How young children use manifest emotions and dominance cues to understand social rules: a registered report

ABSTRACT

Given the complexity of human social life, it is astonishing to observe how quickly children adapt to their social environment. To be accepted by the other members, it is crucial to understand and follow the rules and norms shared by the group. How and from whom do young children learn these social rules? In the experiments, based on the crucial role of affective social learning and dominance hierarchies in simple rule understanding, we showed 15-to-23-month-olds and 3-to-5-year-old children videos where the agents’ body size and affective cues were manipulated. In the dominant rule-maker condition, when a smaller protagonist puts an object in one location, a bigger agent reacts with a positive reaction; on the contrary, when the smaller protagonist puts an object in another location, the bigger agent displays a negative reaction. In the subordinate rule-maker condition, the roles are shifted but the agents differ. Toddlers expect the protagonist to follow the rules (based on anticipatory looks), independent of the dominant status of the rule-making agent. Three-to-five-year-old pre-schoolers overall perform at the chance level but expect the protagonist to disobey a rule in the first trial, and obey the rule in the second trial if the rule-maker is dominant.

KEYWORDS:

• Social rule
• affective social learning
• dominance
• social development
• deontic reasoning

Human societies are unique in their level of complexity. Not only do we live in large groups characterised by dense and intense networks of differentiated relationships, but most of our social behaviours are regulated by complex systems of norms that differ from one group to another. Acting appropriately as a group member is essential not only to coordinate respective behaviours but also because any deviation from norms can risk being ostracised by the other members of the community. Figuring out what these social rules are is therefore crucial for newcomers, and, in particular, the children who have to figure out the rules of the game they will have to play for the rest of their life. In that context, the main question is therefore the following: how do young children learn what is expected from them as future members of a community? How do they make sense of this complex world made of rules that are often somewhat arbitrary (e.g. greeting by shaking hands, bending the upper body, or even touching elbows)?

Affective social learning and social rules

Learning how to behave appropriately in society requires a certain number of abilities. First, young children must take into account others’ behaviour and learn how people react in different contexts. It has been shown that infants do indeed have very early emerging tendencies to detect, prefer, and follow conspecifics surrounding them (Simion et al., 2008). Moreover, very young children show a rudimentary form of gaze following (Farroni et al., 2004; Scaife & Bruner, 1975), being therefore prepared to be attentive to what others consider relevant. This ability is crucial because it helps them to distinguish, among the multitude of stimuli that surround them, those who are visually “selected” by adults.

At around 9–12 months of age, infants begin to share attention with an adult (e.g. declarative pointing, following communicative cues), building a referential triangle between themselves, adults, and objects or events in the external world (Tomasello, 2014; Tomasello & Rakoczy, 2003). Thanks to this triadic relationship, infants’ attention can be oriented to the elements of the environment that adults consider as worth being attentive to. But this is not the whole story: at the same age children also start to consider the emotions expressed by others in their evaluation of the situation. This has been for instance shown in the famous “visual cliff” paradigm, where 12 months old infants utilised facial expressions of their mothers to disambiguate a situation where they were not sure about the potential danger of the situation (Sorce et al., 1985). That is, infants look at their mothers’ faces before crossing the visually ambiguous and “dangerous-looking” side of the plexiglass-covered table where the depth is manipulated as an illusion, and infants crossed the “dangerous-looking” side if their mothers expressed joy or interest in their poses. Such types of social referencing are crucial because they enable the children not only to be sensitive to certain aspects of their environment but to learn, via others’ emotional answers, how to appraise what they are witnessing. Such processes have recently been named affective social learning (Clément & Dukes, 2017; Dukes et al., 2021; Dukes & Clément, 2019) and there is some evidence that infants have early sensitivity to their conspecifics’ affective cues, for instance being able to distinguish some basic emotions of their conspecifics (Bayet & Nelson, 2019). With age, more fine-grained and object-directed behaviour regulation based on others’ affective reactions develops. For instance, while encountering ambiguous objects, 16-months-olds are better than 12-month-olds at regulating their actions based on the emotional reactions of adults (Kim & Kwak, 2011). Moreover, 18-month-olds regulate their actions on novel objects based on observing how others affectively react to those objects (Repacholi & Meltzoff, 2007). These and other studies demonstrate that during ontogeny, infants engage in affective social learning to understand how to behave under uncertainty, and their approach to artefacts, foods, and people are increasingly regulated by affective cues from adults (for a review see, Harris, 2019). For instance, in one study, it was demonstrated that 12 and 16 – month olds change their behavioural reactions (manipulation of objects or looking at the adults more frequently) and affective reactions towards ambiguous toys based on the emotional facial gesture (fearful or happy) of adults (Kim & Kwak, 2011). Fourteen-to-48-months old toddlers and young children shape their approach to novel food based on whether adults tried the food (Harper & Sanders, 1975), and young children’s various behaviours such as approaching a novel object or interacting with a stranger depend on the pre-emotional reactions (whether the novel person or object is equated with positive or negative affect before children approach them) or online emotional reactions (whether the novel person or object is equated with positive or negative affect while children are approaching them) of other adults towards these objects, foods or strangers (Fusaro & Harris, 2008; Harris, 2019; Harris & Nunez, 1996; Skinner et al., 2017, 2020).

Deontic reasoning and social rule-learning

Affective social learning might play an essential role in the process of figuring out what is expected from infants in their social environment, and there is also some evidence showing that infants start to master a kind of deontic reasoning, i.e. reasoning on what may, should, must, or ought to be done (or not to be done) in a social context early in their development (Hilpinen, 2012; Kaufmann & Clément, 2014; Wellman & Miller, 2008). By the end of their second year, they become more sensitive to norms and rules and expect third-party agents to follow the rules (Schmidt et al., 2019; Schmidt & Rakoczy, 2018). But it is generally admitted that it is from around age 3 that children get a richer understanding of social norms as something applying to everyone engaging in a given activity (Kalish & Cornelius, 2007; Rakoczy & Schmidt, 2013; Riggs & Kalish, 2016; Schmidt et al., 2016; Schmidt & Rakoczy, 2018).

The cognitive processes underlying such reasoning are yet not well understood. For many, deontic reasoning is intrinsically related to the theory of mind because children have to imagine how obligations and permissions are represented in social actors to predict their behaviour (Kalish & Cornelius, 2007; Wellman & Miller, 2008). However, others have evoked the possibility that the detection of rules and notably violation of rules are so important in social life that their identification of rules and violations could be linked to simpler and less mentalistic inferences (Cummins, 1996a; Kaufmann & Clément, 2014). One study even showed that pre-schoolers were able to figure out a rule (B) by noticing that a bearer of authority (e.g. a teacher in a kindergarten) displayed an angry face when action A was accomplished by a character (e.g. a child) and a happy face when she was doing B (Clément et al., 2011). However, children in this study were verbally helped by the experimenter who clearly stated that the reason for the teacher’s emotional reaction was due to the character’s action. It is, therefore, possible that language played an important role in this case, as in many experiments where social rules are explicitly stated by the experimenters (Rakoczy et al., 2008; Rossano et al., 2011; Wyman et al., 2009). Moreover, these results were obtained with pre-schoolers. Given the importance of rule-following behaviour from early years on, answering the question of how infants and toddlers utilize others’ non-verbal emotional cues when they have to figure out the social rule underlying a certain situation has theoretical and empirical value.

