Abstract
Theory of mind is one of the most important cognitive factors in social information-processing, and deficits in theory of mind have been linked to aggressive behavior in childhood. The present longitudinal study investigated reciprocal links between theory of mind and two forms of aggression – physical and relational – in middle childhood with three data waves over 3 years. Theory of mind was assessed by participants’ responses to cartoons, and physical and relational aggression were assessed through teacher reports in a community sample of 1657 children (mean age at Time 1: 8 years). Structural equation modeling analyses showed that theory of mind was a negative predictor of subsequent physical and relational aggression, both from Time 1 to Time 2 as well as from Time 2 to Time 3. Moreover, relational aggression was a negative predictor of theory of mind from Time 1 to Time 2. There were no significant gender or age differences in the tested pathways. The results suggest that reciprocal and negative longitudinal relations exist between children’s theory of mind and aggressive behavior. Our study extends current knowledge about the development of such relations across middle childhood.
Keywords
Aggressive behavior is one of the most pressing social problems in our world today, because of its detrimental effects on the lives and health of human beings (Dahlberg & Krug, 2002). Already in childhood, aggression has negative consequences, for instance, in the form of maladjustment of both children who show aggressive behavior (e.g. Crick et al., 2006) and children who are the targets of that behavior (e.g. Crick et al., 2001). Social cognition in general and theory of mind (ToM) in particular have been recognized as core processes involved in the development of aggressive behavior (Huesmann, 1998). However, whether ToM is negatively or positively linked to aggression, and whether ToM is a cause or even a consequence of aggression are issues that have not yet been sufficiently addressed. Thus, the present study aimed at examining potential reciprocal relations between ToM and aggressive behavior in middle childhood.
Aggressive behavior is defined as “any form of behavior directed toward the goal of harming or injuring another live being who is motivated to avoid such treatment” (Baron & Richardson, 1994, p. 7). Harming others can be done by physical means, for example by hitting or kicking, known as physical aggression (PHY-A; e.g. Crick, Casas, & Ku, 1999), or by damaging their friendships or feelings of inclusion in the peer group, for example by spreading rumors, known as relational aggression (REL-A; e.g. Crick & Grotpeter, 1995). Some authors use the terms indirect or social aggression instead of REL-A. However, all three constructs essentially refer to the same form of aggression (Archer & Coyne, 2005), we refer to REL-A, even when the cited authors used an alternative term. Regarding the developmental course of the forms of aggression, the majority of children, girls in particular, reduce their use of PHY-A (Côté, Vaillancourt, Barker, Nagin, & Tremblay, 2007) and tend to increase their use of REL-A (Crick et al., 1999) during middle childhood.
There are typical gender differences in the two forms of aggression, with boys engaging in PHY-A more often than girls (e.g. Lansford et al., 2012). Theoretical explanations for this gender difference in PHY-A include hormonal, evolutionary, and social role accounts (see Krahé, 2013, for review). With regard to REL-A, several studies have found that girls engage in REL-A more often than boys (e.g. Ostrov, Murray-Close, Godleski, & Hart, 2013). One explanation is that girls tend to have more dyadic relationships than boys, which seems to fuel REL-A (Murray-Close, Ostrov, & Crick, 2007). However, other studies revealed either no gender differences in REL-A (e.g. Lansford et al., 2012) or higher scores for boys (e.g. Henington, Hughes, Cavell, & Thompson, 1998). These mixed findings highlight the importance of conducting further research on the developmental courses of PHY-A and REL-A during middle childhood, thereby including both forms of aggression and differentiating developmental patterns for girls and boys.
The Social Information-Processing (SIP) Theory of Aggression
Crick and Dodge’s (1994) social information-processing (SIP) model proposes that deficits and biases in processing social information, especially the misinterpretation or neglect of important social cues, can cause social maladjustment in children, including aggressive behavior. SIP is conceptualized as a sequence of six processing stages: (1) encoding and (2) interpretation of social cues, (3) clarification of individual goals, (4) access to or construction of potential behavioral responses, (5) decision for a response, and (6) behavioral enactment. This stepwise approach has helped to identify differences in SIP patterns in children who show higher as compared to lower levels of aggression. For instance, aggressive children tend to attribute hostile intentions to their peers in ambiguous social situations more often than non-aggressive children (de Castro, Veerman, Koops, Bosch, & Monshouwer, 2002, SIP Steps 1 and 2). Furthermore, aggressive children generate more aggressive responses than their non-aggressive peers (de Castro, Merk, Koops, Veerman, & Bosch, 2005, SIP Step 4).
