Selective Helping in Children:
Are 2 Year Olds Biased to Help Victims over Aggressors? Josje M. de Valk
University of Amsterdam & Uppsala University
Author Note
Josje M. de Valk, Department of Psychology, University of Amsterdam in collaboration with Uppsala University, Sweden.
Correspondence concerning this article should be addressed to Josje M. de Valk, Weezenhof 84-44, 6536BW Nijmegen, the Netherlands.
Abstract
From an evolutionary perspective, helping non-relatives is beneficial because helped ones are likely to return the favor. Consequently, helping should be selective: one should prefer helping individuals who are more likely to return the favor. This study investigated whether children’s helping behavior was selective by examining whether 2 year olds preferred helping victims over aggressors. A new method to measure helping behavior was developed: Two non-human agents (victim and aggressor) dropped tokens and children could help by returning the tokens. To validate that this method measured helping behavior, we examined whether children helped more/faster when the agents needed help (experiment condition) than when the agents did not need help (control condition). To investigate to what extent children’s helping behavior was selective, we examined whether children helped victims more/faster than aggressors. A manipulation check tested if children evaluated the victim and aggressor differently. Results showed that children helped equally often/fast in the experiment and control condition. Two explanations are possible: 1) children misinterpreted the goals of the agents which resulted in returning tokens in both conditions or 2) children did not help but they returned tokens because they wanted the game to continue. Also, children helped both the victim and aggressor equally often/fast. The manipulation check showed that children did not discriminate between the victim and aggressor, hence the absence of a difference between helping the victim and aggressor was probably due to an unsuccessful manipulation.
Key words: prosocial behavior, instrumental helping, selective helping, 2 year olds, non-human agents, victim, aggressor
Selective Helping in Children:
Are 2 Year Olds Biased to Help Victims over Aggressors?
Children start helping others at a very young age. An example is a child who is handing blocks to another child who is building a tower. But is children’s helping behavior selective? Do they prefer helping some individuals over others? This study aimed to answer the question whether children are biased to help some individuals (victims) over others (aggressors). First, the concepts prosocial behavior and helping behavior will be introduced. Then we will elaborate on selective helping in children: some studies suggest helping is not selective while others suggest that it may be. This study is set up to resolve this ambiguity.
Prosocial behavior
Prosocial behavior is defined as behavior that benefits another individual (Warneken & Tomasello, 2009), regardless of whether it benefits or harms the performer. Prosocial behavior is not limited to humans; social animals like dolphins, elephants and great apes behave prosocially too (de Waal, 2008). For example, dolphins have been observed
supporting a sick conspecific to move to the surface to breathe (Connor & Norris, 1982) and apes have been observed helping youngsters to move from one tree to another (de Waal, 2008). Why these animals perform prosocial behavior to non-relatives1 can be explained by taking an evolutionary perspective. Even though it does not necessarily benefit the performer in the short term, in the long term it does; it is thought to have evolved because receivers will return the favor in the future (de Waal, 2008; Warneken & Tomasello, 2009).
At least four different kinds of prosocial behavior have been documented in the literature: sharing, informing, comforting and instrumental helping (Dunfield, Kuhlmeier, O’Connell & Kelley, 2011; Dunfield & Kuhlmeier, 2013; Warneken & Tomasello, 2009). These different kinds of prosocial behavior usually do not correlate which suggests that they are distinct types of behaviors. They are likely to have different underlying mechanisms and developmental trajectories (Dunfield et al., 2011; Dunfield & Kuhlmeier, 2013). Dunfield and colleagues conclude that there is no such thing as ‘prosocial behavior’ in general and
therefore the different kinds of prosocial behavior should be studied separately.
Instrumental helping
During development, the first prosocial behavior to emerge is instrumental helping, which is defined as a tendency to complete another individual’s goal (Kenward & Gredebäck, 2013). Between 14 and 18 months, children start helping other humans (Warneken &
1
Prosocial behavior to relatives can be explained by promoting one’s own genes (kin selection theory) instead of reciprocal mechanisms(Hamilton, 1964).
3
Tomasello, 2006; Warneken & Tomasello, 2007). For instance, if an adult tries to take an object that is out of reach, a child may hand over the object.
Warneken and Tomasello (2006) were amongst the first to experimentally study helping behavior. They studied helping behavior of 18 month old children and chimpanzees. In the experiment condition, experimenters performed goal-directed actions but for some reason they could not finalize the actions (i.e. they needed help). In the control condition,
experimenters performed the same actions but now they did not finalize the actions on purpose (i.e. they did not need help). For example, an adult was hanging the laundry but then he accidentally (experiment condition) or intentionally (control condition) dropped a
clothespin on the floor. Both children and chimpanzees helped the experimenter more in the experiment condition than the control condition. In other words, the intentions of the
experimenter were understood; help was provided more often when the experimenter actually needed help. A follow-up study found helping in even younger children: From approximately 14 months old, children helped others in simple helping tasks (Warneken & Tomasello, 2007).
Warneken and Tomasellos’s study (2006) underlines the relevance of understanding intentions in instrumental helping: one needs to understand someone’s intention to complete a goal in order to help. This is in line with other studies showing that from about 6 months old, infants understand simple actions in terms of their goals (Kochukhova & Gredebäck, 2010; Woodward, 1998). For example, infants are more surprised when the goal of an action changes than when the path of an action changes (Woodward, 1998). Thus, way before children start helping others they already understand other’s actions in terms of their goals. But of course, there is more to helping than merely understanding the goal, one should also recognize how to complete the goal and one should be physically able to complete the goal (Dunfield et al., 2011).
Selective helping in childhood
As illustrated by Warneken & Tomasello (2006; 2007) children frequently help. Nevertheless, the question ´whom do children help?´ remains to be answered. From an
evolutionary perspective, helping non-relatives will become beneficial in the long term only if one favors helping individuals who are likely to return the favor.2 In other words, helping behavior should be selective i.e. biased to individuals who are likely to reciprocate in the
2
Another possibility is that helping is intrinsically rewarding. In that case, helping would immediately benefit the performer. That means that one would not have to favor helping ones who are likely to return the favor. However, Hepach, Vaish and Tomasello (2012) showed that children’s arousal levels were similar when they see someone helped and when they help themselves (in contrast, arousal levels were increased when someone was not helped). This suggests that it is irrelevant whether a child helps himself or not i.e. helping is likely not to be intrinsically rewarding.
4
future (Warneken & Tomasello, 2009; Krebs, 2008). This statement has been supported by research in adults (Batson, Duncan, Ackerman, Buckley, and Birch, 1981; Krebs, 1975). Whether helping behavior in children is selective, is yet unknown.
In the literature, one can find opposing views. Warneken and Tomasello (2009)
suggested that infants initially help indiscriminately (no selective helping). However, they did not provide data to support this statement. The only available data points in the opposite direction: studies on other kinds of prosocial behavior found support for selective prosocial behavior, hence helping behavior may be selective too.
