Seven‐year‐olds’ aggressive choices in a computer game can be predicted in infancy

Abstract Concerns about the relationship between computer games and children's aggression have been expressed for decades, but it is not yet clear whether the content of such games evokes aggression or a prior history of aggression promotes children's interest in aggressive games. Two hundred and sixty‐six 7‐year‐old children from a nationally representative longitudinal sample in the UK played a novel computer game (CAMGAME) in which the child's avatar encountered a series of social challenges that might evoke aggressive, prosocial or neutral behaviour. Aggressive choices during the game were predicted by well‐known risk factors for aggressive conduct problems and the children's own early angry aggressiveness as infants. These findings suggest that children who are predisposed to aggression bring those tendencies to virtual as well as real environments.


| INTRODUCTION
Concerns about the relationship between computer games and children's aggression have been expressed for decades (e.g., Silvern & Williamson, 1987). It is increasingly acknowledged that spending more time playing video games does not necessarily lead to violent behaviour (Cunningham, Engelstatter, & Ward, 2016;DeCamp, 2015), and so the focus for research has now turned to possible links between game-playing and milder forms of aggression (Ferguson, 2015). However, even this claim has been met with scepticism (e.g., McCarthy, Coley, Wagner, Zengel, & Basham, 2016).
Two opposing causal models -exposure effects versus selection effects -have been offered to explain a correlation between gameplaying and aggression. Advocates of exposure effects propose that high levels of game-playing might foster aggressive tendencies via the interplay of arousal within the game context and social learning processes, for example, identification with violent game avatars which might lead to higher rates of aggression in and out of game contexts (e.g., Anderson & Bushman, 2002). For example, young people in a large national cohort sample who played video games with violent content were at a somewhat increased risk for conduct disorder in adolescence (Etchells, Gage, Rutherford, & Munafó, 2016); however, playing violent video games was not associated with violent behaviour in a study that controlled for the adolescents' social environment (DeCamp, 2015). Longitudinal analysis of the UK Millennium Cohort Study revealed that conduct problems were predicted not by the time children spent playing computer games but rather by the hours spent watching television (Parkes, Sweeting, Wight, & Henderson, 2013).
Alternatively, advocates of selection effects propose that a pre-existing disposition to be aggressive, possibly underpinned by This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. neurobiological processes, might heighten an individual's interest in computer games, particularly violent ones (e.g., Elson & Ferguson, 2014). For example, in a representative sample of German adolescent gamers, physical aggression predicted the use of violent video games one year later, but the reverse was not true, even when controlling for sex, education and frequency of video game play (Breuer, Vogelgesang, Quandt, & Festl, 2015). Tests of these alternative models require not just evidence for the influence of exposure to computer games but also suitable controls for potentially confounding factors. In contexts where the child might make different choices in a game, children with a history of anger and aggression may be more likely to enact violent solutions to the dilemmas posed by the game. It is therefore important to obtain pretest information about children's prior aggressiveness before concluding that game-playing makes children more aggressive. Longitudinal studies are required. However, to be convincing, longitudinal analyses must begin very early in development; the pathway to problematic levels of aggression originates in the first 2 years of life (e.g., Alink et al., 2006;Côté, Vaillancourt, LeBlanc, Nagin, & Tremblay, 2006;Lorber, Del Vecchio, & Smith Slep, 2015;NICHD Early Child Care Research Network, 2004), long before children develop the motor skills to play computer games.
Our longitudinal study of a representative British sample has demonstrated that individual differences in anger and the use of physical force emerge in the first year of life and predict aggression in childhood . We hypothesized that individual differences originating in infancy might also predict children's later aggressiveness in virtual environments, even when controlling for other risk factors for early aggression and subsequent exposure to computer games. Therefore, as part of a battery of measures for an assessment of our longitudinal sample at 7 years of age, we created a bespoke first-person perspective game in which the protagonist avatar encounters a number of social challenges from game characters that might evoke aggression or alternative responses. The fact that all the children in the sample were assessed in infancy and tested on a standardized game in middle childhood made it possible to control for early aggressive tendencies, prior to any experience of playing games.
If children's aggressive choices in the game could be predicted by their angry aggressiveness in infancy, long before they could physically play computer games, these analyses would provide strong evidence for selection effects. We controlled for key risk factors that were earlier found to be associated with aggression in infancy, in particular socioeconomic disadvantage in the family, parental aggression and mothers' depression in pregnancy (Dodge, Coie, & Lyman, 2006;Tremblay et al., 2004;Waters, Hay, Simmonds, & van Goozen, 2014), as well as the children's current exposure to computer games. Committee. Based on initial power calculations, 332 primiparous women were recruited from antenatal services in two National Health Service (NHS) Health Care Trusts in Wales, UK (see Hay et al., 2014).