In terms of the chronification of aggressive behavior, Crick and Dodge (1994) assumed that the relation between social experiences and SIP is reciprocal. They proposed that mental representations of social experiences are stored in long-term memory, in a so-called database, and are then accessed in each of the postulated SIP steps. Thus, social knowledge is retrieved and, in turn, shapes behavior at each step. Consequently, social cognition influences the occurrence of aggressive behavior, and, in turn, memories of past aggressive behavior influence social cognition. In one of the few prospective studies that have examined this reciprocal link, Lansford, Malone, Dodge, Pettit, and Bates (2010) demonstrated a dynamic developmental cascade in which SIP and aggressive behavior influenced one another over the course of middle childhood. Crick and Dodge (1994) proposed investigating the impact of social experiences, such as aggressive behavior, on social cognitions. This could foster understanding of possible mechanisms by which children develop outcome expectations for specific social situations, which, in turn, children may use to reach a response decision (SIP Step 5).
ToM, which enables individuals to “predict what others are going to do on the basis of their desires and beliefs” (Frith & Frith, 2010, p. 165), is involved in at least some steps of the SIP model (Harvey, Fletcher, & French, 2001). ToM is typically distinguished into two facets: Making inferences regarding others’ beliefs, intentions, or desires, refers to cognitive ToM, and inferring others’ emotions refers to affective ToM (e.g. Shamay-Tsoory, Harari, Aharon-Peretz, & Levkovitz, 2010). In the last decade, behavioral (Devine & Hughes, 2013) as well as neuroscientific studies (Vetter, Weigelt, Döhnel, Smolka, & Kliegel, 2014) confirmed the further development of ToM after the accomplishment of understanding of false belief during preschool age. The development of cognitive ToM (Devine & Hughes, 2013) as well as the understanding of advanced affective ToM (Vetter et al., 2014) are assumed to continue throughout middle childhood and adolescence.
Both facets of ToM are assumed to influence at least some of the steps postulated in the SIP model. For instance, ToM plays an important role in attributional processes and in forming a mental representation of a social situation (Teufel, Fletcher, & Davis, 2010), both of which are part of encoding and interpretation of social cues (SIP Steps 1 and 2). Furthermore, ToM seems to play a role in reaching a response decision (SIP Step 5), as has been shown in a peer-coordination game, in which 6-year-old children adjusted their response decisions by applying their ToM skills (Grueneisen, Wyman, & Tomasello, 2015).
Theory of Mind and Aggressive Behavior
Numerous studies have examined the relation between ToM and aggressive behavior in children. However, most of these studies yielded mixed results, were situated in preschool age, were cross-sectional, or did not control for earlier levels of behavior. Some studies supported the “social skills deficit” view described above, which assumes that deficits in ToM co-occur with a pattern of SIP that fosters aggressive behavior (Crick & Dodge, 1994). For instance, deficits in ToM were correlated with behavioral problems in 2-year-olds (Hughes & Ensor, 2006), as well as in a clinical sample of 5- to 9-year-olds (Fahie & Symons, 2003). Similarly, preschool-aged children with superior ToM were rated as being less aggressive toward their peers (Capage & Watson, 2001).
However, in contrast to the social skills deficit view, Sutton, Smith, and Swettenham (1999a) proposed that children who show aggressive behavior toward their peers, instead of being “socially inadequate”, are more socially competent than their victims. This reasoning comes primarily from research on bullying, as a specific form of aggressive behavior that is ongoing and involves a power asymmetry between perpetrator and victim (Olweus, 2013). In children with low ToM abilities, bullying may result from deficits in SIP (e.g. misinterpretation of social cues), as described above. However, bullies also may have superior ToM abilities that enable them to manipulate others, which was demonstrated particularly for ringleader-bullies in middle childhood (Sutton, Smith, & Swettenham, 1999b). Sutton et al. (1999a) further argue that ToM abilities may be particularly relevant for the use of relational forms of bullying, because children with higher ToM abilities are better able to anticipate the victim’s reactions to REL-A, and by which means they can best harm their victim. Although this assumption was formulated for relational bullying, it may be applicable to REL-A in general, because most relationally aggressive acts require the involvement of others. In line with this reasoning, in one of the few longitudinal studies addressing the link between ToM and aggression, 5-year-old children’s ToM skills were positively related to REL-A one year later, but only in children with average or low levels of prosocial behavior (Renouf et al., 2010). In this study, PHY-A was unrelated to ToM.