Even though there might be some norms that might require understanding other’s mental states, what is meant here by “social rule” here is not purely mentalistic, since toddlers and young children do not always or necessarily need to understand other ones’ perspectives to understand most of the social rules, especially very simple rules (Clément et al., 2011; Rakoczy & Schmidt, 2013). Simple social rules require to make the mapping between a particular situation (e.g. a toddler approaches a dangerous object) and the emotional output given by an agent (or agents) following this particular situation (e.g. the mother reacts negatively). Nevertheless, social rule understanding might still require going beyond the negativity bias, namely “the tendency for negativity to have a stronger impact than positivity (p. 68; Norris, 2021)”, since there is a necessity to evaluate not only what is not allowed, but also what is allowed, and what is allowed may be followed by a positive affect rather than negative affect. It is known that even young infants have a negativity bias (for a review see, Vaish et al., 2008). In a nutshell, affective cues are valuable sources of learning social rules.

Dominance hierarchies, deontic reasoning, and social-rule learning

As indicated by Denise Cummins (p. 35), “dominance hierarchies, as a social organization, constitute evidence that members are capable (at least) of distinguishing between what is forbidden and what is permitted under which circumstances” within the same group (Cummins, 1998). Recent research indicates that children possess early emerging tendencies to represent dominance relations (Pun et al., 2017). Following some other researchers, here we position dominance hierarchies as a part of asymmetric relation (Fiske & Fiske, 2007; Spokes & Spelke, 2017) – such as parent–child, teacher-student, leader-follower – in which “higher ranking people are entitled to deference and respect from subordinates; subordinates are entitled to pastoral protection (Fiske & Fiske, 2007, p. 291)”. Infants as young as 6-to-10 months can represent social dominance relations based on group size (Pun et al., 2016), and 10-to-13-month-old infants use relative body size as a cue to represent dominance relation (Thomsen et al., 2011). Infants expect a larger agent to prevail over a smaller agent in a conflicting situation (Thomsen et al., 2011). Preschool children (3-to-5-year old) use numerical age in their evaluations of contests (who would win?) (Pellegrini et al., 2007) and power (who is the boss?) (Charafeddine et al., 2015). Beyond the size information, young children (and infants in some cases) use other affective cues to infer social dominance, such as facial gestures and emotions (lowered eyebrow, tense mouth), body postures, voice pitches, and affective vocalizations, similar to some other species, especially nonhuman primates (Keating, 1985). At 12 months old, they can make multimodal integration of emotion cues considering interactions between a dyad and make predictions based on asymmetric cues (size of the agent). That is when a small and large agent are separated after some close activity, and the small agent cries following the separation, infants look at the larger agent more and expect the larger agent to return to the smaller agent (Biro et al., 2014). Three-to-10-year-old children judge the dominant faces as “strong” and “mean”, and the striking similarity between different age groups’ judgments suggests that from the early years onwards, humans are sensitive to dominance cues in their judgments from the facial features (Cogsdill et al., 2014). Relatedly, 3-to-7-year-old children systematically select faces with a tense mouth (compared to relaxed mouth) or lowered eye-brows (compared to raised eye-brows) as the dominant actor in some stories requiring defining the roles such as the winner, boss, rule maker, or deferrer in conflicts or games (Keating & Bai, 1986). In line with these findings, preschool children use affective and ostensive cues to evaluate the dominance status of individuals (Brey & Shutts, 2015; Charafeddine et al., 2015; Keating et al., 1977; Terrizzi et al., 2019). In summary, infants and children use non-verbal cues to identify dominance, and they have specific behavioural expectations from the dominant characters.

Independent of understanding of dominance, it has been shown that 18-month-olds expect others to follow the conforming actions, and correct them if they do not do so (Schmidt et al., 2019). Moreover, 3-to-5-year-old children understand rules, permissions, and violations in varying scenarios, and are specifically sensitive to detecting rule violators (Cummins, 1996c; Dack & Astington, 2011; Harris & Nunez, 1996; Riggs & Kalish, 2016). Not only children are to be good at detecting norms, but certain research indicates a link between deontic reasoning and dominance (Cummins, 1996b; Kaufmann & Clément, 2014). For example, 21-month-olds follow a social rule indicated by the leader (the agent’s dominant status is shaped by respect and the consensus of the dominant status of the agent by others) – but not the bully (the agent’s dominant status is shaped by physical power and aggression towards other agents) – and follow the rule later in the absence of the dominant leader (Margoni et al., 2018). Regarding understanding authority, when 3-to-9-year-old children listen to scenarios describing different power relations, they overall ascribe dominance to the figure who takes more resources, wins a zero-sum conflict, assigns the permissions, gives orders, and sets norms (Gülgöz & Gelman, 2017). However, the other way around is also true. It is not only the case that children ascribe social dominance to whom has more power, but they also ascribe decision-making and resource control power to the dominant agent: when an agent is dominant (based on individual cues such as size or age, relation history, or as a leader) young preschool children prefer and follow the testimony of this agent in the absence of clear information (Bernard et al., 2016; Castelain et al., 2016), ascribe more responsibility (Stavans & Diesendruck, 2021), allocate more resources, and judge them as the powerful and the mature ones (Charafeddine et al., 2015, 2016).

This study

Based on the existing literature, we propose a scenario to test young children’s ability to understand social rules. We have two research questions: (i) Do young children ascribe the social rule-making role to the authority figure and predict another agent to follow the social rule in third-party interactions? (ii) Do children use affective cues to predict social rules in third-party interactions? The former question is the core research question of this study. The novelty of this study lies in systematically exploring pre-schoolers and toddlers’ use of asymmetric and affective cues in social-rule learning in third-party interactions.

We conducted two studies to answer our core questions, one with explicit (pointing and verbal responses) and another with implicit measures (anticipatory gazes and looking time), with 3-to-5-year-old children and 15-to-23-months toddlers, respectively. We believe that it might be crucial to use various implicit measures, considering the current replication problems of studies using especially anticipatory gaze as a measure and the mismatch between the results of looking-time and anticipatory gaze measures in some other domains such as the implicit theory of mind and goal-understanding (Barone et al., 2019; Byers-Heinlein et al., 2022; Daum et al., 2012; Dörrenberg et al., 2018; Ganglmayer et al., 2019; Havron, 2022; Paulus, 2022). Following the deontic reasoning account, we argue that deontic reasoning capacity develops in the early years of childhood, already in the second year. Considering toddlers’ robust understanding of norm violations, and their capacity to generalise asymmetric relations including dominance relations, the target age group for toddlers was 15-to-23-months old, based on the findings in the literature highlighted in the introduction.

In Experiment 1, 3-to-5-year-old children, and in Experiment 2, 15-to-23-month-old infants were presented with two blocks of animations (dominant rule-maker condition and subordinate rule-maker condition), each block in three steps: (1) goal/action induction, (2) rule-induction, and (3) and test trials. In the dominant rule-maker condition, a protagonist (subordinate – S) with a neutral facial gesture puts some objects into a blue or green box. In the rule-induction familiarisation trial, a bigger agent (dominant-D) enters the stage. While S is heading to one of the boxes, D demonstrates an angry facial gesture and makes an angry vocalisation after S puts the object in that specific box (unapproved box). On the other hand, while is S heading to another box, D’s facial gesture changes into a smiling facial gesture, and D makes approving vocalisation after S puts the object in that specific box (approved box). In the subordinate rule-maker condition (with different agents), the structure and the sequence of the animations are the same as in the dominant rule-maker condition, but the roles of the agents (the rule maker is smaller – subordinate, the protagonist is bigger – dominant) are shifted.