Three studies found support for selective prosociality. First, 21 month olds handed over a toy more often to a social than to an antisocial actor (Dunfield & Kuhlmeier, 2010). Second, 3 year olds handed over a toy less often to an aggressor than to a helper or neutral person (Vaish, Carpenter & Tomasello, 2010). A third study found that 18 and 25 month olds helped and shared more with a victim than a neutral person (Vaish, Carpenter & Tomasello, 2009). The first two studies measured whether children preferred to hand over a toy to one individual over another (Dunfield & Kuhlmeier, 2010; Vaish, Carpenter & Tomasello, 2010). This might reflect selective helping. However, the toy could be seen as a reward hence these studies might have tapped into reward/punishment processes rather than instrumental helping. On the other hand, Vaish and colleagues (2009) did measure helping behavior by coding the attempts of a child to get an experimenter’s lost balloon back. However, they merged sharing and helping behaviors in one category and did not report data on helping behavior
specifically.
In sum, these studies showed that children behave more prosocially to some individuals over others: prosocial behavior seems to be selective. However, none of these studies focused on instrumental helping. Thus, even though children’s prosocial behavior appears selective, whether children’s instrumental helping behavior is selective is yet unclear. The goal of the present study was to find out whether children’s helping behavior is selective, i.e. whether they are biased to help some individuals over others.
The first step in responding differentially to different persons is being able to
distinguish between their characters. Kanakogi, Okumura, Inoue, Kitazaki and Itakura (2013) illustrated that 10 month old infants distinguished between victims and aggressors: infants preferred a victim over a neutral agent and a neutral agent over an aggressor. Besides, children behaved more prosocially to victims and less to aggressors (Vaish, Carpenter & Tomasello, 2009; Vaish, Carpenter & Tomasello, 2010). Therefore the key question of this study was whether children preferred to show instrumental helping behavior toward victims compared to aggressors.
The present study
To answer the question whether children prefer helping victims over aggressors, a new experimental setup was developed in which two non-human agents (a victim and aggressor) could collect and drop tokens. Children could help by picking up the token and returning it to the agents.
A non-human agent is a geometric shape that lacks human-like characteristics. Kenward and Gredebäck (2013) illustrated that children help non-human agents. Besides, Kanakogi et al. (2013) found that children preferred non-human victims over non-human aggressors. Based on these studies, non-human agents were suitable for measuring selective helping in children. In addition, the behavior of non-human agents could be manipulated without interference of other behaviors associated with human agents (e.g. looking behavior, facial expressions). Therefore, it allowed for a more ‘pure’ manipulation.
Because of the novelty of the procedure, we needed to assess whether we were actually measuring helping behavior. As in other helping studies, two conditions were included: an experiment and control condition. In both conditions, there were two agents (a victim and an aggressor) who knocked down a token. The agents accidentally dropped the token in the experiment condition (i.e. the agents needed help) yet they intentionally dropped the token in the control condition (i.e. the agents did not need help). If our procedure would elicit helping behavior, then children should return the tokens more often or faster when the agents needed help (experiment condition) than when the agents did not need help (control condition) (Hypothesis 1).
Second, we predicted that children would help the victim more often or faster than the aggressor (Hypothesis 2). We manipulated the identity of the agents as follows: the aggressor hit the victim multiple times as in Kanakogi et al. (2013) and Premack and Premack (1997). To make the manipulation even stronger, the aggressor took tokens that belonged to the victim. Within their second year, children begin to relate persons to objects and they begin to master the concept of ‘possession’ (Rochat, 2011). Two year olds tend to quickly infer the possessor of an object and they have a first possessor bias: the one who has the object first, is thought to be the possessor (Blake & Harris, 2011). The present study builds on these
concepts; if they can infer the possessor of an object, they are likely to detect violations of possession too. Therefore, the fact that the aggressor takes tokens that belong to the victim should add to the understanding of different agents being present in the scene (i.e. agressor vs. victim). Accordingly, it was expected that children would help the victim more often or faster than the agressor.
Research shows that helping behavior can be observed from about 14 months old (Warneken & Tomasello, 2007). However, that applies only to very simple helping situations. Our paradigm was relatively complex because children had to have a basic concept of
possession, hence 2 year olds were tested in the present study.
In addition to validating the new experimental set-up and answering the question whether children are biased to help victims over aggressors, we tried to explain individual differences in helping behavior. Previous studies have found large individual differences: For instance, only half of the sample in Warneken and Tomasello (2006) helped in the more complex tasks. In addition, only 40% of the sample in Kenward and Gredebäck (2013) helped non-human agents. As pure exploration, we looked at age, gender and temperament to try to explain why some children help while others do not. About half of our sample participated in an earlier study in which temperament was measured with the Infant Behavior Questionnaire Revised (Gartstein & Rothbart, 2003), which allowed us to relate temperament to helping behavior.
Methods
Participants
59 children participated in this study of which 34 were female. The average age was 25.2 months (age range from 23.9 to 28.7 months). The experiment was partly conducted as part of a longitudinal project on prosocial behavior; about half of the sample participated in an earlier study while they were still nine months old (n=33).
The study was approved by Uppsala Regional Ethics Committee. All participants were registered in the database of the Child and Baby Lab. Informed consent was given by the parents. The families received a gift token in return for participating.
Design
The design was a 2 (condition: experiment/control) by 2 (what agent drops a token: victim/aggressor) within subjects design. The experiment consisted of two blocks (one block for each condition) and each block consisted of two trials (one trial for each agent). Thus, there were four trials in total: two trials in which a victim and aggressor dropped a token in the experiment condition and two trials in which a victim and aggressor dropped a token in the control condition. The order of agents was counterbalanced within experimental and control blocks (Table 1).
Table 1 Overview counterbalancing Block 1 Block 2 N Experiment condition Trial 1: victim Trial 2: aggressor Manipulation check Control condition Trial 1: victim Trial 2: aggressor Manipulation check 17 Experiment condition Trial 1: aggressor Trial 2: victim Manipulation check Control condition Trial 1: aggressor Trial 2: victim Manipulation check 13 Control condition Trial 1: victim Trial 2: aggressor Manipulation check Experiment condition Trial 1: victim Trial 2: aggressor Manipulation check 15 Control condition Trial 1: aggressor Trial 2: victim Manipulation check Experiment condition Trial 1: aggressor Trial 2: victim Manipulation check 14
Note. Four possible sequences of the four trials that each participant completed.
The agents had a different goal in both conditions hence completing the goal (helping) should result in different behaviors. In the experiment condition, the goal of both agents was to collect tokens. In the control condition, the goal of both agents was to get rid of tokens. In both conditions, the identity of the agents was manipulated by letting one of the agents take tokens from and bash the other agent. At the end of each condition, both agents took turns dropping a token. The different contexts of dropping the token (collecting or throwing away tokens) were intended to clarify whether dropping the token was accidentally (experiment condition) or intentionally (control condition). After each condition, children were asked to pick one of the agents (manipulation check). (More detailed information on both conditions and on the manipulation check can be found in the Procedure section.)