Because most previous studies were cross-sectional (Gomez-Garibello & Talwar, 2015), or did not control for earlier levels of behavior, not much is known about the direction of effects between ToM and different forms of aggression, PHY-A and REL-A, particularly in middle childhood. There is some evidence that ToM influences later aggression, for instance in preschool age (Renouf et al., 2010). However, the reverse path is also possible, as has been found in preschoolers whose aggressive behavior at 2.5 years negatively predicted their ToM skills 4 months later (Song, Volling, Lane, & Wellman, 2016). The SIP model proposes reciprocal relations between social cognition (i.e. ToM) and social behavior, based on a continuous retrieval of stored mental representations of social experiences from one’s database (Crick & Dodge, 1994). In addition, not much is known about potential changes in this link during middle childhood, when the further development of ToM abilities (e.g. Hughes & Devine, 2015) coincides with changes in PHY-A and REL-A (e.g. Côté et al., 2007). After the transition to school, a variety of social experiences, such as involvement in aggressive behavior, could influence the further development of ToM (see Hughes & Leekam, 2004, for a review on the links between ToM and social relationships). Yet, to our knowledge there are no longitudinal studies that examined both the potential path from ToM to PHY-A and REL-A, and the reverse path from these aggression forms to ToM at the same time over the course of middle childhood.
Additionally, the roles of gender and age as potential moderators of the links between ToM and aggression require further study. As explained above, girls and boys differ in their expression of REL-A and PHY-A (e.g. Lansford et al., 2012), therefore the links to ToM could also differ between genders. Until now, the very few studies that considered gender differences in the relation between ToM and aggression produced inconsistent results (e.g. Kokkinos, Voulgaridou, Mandrali, & Parousidou, 2016; Renouf et al., 2010). With regard to age, only a few studies have considered age as a moderator in the relation between ToM and aggression. As this link was mainly examined in children of preschool age, even less is known about possible age differences in middle childhood. One study found a moderation by age in the relation between REL-A and ToM in South-American children (Gomez-Garibello & Talwar, 2015).
The Current Study
In our three-wave longitudinal study, we examined the reciprocal relations of ToM with PHY-A and REL-A in middle childhood in a large community sample over a period of 3 years. At the first data wave (T1), participants had a mean age of 8 years, the second wave (T2) was conducted about 9 months later, and the third wave (T3) another 24 months later. Using structural equation modeling, we investigated the reciprocal relations while controlling for earlier levels of ToM and partialling out stable between-person differences in aggressive behavior, respectively.
On the basis of the SIP model (Crick & Dodge, 1994) and in line with earlier empirical evidence (e.g. Capage & Watson, 2001), we expected a negative path from ToM to subsequent PHY-A. Regarding the path from ToM to later REL-A, the SIP model also suggests a negative path, whereas other research, particularly in the bullying tradition, suggests a positive path (e.g. Renouf et al., 2010). We examined these competing predictions in our analyses. Additionally, in line with the SIP approach, we postulated prospective paths from aggressive behavior to ToM. Past social experiences are stored in long-term memory and should be retrieved in every step of SIP. Since ToM is involved in at least some of the SIP steps, it might be influenced by earlier aggression. We therefore hypothesized negative paths from both PHY-A and REL-A to subsequent ToM.
In addition, we explored possible gender and age differences in the reciprocal associations between aggression and ToM. This was deemed important, because the inconsistent findings across previous studies point toward the need for further investigations to clarify the potential moderating influence of gender and age on the paths between ToM and aggression. To examine whether the proposed relations could be confirmed for all subsamples of our study, we conducted separate multi-group analyses with the gender groups and with three age-groups, respectively.
Method
Participants
At the first data wave (T1) 1657 children (52.1% girls) between 6 and 11 years (M = 8.36, SD = 0.95, range 6.23–11.33) participated. At the second wave (T2), about 9 months after T1, 1611 children (51.8% girls) participated again (M = 9.12, SD = 0.93, range 7.12–11.90). At the final wave (T3), about 24 months after T2, 1501 children (51.7% girls) participated again (M = 11.07, SD = 0.92, range 9.12–13.76). All children who took part at T1 were included in our analyses, and missing data due to dropout or incomplete measures were handled by the full information maximum likelihood procedure (FIML; Enders & Bandalos, 2001), resulting in a sample size of N = 1657 children. The children who no longer participated at T3 had significantly higher PHY-A and REL-A, and lower ToM and verbal ability scores at T1 than those children who completed T3.