In Experiment 1, we hypothesise the following, which will be tested against the null hypotheses:

H1 (action prediction, and the rule-making function of affective expressions): Three – to five-year-old children predict the protagonist to go to the approved box rather than the unapproved box more than the chance level.

H2 (action prediction and the rule-making function of the dominant agent): Three – to five-year-old children’s action prediction performance (correctly selecting the approved box, namely predicting the rule) is better in the dominant rule-maker condition compared to the subordinate rule-maker condition.

In Experiment 2, toddlers watched the same animations. After the first two steps of familiarisation (action/goal induction and rule induction), infants got the test trial coherently or incoherently. In the coherent test trial, the protagonist followed the rule (go to the approved box), and in the incoherent test trial, the protagonist did not follow the rule (go to the unapproved box). In Experiment 2, we have four hypotheses, which will be tested against the null hypotheses:

H3 (action anticipation and the rule-making function of affective expressions): Fifteen – to twenty-three-month-old toddlers anticipate the protagonist to go to the approved box rather than the unapproved box more than the chance level.

H4 (action anticipation and the rule-making function of the dominant agent): Fifteen – to twenty-three-month-old toddlers’ action prediction performance (correctly selecting the approved box, namely predicting the rule) is better in the dominant rule-maker condition compared to the subordinate rule-maker condition.

H5 (looking time and the rule-making function of affective expressions): Following the affective social learning account, we hypothesize that 15-to-23-month-old toddlers get surprised (look more) if the protagonist selects the unapproved box (incoherent test trial) compared to the approved box (coherent test trial).

H6 (looking time and the rule-making function of the dominant agent): Following the deontic reasoning account and the findings in the dominance understanding literature, we hypothesize that 15-to-23-month-old toddlers get surprised (look more) more if the protagonist selects the unapproved box (incoherent test trial) compared to the approved box (coherent test trial) in the dominant rule-making condition compared to the subordinate rule-maker condition.

The study protocol was approved by the Research Ethics Commission of the University of Neuchâtel. Informed consent was obtained from the parents of the children, and 3-to 5-year-old children were also asked whether they would like to participate in the studies before the experiment started. As compensation, all the participants got a participation certificate indicating their participation in the study and a book or a small toy.

Experiment 1

Methods

Participants

Fifty-four normally developing 3-to-5-year-olds were recruited for this study (23 girls, M_age = 52.5 months, SD_age = 10.3 months, Range_age = 52–71 months). One participant’s exact age in months information was missing. There were 19 three-year-olds, 22 four-year-olds, and 13 five-year-olds. All participants in both studies were recruited from two small cities (Neuchâtel and Bienne) and their surroundings in Switzerland. Children were monolingual or bilingual speakers (French, Swiss German, English, etc.) but all of them could speak in French fluently. Please see the OSF link and the initial registered report for sample size calculation details and the sampling strategy.

Set-up and stimuli

Participants watched some animations from a computer monitor (∼53 cm × 30 cm) in a lab, or a 15-inch computer in a quiet place in kindergarten or school, and the distance between the participant and the screen was around ∼60 cm. After explaining the study, the experimenter asked parents of the 3-to-5-year-old children to wait outside of the lab room or behind their children, at around a 1-metre distance (if parents sat behind, they were warned not to interfere). In the schools/kindergartens, the testing took place in a quiet room in the target institution.

Animations were prepared with Adobe Animate CC 2022, and the sound effects were selected from the Adobe Audition 2022 sound effects library, including the approving voice. The angry voice was selected from a set of validated emotional vocalizations (Sauter et al., 2010). The stimuli were presented with a presentation programme, but interactively (Dack & Astington, 2011). The shapes of the agents and other content in the video stimuli are similar to the ones used in other developmental studies (Kajanus et al., 2020; Mascaro & Csibra, 2012; Pun et al., 2016; Thomsen et al., 2011). For some examples, please see supplementary videos.

Design

The design of the experiment is similar to the preliminary studies on the early development of social dominance (Gazes et al., 2017; Mascaro & Csibra, 2012; Thomsen et al., 2011), and the design of Experiment 1 is built upon Experiment 2 considering the findings in 3-to-5-year-old children in the literature (Castelain et al., 2016; Charafeddine et al., 2015). Each participant was presented with two conditions (within-subject), each condition in three steps: (1) goal/action induction, (2) rule induction, and (3) test trials, in which the first two steps are the familiarisation trials, and the third step is the test trials. The two conditions are as follows: the dominant rule-maker condition and the subordinate rule-maker condition.

Dominant rule-maker condition

In the goal/action familiarisation trials, there is a bucket in the middle of the screen, and there are two boxes (a blue box and a green box). Each box is located close to the right or left lower corner of the screen. During the goal/action induction familiarisation, a protagonist (subordinate – S) with a neutral facial gesture gets an object from the bucket and puts it into the blue or green box, after a brief bell-ring sound (see Movie S1). This action is repeated four times. In the rule-induction familiarisation, a bigger protagonist with a neutral face (dominant-D) enters the stage from the left upper part of the screen. While S is heading to one of the boxes, D demonstrates angry facial gestures, and angry vocalization after S puts the object in that specific box (unapproved box). On the other hand, while is S heading to another box, D’s facial gesture changes into a smiling facial gesture, and D makes an approving vocalization after S puts the object in that specific box (approved box) (see Movie S2). The affective vocalizations are always played on the same side where the dominant agent stays (e.g. the dominant agent is on the left side, and affective sounds come from the left speaker). In the test trial of Experiment 1 (with 3-to-5-year-olds), after S gets an object from the bucket and moves in front of the bucket, children were asked where would the agent go (see Movie 3). To be sure that the voices that were used in the study (angry and approving vocalizations) to convey approving or non-approving information were appropriate for the study, a preparatory study was conducted with 28 adult participants (16 Females, M_age = 29.32), before the Experiment 1 and Experiment 2. Participants evaluated these two vocalizations on various dimensions (arousal, valence, approval, appreciation) on a 9-point Likert scale. Please see Supplementary Material 1 for more information on the preparatory study.

Subordinate rule-maker condition

The executed actions in this condition are the same as the dominant rule-maker condition. However, bigger (D) and smaller (S) agents’ roles are shifted in this condition. That is, D puts the objects into the boxes, and S makes the affective expressions during the rule-induction trials. Note that in both conditions, the colour of the agents, objects, and background, and the shape of the agents and the objects are different, namely, the agents are not the same between conditions, to prevent the carry-over effect. Nevertheless, there is a clear size difference between D and S in both conditions. See Movie 5–10, for the videos in this condition. Please note that the rule is the same in both conditions, namely the same box is approved or unapproved by the rule-making agent.

