Measurement Invariance and Psychometric Properties of the Reactive and Proactive Aggression Questionnaire (RPQ) Across Genders
Abstract
Abstract. Aggression is an important risk factor predisposing adolescents to disruptive and criminal behaviors. The present study aimed to assess the psychometric properties and measurement invariance of the Persian version of the Reactive–Proactive Aggression Questionnaire (RPQ) across genders. A sample of 450 students (Mage = 14 years, SD = 0.81) were recruited randomly and completed the Persian version of the RPQ, Child Behavior Checklist, and Strengths and Difficulties Questionnaire. The 2-factor model demonstrated an acceptable fit to the data, supporting the distinction between reactive and proactive aggression among Iranian adolescents. Although the result indicated the equivalence of factor structure of the RPQ across genders, the invariance of the metric model was not confirmed in this study. Suggestions for future research and a more accurate assessment of these two kinds of aggression are further discussed.
Previous studies suggest that aggression is a severe and pervasive problem in adolescence, contributing to many crimes (Bobadilla et al., 2012; Cima & Raine, 2009; Snyder & Sickmund, 1999; Vitaro et al., 2006), aimed to harm people and/or objects (Dodge, 1991). Research on aggressive behavior has introduced various categories of aggression (Cenkseven-Önder et al., 2016) based on its different forms and functions (Hubbard et al., 2010; Kempes et al., 2005).
Dodge and Coie (1987) first proposed a frequently used dichotomy classification, resulting in two subtypes of aggression, , namely, reactive and proactive (Cima & Raine, 2009; Conaty, 2006). Reactive aggression is defined by impulsive behaviors that occur in response to a provocation, frustration, or perceived threat which is best described by frustration–aggression theory (Polman et al., 2007). It is the retaliatory form of aggression motivated by anger (Hubbard et al., 2010; Polman et al., 2007). In contrast, proactive aggression is the cold-blooded and goal-oriented form of aggression (Dodge, 1991; Hubbard et al., 2010; Raine et al., 2006). This instrumental kind of aggression is best conceptualized by the social learning theory that emphasizes the role of operant conditioning and vicarious learning (Bandura, 1973).
Although there is a controversial debate on the usefulness of this distinction (Bushman & Anderson, 2001; Euler et al., 2017), several studies demonstrated the unique associations between these two subtypes of aggression and external criteria (Dodge, 1991; Dodge & Coie, 1987; Hubbard et al., 2010; Kempes et al., 2005; Vitaro et al., 2006), suggesting different causal pathway of reactive and proactive aggression. A literature review showed different behavioral and neurocognitive profiles among these groups (Smeets et al., 2017). Proactive aggression was found to be associated with narcissism (Seah & Ang, 2008), poor peer relationships in early childhood, blunted affect (Raine et al., 2006), higher levels of working memory (Hecht & Latzman, 2018), and increased expectation of positive outcomes (Poulin & Boivin, 2000). While past research indicated that correlates of reactive aggression included information processing deficits (Dodge & Coie, 1987), physical violence (Brendgen et al., 2001), lack of self-control, increased impulsivity, problem-solving deficiency (Atkins et al., 1993), social anxiety, lack of close friends (Raine et al., 2006), narcissistic traits (Bobadilla et al., 2012), increased emotion dysregulation (Hecht & Latzman, 2018), schizotypal traits, poor interpersonal relations (Seah & Ang, 2008), social maladjustment, internalizing behavior, and peer rejection (Card & Little, 2006).
However, there are some inconsistent findings in the correlates of these subtypes of aggression. For instance, while proactive aggression has been associated with delinquency in some studies (Atkins et al., 1993; Brendgen et al., 2001; Raine et al., 2006; Vitaro et al., 1998) there are results indicating a strong correlation between delinquency and reactive aggression (Card & Little, 2006).
The conflicting outcomes could be justified by different sex or ages of the participants (Bobadilla et al., 2012; Connor et al., 2003; Kempes et al., 2005). Although there are many studies on the development of aggression across age and gender, very few considered these subtypes of aggression. Regarding gender comparison, one study showed that boys got higher scores in reactive aggression, but no significant differences have been reported for proactive aggression across genders (Kempes et al., 2006; Lickley & Sebastian, 2018).