Children were recruited at 33 community primary schools representing a variety of rural and urban areas. About 33.7% of the mothers and 36.9% of the fathers reported holding a university degree, 21.6% and 13.5% a university entrance qualification, 42.9% and 48.0% a vocational level qualification, and 1.8% and 1.7% no or a low level of school graduation.
Materials and Procedure
The study was part of a larger research project on intrapersonal developmental risk factors in childhood and adolescence from a longitudinal perspective. Data were collected in individual sessions at the children’s schools in a private room by a trained research assistant. Class teachers received individual questionnaires for each participating child. All procedures were approved by the Ethics Committee at the University of Potsdam and by the Ministry of Education, Youth and Sport of the Federal State of Brandenburg, Germany. For each child, informed consent was obtained from the primary caregiver, and the children gave verbal consent.
Theory of mind
At T1 and T2, cartoons were used to assess ToM (Sebastian et al., 2012). Based on a pilot study, 12 cartoons (six for cognitive, six for affective ToM) of the original 20 cartoons were chosen as representing an average difficulty level for the studied age group. The cartoons were presented to the children on netbooks using E-Prime 2.0 Professional (Psychology Software Tools, 2012). In each trial, the first three pictures depicted a little story with two characters (A and B). In the affective stories, character A displayed an emotion, for example, sadness after losing a boat on a river (Figure 1); in the cognitive stories, character A displayed an intention or desire, for instance, wanting to pick apples from a tree. Then, two further pictures were presented simultaneously, each displaying a possible ending. The “correct” ending implied that character B understood A’s mental state (e.g. by comforting or helping); in the “incorrect” ending, character B showed uncaring behavior. The positions of the types of endings were varied across trials, and children were asked to select the picture they thought represented the correct ending by pressing the F-key or the J-key (on a QWERTZ keyboard) for the left or right picture, respectively. Children received feedback only in two initial exercise trials, but not when they worked through the 12 trials. The trials were presented in random order.
Example cartoon story ‘Boat’ for assessing affective ToM.
The number of correct responses were summed up separately, resulting in one score for affective ToM and one for cognitive ToM, both with a range of 0–6. In the analyses, we included ToM as a latent factor with the affective and the cognitive ToM score as indicators. Ordinal alpha, which is appropriate for binary items (Gadermann, Guhn, & Zumbo, 2012), indicated good internal consistency for our ToM measure at both time points (αT1 = .81, αT2 = .82).
Aggressive behavior
At all three time points, aggressive behavior was assessed using six items adapted from the Children’s Social Behavior Scale-Teacher Form (CSBS-T; Crick, 1996). Classroom teachers indicated how often during the last 6 months the child had shown behaviors of PHY-A (three items, e.g. “hit, shove, or push peers”) and REL-A (three items, e.g. “try to exclude certain peers from peer group activities”), using a response scale ranging from 1 (never), 2 (once a month or rarely), 3 (several times a month), 4 (several times a week), and 5 (every day). In the hypothesis-testing analysis, we used manifest mean scores of PHY-A and REL-A, and included random intercepts for both forms, as described below. Internal consistency was excellent (PHY-A: αT1 = .93, αT2 = .94, αT3 = .93; REL-A: αT1 = .91, αT2 = .92, αT3 = .90).
Verbal ability
Because children’s ToM skills can be confounded with their verbal ability (Milligan, Astington, & Dack, 2007), it is essential to control for the latter. At T1, we used a vocabulary test from the Potsdam-Illinois Test for Psycholinguistic Abilities (P-ITPA; Esser, Wyschkon, Ballaschk, & Hänsch, 2010) to assess children’s word knowledge and ability to detect relations between words. Scores were normed for children’s age.