Materials
The experimental setup can be seen in Figure 1. The agents were geometric shapes made of wood and had no human-like characteristics (e.g. no eyes). Magnets were attached to the agents which allowed the experimenter to move the agents from underneath the table. To discriminate between conditions, the agents and tokens had a different shape and color in each condition. Both agents had their own garden where they collected tokens in the experiment condition.
The experimenter wore a visor so the participant could not see the experimenter’s gaze direction. The child was placed in front of the stage and the parent was seated behind the child on a separate chair. The parent wore blinded sunglasses to prevent cueing the child
systematically.
Plexiglas was attached to the front of the stage to decrease the tendency of the children to interfere with the procedure. Helping behavior was still possible by returning the token through the small gap between the stage and the Plexiglas.
The stage was placed behind a curtain, which could be raised and lowered, so the experimenter could prepare the experiment without the child noticing.
Figure 1. The experimental setup from the child’s perspective. The experimenter used magnets to operate the agents from underneath the stage. The agents could drop tokens through the small gap between the table and Plexiglass and children could help by returning the token through this gap.
Procedure
The experimenter and the child played an informal warm-up game to make the child feel comfortable. After parents gave consent, the camera was turned on and the child and parent were seated in their own chair. The parent was asked to wear the blinded sunglasses. The experimenter raised the curtain and the first condition started.
The experiment and control condition were as similar as possible, the only difference being the context of dropping the tokens: in the experiment condition, the goal was to collect the tokens hence dropping the two last tokens should be perceived as accidental (which could have led to helping). In the control condition, the goal was to knock down tokens hence dropping the two last tokens should be perceived as done intentionally (which could have led to less/no helping. The movement of the agents when they dropped the tokens was identical in both conditions.
In the experiment condition (Figure 2), both agents took tokens to their gardens i.e. the goal of both agents was to collect tokens (familiarization phase). To manipulate the identity of the agents, the aggressor subsequently stole tokens, moved them to its own garden and bashed the victim (aggression phase). The familiarization and aggression phase took approximately 3 minutes. After these phases, the helping phase started: either the victim or the aggressor dropped a token (trial 1), subsequently the experimenter waited 30 seconds for the child to return the token while jiggling the agent in a standardized way. Then, the other agent dropped the last token (trial 2). Again, the experimenter waited 30 seconds for the child to return the token. If the child returned a token, the agent took the token to its garden. Note that the child could put the token anywhere at the edge of the table, so the child did not necessarily have to return the token to the agent who dropped it.
In the control condition (Figure 3), both agents took tokens and knocked them off the table i.e. the goal of both agents was to get rid of tokens (familiarization phase). To
manipulate the identity of the agents, the aggressor subsequently took tokens from the victim, threw them off the table and bashed the victim (aggression phase). The helping phase was identical to the experiment condition. If the child returned the token, the agent returned to its garden without the token since he did not want the token in the first place (if the agent would have taken the token, this might have obscured the goal of the agent which could affect the subsequent trial).
In both conditions, a voice-over accompanied the familiarization and stealing phases to ensure that children would have the right interpretation of the agents’ actions. Also, pilot studies showed that the voice-over increased children’s attention. As mentioned before, the agents and tokens had a different color and shape in each condition. To discriminate between conditions even more, the agents and tokens were also named differently in the voice-over. Please refer to Appendix 1 for (a translation of) the voice-over.
After each condition, the curtain was lowered and the manipulation check task was done. The task was performed by the parent because the experimenter was not blind to condition. The experimenter handed a tray with the two agents over to the parent and the parent was asked to present the tray to the child. A transparent lid covered the two agents to prevent the child from grasping them immediately. The parent asked the child ‘which one do you take?’ (as in Kanakogi et al., 2013), then the lid was removed and the child could take (one of) the agents. The manipulation check was done twice: for both conditions, the child had to express his preference for one of the agents. After the first manipulation check task, the curtain was lowered and the experimenter prepared the stage for the remaining condition.
Coding
Coding of the video data was performed by the experimenter. Ambiguous situations were discussed with the supervisor until they came to a consensus.
Helping/no helping. After an agent dropped a token, the experimenter waited 30 seconds for the child to respond. If the child returned the token within those 30 seconds, it was coded as helping. Also, if the child tried to return the token but did not manage (e.g. because it was too difficult to get the token through the small gap between the Plexiglas and the paradigm), it was coded as helping.
Helping latency. Helping latencies were computed by subtracting the time when the token dropped from the time when the child put the token back on the table. When children had difficulties picking up or returning the token, the timing of the first attempt was coded. Distraction. Per condition, there were 3 stealing and 3 bashing actions to emphasize the identity of the agent (i.e. aggressor vs. victim). When participants did not focus on the stage during any of the bashings (e.g. they turned around to look at their parents or they looked around in the experiment room), they were labeled as ‘distracted’ and were excluded
from analyses.
Manipulation check. The first agent that the child took was coded as the agent that they preferred. Whenever the child pointed at the agent(s), this was discarded because the question was ‘which one do you take’.
1. Familiarization phase
Both agents took turns to take tokens and bring them to their garden. The goal of both agents was to collect tokens.
At the end of the familiarization phase, both agents had three tokens in their gardens and there were still 2 tokens left on the table.
2. Aggression phase
The aggressor approached the victim, bashed it three times, stole a token, bashed the victim three times again and took the token to its own garden. This happened three times until the victim did not have any tokens left.
3. Helping phase
Both agents took turns to take one of last tokens that were on the edge of the table. They accidentally dropped the token and asked for the token by jiggling in a standardized way.
4. Manipulation check
Both agents were presented and the child was asked to take one.
Figure 2. A visualization of the experiment condition.
1. Familiarization phase
Both agents took turns to take two tokens and throw them off the table. The goal of both agents was to get rid of the tokens. At the end of the familiarization phase, the victim had three tokens left in its garden and there were two tokens left at the edge of the table (note: this was different from the experiment condition to enable the same number of bashings).
2. Aggression phase
The aggressor bashed the victim three times, took a token from the victim’s garden, threw away the token and bashed the victim again. This happened three times until the victim did not have any more tokens in its garden.
3. Helping phase
Both agents took turns to throw away one of the last tokens that were on the table. They dropped the token and jiggled in a standardized way to make sure the behavior in the helping phase was identical to the experiment condition. .
4. Manipulation check
Both agents were presented and the child was asked to pick one.
Figure 3. A visualization of the control condition.
Results
Two different dependent variables were analyzed: the binary variable helping (yes/no) and the continuous variable helping latency. Both variables were analyzed separately, hence results are discussed separately too. All reported p-values are two-sided.
Exclusion criteria
The following participants were excluded: a) Participants who did not pay attention during the bashing and stealing part in at least one of the conditions (n=11), b) participants whose parents interfered in at least one trial (n=3) and c) participants whose data could not be used due to material failure or experimenter error (e.g. agent fell on its side) (n=4). In total, 16 participants met at least one of the exclusion criteria, which resulted in a total sample size of 43.
