel for children, which had the advantage of eliminating children's ability to rely on preexisting knowledge of a specific game, thereby allowing us to capture children's behavior as per their sensitivity to cooperative and competitive norms and partner-related cues. Another consideration is that we designed this study to look at distributions prior to a game play context. Thus, results may not be representative of the dynamic offering and withholding of resources that might occur within a game.
Future work would also benefit from the inclusion of a more diverse sample of children, in terms of ethnicity, parental education and/or SES, and age given that a child's broader socio-cultural context may play a role in their resource allocation with respect to willingness, frequency, spontaneity, and generosity. For instance, individual socio-contextual factors that are most often examined in Western research studies of resource distribution include allocator-recipient relationship, recipient behavior, gender, and the allocator's previous experiences, all of which may be differentially influenced by the social norms embedded within other cultures (e.g., personal goals vs. group goals, boundaries vs. connectedness; Rao & Stewart, 1999). It follows, then, that the task goals and player-message cues presented in the current study may be perceived or acted upon differently through the lens of one's culturally valued principles. As a result, it would be important to extend the current work to include a larger and more diverse sample of children to ensure that a variety of cultural backgrounds and perspectives are represented. Moreover, with a larger and more widespread sample, one would also be able to examine if there are important interplays between a child's cognitive skills and broader socio-cultural context in predicting social behavior, and, further, whether this differs across development. Previous work has found that ToM and EF skills show different relations to behavior depending on age (Im-Bolter, Agostino, & Owens-Jaffray, 2016) and therefore additional age groups could be included to determine if there are developmental patterns.
Separately, parent education was quite high in the current sample, and as such, these results may not be representative of children coming from other socio-economic backgrounds. Indeed, past work has found that resource availability impacts children's approach to working with others (Brown, 1996). For example, Safra and colleagues (2016) found that children from lower SES environments (e.g., income deprivation) gave fewer toys away to an unknown child compared to children from middle SES environments. Thus, children who have historically needed to safeguard scarce or less secure resources may continue to protect available resources, even in times of greater abundance.
Finally, a measure of verbal ability should be included in future work as a control variable given that past work has demonstrated associations between ToM and verbal skills (e.g., Astington & Baird, 2005). Thus, controlling for verbal ability would allow for more clearly elucidating the unique role that ToM plays in children's partner-directed communicative behavior.

Conclusions
In conclusion, context goals, but not messages from other players, impacted children's resource distributions. Girls and boys both provided other players with more gameplay items in a cooperative (versus a competitive environment), though, in competitive contexts, girls gave more resources to future opponents than boys. Moreover, children with better ToM provided potential teammates with more game play resources. While not affecting their distributions, the type of comments made by (virtual) players did not go unnoticed by children: when children received self-focused messages within a competitive context, they generated fewer prosocial and more antagonistic verbal responses for peers. Children with better EF generated a greater number of prosocial verbal responses within a cooperative context.
Together, the findings serve to highlight how contextual cues influence children's resource distributions whereas a combination of contextual and interpersonal cues affect their partner-directed communication within such distributions. Further, findings also offer greater insight into the way in which preschool-age children's cognitive abilities are associated with social behaviors but within specific situations, in our work, primarily within cooperative contexts.