Procedure

All the studies were conducted in French. Three-to-five-year-old children were asked to sit in front of the computer screen, and the experimenter sat on the right side of the participant. The experimenter said “Today, we will watch some videos together and I will ask you some questions. Let’s watch the videos together”. Then the experimenter started the action/goal familiarisation trial. After an agent enters the stage and puts a bucket in the middle of the stage, the experimenter stops the video and says, “I want to introduce you to someone”, and while pointing to the agent, the experimenter says “Its name is Loulou. Let’s watch what Loulou is doing”. After the action/goal familiarisation trial, the experimenter asked the first control question (“What was Loulou doing?”), to be sure that children understood what was going on in the trials (goal/action induction control question). Independent of the answers, the experimenter said: “Loulou was putting some objects into boxes, right?” After goal/action induction trials, the experimenter said “Now, I want to show you another video”., and the rule-induction familiarisation trial was shown. Right after the rule-induction familiarisation video, the test trial video was shown. In the test video, after the agent gets an object from the bucket and moves in front of the bucket, children were asked the following question: “What do you think (CHILD’S NAME)? Where will Loulou put the object?” This question is the action prediction question. If children did not answer the question, they were motivated to select one of the boxes; “Will Loulou put it here or there?” After the children’s answers, they will be asked a justification question: “Why will Loulou put the object there?” This justification question aims to further explore what motivates children to make a particular prediction and to understand their reasoning behind their own action selection (Gönül & Paulus, 2021). Children will be also asked the following control question for each box: “Is it OK for Loulou to put objects into this box?”, while pointing to the left or right box, to check whether children understood the rule. The reason why we ask these justification and control questions is to be sure that children go beyond the rather simple mapping between boxes and affective sound (e.g. box A → negative, box B → positive), or that children do not make their predictions only based on merely negativity bias (e.g. avoiding the unapproved box since it is matched with negative emotion, independent of the positive emotion matched with the approved box).

Even though it has been repeatedly shown in the literature, to be sure that children understand the dominance hierarchy between agents (based on size cues), we also asked them exploratory questions as to which agent was the boss or the strong one. Thus, lastly, the experimenter presented a still image showing S and D side by side, in the middle of the screen, and asked the following questions, in the following order: “Do you think that one of the two characters is the boss?”, [If the answer to the previous question is positive] “According to you, who is the boss?”, [If the answer to the previous question is negative] “If you had to choose between the two characters, which one would be the boss?”, “According to you, who is stronger here?”. Even though the literal translation is “boss”, the actual wording was chief, since it means “boss” for Swiss children. These questions are used to evaluate the decision and physical power and are used to evaluate social dominance in the literature (Castelain et al., 2016; Charafeddine et al., 2015). If they did not answer these questions, the questions were repeated, and children were motivated to select one of the agents. If they pointed to both agents, they were asked to select one (forced-choice). Children were also asked: “Who is the boss in your classroom?” This question aimed to understand whether children understand a relatively abstract concept of being a boss (being in charge) (Castelain et al., 2016; Charafeddine et al., 2015), and were asked at the very end of the experiment. All other questions were asked in both conditions. In both conditions (the dominant rule-maker and the subordinate rule-maker), the same procedure was followed (the name of the protagonist putting objects into the boxes was Zuzu in the second block). All the children got both conditions (within-subject). See Figure 1 for the trial sequences and questions.

Figure 1. The flow of trails and questions, with the still-images from animations in Experiment 1. The sound icon represents the scenes in which the dominant agent vocalised an approved or unapproved voice, and the tick and stop icons represent the approving or disapproving nature of the vocalisation, respectively.

Order of the conditions (dominant rule-maker first or subordinate rule-maker first), object-putting-order in the goal/action induction familiarisation trial (Left-Right-Right-Left or Right-Left-Left-Right), towards which box the agent goes in the rule induction familiarisation trial (right or left box), the box that is approved by the rule-maker (right or left box) in the rule induction familiarisation trial, and the order of pointing – by the experimenter – for the “Is A allowed to put objects into this box?” question in the rule induction familiarisation trial was counterbalanced (Latin-square) between participants (32 versions). These counterbalancing versions were the same in both conditions. Please see the OSF link for all counterbalancing versions of Experiment 1 in an Excel sheet. Children got one of the versions randomly, with a priory random order that was generated by a simple R code (please see the OSF link). All the sessions were video-recorded by a video camera for later coding. The experimenter also wrote down the answers of the participants during the experiments.

Coding and data analysis plan

Children’s answers to the questions were coded in the following way. For the goal/action induction control question (“What was Loulou doing?”), children’s answers were transcribed first. If they mentioned the agent’s action or the goal action (e.g. “going to the boxes”, “putting toys into the boxes”), they got a score of 1, and if they gave irrelevant answers or did not answer, they got a score of 0.

Children's answers or pointing to the action prediction question (“What do you think, [child’s name]? Where will Loulou put the object?”) (test trial) were coded. Note that this was a forced-choice question, and if children did not answer the action prediction question, they were asked another question (“Will Loulou put it here or there?”). If children selected the approved box, they got a score of 1, and a score of 0 if they selected the unapproved box.

For the justification question (“Why will Loulou put the object there?”), children’s answers were planned to be coded post-hoc (Gönül & Paulus, 2021), considering the explorative nature of this question. However, most of the children did not answer this question, thus, this coding was skipped (see the dataset in the OSF link for the answers). For the other control question regarding the rule learning, children's answers to the “Is Loulou allowed to put objects into this box?” question were coded (“correct” or “incorrect”). If children answer correctly, they get 1, and 0 if they give the incorrect answer. Nevertheless, the majority of children perhaps found this question ambiguous and responded “yes” to both boxes. Thus, this coding was dropped from the analyses. For the social dominance understanding control questions “ … who is the boss?” and “ … who is stronger here?” (forced-choice), children’s verbal indications or pointing towards the agents (D or S) were coded and given scores (D = 1, S = 0) for each question. Thus, children could get max 4, min 0, and the chance level was 2 for two questions and two conditions (dominance understanding scores). Note that the “Who is the boss in your classroom?” questions were coded as correct and incorrect. The correct answer was an adult name from the school team (teacher, assistant, or director), which was double-checked by asking the parents after the experimental session.

For both experiments, we examined the dataset in three stages: (1) preliminary analysis, (2) confirmatory analysis, and (3) exploratory analysis. For all the details of these analyses, see supplementary material 2. For the complete data analysis strategy, please see the conditionally accepted registered report in the OSF link.

Preliminary analysis plan. In this section, the possible relation between the answers to the control questions and the action prediction question was explored. Moreover, the relation between action prediction performance and various variables (sex, condition order, age in months, and dominance understanding score) was explored. Children’s dominance understanding was also explored in this section. Please see Supplementary Material 3 for the results of the preliminary analyses.

Confirmatory analysis plan. For H1 (action prediction, and the rule-making function of affective expressions), a binomial test for each condition was conducted, and the alpha values were Bonferroni corrected (p = .025). For H2 (action prediction and the rule-making function of the dominant agent), children’s action prediction results were compared between two conditions (dominant rule-maker condition and subordinate rule-maker condition) with a Generalised Estimating Equations (GEE) model considering that the independent variable was a within-subject variable, and the outcome variable was binary. GEE model with binomial error distribution and logit link function and robust estimation results were calculated in R.

Exploratory analysis plan. If gender, age in months, or condition order turned out to be significant in the preliminary analysis, a full factorial exploratory GEE model was created including the respective effect (all exploratory GEE models include condition). All the data for both experiments were analysed in R for consistency (R Core Team, 2021).

Results

Preliminary analyses

To see whether children understand dominance from physical features, the scores of big (dominant-D) and small (subordinate-S) agents in two trials based on the answers to the follow-up questions on dominance understanding (“ … who is the boss/chief?” and “ … who is stronger here?”) were compared to the chance level (D = 1, S = 0, maximum score = 4, min score = 0, chance level = 2). Children selected the big agent as the dominant one significantly above the chance level, V = 454, z = 4.186, p < .001, r = .570. For the other preliminary analyses results, please see Supplementary Material 2.

Confirmatory analyses

H1 (action prediction, and the rule-making function of affective expressions):

Children did not predict the protagonist to go to the approved box rather than the unapproved box more than the chance level, V = 224, p = .598. Based on the follow-up binomial tests, this was the case for both the dominant rule-maker condition, p = .608, and the subordinate rule-maker condition, p = .752.