22 participants were in the experiment condition first. Of those participants, 15 (7) were in the condition where the victim (aggressor) needed help first. 21 participants were in the control condition first. Of those participants, 12 (9) were in the condition where the victim (aggressor) needed help first. Participants were equally distributed over conditions (Fisher’s exact test, N=43, p=.54).
All analyses were run for the total sample (N=59) and the excluded sample (n=43). Results did not differ hence only the results of data with exclusions are reported.
Binary data: helping yes/no
Because of the two within subjects factors Condition and Agent, there was a risk of carry-over effects: Having completed one condition may have affected performance in another condition. If there were carry-over effects, not all four trials per participant should have been included in the analyses. Therefore, the analyses on carry-over effects are reported first. Subsequently, the analyses to test our hypotheses are reported: first, we investigated whether there was more helping in the experiment versus control condition and second, whether there was more helping of the victim than the aggressor. Then, results of the manipulation check are reported and finally, results of some exploratory analyses on individual differences will be presented. Carry-over effects. Both within subject factors Condition and Agent were
counterbalanced. However, helping in the latter trials may have been affected by the first trial(s). More specifically, there could have been a carry-over effect of condition (1) and/or agent (2).
Considering the carry-over effect of condition, there might have been more helping in the control condition when this condition came second because participants were primed by
the experiment condition to help. This hypothesis was not supported by the data (Fisher’s exact test, N=43, p=.20) (Table 2).
Second, there might have been less helping in the experiment condition when this condition came second because participants would be primed by the control condition not to help. This hypothesis was not supported by the data (Fisher’s exact test, N=43, p=.54) (Table 3).
Table 2
Control Condition
Control first Control second
No help 10 5
Help at least once 12 16
Note. Helping in the control condition split by order of conditions Table 3
Experiment Condition
Experiment first Experiment second
No help 7 10
Help at least once 14 12
Note. Helping in the experiment condition split by order of conditions.
Considering the carry-over effect of agent, there might have been more helping of the aggressor when the victim came first because participants were primed by the victim to help. This hypothesis was not supported by the data (Fisher’s exact test, N=43, p=1) (Table 4). Second, there might have been less helping of the victim when the aggressor came first because participants were primed by the aggressor not to help. This hypothesis was not supported by the data either (Fisher’s exact test, N=43, p=1) (Table 5).
Table 4
Helping the Aggressor
Aggressor first Aggressor second
No help 6 10
Help at least once 10 17
Note. Helping of the aggressor split by order of agents.
Table 5
Helping the Victim
Victim first Victim second
No help 9 6
Help at least once 18 10
Note. Helping of the victim split by order of agents.
In sum, there was no support for carry-over effects of condition nor for carry-over effects of agent. Therefore, all four trials (Block 1 and 2) will be included in the analyses. Because of the lack of carry-over effects it was not necessary to analyze the first two trials only (Block 1) or the first trial only (Trial 1). However, for the sake of completeness, these analyses were run as well and results are reported in Appendix 2.
Helping in experiment and control condition. To test whether there was more helping in the experiment versus control condition, what agent was being helped was not taken into account. The dependent variable was whether there was helping (at least one of the agents was helped) or no helping at all. Results of a McNemar’s test show that there was as much overall helping in the experiment as in the control condition (N=43, p=.62) (Table 6). The odds ratio was .33 with a 95% confidence interval from .01 – 4.2. The confidence interval contained the value ‘1’ which suggests that the odds of helping was the same in the
experiment and control condition.
Table 6
Both Conditions
No help in experiment condition
Help at least once in experiment condition
No help in control condition 14 1
Help at least once in control condition
3 25
Note. Helping in the experiment and control condition.
Helping victims and aggressors. To test the hypothesis that the victim would be helped more often than the aggressor, there were McNemar’s test performed for each condition separately.
Inconsistent with the hypothesis, the victim and aggressor were helped equally often in the experiment condition (N=43, p=.46) (Table 7). The odds ratio was 2.5 with a confidence interval from .41 – 26.25. The confidence interval contained the value ‘1’ which suggested that the odds for helping the aggressor and the victim were the same.
It was found that the victim and aggressor were also helped equally often in the
control condition (N=43, p=1) (Table 8). The odds ratio was 1 with a confidence interval from
.13 – 7.47. The confidence interval contained the value ‘1’ which suggested that the odds for helping the aggressor and victim were the same.
Table 7
Experiment Condition
Not helping the victim Helping the victim
Not helping the aggressor 15 5
Helping the aggressor 2 21
Note. Helping the two agents in the experiment condition
Table 8
Control Condition
Not helping the victim Helping the victim
Not helping the aggressor 17 3
Helping the aggressor 3 20
Note. Helping the two agents in the control condition
Manipulation check. Inconsistent with the hypothesis, the participants chose the victim and aggressor equally often.
In the experiment condition, 17 participants chose the victim, 22 chose the aggressor, 3 chose both and there was 1 missing value. A binomial test comparing frequency of
choosing the victim and aggressor yielded no significant difference (N=39, observed proportion =.44, p=52).
In the control condition, 18 participants chose the victim, 19 chose the aggressor, 4 chose both and there were 2 missing values. A binomial test comparing the frequency of choosing the victim and aggressor yielded no significant difference (N=37, observed proportion =.49, p=1).
Furthermore, every participant did the manipulation check twice so consistency of choices was studied. Please refer to Table 9 for an overview of the choices in both conditions (note: only participants who never chose ‘both agents’ and who did not have missing data were included). The choice of the agent in the control condition was independent of the choice of the agent in the experiment condition (Mc Nemar’s test, N=35, p=1), indicating that there was no consistency in choices over the two conditions. In fact, more children chose inconsistently than consistently (Binomial test, N=35, observed proportion=.66, p=.02).
Table 9 Choice Consistency Chose victim in experiment condition Chose aggressor in experiment condition Chose victim in control condition 4 12 Chose aggressor in control condition 11 8
Note. Consistency in choice of agents in the manipulation check, split by condition.
Summary binary data. Both hypotheses have not been supported by the data; children helped equally often in the experiment and control condition. Besides, children helped the victim equally often as the aggressor. The manipulation check did not match expectations either; children chose the victim and aggressor equally often.
Remarkably, there seemed to be a group of children who were very eager to help and a group of children who did not help at all. Some exploratory analyses on individual differences in helping are reported below.
Individual differences (exploratory). Results of the binary data analyses showed that a large group of participants did not help at all (n=14). In contrast, another group of
participants was very eager to help: 29 participants helped at least once and the majority of this group helped more than once (n=26). Thus, participants seem to be divided in 2 groups: a helpful and an unhelpful group.