Although these subtypes of aggression co-occur in children (Merk et al., 2005) and aggression is better conceptualized on a continuous dimension, the distinction between reactive and proactive aggression can still enhance our understanding of their different causes and outcomes (Dodge, 1991). In a similar way, this knowledge would contribute to the development of more accurate preventive and therapeutic interventions (Hubbard et al., 2010; Merk et al., 2005; Pechorro et al., 2015). As such, reliable and valid measures that could distinguish between these two subtypes of aggression are vital. Dodge and Coie (1987) discriminated these two subtypes of aggression among children aged 3 years and above. Their 6-item rating scale can be completed by children, their teachers/parents (Kaat et al., 2015), or correctional facility staff (Hubbard et al., 2010). However, self-report measures are privileged over others by accurately recognizing the inherent motivation of aggression, the central feature of this distinction (Raine et al., 2006). The Reactive–Proactive Aggression Questionnaire (RPQ) is a brief self-report measure developed by Raine et al. (2006). The psychometric properties of this measure have been investigated by several studies across different cultures, supporting its reliability and validity (Cenkseven-Önder et al., 2016; Fossati et al., 2009; Fung et al., 2009; Pechorro et al., 2015; Seah & Ang, 2008). The Cronbach's α values in the initial validation of the RPQ among schoolboys aged 7 and 16 years were .84 for the reactive scale, .86 for the proactive scale, and .90 for the total score (Raine et al., 2006). The Portuguese version of the RPQ showed good psychometric properties, including factor structure, internal consistency, convergent validity, discriminant validity, and criterion-related validity (Pechorro et al., 2018). Cima et al. (2013) have identified Cronbach's α of the Dutch version as .83 for the reactive scale and .87 for the proactive scale, supporting the RPQ as a reliable and valid instrument to identify these subtypes of aggression. Its criterion validity was also established due to the significantly lower scores of nonoffenders compared to offenders (Cima et al., 2013). The study conducted on Turkish children and adolescents indicated that the item–total correlation ranged from .40 to .74, and an alpha coefficient of both subscales and the total score were above .81, suggesting a relatively high internal consistency (Cenkseven-Önder et al., 2016).
There are studies in favor of the original 2-factor model across various cultures. Fung et al. (2009) have examined the generalizability of the RPQ in an East Asian population. Despite the significant correlation between reactive and proactive aggression (r > .54), the original 2-factor model was superior to the 1-factor model, and the item loadings were greater than .45. A better fit of the 2-factor structure was replicated in a study among Portuguese youth (Pechorro et al., 2018); however, the measurement invariance across genders was supported after excluding Item 21. A 2-factor model of the RPQ best fit the data in a Spanish version while it was variant across genders and ages, with the best fit for males younger than 25 years of age (Toro et al., 2020). The original 2-factor model of the RPQ also obtained the best fit in Dutch (Cima et al., 2013) and Turkish samples (Cenkseven-Önder et al., 2016).
Despite these findings supporting the 2-factor structure of the RPQ, some studies failed to confirm the original model. Using a person-based approach, Brugman et al. (2017) revealed a 3-factor structure based on the severity of aggression in forensic and nonforensic samples. Moreover, the 4-factor structure has been explored, combining two subtypes of aggression and various contexts for reactive aggression (playing games and a defensive form) based on a variable-based approach. However, the 4-factor solution was considered less solid since items were more uniquely related to each subscale of aggression in the 2-factor model of the RPQ. In the same way, Smeets et al. (2017) have extracted three factors in the clinical sample, including reactive aggression due to external provocation, reactive aggression due to internal frustration, and proactive aggression. Pang et al. (2013) also identified three distinct clusters of aggression (including high reactive and low proactive aggression, low on both subscales, and high on both subscales) in a Singaporean sample.