H2 (action prediction and the rule-making function of the dominant agent):

GEE results indicated that children’s action prediction performance was not significantly higher in the dominant rule-maker condition compared to the subordinate rule-maker condition, Wald χ²(1) = 0.111, p = .74. Thus, both H1 and H2 were not confirmed with the current data.

Exploratory analyses

We explored the potential effect of trial or condition order on action prediction. A full factorial GEE model with condition and trial revealed a significant effect of trial, Wald χ²(1) = 6.76, p = .009, but not condition, Wald χ²(1) = 0.11, p = .739, and condition and trial interaction, Wald χ²(1) = 0.42, p = .519. A second full factorial GEE model with condition and condition order (dominant or subordinate rule-maker condition first) revealed a significant interaction between condition and condition order, Wald χ²(1) = 6.77, p = .009, but no main effect of condition, Wald χ²(1) = 0.11, p = .739, and condition order, Wald χ²(1) = 0.39, p = .533. Follow-up test with Tukey corrected results demonstrated that when children got the subordinate rule-maker condition first, there was no significant effect of condition, z = 1.108, p = .160. However, when children got the dominant rule-maker condition first, there was a significant effect of condition, z = 2.342, p = .019. This means that in the first trial, both in subordinate and dominant rule-maker conditions, they gave more incorrect responses compared to correct responses, namely they opposed the rule by selecting the unapproved box. On the other hand, in the second trial, children followed the rule by selecting the approved box only in the dominant rule-maker condition (See Figure 2).

Figure 2. The proportion of correct (selecting approved box) and incorrect (selecting unapproved box) responses is based on condition and condition order.

Discussion

In this study, we aimed to examine the role of affective expressions (anger and approval), and dominance hierarchies on social rule learning in 3-to-5-year-old children. Children were familiarised with a situation in a set of scenes: a protagonist put some objects in one of two boxes repeatedly, but then an agent (rule-maker agent) came and demonstrated approval signs (smiling and vocalising “hmmm, hmmm” sound) when the first protagonist put an object into one box and demonstrated unapproving signs (angry face and angry vocalisation) when the first protagonist put an object into another box. In one condition, the body size of the rule-making agent was almost twice bigger than that of the protagonist (dominant rule-maker condition), while in another condition the rule-making agent’s body was almost twice smaller than the other protagonist.

The confirmatory analysis results demonstrated that children overall perform at chance level, a finding that does not provide direct evidence for the affective social learning hypothesis. Secondly, contrary to our expectations, there was no overall significant difference between dominant rule-maker and subordinate rule-maker conditions, which challenges both deontic reasoning and affective social learning accounts (Clément & Dukes, 2017; Cummins, 1996a). However, the preliminary and exploratory analysis results may indicate the influence of certain contextual factors. Specifically, exploratory results suggest that children expect the protagonist to disobey the rule in the first trial and to follow the rule in the second trial only when the rule-making agent is dominant. We will further discuss the theoretical interpretation of all the results in the general discussion section.

The results in Experiment 1 may be attributed to the age range of the children. Children undergo significant developmental changes concerning their socio-cognitive abilities, and understanding of norms and dominance hierarchies, and they engage in more varied social interactions as they grow (Pun et al., 2017; Schmidt & Rakoczy, 2018; Tomasello & Rakoczy, 2003). Therefore, children may have interpreted the trials differently based on their experience. As indicated in the introduction, studies suggest that toddlers may learn rules by considering affective expressions and dominance hierarchies, and expect others to follow the rules (or be surprised when they do not do so). To address this, we conducted a second study with toddlers to test the affective social learning and deontic reasoning hypotheses using an implicit paradigm.

Experiment 2

Methods

Participants

The final sample included 42 fifteen-to-twenty-three-month-old toddlers with ∓ 15 days for the minimum and maximum age in months (22 girls, M_age = 18.07 months, SD_age = 2.49 months). Three toddlers were excluded because they did not want to watch the videos from the beginning at all.

The sample size rationale was based on the relevant studies on the topic (Enright et al., 2017; Margoni et al., 2018; Mascaro & Csibra, 2012; Pun et al., 2016; Schmidt et al., 2019; Thomas et al., 2018; Thomsen et al., 2011). At least 35 toddlers were required. For a detailed explanation of our sample size rationale, please see the OSF link and the initial registered report. Even though we aimed to collect data from 54 participants in the conditionally accepted registered report (just to be able to make the sample size of the experiments the same between 2 experiments), we could not reach this number because of various factors such as difficulty in finding participants. Considering the time limit of the project (31 May 2024), we terminated data collection at 42 participants.

Set-up and stimuli

Toddlers sat on their parent's or teacher’s lap, or a baby chair and watched the video sequences on a 24-inch screen with an eye tracker, and the parents were instructed to remain silent (they wore an eye mask, and listened to music with headphones). If the participant sat on a baby chair, the mother or the teacher sat on another chair behind the baby and remained silent. The stimuli were presented with Tobii Pro Lab software, and a Tobii Pro Spectrum eye-tracker was installed right underneath the screen. Infants’ eye movements were recorded by the eye-tracker with a sampling rate of 1200 Hz. Children watched the same animations that were described in Experiment 1. One additional point should be emphasised here. During the familiarisation trials, after the protagonist takes an object from the bucket and then walks in front of the bucket (in the middle of the screen), participants hear a bell-ring, right before the protagonist heads to one of the boxes. This bell ring aims to provide a cue that the agent is going to head to one of the boxes and increase the chance of anticipatory looks of the participants during the test trials. The test phase was slightly different for Experiment 2.

Design and procedure

The design of Experiment 2 is similar to the preliminary studies on the early development of social dominance (Gazes et al., 2017; Mascaro & Csibra, 2012; Thomsen et al., 2011). The experiment started with a 5-point eye-tracker calibration (growing animated agents as points during calibration). Afterwards, toddlers watched two blocks (conditions) of videos (within-subject), as described in Experiment 1, each condition included three videos: (1) a goal/action induction familiarisation trial, (2) a rule-induction familiarisation trial, and (3) and test phase. After watching the goal/action induction familiarisation (see Movie S1 and Movie S6) and rule-induction familiarisation (see Movie S2 and S7), toddlers got both a coherent test trial (Movie S4 or S9) and an incoherent test trial (Movie S5 and S10) at the end of each condition. In both types of test trials, a protagonist stops in front of the bucket, a bell sound rings (as an action anticipation cue for the participant), and the protagonist waits for 3 sec before going to the box that is approved (coherent test trial) or unapproved (incoherent test trial) by the rule-making agent. Half of the participants got the coherent test trial first and another half – second (between subjects). After the protagonist executes the action, the animation is frozen for 40 s. Thus, Experiment 2 has a 2X2 repeated measure design with condition (dominant rule-maker and subordinate rule-maker) and coherence of the test trial (coherent and incoherent).

Order of the conditions (dominant rule-maker first or subordinate rule-maker first), test trial order (coherent-incoherent or incoherent-coherent), object-putting-order in the goal/action induction familiarisation trial (Left-Right-Right-Left or Right-Left-Left-Right), towards which box the agent goes (right or left box) in the rule induction familiarisation trial, the box that is approved by the rule-maker (right or left box) in the rule induction familiarisation trial, were counterbalanced (Latin-square) between participants (32 versions). Version randomisation was the same as Experiment 1. All the sessions were video-recorded by the video camera of the eye-tracker.