In an exploratory analysis it was investigated why some children did or did not help consistently. There were four trials in total so each child could have helped between 0 and 4 times, which we called number of helping attempts. We correlated number of helping
attempts to gender, age and temperament scores. Furthermore, we investigated whether there were gender, age and temperament differences between the helpful and unhelpful group. First of all, there was no significant point-biserial correlation between age and number of helping attempts (r(41)=-.19, p=.22). However, there was a marginal significant age difference between the unhelpful group (M=2.14, SD=.13) and the helpful group (M=2.09, SD=.07) (F(1,42)=2.95, p=.09): the children in the unhelpful group were older than in the helpful group. However, categorizing children in ‘helpful’ and ‘not helpful’ groups is rather artificial while the correlation coefficient is a more realistic reflection of the relationship. Based on the fact that this more realistic correlation was not significant and only a marginal significant age difference was found between 2 artificially created groups, this result should be interpreted with caution.
Second, there was no significant point-biserial correlation between gender and number of helping attempts (r(41)=-.03, p=.86). In the unhelpful group, 5 were male and 9 were female. In the helpful group, 14 were male and 15 were female. A Fisher’s exact test showed that the distribution of males and females was the same in both groups (N=43, p=.53), hence gender does not explain why some children helped consistently and others did not help at all. Third, temperament may explain individual differences. As mentioned in the
description of the participants, a part of our sample also participated in an earlier study when they were approximately nine months old3. Temperament was summarized in three subscales: extraversion, negative affectivity and regulation (Gartstein & Rothbart, 2003). There was no significant correlation between number of helping attempts and the three temperament subscales Extraversion (r(20)=.013, p=.95), Negative Affectivity (r(20)=.22, p=.34) and Regulation (r(20)=.23, p=.31). Extraversion scores of the helpful group (M=4.69, SD=.58) were not different from the unhelpful group (M=4.76, SD=.44) (F(1,21)=.09, p=.77). Negative Affectivity scores of the helpful group (M=3.42, SD=.45) were not different from the
unhelpful group (M=3.25, SD=.62) (F(1,21)=.59, p=.45). Regulation scores of the helpful group (M=4.52, SD=.46) were not different from the unhelpful group (M=4.41, SD=.61) (F(1,21)=.21, p=.66).
In sum, neither age, gender nor temperament could explain why some children were eager to help while others were not.
Whom did they help (exploratory). There was a gap between the Plexiglass and the table across the whole length of the table, hence children could return the token anywhere they wanted. In 10.9% of the trials children returned the token somewhere in between the two agents and in another 7.6% of the trials children returned the token to the wrong agent.
Strikingly, children returned the token to the agent who dropped it in the majority of the trials (81.5%) . A binomial test comparing frequency of returning to the agent who dropped it
versus returning it somewhere else was significant (N=92, observed proportion=81.5, p<.001).
Helping latencies
Two different tests have been used to analyze helping latencies: a Repeated Measures ANOVA and Wilcoxon Signed Ranks Tests. The Repeated Measures ANOVA leaves out all missing values which would exclude all participants who did not help. To circumvent this bias, a latency of 40 seconds was artificially assigned to those who did not help. This resulted in data unsuitable for parametric analysis because it was heavily skewed with many data
3 As mentioned before, 33 participants participated in an earlier study; 11 of these 33 participants met at least one of the exclusion criteria hence the temperament analyses are performed on a sample of 22 participants.
19
points on this arbitrary cut-off. These tests did not yield significant results, hence results can be found in Appendix 3. Below are the results of the Repeated Measures ANOVA. Again, analyses of carry-over effects are reported first in order to know what variables to include in the Repeated Measures ANOVA.
Carry-over effects. Again, helping in the latter trials may have been affected by the first trial(s). More specifically, there could have been a carry-over effect of condition (1) and/or agent (2).
Considering the carry-over effect of condition, there might have been faster helping in the control condition when this condition came second because participants were primed with helping by the experiment condition. The results of a One Way ANOVA did not support this hypothesis (F(1,25)=.1.03, p=.32). Second, there might have been slower helping in the experiment condition when this condition came second because participants were primed with not helping by the control condition. The results of a One Way ANOVA did not support this hypothesis (F(1,24)=.006, p=.94). In sum, there were no carry-over effects of condition. Considering the carry-over effect of agent, there might have been slower helping of the victim when the aggressor came first, because participants were primed not to help by the previous trial in which the aggressor needed help. The results of a One Way ANOVA did not support the hypothesis (F(1,26)=1.89, p=.18). Second, there might have been faster helping of the aggressor when the victim came first, because participants were primed to help by the previous trial in which the victim needed help. The results of a One Way ANOVA supported this hypothesis (F(1,25)=12.29, p=.002) (Table 10).
In sum, there was no carry-over effect of condition, but there was a carry-over effect of agent. Therefore, ‘order of agent’ was included in the Repeated Measures analyses reported below.
Table 10
Descriptive Statistics
Agent first Mean latency (s) SD N
Aggressor first Latency aggressor 15.45 5.20 10
Latency victim 8.25 4.37 10
Victim first Latency aggressor 8.74 4.57 17
Latency victim 11.5 6.69 18
Note. Mean latencies for helping the aggressor and victim, split by what agent needed help first.
Because of the carry-over effect of agent, order of agents was included in the model. A Repeated Measures ANOVA was performed with within subjects factors Condition and Agent, between subjects factor Order of agents and dependent variable helping latency
(normal latencies) (all possible interaction effects were included). As the results from the ANOVA do not warrant conclusions in support for the null hypotheses, we also conducted Bayesian analyses.
Helping in experiment and control condition. None of the main effects were significant. First, the main effect of condition was not significant (F(1,11)=.69, p=.42). A Bayesian analysis found weak support for the null hypothesis (p(H0|D)=.66, p(H1|D)=.34). Both analyses show that participants helped equally fast in the experiment and control condition.
Helping victims and aggressors. Second, the main effect of agent was not significant (F(1,11)=2.09, p=.18) suggesting that participants helped the victim as fast as the aggressor. In contrast, a Bayesian analysis found positive support for the alternative hypothesis
(p(H0|D)=.17, p(H1|D)=.83), which suggests that there is a difference in latency between the victim and aggressor. However, due to a significant interaction between Agent and Order of agent (see below), no conclusions can be drawn from this result.
Third, the main effect of order of agents was not significant (F(1,11)=.86, p=.38) suggesting that participants helped equally fast when the victim or aggressor needed help first. In contrast, a Bayesian analysis found strong support for the alternative hypothesis
(p(H0|D)=.06, p(H1|D=.94)), which suggests that there is a difference in latency when the victim or aggressor needed help first. Again, due to a significant interaction between Agent and Order of agent (see below), no conclusions can be drawn from this result.