| Participants
The recruitment strategy yielded a nationally representative sample, not significantly different on socio demographic characteristics from the families of firstborn children in the most recent UK national cohort study (Connolly & Platt, 2014;Kiernan, personal communication, 2009). By 7 years of age, 22 families had withdrawn from the study and one had never been traced, leaving 309 (93%) available for study. Of those, 287 (93%) were assessed at age 7. The current analysis focuses on 266 children (93% of those assessed at 7 years) who played a bespoke computer game designed to allow aggressive or alternative choices in response to social dilemmas. Sociodemographic characteristics are reported in Table 1.

| Materials
The Castell Arth Mawr Adventure Game (CAMGAME) is a first-person perspective game inspired by the classic Robbers Cave experiment (Sherif, Harvey, White, Hood, & Sherif, 1949) which induced conflict between two groups of children attending a summer camp. The game script, which was approved by the Cardiff University School of Psychology Research Ethics Committee, was instantiated in a modified version of the game The Elder Scrolls V: Skyrim (Bethesda, 2011) using freely available development tools for that game (see Supplementary Materials). Children were asked to imagine they were on a school trip to a Welsh castle, accompanied by a teacher and classmates who were wearing red sweatshirts with their school logo. As they explored the castle, they discovered that strange children from another school, wearing blue sweatshirts, were also exploring the castle. The red-shirted and blue-shirted school groups were competing to find treasure hidden by the Bear King somewhere in the castle. Note. Basic educational qualifications are defined as 5 or more pass marks on the national GCSE examinations in the UK or their equivalents. The socioeconomic adversity score for the subsample did not differ significantly from the overall sample recruited in pregnancy.

Several challenges that might provoke aggression (taunts and shoving
by the unfamiliar peers) and empathy (distress or injury shown by another person) were embedded into the script.
In this first-person game, each player saw his/her own pair of hands picking up a mallet that could pound things or people. The mallet could be used prosocially to respond to requests for help made by characters in the game or aggressively to hit other people.

| Procedure
Parents were interviewed at home and completed demographic questionnaires during the third trimester of pregnancy. The children were assessed at mean ages of 6, 12, 21 and 33 months in an alternating sequence of home and laboratory visits, with two home visits at age 7. One postgraduate researcher interviewed the parent and another tested the child on measures of cognition, language and social problem-solving, as well as the CAMGAME.

| Socioeconomic adversity
A general index of child's exposure to maternal factors known to be associated with risk for social adversity was created using polychoric principal component analysis. The maternal experiences that contributed to this index were: (1) the mother not having achieved basic educational attainments (i.e., the mother having no qualifications or fewer than five GCSEs or equivalent attainments); (2) the mother being 19 years of age or under at the time of child's birth; (3) the mother not being legally married during the pregnancy; (4) the mother not being in a stable couple relationship during the pregnancy; and (5)

| Prenatal depression
The Wave 1 prenatal interview incorporated the mood disorder and anxiety disorder sections of the Schedules for Clinical Assessment in Neuropsychiatry (SCAN; Wing et al., 1990), with an additional screen

| Angry aggressiveness in infancy
At 6  into a car seat and observed tendencies to strike out at or grab toys from peers (Hay et al., 2010) and predicted the children's later aggressive conduct problems at age 3 years .
CICS factor scores were derived through a measurement model using Mplus 7 whereby a latent dimensional construct was estimated using the maternal, paternal and third informant's assessments as indicators. These factor scores, analogous to standardized scores, were constrained to have a mean of 0 and SD = 1. Mplus 7 (Muthén & Muthén, 2012) was used to implement this measurement model and calculate a factor score.

| Children's aggressive choices in the CAMGAME
Five events that might evoke aggressive choices were embedded into the game (Table 2). In each case, the child's use of the mallet against game characters was used as an operational definition of an aggressive choice, which was given a score of 1. Testers recorded the child's response to each event on an IPad. Scores from each challenge were summed (range, 0 to 5), yielding a scale of aggressive response to challenges in the game that showed internal consistency (α = .69) across five vignettes in the game. The proportion of aggressive choices was computed for each child and an arc sine transformation of those proportion scores was used.
A further check on the validity of the child's use of the mallet as a measure of aggression against game characters was made by examining mallet use in two vignettes that were directed to inanimate objects (smashing bottles in the absence of any requests for help), not child or adult characters. A principal components analysis showed that use of the mallet for bottle-smashing loaded on a separate factor from the five challenges that entailed aggression against human characters.