Coding and data analysis plan

Coding

In Experiment 2, Anticipatory Looking Score (ALS) and Differential Looking Score (DLS) were coded as action anticipation measures. Considering infants’ possible difficulty in gaze control in action anticipation, we defined two areas of interest (AOI) for each box with Tobii Pro Lab: small AOI and large AOI (Biro et al., 2014) (see Figure 3). These measures were taken within the 3-sec frame when the agent stopped in front of the bucket before heading to one of the boxes. The anticipatory-looking score was based on toddlers’ first fixation on one of the boxes (small AOI) or the right/left lower side of the screen (large AOI). Nevertheless, only big AOIs used in the analysis section considering that there were a lot of cases in which children did not make anticipatory looks on the small AOIs. First-look at the approved and unapproved box areas were coded as 1 and – 1, respectively. If they did not make any anticipatory looks on the AOIs, it was coded as 0. Thus, each participant could get a max of 4, min – 4, and the chance level was 0 (in two blocks, two test trials in each block).

Figure 3. Small and larger AOIs. Two AOIs defined for each box. An example of small AOI (right) and large AOI (left) is represented by semi-transparent purple shapes on the boxes.

Another action anticipation measure was DLS, which was calculated based on the looking time difference between the AOIs on approved and unapproved boxes (or right/left side of the screen) within 3 s before the agents head one of the boxes. DLS was calculated based on the following formula: $\mathrm{DLS}={\displaystyle \frac{t(\mathrm{approved}\mathrm{box})-t(\mathrm{unapproved})}{t(\mathrm{approved}\mathrm{box})+t(\mathrm{unapproved})}}$where t indicates the total fixation duration (including partial fixations) on the target box. The mean score of DLS was calculated for each condition. Thus, the DLS value was between – 1 and 1 for each condition. While positive values indicate a preference for the approved box, negative values indicate a preference for the unapproved box. Zero is the chance level and indicates absolute no-preference.

Beyond the anticipatory-looking score and DLS measures, we also defined an AOI around the rule-making agent in the rule-induction trial. With the help of the eye-tracker, we coded the number of starting fixations counts on this AOI.

Toddlers’ looking time was coded beginning from the time that the actor finishes executing the action (put the object in one of the boxes) to the time that infants look away from the screen for more than ∼2 consecutive seconds or 40 s was elapsed, whichever came first, in line with a study that we adopted its design for the current study (Mascaro & Csibra, 2012). If a participant looked away for ∼2 s or more, the experimenter pressed a button to proceed with the following stimulus. The minimum criterion for the looking time coding was that the participant needed to look at the agent’s action when the agent put the object in one of the boxes. Log-transformed looking time values were used as the dependent variable (Csibra et al., 2016), unless indicated.

Analysis

The same stepwise method that was described to analyse the dataset of Experiment 1 was also used in Experiment 2: (1) preliminary analysis, (2) confirmatory analysis, and (3) exploratory analysis. For further details of these analyses, see supplementary material 2 and the conditionally accepted registered report in the OSF link.

Preliminary analysis. In this section, first, the possible relation between the fixation count on the rule-making agent during the rule-induction trial and the dependent variables (anticipatory looking score, DLS, and looking time) was investigated. Beyond that, we also examined the possible effect of condition order, gender, and age in months on the dependent variables. Regarding the looking time, we also examined whether the test trial order (coherent test trial first or incoherent test trial first) had any effect on looking time. Finally, we explored the correlations between the dependent variables. For the details of the preliminary analyses, see Supplementary Material 3.

Confirmatory analysis. To test H3 (action anticipation and the rule-making function of affective expressions), for the anticipatory-looking score data, we used a one-sample Wilcoxon signed-rank test (considering the scaled nature of this data) to be able to check whether anticipatory-looking scores were different from the chance level of 0. Second, for the DLS data, we used a one-sample t-test (considering the continuous nature of the data), to check whether DLSs are different from 0. To test H4 (action anticipation and the rule-making function of the dominant agent), for the anticipatory looking score data, we used a paired-sample Wilcoxon signed-rank test, to be able to compare the within-subject conditions: dominant and subordinate rule-making conditions. Second, for the DLS data, we used a paired-sample t-test to see whether the within-subject conditions differed in DLS. If the normality assumption was violated, non-parametric tests were preferred instead of t-tests, following Experiment 1.

To test H5 and H6 (looking time, and the rule-making role of affective expressions and the dominance agent), we planned to run a full-factorial mixed ANOVA model with condition and test type as the main effects and their interaction. Considering that the normal distribution of errors assumptions was violated for ANOVA, we used Generalised Linear Mixed Models (GLMM). For GLMMs, not the log-transformed but the untransformed-looking data were used, as it is possible to log-transform the link function in GLMM. We compared Gaussian, Gamma, and inverse Gaussian error distributions with log-link functions, and the best fitting model was selected (Gamma with log link).

Exploratory analysis plan. If any of the variables examined in the preliminary analysis (gender, age in months, condition order, fixation count on the rule-maker agent, or test trial order) turned out to be significant, we included them in the exploratory analysis. For further details, see the initial registered report.

Integrative-analysis

In the registered report, we promised to compare predictive measures in both studies (action prediction in Experiment 1 and action anticipation in Experiment 2) by coding whether the action prediction or the action anticipation was correct or incorrect for each participant. Considering that 26% of the toddlers did not make any predictive looks on the boxes in all of the 4 test trials, we skipped the integrative analysis.

Results

Confirmatory analyses

H3 (action anticipation and the rule-making function of affective expressions):

In total, 32 children made anticipatory looks in at least 1 of 4 trials (76%). In other words, of 168 trials for all toddlers (42 participants, 4 trials), toddlers made action anticipations in 58 trials (%34.5). Anticipatory Looking Scores (ALS) demonstrated that toddlers expected the protagonist to go to the approved box than the unapproved box significantly more than the chance level zero, V = 246.5, p = .0309, r = .33, please see Figure 4. However, this does not imply a substantial preference among toddlers for the approved box, as suggested by the Differential Looking Scores (DLS). The DLS did not exhibit a statistically significant deviation from the chance level (V = 290, p = .3124).

Figure 4. Boxplot and the violin-plot of the Anticipatory Looking Scores (ALS). Blue dots represent the sum of ALS over 4 trials for each toddler. The dotted red line represents the hypothetical chance level. For each trial, toddlers got a score of 1, –1, or 0, if they made an anticipatory look towards the expected box, unexpected box, or no anticipatory looks at either box, respectively.

H4 (action anticipation and the rule-making function of the dominant agent):

The difference between dominant and subordinate rule-maker conditions was not significantly different with regards to ALS, (V = 181, p = .8929), and DLS (V = 269, p = .2604).

H5 (looking time and the rule-making function of affective expressions) and H6 (looking time and the rule-making function of the dominant agent):

Only in 4 cases in a total of 164 trials, the participant did not look at the direction that the agent was heading in the test trials, so those trials were excluded from subsequent analyses. A full factorial GLMM (Gamma distribution family with log link function) with condition (dominant or subordinate rule-maker) and test type (congruent and incongruent) revealed that neither the main effects of condition, Wald χ²(1) = 0.354, p = .5517, nor test type, Wald χ²(1) = 0.0273, p = .8687, nor their interaction, Wald χ²(1) = 0.1194, p = .7297 were statistically significant.