There was a significant interaction between Agent x Order of Agent (F(1,11)=14.9, p=.003). In accordance with this result, a Bayesian analysis found strong support for the alternative hypothesis (p(H0|D)=.001, p(H1|D=.999)). Post hoc tests were performed to take a closer look at the significant interaction between Agent and Order of agent (3 tests,
Bonferroni corrected alfa: .017). When the aggressor needed help first, the victim was helped faster than the aggressor i.e. participants were faster in the second trial. The mean latency for helping the aggressor was 15.45 seconds, the mean latency for helping the victim was 8.74 seconds (F(1,25)=12.29, p=.002). When the victim needed help first, there was no significant difference between helping the aggressor and victim (F(1,26)=1.89, p=.18) (Table 11), yet the difference was in the same direction (shorter latencies in the second trial) and the difference became marginally significant when no data was excluded (F(1,35)=3.2, p=.08).
All other interaction effects were not significant (Condition x Agent (F(1,11)=.70, p=.42), Condition x Order of Agents (F(1,11)=.29, p=.6, Condition x Agent x Order of Agent (F(1,11)=.70, p=.42).
Table 11
Descriptive Statistics
Agent: Aggressor Agent: Victim
Aggressor first M=15.45, SD=5.2, n=10 M=8.25, SD=4.37, n=10
Victim first M=8.74, SD=4.57, n=17 M=11.5, SD=6.59, n=18,
Note. Mean latency for both agents in block 1 and 2, split by order or agents. When the aggressor
needed help first, participants helped significantly faster in the second trial. When the victim needed help first, participants also helped faster in the second trial although this was not significant. (No significant difference was found between helping the aggressor when the aggressor came first and helping the victim when the victim came first (upper left versus lower right cell in Table 11) (F(1,26)=2.60, p=.12)).
Summary latencies. The hypothesis that there would be faster helping in the experiment than control condition was not supported; children helped equally fast in both conditions. Furthermore, there was no support for the hypothesis that children would help the victim faster than the aggressor.
Discussion
A new procedure to measure children’s helping behavior was developed. To check if helping behavior was actually measured, we examined 2 year olds’ helping behavior in response to agents who needed help (experiment condition) and agents who did not need help (control condition). Besides, to find out to what extent children’s helping behavior was selective, we examined 2 year olds helping behavior in response to victims and aggressors. First, we will discuss the effects of our experimental manipulation (experiment vs. control) on helping behavior, next we will present results regarding the response to victims and
aggressors. Finally, we will briefly discuss individual differences in helping behavior and suggestions for future research.
Did children help?
No difference was found between the experiment and control condition, both when analyzing the binary variable (helping yes/no) and when analyzing helping latencies. That is, children returned the tokens equally often and equally fast to the agent, irrespective of whether the context of the experiment was collecting tokens (experiment condition) or throwing away tokens (control condition). The goal of the agents was different in both conditions, hence the absence of a difference in helping behavior suggests that children did not perceive and respond to the agents’ goals.
There are a few different explanations for the absence of a difference between the two conditions. First, two explanations will be described that assume that children were in fact helping but they misinterpreted the goals of the two conditions. Because of this
misinterpretation, completing the goals of the agents (i.e. helping) resulted in returning tokens in both conditions which is why no condition difference was found. Then, another explanation will be described that assumes that children were returning tokens for other reasons than helping, hence no difference between conditions was found.
First, the way the control condition was designed may have caused the absence of a difference between the experiment and control condition. The intended goal of the agents in the control condition was to get rid of tokens. However, the goal of the agents could have been interpreted differently. Children might have thought the goal of the agents was the activity of throwing away the tokens through the little slots. If that was the children’s
interpretation, children should also return tokens in the control condition so the agents could continue throwing away tokens. This means that children may have helped in both conditions, resulting in no difference between conditions.
Second, another potential reason why we did not find a difference between the experiment and control condition is that the goals of the agents in the current study were rather abstract (collecting tokens and throwing away tokens) while in other helping studies the goals were very specific (e.g. handing over an object that is out of reach). Two year olds begin to master the concept of possession (Rochat, 2011), hence they likely have understood that the agents wanted the tokens (experiment condition) or not (control condition). However, even if they did fully grasp the abstract goals of the agents, completing such an abstract goal (i.e. helping) might still have been too difficult. Usually, returning an object that is dropped is completing a goal on its own. Children from 14 months old are well capable of helping in this situation (Warneken & Tomasello 2007). However, in our setup returning an object
contributed to the higher order goals of ‘collecting tokens’ or ‘getting rid of tokens’. Children may have failed to connect the concrete situation of a dropping token to these higher order goals. This would suggest that children were helping but they were completing a subgoal (i.e. returning an object to someone who dropped it) instead of the higher order goal (abstract goal hypothesis). In this case, children may have helped in both conditions, resulting in no
difference between conditions.
In sum, the two explanations above suggest that children were helping but they misinterpreted the goals of the agents. Completing the agents’ goals therefore resulted in returning tokens in both conditions, hence no difference between the experiment and control condition was found.
However, it is also possible that no helping behavior was measured at all, but children were returning tokens for other reasons. For instance, children might have returned tokens because they thought the tokens belonged on the table. However, in that case, one would expect children to have put the tokens randomly on the table, i.e. they would not specifically return the token to the agent who dropped it (note that there was a gap over the entire length of the table, children could return the token anywhere they liked). However, in the majority of the trials, children returned the token specifically to the agent who dropped it. This suggests that children did not return the token because they thought the tokens belonged on the table.
A more plausible alternative explanation for why children returned tokens is that children just wanted ‘the show’ to continue. In this case, the absence of a difference between conditions was due to the fact that children were not helping. This explanation is consistent with our data since the best way to make the show continue is to return the token to the agent who dropped it. 4
4 A low level priming explanation for why children returned tokens specifically to the agent who dropped it does not seem plausible. If children would return the token to the location where it came from because that location was still activated in memory, children would have acted without any understanding of the situation. This did not seem to be the case: children were often commenting when the token dropped (e.g. ‘Oh! It dropped!’) which suggests some understanding of the situation.
24
In sum, no difference was found between the experiment and control condition. This either reflects that children misinterpreted the goal of the agents, in which case they were helping in both conditions which led to no difference between conditions. Another possibility is that children were not helping but they were returning the tokens because they wanted the show to continue.
Helping victims and aggressors
No difference was found between helping the aggressor and helping the victim, both when analyzing the binary variable (helping yes/no) and when analyzing helping latencies. That is, both agents were helped equally often and equally fast in both conditions. Also, the manipulation check revealed that children did not prefer one of the agents, which suggests that the agent manipulation did not work.
Only when analyzing helping latencies, there was a significant interaction between Agent and Order of Agent. This likely reflects a block order effect. When the victim needed help first, the aggressor was helped significantly more slowly i.e. helping was faster in the second trial. When the aggressor needed help first, helping was also faster in the second trial, although the difference in latencies was not significant. Informal observations suggest this effect was due to the distance between child and experimental setup. Two year olds often move around a lot, which makes it difficult (if not impossible) to keep them in one place during the whole experiment. In this study, children got up from their chair to help in trial 1 but they often did not return to their chair before the second trial started. Therefore they were closer to the dropped token in trial 2 which explains the block order effect. Additionally, some children had trouble returning the token through the small gap between Plexiglass and the table; repeating the action speeded up the process.