Statistical Analyses
Structural equation models were conducted with Mplus (Version 7.4; Muthén & Muthén, 1998–2015), and descriptive analyses were computed with SPSS (Version 23). In all structural equation models, the robust maximum likelihood estimator (MLR) was used to account for deviations from normality of the data. As recommended by Hamaker, Kuiper, and Grasman (2015) we included a latent random intercept for both forms of aggressive behavior to control for stable between-person differences. This procedure provides a more accurate examination of within-person changes. As ToM was measured only at two time points, we were not able to include a random intercept for this construct (random intercepts require at least three time points). Instead, we included ToM as a latent factor with the affective and the cognitive ToM scores as indicators. Metric measurement invariance over time was established for ToM.
Age and gender were covariates for all constructs, and verbal ability assessed at T1 was an additional covariate for ToM. In addition, separate multi-group analyses were conducted to examine potential gender and age differences. A model in which all paths were restricted to be equal between the groups (fully constrained) was compared with a model in which all paths were freed between the groups (fully unconstrained). Model comparisons were based on χ2 differences used with scale corrections as proposed by Satorra and Bentler (2001) due to the application of the MLR estimator. Age-groups (“younger” vs. “middle” vs. “older”) were defined based on splitting the sample by age at T1 at the 33th and 66th percentile (Md 33% = 7.83 years, Md 66% = 8.86 years).
The percentage of missing values ranged from 1.6% to 33.3% (for the sample size of each measure see Table 1). The rate of missing data was highest on the teacher reports of aggression, due to not-returned questionnaires. The FIML procedure was implemented in all analyses to handle missing data (Enders & Bandalos, 2001).
Descriptive statistics and gender differences.
Note. a mean score; b sum score. *p < .05. **p < .01. ***p < .001.
Overall model fit was evaluated according to Hu and Bentler (1999), with RMSEA ≤ .06, CFI ≥ .95, TLI ≥ .95, and SRMR ≤ .08 indicating a good fit. The χ2 statistic is reported to provide complete model fit information, but it was not evaluated as an absolute index of model fit because it is biased for large samples (Schermelleh-Engel, Moosbrugger, & Müller, 2003). Effect sizes for t-tests are reported as Cohen’s d with the conventional interpretation of d = 0.20 as small, d = 0.50 as medium, and d = 0.80 as large effects (Cohen, 1988). For our structural equation model, we report standardized regression coefficients that α can be interpreted in a similar way as Cohen’s d.
Results
Descriptive Statistics, Gender Differences, and Correlations
Table 1 presents the descriptive statistics and the results of statistical tests for gender differences on the aggressive behavior measures as well as the ToM scores. Teacher-rated PHY-A scores were significantly higher for boys than for girls at each data wave, with medium to large effect sizes (ds ≥ 0.66). For REL-A, the gender difference was small (ds ≤ 0.14), but boys were rated as significantly more relationally aggressive than girls at T2 and T3. With regard to ToM, no significant gender differences emerged.
PHY-A scores decreased significantly from T1 to T2, t(1117) = -2.68, p < .01, d = 0.06, but were stable between T2 and T3, t(843) = -0.99, p = .32, d = 0.03. REL-A scores did not vary significantly between T1 and T2, t(1115) = 1.86, p = .06, d = 0.05, or between T2 and T3, t(840) = -1.66, p = .10, d = 0.06. ToM scores increased significantly from T1 to T2, t(1543) = 16.22, p < .001, d = 0.48.
Bivariate correlations between all variables are displayed in Table 2. Age was positively correlated with ToM at T1 and T2, but uncorrelated with aggressive behavior. All correlations between ToM and aggressive behavior were negative. Stability of ToM between T1 and T2 was relatively low, whereas PHY-A and REL-A showed moderate to high stabilities between the time points. PHY-A and REL-A were strongly positively associated with each other at all time points.
Intercorrelations of study variables.
Note. Stabilities are highlighted in italics; N = 1657. *p < .05. **p < .01. ***p < .001.
Hypothesis Testing
To test our first set of hypotheses, we computed a structural equation model with latent factors for ToM at T1 and T2, and manifest mean scores for PHY-A and REL-A at T1, T2, and T3, including random intercepts for both REL-A and PHY-A. In doing so, we investigated the longitudinal reciprocal relations between ToM and PHY-A and REL-A, respectively, controlling for earlier levels of ToM and for stable between-person differences in aggressive behavior. Furthermore, all constructs were controlled for age and gender, and ToM was additionally controlled for T1 verbal ability. Metric measurement invariance over time was established in a step-up approach for PHY-A, REL-A, and ToM from T1 to T2. The overall fit of our final model was good, RMSEA = .03, 90% CI [.02, .04]; CFI = .99; TLI = .98; SRMR = .03; χ2(33) = 71.38, p < .001. The resulting model is presented in Figure 2.