Exploratory analyses

We explored the potential effect of trial, condition order, and test trial order on toddlers’ looking time. A full factorial GLMM with the condition, test type, and trial revealed a significant effect of trial only, Wald χ²(1) = 18.883, p < .001, in which toddlers looked longer in Trial 1 (m = 12260 ms) compared to Trial 2 (m = 7961 ms). None of the other main and interaction effects were significant (all ps > .7718). A second full factorial GLMM with condition, test type (coherent or incoherent), condition order (dominant or subordinate rule maker first) and test trial order (coherent or incoherent test trial first) revealed significant two-way interaction between condition and condition order effect, Wald χ²(1) = 21.819, p < .000, and test type and test trial order, Wald χ²(1) = 6.022, p = .0143, and, and a significant three-way interaction effect between condition, test type, and test trial order, Wald χ²(1) = 4.028, p = .0448, and, but no other effects (all ps > .1124). The two-way condition and condition order effect, which is related to the trial effect, indicates that toddlers looked at the dominant rule-maker test trials overall longer compared to the subordinate rule-maker trials in both trials, while looked at subordinate rule-making trials longer only when it is the first trial. To explore the intricate three-way interaction effect (condition, test type, and test trial order), we split the data based on the condition, and Bonferroni corrected the results (p = .025). In the dominant rule-maker condition, there was no significant effect of test trial order and test type (p > .1634). On the other hand, in the subordinate rule-maker condition, there was a significant interaction between test trial order and test type, Wald χ²(1) = 13.781, p = .0002. Subsequent pairwise comparisons, employing Tukey's correction, unveiled that within the subordinate rule-making condition, toddlers demonstrated heightened surprise levels when the initial trial depicted the agent adhering to the rule, specifically during the coherent-first test trial as opposed to the incoherent trial (z = 2.73, p = .0318).

Discussion

This study aimed to examine the effect of affective expressions (affective social learning hypothesis) and dominance hierarchies (deontic reasoning hypotheses) on toddlers’ understanding of social rules using an implicit paradigm. Similar to Experiment 1, toddlers observed a scenario in which a protagonist puts some objects into one of two boxes, and then a rule-making agent approves the other agent’s putting the object in one box but not the other. In the dominant rule-maker condition, the rule-making agent was bigger than the protagonist, and vice versa in the subordinate rule-maker condition. Toddlers got two test trials after each condition: one test with a coherent (expected) action in which the protagonist goes to the box that was approved by the rule-making agent, and another test with an incoherent (unexpected) action in which the protagonist goes to the box that was unapproved by the rule-making agent. In each trial, infants’ action anticipation was calculated before the protagonist headed to one of the boxes, and their looking time was calculated from the time that the protagonist put the object into one of the boxes to the time that infants looked away from the screen for more than ∼2 consecutive seconds or 40 s was elapsed.

Of the toddlers who made anticipatory looks, their gazes were significantly more often directed towards the box consistent with the given implicit rule, specifically the expected or coherent box, supporting hypothesis H3 (action anticipation and the rule-making function of affective expressions). These findings suggest that toddlers learn social rules from affective expressions and make predictions based on these rules. However, concerning hypothesis H4 (action anticipation and the rule-making function of the dominant agent), action anticipations were not influenced by the dominant status of the rule-making agent. Additionally, hypotheses H5 (looking time and the rule-making function of affective expressions) and H6 (looking time and the rule-making function of the dominant agent) were not confirmed. Exploratory follow-up results reveal some interesting contextual effects, such as the order of test trials. For example, when toddlers observe the large protagonist following the rule set by the small subordinate agent in the first test trial of the subordinate rule maker condition, they find it quite unexpected. Similar effects of test trial order have been observed in other studies with infants (Mascaro & Csibra, 2012; Pun et al., 2016; Thomsen et al., 2011).

General discussion

Children go through significant social and cognitive changes in their first five years of life. They not only engage in complex social interactions with their parents, peers, and other people in their surroundings, but also learn the social rules that they should maintain to be able to become members of “cultures”. But how do they learn these social rules, which can be quite arbitrary from time to time? In this study, we examined toddlers’ and 3-to-5-year-old children’s understanding of a simple rule from affective expressions and dominance hierarchies in third-party interactions, following the affective social learning and deontic reasoning literature. Confirmatory analysis results of Experiment 1 show that preschool children do not systematically use affective cues and dominance hierarchies in their explicit predictions of others’ actions, even though they understand that the big agent is the dominant one. Experiment 2 results demonstrate that toddlers expect others to follow the social rules (based on their anticipatory looks) independent of the dominant status of the rule-making agent, but do not get surprised (based on looking time results) if the agent breaks the rule.

In Experiment 1, contrary to the affective social learning account and our hypothesis H1, children did not expect others to follow the rules indicated by affective expressions. While this could be a genuine null result, alternative explanations should be considered, particularly given that children are easily affected by the contextual details in the stimuli. First, substantial evidence shows that young children pay attention to people’s expressions when evaluating novel situations (e.g. Fusaro & Harris, 2008; Sorce et al., 1985). Second, in our stimuli, we had the goal-induction trial where the agent put objects in the boxes four times, followed by the one-time rule-induction trial. During the test trials, when children were asked in which box Loulou or Zouzou would put the object, they might have selected one of the boxes based on the agent’s equal visits to both boxes during the goal induction trials. Children might have prioritised the frequency of the action, interpreting the test trial question based on this frequency information. This claim is partially supported by the allowance question (“Is Loulou allowed to put objects into this box?”). Regardless of the box, most children said “yes” for both, indicating they prioritised frequency information and might have thought that since the protagonist went to the boxes several times, she was allowed to go to both boxes. A study with only the rule-induction trial (either once or several times) may help to understand the actual role of the social rule. The current design may actually be testing the contrast between habit and one-shot rule learning based on affective expressions. Additionally, in the test trial, the rule-making agent was not present, making the test stimuli visually more similar to the goal-induction trials. In summary, even though children may be able to readily infer rules and norms (Clément et al., 2011; Harris & Nunez, 1996; Riggs & Kalish, 2016; Schmidt & Rakoczy, 2018), they may also be sensitive to habits or repeated/scripted actions (Taylor et al., 2023) in predicting others actions (Gönül et al., 2024; Gönül & Paulus, 2021).

In Experiment 1, contrary to our expectations based on the deontic reasoning hypothesis (H2), children did not systematically expect others to follow the social rules based on the dominant status of the rule-maker. This may be again related to the goal-induction familiarisation trial which was repeated several times before the one-time rule induction trial. For instance, children in this study did not have difficulty understanding dominance relations based on body size, consistent with the literature (Lourenco et al., 2016; Thomsen et al., 2011). When shown a big and a small agent side by side after a dominant or subordinate rule-maker condition, children identify the big agent as the stronger one. However, while children chose the big agent as the boss/chief after the dominant rule-maker condition, this was not the case after the subordinate rule-maker condition. They did not systematically choose the big agent as the boss/chief after the condition that the rule-maker was subordinate. This finding aligns with the literature, demonstrating different types of dominance relations: those based on physical size and power, and those based on decisional power (Castelain et al., 2016; Charafeddine et al., 2015; Keltner et al., 2003; Margoni et al., 2018; Pun et al., 2017; Thomsen, 2020).