Therefore, the contradictory results of prior research and the lack of study on the convergent and criterion validity of the RPQ in the Iranian population suggest the need for further research on the structural validity of the RPQ to distinguish reactive from proactive aggression. Hence, this research aimed to investigate the replication of the original 2-factor model of the RPQ in a different culture (Raine et al., 2006) and its psychometric properties across genders. Therefore, the first hypothesis was that the original 2-factor model of the RPQ would achieve adequate fit in a sample of Iranian adolescents while the model form, factor loadings, and item intercepts would be invariant across genders. Second, the RPQ was predicted to show good internal consistency as measured by the omega and alpha coefficient, mean inter-item correlation (MIIC), and corrected item-total correlation in both groups. Third, we hypothesized that both RPQ subscales would relate positively to criterion-related variables, including rule-breaking behaviors, aggressive behaviors, and conduct problems, in a similar way in both groups. Finally, to support that proactive aggression is associated with delinquency (Bobadilla et al., 2012), we examined whether detained individuals could be distinguished based on their scores on the proactive aggression subscale. It was predicted that incarcerated individuals would score higher than nonincarcerated adolescents on the proactive subscale.
Method
Participants
A total of 450 students, including 217 boys (Mage = 14.29 years; SD = 0.80) and 233 girls (Mage = 13.89 years; SD = 0.78), participated in this study (Mage = 14 years; SD = 0.81). There were 133 students from Grade 7, 146 students from Grade 8, and 171 students from Grade 9. The participants were recruited via a stratified sampling method from six secondary schools in Yazd, a city in the middle of Iran, considering various socioeconomic statuses.
A subsample of 145 students, including 69 girls (Mage = 13.6 years; SD = 0.93) and 76 boys (Mage = 13.8 years; SD = 1.05), was chosen randomly to measure the convergent validity. This sample comprised the minimum participants needed (N = 67) to achieve the desired statistical power (i.e., .80) recommended by Cohen (1988) with the medium effect size (i.e., r = .30).
The RPQ was administered in a forensic context to establish criterion validity. Since there was no girl in imprisonment, all the incarcerated boys (N = 31) aged 15–19 years old (Mage = 17.4 years; SD = 1.01) in the juvenile detention center of Yazd were recruited. This sample size was enough to achieve an acceptable level of statistical power (Cohen, 1988) with a medium effect size (i.e., r = 0.50).
Measures
Reactive–Proactive Aggression Questionnaire (RPQ)
Participants completed the RPQ, a 23-item questionnaire that is suitable for a wide age range and evaluates two functions of physically and verbally aggressive behaviors (Raine et al., 2006). There are 11 items in the proactive and 12 items in the reactive subscale. Each item is rated on a 3-point Likert-type scale with 0 = never, 1 = sometimes, and 2 = often.
Child Behavior Checklist (CBCL)
The emotional and behavioral problems were assessed by the youth self-report version of the Child Behavior Checklist for adolescents aged 11–18 years (Achenbach, 1991). This measure contains 113 items rated on a 3-point Likert-type scale (0 = absent, 1 = occurs sometimes, 2 = occurs often). It includes subscales that assess anxiety, depression, somatic complaints, social problems, thought problems, attention problems, rule-breaking behaviors, and aggressive behaviors. Symptoms of rule-breaking behaviors and aggressive behaviors make externalizing problems score. There is a large body of literature in support of the psychometric properties of this scale in different countries (Achenbach & Rescorla, 2001; Minaee, 2006). This study used two subscales of aggressive and rule-breaking behaviors to evaluate convergent validity.
Strengths and Difficulties Questionnaire (SDQ)
A self-report version of the SDQ (Goodman et al., 1998) was used to measure emotional and behavioral disorders. It is composed of 25 items assessing positive and negative attributes of children and adolescents across five subscales: conduct problems, inattention–hyperactivity, emotional symptoms, peer problems, and prosocial behavior. Items are rated on a 3-point Likert-type scale with 0 = not true, 1 = somewhat true, and 2 = certainly true. The Persian version of SDQ demonstrated high validity and reliability in several studies (Tehranidoust et al., 2007). Only the conduct problem subscale (α = .72) was used in this study.
Procedure
Permission was asked from the developer of the RPQ before the initial questionnaire was translated into Persian. The face and content validity were assessed by two experts, causing a minor change in the wording. After getting approval from the ethical committee of the university, participants were recruited from all educational regions in Yazd, Iran. After reading the instruction, randomly selected participants completed the questionnaires in a group setting. All participants were informed that they could withdraw at any time without any negative consequences for them.
Authorization to attend the juvenile detention center was obtained from the General Administration of Prisons in Yazd. The same procedures were done for incarcerated boys, except that the questionnaire was read for five illiterate participants. Data were analyzed using IBM SPSS Statistics v27 and LISREL v10.20.