Structural equation model for physical aggression, relational aggression and theory of mind from T1 to T3.
Reciprocal effects of ToM, REL-A, and PHY-A
In line with our hypotheses, we found significant negative paths from ToM at T1 to PHY-A at T2 (β = -.08, p < .05), and from ToM at T2 to PHY-A at T3 (β = -.12, p < .01). Similarly, the hypothesized paths from ToM at T1 to REL-A at T2 (β = -.09, p < .05) as well as from ToM at T2 to REL-A at T3 (β = -.11, p < .05) were significant and negative. The reverse negative path was confirmed only from REL-A at T1 to ToM at T2 (β = -.12, p < .05), but not from PHY-A at T1 to ToM at T2 (β = .09, p = .13). In addition to the expected paths, a negative path from PHY-A at T2 to REL-A at T3 was found (β = -.16, p < .01).
Gender differences
To examine gender differences, we compared the fully constrained multi-group model by gender with the fully unconstrained model (fit constrained model: RMSEA = .05, 90% CI [.04, .05]; CFI = .95; TLI = .93; SRMR = .06; χ2(95) = 256.03, p < .001; fit unconstrained model: RMSEA = .03, 90% CI [.03, .04]; CFI = .98; TLI = .96; SRMR = .04; χ2(65) = 127.14, p < .001). The difference between these two models was significant, Δχ2(30) = 119.43, p < .001. In the next step, we computed a revised and final model, in which we constrained all paths to be equal, except the paths that were found to differ significantly between boys and girls in the fully unconstrained model. This revised model showed a good model fit, RMSEA = .03, 90% CI [.02, .04]; CFI = .98; TLI = .97; SRMR = .04; χ2(86) = 153.49, p < .001, which was significantly better than that of the fully constrained model, Δχ2(9) = 79.87, p < .001, and not significantly worse than that of the fully unconstrained model, Δχ2(21) = 27.83, p = .145. Based on the revised model, there were no significant gender differences in the hypothesized paths (ps ≥ .12). The only significant gender differences were found in the concurrent correlations between PHY-A and REL-A, which were stronger for boys (rs ranging from .54 to .69, ps < .001) than for girls (rs ranging from .42 to .45, ps < .001). In addition, on a within-person level, REL-A at T1 predicted REL-A at T2 only in girls (β = .26, p < .001), but not in boys (β = .09, p = .17).
Age differences
To examine age differences, we followed the same procedure with the three age-groups; fit constrained model: RMSEA = .03, 90% CI [.02, .03]; CFI = .98; TLI = .98; SRMR = .06; χ2(164) = 227.32, p < .001; fit unconstrained model: RMSEA = .03, 90% CI [.02, .04]; CFI = .99; TLI = .97; SRMR = .05; χ2(104) = 154.20, p < .001). The fully constrained model did not fit worse than the fully unconstrained model, Δχ2(60) = 75.41, p = .09, indicating that there was no significant moderation by age.
Discussion
The primary aim of this research was to study the reciprocal relations between ToM (assessed by children’s responses to a cartoon task) and teacher-reported PHY-A and REL-A over time in a large community sample of girls and boys in middle childhood. Our longitudinal study with three waves of measurement covered a time period of about 3 years. Past studies on the association of ToM and aggression were mostly cross-sectional, were conducted at preschool age, and did not separate distinct forms of aggression. Therefore, we investigated the reciprocal relations between ToM and PHY-A and REL-A during middle childhood in a prospective design using structural equation modeling, controlling for age, gender, and verbal ability as well as earlier levels of ToM and partialling out stable between-person differences in aggressive behavior.
Consistent with our predictions, we found that ToM was a significant negative predictor of both PHY-A and REL-A. This was true for the paths from T1 to T2 as well as from T2 to T3. These results indicate that children with lower ToM scores were rated by their teachers to be more physically and relationally aggressive at a later occasion, while controlling for stable between-person differences in aggressive behavior. Even though we did not examine mediating processes in our study, these results are in line with the predictions derived from the SIP model (Crick & Dodge, 1994; Harvey et al., 2001) that deficits in ToM may lead to biased or deficient SIP, which in turn may lead to more aggressive behavior. For example, previous research revealed that the encoding and interpretation of situational cues were less accurate in children with early deficits in ToM (e.g. Choe, Lane, Grabell, & Olson, 2013). Our findings do not support the proposition by Sutton et al. (1999a) that children may use their proficient ToM to manipulate others, joining other studies that also failed to find differences in ToM between bullies and noninvolved children (e.g. Gini, 2006). However, the consideration of potential moderating variables, such as prosocial behavior (e.g. Renouf et al., 2010), may be important to identify those children that exploit their ToM skills in an aggressive fashion.