The results of Experiment 2, assessed by the Anticipatory Looking Scores (ALS), revealed that toddlers consistently exhibited anticipatory looks towards the anticipated box. Each test trial elicited a score of 1 for the initial anticipatory look towards the expected box, –1 for the unexpected box, or 0 if no anticipatory looks were observed within a 3-second window before the agent’s approach to either box. This finding provides support for H3 (action anticipation and the rule-making function of affective expressions), suggesting that toddlers expect others to follow rules. However, despite their anticipation of expected actions from others, the differential looking score (DLS) results indicate an equal duration of gaze towards both the expected (or “correct”) and unexpected (or “incorrect”) boxes. This outcome aligns with our interpretation of Experiment 1, regarding the potential influence of the goal-induction trial on children’s predictions. While toddlers may implicitly anticipate the protagonist to follow the rule (evident from ALS results), they may still shift their gaze between both boxes since the protagonist visited both boxes without any established “rule”. This phenomenon might explain toddlers did not exhibit surprise or prolonged gaze (H5, looking time and the rule-making function of affective expressions) when the protagonist violated the social rule. The results of Experiment 2 also failed to reject the null hypotheses for H4 (action anticipation and the rule-making function of the dominant agent) and H6 (looking time and the rule-making function of the dominant agent). While it is conceivable that these are genuine null effects, further research is necessary to elucidate when and how toddlers actually commence learning rules from authority figures. Although infants and toddlers may form expectations for others to conform to the rules set by the dominant status of agents, such as their size (Thomsen et al., 2011) or history of success (Mascaro & Csibra, 2012), this doesn’t necessarily imply an immediate expectation for others to conform to the rules set by a dominant agent solely based on their physical stature as the “rule-maker”, even though they may exhibit heightened attention towards dominant agents. It may be yet early to draw strong conclusions about the role of dominance status in social rule learning based on the results of Experiment 2. In addition to the habit-rule contrast highlighted in previous sections, the exploratory results of Experiment 2 indicate the presence of trial effects and various interaction effects. Interestingly, our exploratory analysis revealed that toddlers displayed the highest level of surprise when initially encountering a condition where a small (subordinate) agent was the rule maker, particularly when the coherent test trial (where a large protagonist follows the rule of a small agent) was presented first. This suggests that toddlers do not expect a dominant protagonist to follow the rule of a subordinate agent at first, but may adjust their expectations based on various contextual factors, such as the order of events.

Moreover, toddlers exhibited longer gaze durations towards the test trials featuring dominant rule-makers overall, compared to those featuring subordinate rule-makers, across both trials. However, they displayed longer gaze durations towards subordinate rule-making trials only when they occurred as the first trial. While caution is warranted in interpreting these effects given the exploratory nature of the analysis (including the fact that sample sizes were not specifically determined to detect these complex interaction effects), it is plausible that toddlers may allocate greater attention when the rule-maker is dominant, without yet forming specific expectations regarding whether others should adhere to the rules established by dominant agents based on a single instance.

Future directions and conclusion

Both experiments have crucial implications for the affective social learning and deontic reasoning research, and potential for future research. Based on explicit action predictions of preschool children in Experiment 1, there was no strong evidence to suggest that children readily learn social rules from affective expressions and dominant agents readily, and toddler expect others to follow the rules of authority figures. However, exploratory results give interesting insights for future research. Exploratory results indicate that children expect others to disobey a rule in the first trial, but to follow the rule of a dominant agent in the second trial. On the other hand, toddlers expect others to follow the rules (based on anticipatory looks), but are surprised only when in the first trial the dominant protagonist follows the rule of a subordinate agent, similar to the other studies where infants and toddlers have different expectations based on dominance hierarchies (Enright et al., 2017; Mascaro & Csibra, 2012; Pun et al., 2016; Thomas et al., 2018, 2011). Our null and exploratory results shed some light on future research.

There is an interesting Trial effect, especially in Experiment 1, where children disobey the rule in Trial 1, but obey the rule in Trial 2 if the rule-maker is dominant. We further discuss this point in Supplementary Material 4. Beyond the other points raised above, several studies have already indicated that 3-to-11-years-old children from different cultures demonstrate varying judgments towards and expectations from authority figures (Bernard et al., 2016; Castelain et al., 2016; Fonn et al., 2022; Kim, 1998), even though there are also similarities based on the cultural similarities such as children from Western cultures (e.g. Charafeddine et al., 2016; Cogsdill et al., 2014). Various socio-cognitive abilities are affected by culture (Green et al., 2016; Kaufmann & Clément, 2014; Rogoff, 2003; Selcuk et al., 2023), and observational results indicate that children living in remote societies where there is no systematic and institutionalised educational system are better at gaining social information from observational learning (Clément & Dukes, 2019; Rogoff, 2003). Thus, one can argue that affective observational learning may be more effective in children living in some non-WEIRD (Western, Educated, Industrialised, Rich, and Democratic) societies (Henrich et al., 2010; Rogoff, 2003), and one can get different results if the current study is conducted in a non-WEIRD culture. While infants and toddlers may use affective and dominance cues more directly in their social rule learning, children may demonstrate more variation depending on their own dominance status and understanding, and their cultural knowledge of dominance hierarchies. Future studies should consider the potential effects of culture and children’s own dominant status on their social rule learning from affective expressions and authority figures.

Beyond that, even though the majority of toddlers made at least one anticipatory looks in four trials, overall the number the anticipatory looks were not very high. Despite using a ring bell during the familiarisation trials before the acting agent moved towards one of the boxes, this may have not been enough for toddlers to make predictive looks. Future studies should include various other cues such as lighting up the boxes to motivate toddlers to make predictive looks. Beyond all these findings, future studies should also include a scenario in which two agents have the same body size but one of them is the rule-maker. Such a condition may help to disentangle the intricate relation between affective expressions and dominance hierarchies in social rule learning. Additionally, in the current design, children could have had difficulty understanding the “rule”. For example, the rule may state either all items must go in a specific box or items of a certain type must go in a specific box. Even though we used the same type of object in each trial, clearer rules could make a difference in outcomes (for example, by making the types/categories of objects different but bound with a rule – object A with a blue box and object B with a green box), which should be explored in the future.

In conclusion, this study aimed to examine the contribution of affective and dominance cues on toddlers’ and 3-to-5-year-olds social rule learning in third-party scenarios. Participants watched scenarios where a subordinate agent follows the rule of a dominant agent or vice versa. In Experiment 2, while toddlers expected others to follow the rules (based on anticipatory looks), they were not surprised (based on looking time) if others did not follow the rules. In Experiment 1, older children did not expect others to follow rules based on affective and dominance cues. Both experiments revealed that children are easily affected by contextual factors (such as trial) in their rule-learning. Future studies should explore the cross-dependent effects of development, culture, and individual differences on social rule learning.

Acknowledgments

We thank Albulenë Xhemaili Meraku, Nawel Bourahla, and Alexandra Rossy for their great help and assistance in the data collection process. Most importantly, we would like to thank all the parents and children who participated in our study. Data collection would not have been possible without the invaluable support of the Neuchâtel Natural History Museum, the “Ville de Neuchâtel, Service de la Famille, les structures d’accueil extrafamiliales”, and all the collaborating kindergartens and schools in Neuchâtel and Bienne and teachers (Babydream Club in Bienne, Sorimont in Neuchâtel and others). As this paper was a registered report, we received invaluable feedback from various reviewers. We sincerely appreciate their constructive comments throughout the multiple rounds of the review process.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data, sampling strategy and sample size calculations, R files for the analysis syntax, example video stimuli, and other related files can be found in the following link: https://doi.org/10.17605/OSF.IO/MZFPU

 Google Scholar

.

Additional information

Funding

The study procedure was approved by the “Commission d’éthique de la recherche de l’Université de Neuchâtel” on 11 December 2020. This research received funding from the NCCR Evolving Language, Swiss National Science Foundation Agreement #51NF40_180888.