Statistical Analyses
Confirmatory factor analysis was conducted to examine the factor structure of the RPQ. The diagonally weighted least squares estimation was used considering the 3-point Likert-type scale of the RPQ which produces ordinal data, and non-normality of the items at the multivariate level. Model fit of different factor structures, including the 1-factor, the original 2-factor, 3-factor (Smeets et al., 2017), and second-order factorial models, were assessed. The second-order factorial model was based on the 2-factor model in which a second-order latent variable (general aggression) was supposed to account for the association between first-order factors, i.e., reactive and proactive aggression. All items are loaded by one latent factor named general aggression in the 1-factor model, while the factor variance was constrained at 1.00.
Various goodness-of-fit indices were used for the investigation of model fit, including Satorra–Bentler χ2/degrees of freedom, comparative fit index (CFI), incremental fit index (IFI), the goodness-of-fit index (GFI), nonnormed fit index (NNFI), and the root-mean-square error of approximation (RMSEA). The acceptable cut-off value for CFI and NNFI is .90, indicating an adequate fit (Hoyle, 1995). Hu and Bentler (1999) recommended that CFI > .95 and RMSEA < .06 indicate a good-fitting model. χ2 divided by degrees of freedom (χ2/df) was also calculated; therefore, values less than two and five were considered good and acceptable, respectively (Schumacker & Lomax, 2004). Values of IFI that exceed .90 were considered good (Pechorro et al., 2015). The difference χ2 (Δχ2) test was computed to compare the models in a hierarchical relationship (Raine et al., 2006). The expected cross-validation index was used for nonnested models' comparison so that the lowest value indicates a better fit (Brown, 2015; Kline, 2015). Since factor loadings are considered meaningful when they are above .30, items with standardized loading lower than that value would be removed (Pechorro et al., 2018). In this study, the model fit was not improved based on the modification indices to avoid a data-driven strategy. Then, measurement invariance of the primary model was assessed using multigroup CFA (Milfont & Fischer, 2010). In this approach the psychometric equivalence of a construct is assessed across various groups. Therefore, in the first stage, the configural model is defined in which all parameters are allowed to be freely estimated across groups. This model is the baseline for comparison and estimated the equivalence of the overall factor structure in groups. At the next step, metric invariance would be assessed to investigate whether the same observations indicate the same latent constructs across different groups. Therefore, while item intercepts are still allowed to be freely estimated, the factor loadings are constrained to be equivalent across groups. In the final step, scalar invariance examine the similarity of item intercepts across groups. The scalar invariance is considered a precondition for comparing the factor means between groups (Putnick & Bornstein, 2016; Zager Kocjan et al., 2021). In each step, the baseline model is compared with the constrained model. The difference between these nested models is assessed based on the significance of the χ2 difference test as well as ΔCFI and ΔRMSEA (Xu & Tracey, 2017). Therefore, a significant deterioration in the fitness of the more constrained model alongside ΔCFI ≤ −.005 and ΔRMSEA ≥ .010 indicates measurement noninvariance (Cheung & Rensvold, 2002), taking into account the non-equality of groups' sample sizes (Chen, 2007).
In the last step, the psychometric properties of the 2-factor model were investigated across genders. Internal consistency was assessed using omega and alpha coefficients, MIIC, and corrected item–total correlations (CITC). Omega and α values ≥ .70, the minimum mean inter-item correlations within the range of .15–.20, and CITC's above .20 were considered the indicators of adequate internal consistency (Clark & Watson, 2016; Hair et al., 2009; Pechorro et al., 2018). Convergent validity was examined using correlation analysis between the scores of the RPQ subscales and conduct problem, aggressive and rule-breaking behaviors. The residualized scores were used alongside the raw scores to evaluate the differential correlates of reactive and proactive aggression (Pechorro et al., 2015; Raine et al., 2006). Criterion validity was examined by comparing the reactive and proactive scores between individuals with and without delinquency.