The finding that children did not discriminate between the agents in the helping test and in the manipulation check is in contrast to previous studies where children preferred helpers over hinderers (Hamlin, Wynn & Bloom, 2007; Hamlin & Wynn, 2011) and victims over aggressors (Kanakogi et al., 2013). These studies all tested children younger than 1 year while we tested 2 year old children. Possibly, this specific manipulation works for younger children only. When children grow older, they can become attracted to aggression: 3 to 6 year olds who used aggression to achieve a goal were seen as more likeable (Hawley, 2002) and socially competent (Hawley & Vaughn, 2003). In the present study, aggression was also used to achieve a goal - that is collecting or throwing away tokens. The bashing and stealing behavior of the aggressor might have been interpreted as necessary behavior to achieve a goal and may therefore have not been evaluated negatively.
The children in our sample were right between the age where they prefer the victim (younger than 1 year) and the age where they become attracted to aggression (around 3 years). This might have caused the null results; some preferred the victim, others the aggressor and most of them chose inconsistently implicating that they did not show a strong preference for any of the agents. This is in accordance with our data: some children were clearly amused by the aggression phase (attracted to aggression), while others were slightly concerned or seemed indifferent.
In sum, the absence of a difference between helping the victim and aggressor may be due to the manipulation that did not yield a consistent negative (positive) social evaluation of the aggressor (victim). Whether children’s helping behavior is selective is still unclear: it is still possible that children are biased to help victims over aggressors, provided that the aggressor is evaluated negatively.
Individual differences
Remarkably, our sample was divided in children who returned tokens and children who did not return tokens. Exploratory analyses revealed that nor gender, nor age nor temperament at an age of nine months explained why some children were eager to return tokens while others were not.
First, we investigated age differences because the agents’ actions might have been difficult to understand for children. Older children may have been more likely to understand the agents’ actions which may have led to more helping. No such correlation between age and returning was found, although this could be due to the small age range.
Second, we investigated gender differences because some studies suggested that girls were more prosocial than boys (Abramovitch, Corter & Pepler, 1980) and girls were more helpful than boys (Rheingold, 1982). No gender differences were found. There was no
correlation between gender and returning tokens; boys and girls were equally distributed over the group who was eager to return tokens and the group who was not. This is in accordance with recent helping studies in children that did not report gender differences either (Kenward & Gredebäck, 2013; Warneken & Tomasello, 2006; Warneken & Tomasello, 2007).
Third, we investigated temperament differences. At an age of nine months,
temperament was documented of half the children in our sample. We linked temperament to returning tokens, yet no relation was found. This is in contrast to other studies that found a relation between temperament and prosocial behavior (Stanhope, Bell & Parker-Cohen, 1987; Suda & Fonts, 1980). However, some subscales of the temperament measure do change slightly over time (Gartstein, Rothbart, 2003), which may explain why we did not find a relation between temperament and helping behavior. Besides, the absence of a significant
difference between the experiment and control condition suggests that we may not have found a relation between temperament and returning tokens because returning tokens did not reflect helping behavior.
Another variable that may explain individual differences in helping behavior is the family of the child. First of all, children whose parents are warm, supportive and respond to their children’s needs tend to be more prosocial (Knafo et al., 2008). Second, children with younger siblings tend to be more prosocial (Hastings, Utendale, & Sullivan, 2007) Also, the amount parents talk about other’s emotions and mental states is related to prosocial behavior. Brownell, Svetlova, Anderson, Nichols and Drummond (2013) analyzed the way parents read books to their 18 and 24 month old children. Parents differed in how much they asked their children about the mental and emotional state of the book characters. The more parents elicited emotion related talk, the faster and more often children helped.
These studies illustrate that environmental factors like parenting style affect children’s helping behavior, hence they may explain individual differences in helping behavior.
Unfortunately, we did not measure these variables hence we could not test whether these variables explained why some children in our sample were eager to return tokens while others were not.
Future research
Recapitulating, the absence of a difference between the experiment and control condition may be due to the abstractness of the agents’ goals. Potentially, older children would understand these abstract goals which would allow us to test whether this new experimental setup is appropriate for measuring helping behavior. Additionally, approximately 4 year olds are very sensitive to violations of fairness (LoBue, Nishida, Chiong, DeLoache, & Haidt, 2011; Olson & Spelke, 2008; Warneken, Lohse, Melis, & Tomasello, 2011) which increases the chance of a succesful victim/aggressor manipulation. On the other hand, the setup could also be adjusted to make it appropriate for
approximately 2 year olds. This would allow us to test whether children’s helping behavior is selective right from the moment when they start helping others. To make the setup easier, one can make the setup similar to other helping studies where the way an object is dropped clarifies whether the action is intentional or accidental. In that case, the goal of the agent is very simple: he wants the token or he does not. Children from about 14 months old are already capable of helping when an object is dropped. Therefore, the proposed method is likely to work for very young children. The younger the children, the more likely it is that the bashing manipulation will yield a consistent preference for the victim over the aggressor.
Conclusion
A new procedure for measuring helping behavior was developed. To validate if this method measured helping behavior, we examined 2 year olds’ helping behavior in response to agents who needed help (experiment condition) and agents who did not need help (control condition). Children returned tokens equally fast and equally often in both conditions. This may reflect that the goals of the agents were misinterpreted which led to helping in both conditions hence no condition difference was found. Another possibility is that children did not help but rather returned tokens because they wanted the show to continue.
To find out whether children’s helping behavior was selective, we examined children’s helping behavior in response to victims and aggressors. Children helped both agents equally fast and equally often. The manipulation check yielded no result either: children did not prefer one of the agents. In hindsight, we may have found no difference in helping behavior because the manipulation was not strong enough: Children may not have negatively evaluated the behavior of the aggressor hence the aggressor (victim) was not seen as particularly negative (positive). In that case, both agents were evaluated similarly which explains why there was no effect on helping behavior. Results of this study do not necessarily mean that children’s helping behavior is not selective, it merely means that the identity of the two agents was not manipulated well.
From an evolutionary perspective, prosocial behavior to non-relatives benefits the performer in the long term because favors will be returned. Consequently, one needs to behave more prosocially towards people who are likely to return the favor. The same applies to helping behavior: people should be biased to help some individuals over others. This is the first study that tests whether this bias is present in children, but unfortunately the answer is yet to be found.
References
Abramovitch, R., Corter, C., & Pepler, D. J. (1980). Observations of mixed-sex sibling dyads. Child Development, 1268-1271.
Batson, C. D., Duncan, B. D., Ackerman, P., Buckley, T., & Birch, K. (1981). Is empathic emotion a source of altruistic motivation? Journal of personality and Social
Psychology, 40, 290.
Blake, P. R., & Harris, P. L. (2011). Early representations of ownership. New directions for child an adolescent development, 2011, 39-51.