| Teachers' ratings of children's aggressive behaviour problems
Children's current levels of aggression at age 7 were assessed by classroom teachers, using the 19-item aggressive syndrome scale of Teacher Report Form (TRF: Achenbach & Edelbrock, 1986).
Coefficient alpha for the TRF aggression scale was .95 in this sample.

| RESULTS
Descriptive statistics and univariate correlations between the study variables are presented in Table 3. Correlations for girls are above the diagonal; correlations for boys are below the diagonal. Cave Scene X enters the scene and is pushed back by the blue school children, who say, 'You red school loser!' and 'I pushed you that means go away!' X's friends from the red school are on the other side of the room. One of the children is on the floor and says, 'Ow! They pushed me!' If X approaches the blue school children, X is further taunted, 'Yeah we pushed you and your friends, so what? We're gonna get the treasure before you.' If X approaches the red school friend, X hears, 'Oh thank you, I'll be okay.' X uses the mallet to hit the blue school child X walks over to the injured red school friend Ditch Scene X walks along the path and hears the red school friends say, 'Those mean bullies pushed us down here.' The blue school children are at the end of the path whilst the red school friends are at the bottom of a staircase. As X moves closer to the blue school children they say, 'Oh look, it's stupid again.' X uses the mallet to hit the blue school child X walks to injured red school friend Racing to the Treasure Scene In this final scene, X is trying to find the entrance to the castle. X encounters the red school friends and the blue school children who are also 'racing' to the entrance. There is no taunting speech from the blue school children in this scene.
X uses the mallet to hit the blue school child N/A

| Aggressive choices in the game context
The percentage of children who hit human characters ranged from 34% in response to the unfamiliar children's taunts or pushing to 13% in response to the request for help from an old woman. A slight majority of children (51%) never made an aggressive choice.
The mean proportion of aggressive choices across the chal-

| Early prediction of aggressive choices in the game
The first set of analyses tested for links between angry aggressiveness in infancy and aggressive choices in CAMGAME, while controlling for the key risk factors of socioeconomic adversity, maternal aggressiveness, prenatal depression, and male sex, which at the univariate level had been found to be correlated with the infants' angry aggressiveness . These risk factors were entered at the first step of a logistic regression model, with the child's angry aggressiveness in infancy added at the second step. The analysis revealed that, even when the risk factors were taken into account, angry aggressiveness in infancy predicted the children's aggressive choices in the computer game at age 7 (Table 4).

| DISCUSSION
The aim of our study was to examine children's aggressive choices in a computer game, as a way to tap into the moment-to-moment decisions they make in response to social dilemmas that could provoke aggression. Our findings suggest that those children in a representative community sample who showed early signs of anger and aggression were subsequently likely to spend more time playing computer games and to make aggressive choices in those games. Our findings are relevant to the hypothesis that aggressive children opt to play aggressive games, rather than the games themselves making children more aggressive (Elson & Ferguson, 2014). Our longitudinal design and assessment of infants' signs of anger and use of force prior to developing the motor skills needed for the playing of computer games provided a unique opportunity to conduct a strong test for selection versus exposure effects.
As expected in middle childhood, boys made more aggressive choices in the game than girls did. However, when gender was taken into account, prenatal exposure to the mothers' depression was the key family risk factor for aggressive choices in the game, just as it is a known risk factor for anger, aggression and violent crime .
In this study, the children played a bespoke game that was not designed to be violent and therefore extremely aggressive behaviour was not modelled by characters in the game. Rather, the children were responding to fairly mild provocations -for example, being bumped into or taunted by peer avatars. In some instances, they responded with aggression rather than prosocial behaviour to game characters' requests for assistance. It would be of interest to see whether prior risk factors and early aggressiveness also predicted aggressive choices in a more violent game where aggressive actions were more explicitly modelled.
Our study has limitations. Assessment of the children's exposure to games relies on one parent's report. Only one game was played. As in much of the literature on computer games (Ferguson, 2015), effect sizes are relatively small. Situational factors, motoric competence, and expertise in the playing of computer games are likely to influence the children's playing of the game. However, the findings are in line with other evidence for selection effects in the literature on gameplaying (e.g., Breuer et al., 2015) and extend other research that draws attention to the early origins of an aggressive approach to the social world (e.g., Alink et al., 2006;Baillargeon et al., 2007;Breuer et al., 2015;Côté et al., 2006;NICHD Early Child Care Network, 2004).
Early aggressive tendencies predict behaviour in virtual as well as real environments.
Our findings suggest that the emotional and decision-making processes underlying the deployment of aggression might be probed using such virtual environments, such as the novel game paradigm created for this research. Furthermore, aggressive children's attraction to computer games opens up possibilities for future game-based intervention strategies to reduce aggressive decision-making.