Although the effect sizes were small, the present findings are consistent with previous studies that examined relations between ToM and aggression (e.g. Kokkinos et al., 2016). Considering the complex nature of the emergence of aggressive behavior as postulated by the SIP model (Crick & Dodge, 1994), many intra- and interpersonal factors can have an impact (e.g. see Krahé, 2013, for a review). Our study demonstrated ToM to be an important, but certainly not the only intrapersonal predictor for the development of aggressive behavior during middle childhood.
Turning to the postulated reverse paths from aggression to ToM, higher scores in REL-A, but not PHY-A, at T1 predicted lower ToM skills at T2. The negative path from REL-A to ToM is in line with the SIP model, postulating reciprocal paths between social cognition and aggressive behavior (Crick & Dodge, 1994). The reason for the missing path from PHY-A to ToM may be that the different functions of aggression in terms of proactive versus reactive aggression were not considered in our study. In past research, only reactive PHY-A was linked to low socio-cognitive abilities (Crick & Dodge, 1996). Proactive PHY-A, however, is an instrumental and planned behavior (Vitaro & Brendgen, 2005) and may be positively linked to ToM. These opposing relations could have cancelled each other out to eliminate the path from PHY-A to ToM. Nevertheless, it remains unclear why this would have been the case only for PHY-A and not for REL-A.
In addition to the hypothesized paths, the analyses revealed a significant negative path from PHY-A at T2 to REL-A at T3. This result indicates that physically aggressive children tend to reduce their relationally aggressive behavior at a later occasion. This is in contrast to previous longitudinal studies that have found a positive relation or no relation between these two forms of aggression over time (e.g. Card, Stucky, Sawalani, & Little, 2008). However, these studies usually examined aggression on a population level (e.g. by applying traditional cross-lagged panel models), whereas our study examined aggression on the level of the individual (by including random intercepts; Hamaker et al., 2015). The divergence in findings may be an expression of Simpson’s paradox (Kievit, Frankenhuis, Waldorp, & Borsboom, 2013), that is, the direction of a link depends on the level of analysis (i.e. population vs. individual). To our knowledge, there are only a few studies to date that examined PHY-A and REL-A from a person-centered perspective. These studies examined trajectory groups of PHY-A and REL-A as well as the co-occurrence of these groups (e.g. Côté et al., 2007; Ettekal, & Ladd, 2015), and found substantial proportions of children who – on an individual level – showed different trajectories of PHY-A and REL-A (e.g. increasing in PHY-A, but decreasing in REL-A). Therefore, future studies should further investigate the interplay of PHY-A and REL-A longitudinally on an intrapersonal as well as on an interpersonal level.
In further analyses, we explored potential age and gender differences in the reciprocal relations between ToM, PHY-A, and REL-A in middle childhood. We did not find any differences in the reciprocal relations of ToM and the forms of aggression between girls and boys, and between age-groups. However, we found that the correlation between REL-A and PHY-A was higher for boys than for girls, which is consistent with previous research (Card et al., 2008). In addition, we discovered significant gender differences in the mean level of aggressive behavior. At each wave, boys were rated by their teachers to show significantly more PHY-A than girls. This difference was of medium to large size and is in accordance with earlier studies (e.g. Lansford et al., 2012). Similarly, boys were rated to show more REL-A than girls at T2 and T3. This difference was small in size and is in contrast to other studies that found girls as compared to boys to exhibit more REL-A (e.g. Card et al., 2008; Ostrov et al., 2013); albeit, some studies also found no gender differences in REL-A (e.g. Lansford et al., 2012), or higher scores for boys (e.g. Henington et al., 1998). The present results may have been due to characteristics of our male or female subsample or to population differences, which we could not examine in our study. A more fine-grained analysis of gender differences in REL-A remains an important topic for future research. Regarding age, we did not find differences in the reciprocal relations between ToM and the two forms of aggression across the age-groups of our sample. There were no correlations of age and PHY-A or REL-A, but there were positive correlations of age and ToM. In line with previous research (e.g. Devine & Hughes, 2013), this indicates that ToM is still improving with increasing age in middle childhood. Altogether, we conclude that apart from gender differences in the absolute level of aggressive behavior, in the concurrent correlations between REL-A and PHY-A, and bivariate correlations between age and ToM skills, the prospective relations between aggression and ToM are not moderated by gender or age.