Brownell, C. A., Svetlova, M., Anderson, R., Nichols, S. R., & Drummond, J. (2013). Socialization of early prosocial behavior: parents’ talk about emotions is associated with sharing and helping in toddlers. Infancy, 18, 91-119.
Connor, R.C. & Norris, K.S. (1982). Are dolphins reciprocal altruists? Am. Natural. (199), 358–72.
De Waal, F. B. (2008). Putting the altruism back into altruism: The evolution of empathy. Annual Review Psychology, 59, 279-300.
Dunfield, K. A., & Kuhlmeier, V. A. (2010). Intention-mediated selective helping in infancy. Psychological Science, 21, 523 – 527.
Dunfield, K. A., & Kuhlmeier, V. A. (2013). Classifying prosocial behavior: Children's responses to instrumental need, emotional distress, and material desire. Child development, 84, 1766-1776.
Dunfield, K., Kuhlmeier, V. A., O’Connell, L., & Kelley, E. (2011). Examining the diversity of prosocial behavior: Helping, sharing, and comforting in infancy. Infancy, 16, 227-247.
Gartstein, M. A., & Rothbart, M. K. (2003). Studying infant temperament via the Revised Infant Behavior Questionnaire. Infant Behavior and Development, 26, 64-86.
Hamilton, W.D. (1964). The genetical evolution of social behaviour: I & II. Journal of Theoretical Biology, 7, 1-52.
Hamlin, J. K., Wynn, K., & Bloom, P. (2007). Social evaluation by preverbal infants. Nature, 450, 557-559.
Hamlin, J. K., & Wynn, K. (2011)Young infants prefer prosocial to antisocial others. Cognitive Development, 26, 30-39.
Hastings, P. D., Utendale, W. T., & Sullivan, C. (2007). The socialization of prosocial development. Handbook of socialization: Theory and research, 638-664.
Hawley, P. H. (2002). Social dominance and prosocial and coercive strategies of resource control in preschoolers. International Journal of Behavioral Development, 26, 167-176.
Hawley, P. H., & Vaughn, B. E. (2003). Aggression and adaptive functioning: The bright side to bad behavior. Merrill-Palmer Quarterly, 49(3), 239-242.
Hepach, R., Vaish, A., & Tomasello, M. (2012). Young children are intrinsically motivated to see others helped. Psychological science, 23(9), 967-972.
Kanakogi, Y., Okumura, Y., Inoue, Y., Kitazaki, M., & Itakura, S. (2013). Rudimentary sympathy in preverbal infants: preference for others in distress. PloS one, 8, e65292.
Kenward, B., & Gredebäck, G. (2013). Infants help a non-human agent. PloS one, 8, e75130.
Knafo, A., Israel, S., Darvasi, A., Bachner‐Melman, R., Uzefovsky, F., Cohen, L., ... & Ebstein, R. P. (2008). Individual differences in allocation of funds in the dictator game associated with length of the arginine vasopressin 1a receptor RS3
promoter region and correlation between RS3 length and hippocampal mRNA. Genes, brain and behavior, 7, 266-275.
Kochukhova, O., Gredebäck, G. (2010) Preverbal Infants Anticipate that Food will be Brought to the Mouth: An Eye Tracking Study of Manual Feeding and Flying Spoons. Child Development, 81, 1729-1738.
Krebs, D. L. (1975). Empathy and Altruism. Journal of Personality and Social Psychology, 32, 1134 – 1146.
Krebs, D. L. (2008). Morality: an evolutionary account. Perspectives on Psychological Science, 3, 149-172.
LoBue, V., Nishida, T., Chiong, C., DeLoache, J. S., & Haidt, J. (2011). When getting something good is bad: Even three‐year‐olds react to inequality. Social
Development, 20, 154-170.
Olson, K. R., & Spelke, E. S. (2008). Foundations of cooperation in young children. Cognition, 108, 222-231.
Premack, D., & Premack, A. J. (1997). Infants attribute value±to the goal-directed actions of self propelled objects. Journal of cognitive neuroscience,9(6), 848-856.
Rheingold, H.L. (1982) Little children's participation in the work of adults, a nascent prosocial behavior. Child Development. 53, 114–125.
Rochat, P. (2011). Possession and morality in early development. New directions for child and adolescent development, 2011(132), 23-38.
Stanhope, L., Bell, R. Q., & Parker-Cohen, N. Y. (1987). Temperament and helping behavior in preschool children. Developmental Psychology, 23, 347.
Suda, W., & Fonts, G. (1980). Effects of peer presence on helping in introverted and extroverted children. Child Development, 51, 1272-1275.
Vaish, A., Carpenter, M., & Tomasello, M. (2009). Sympathy through affective perspective taking and its relation to prosocial behavior in toddlers. Developmental psychology, 45, 534.
Vaish, A., Carpenter, M. & Tomasello, M. (2010). Young Children Selectively Avoid Helping People With Harmful Intentions. Child Development, 81, 1661–1669.
Warneken, F., Lohse, K., Melis, A. P., & Tomasello, M. (2011). Young children share the spoils after collaboration. Psychological Science, 22, 267-273.
Warneken, F., & Tomasello, M. (2006). Altruistic helping in human infants and young chimpanzees. Science, 311, 1301-1303.
Warneken, F., & Tomasello, M. (2007). Helping and cooperation at 14 months of age. Infancy, 11, 271-294.
Warneken, F., & Tomasello, M. (2009). The roots of human altruism. British Journal of Psychology, 100, 455-471.
Woodward, A. L. (1998). Infants selectively encode the goal object of an actor’s reach. Cognition, 69, 1–34.
Appendix 1 – Translation of the voice-over
Experiment condition
Look at these two creatures, there is a yellow creature (yellow creature wiggled) and a red creature (red creature wiggled). The creatures want tokens!
Look at the yellow creature, it takes a token. Look at the red creature, it takes a token.
Look at the yellow creature, it takes another token. Look at the red creature, it takes another token.
But what’s happening now? The creature hits the other creature and steals a token. And bashes the other creature again! (Identity of aggressor should not be revealed to parent, hence the voice-over does not say what creature is hitting the other one.)
Again, the creature hits the other creature and steals a token. And bashes the other creature again!
(Third time bashing was not accompanied by voice-over to let children get used to moving agents without voice-over, since there was also no voice-over during the helping phase.)
Control condition
Look at these two creatures, there is a yellow creature (yellow creature wiggled) and a red creature (red creature wiggled). The creatures don’t want tokens!
Look at the yellow creature, it throws away a token. Look at the red creature, it throws away a token.
Look at the yellow creature, it throws away another token. Look at the red creature, it throws away another token.
But what’s happening now? The creature hits the other creature and throws away a token. And bashes the other creature again! (Identity of aggressor should not be revealed to parent, hence the voice-over does not say what creature is hitting the other one.)
Again, the creature hits the other creature and throws away a token. And bashes the other creature again!
(Third time bashing was not accompanied by voice-over to let children get used to moving agents without voice-over, since there was also no voice-over during the helping phase.)