In conclusion, the present findings provided the first longitudinal evidence on reciprocal relations between ToM and PHY-A, as well as REL-A, as distinct forms of aggression, in middle childhood. Our results of negative reciprocal relations between ToM and REL-A or PHY-A, respectively, are consistent with the social skills deficit view of Crick and Dodge (1994), and they do not support the proposition by Sutton et al. (1999a) that aggressive children have advanced ToM abilities. This suggests that improving ToM abilities may reduce rather than promote PHY-A and REL-A in middle childhood. The impact of REL-A on children’s well-being is usually evaluated by teachers and parents as being less severe than that of PHY-A (e.g. Hurd & Gettinger, 2011). In contrast, our study showed that trivialization of REL-A may have detrimental effects on socio-cognitive development. Therefore, interventions that aim at reducing REL-A and/or fostering ToM would be indicated for interrupting the vicious circle of deficient socio-cognitive abilities promoting aggressive behavior and vice versa. Following our results, gender differences do not seem to play a prominent role in the examined processes. Therefore, interventions should focus equally on girls and boys.
Strengths and Limitations
We believe our study has several strengths. It is based on a large sample of children attending community elementary schools, and it includes three data waves covering middle childhood over a total of about 3 years. We controlled for stable between-person differences in aggression and for past levels of ToM (construct stability), fostering a causal interpretation of the paths (Marmor & Montemayor, 1977). Further, we distinguished between two forms of aggression, PHY-A and REL-A, which have been found in some previous studies to differ between girls and boys (e.g. Lansford et al., 2012), and we explored potential gender and age differences. Finally, as one of the first studies, we investigated not only the path from ToM to the two forms of aggression, but also the reverse paths. Crick and Dodge (1994) already emphasized the importance of studying the effect of social experiences, such as aggressive behavior, on social cognitions. This can promote understanding of how children develop outcome expectations for social situations, which they then apply in reaching response decisions.
At the same time, there are some limitations to our study. First, we used the prominent SIP model by Crick and Dodge (1994) as a framework for our study, but we did not include measures specifically designed for SIP, for example video vignettes (Lansford et al., 2006). Consequently, the specific underlying mechanisms by which ToM affects SIP cannot be inferred. Further, we did not examine other mediating mechanisms, such as empathy or moral disengagement (e.g. Kokkinos et al., 2016). Future studies should therefore integrate measures for specific cognitive processes that are assumed to influence aggressive behavior.
Second, the time intervals differed between T1 and T2 (about 9 months), and T2 and T3 (about 24 months). Consequently, prospective effects were compared between unequal time periods; however, it is even more remarkable that ToM predicted PHY-A and REL-A at both time intervals. Another limitation is that our ToM measure appeared somewhat too easy for our sample, especially at T2. These ceiling effects reduced the variance in the ToM measures, making it harder to detect potential effects and to identify children that are very skilled in ToM. Future studies should continue to develop time- and resource-efficient ToM measures that create sufficient variability in middle childhood. Finally, the use of teacher reports to assess aggressive behavior can be considered as a limitation of our study. Teachers may not be fully aware of the extent to which children engage in aggressive behavior, in particular when it comes to REL-A, which includes more covert behavior (e.g. spreading rumors).
Despite these limitations, our findings highlight the consideration of reciprocal relations between ToM and aggressive behavior over the course of middle childhood. Future research should further investigate how aggressive behavior and ToM develop over the whole course of childhood and how this feeds back into the development of the respective other part over time. In the long run, this information can help to reduce the detrimental effects of aggressive behavior on children’s lives and health.
Footnotes
Funding
The authors declared receipt of the following financial support for the research, authorship, and/or publication of this article: The research reported in this paper was funded by the German Research Foundation as part of the Graduate College “Intrapersonal developmental risk factors in childhood and adolescence: A longitudinal perspective” (GRK 1668).