Results
Factor Structure and Invariance Assessment
As can be seen in Table 1, the 2-factor model demonstrated acceptable fit to the data (χ2/df = 1.559, CFI = .920, GFI = .964, IFI = .921, NNFI = .911, RMSEA = .045). However, the second-order factorial model (χ2/df = 1.518, CFI = .928, GFI = .965, IFI = .929, NNFI = .920, RMSEA = .044) fit the data better than the 1-factor model (χ2/df = 1.705, CFI = .898, GFI = .959, IFI = .900, NNFI = .888, RMSEA = .049), 3-factor model (χ2/df = 1.547, CFI = .922, GFI = .965, IFI = .923, NNFI = .913, RMSEA = .044), and the 2-factor model. All item loadings exceeded the recommended value of .30 (Pechorro et al., 2018), except for Item 15.
Since the 2-factor intercorrelated structures provided an acceptable fit to the data, we evaluated the measurement invariance of the original factor structure of the RPQ, which was statistically and theoretically supported in previous studies (Brugman et al., 2017). Therefore, in the next stage, two groups were defined based on gender to examine the measurement invariance of the RPQ and the impact of gender on the model fit. The result of multigroup CFA showed that the factor structure of the RPQ was equivalent across both groups. However, the result revealed different factor loadings across genders (Table 2) since the deterioration in a model fit of the metric model compared with the configural model was not within the acceptable range regarding ΔCFI value (Chen, 2007; Cheung & Rensvold, 2002; Xu & Tracey, 2017). Although the change in RMSEA was lower than the recommended threshold (ΔRMSEA = .004), the χ2 difference test (ΔSBχ2 = 253.224, df = 23, p < .001) showed a significantly poorer fit for the model with constrained factor loadings across groups. Further assessment was not justified because the least strict model was not confirmed to be equal (Putnick & Bornstein, 2016).
Factor loadings of items were examined across genders to explore the noninvariant loadings (Table 3). The result indicated different items with factor loading less than .30 across genders (Items 4, 15, 18, and 21 among girls and Items 13 and 15 among boys). Confirmatory factor analysis revealed an acceptable fit of the 2-factor intercorrelated structure after removing items with standardized item loadings less than the threshold in both groups (Table 4).
Internal Consistency
The omega coefficients for the RPQ total score, reactive and proactive subscale scores among girls were .84, .78, and .68, respectively. In the other group, the omega coefficients were as follows: total score = .85, reactive subscale = .74, and proactive subscale = .77. The result of internal consistency for the RPQ subscales, estimated by Cronbach's alpha, could be considered acceptable except for the proactive aggression score among girls (Table 5).
CITC for reactive aggression ranged from .33 to .45 for boys (M = 7.86, SD = 3.50) and .33 to .55 for girls (M = 8.76, SD = 3.94). For proactive aggression, CITC ranged from .30 to .44 for girls (M = 2.03, SD = 2.07) and .32 to .56 for boys (M = 3.47, SD = 3.24). Reactive and proactive aggression scores were significantly correlated among girls (r = .61, p < .001) and boys (r = .63, p < .001).
Convergent Validity
In a subsample of 145 students (76 males and 69 females), convergent validity was assessed separately for each gender, evaluating the relationship of one subscale of SDQ (conduct problem) and two subscales of CBCL (aggressive and rule-breaking behaviors) with the two RPQ subscales. Once more, the correlation analysis was re–run, using the residualized scores to consider the substantial correlation between these subtypes of aggression.
In males, residualized score of reactive aggression was significantly correlated with rule-breaking (r = .261, p = .02) and aggressive behaviors (r = .363, p < .001). Similarly, the residualized score of reactive aggression was significantly correlated with aggressive behaviors among girls (r = .292, p = .01). While purely proactive aggression did not show any significant correlation among boys, it was significantly associated with rule-breaking (r = .283, p = .01) and aggressive behaviors (r = .244, p = .04) in the other group. Besides, the result revealed that the raw score of both subscales and the total score of the RPQ were significantly correlated with the scores of conduct problem, rule-breaking, and aggressive behaviors in a positive direction (Table 6). A marginally significant correlation between the residualized score of reactive aggression and conduct problem (r = 221, p = .06) was seen among girls. In both groups, the strongest correlation was between the raw score of reactive aggression and aggressive behaviors (male: r = .645, female: r = .499, p < .001).
Criterion Validity
An independent samples t test was conducted to compare the scores of 50 students aged 15 years with a sample of incarcerated boys (N = 31). As shown in Table 7, there was a small but nonsignificant difference in reactive aggression scores between incarcerated individuals (M = 8.93, SD = 3.66) and nonincarcerated sample (M = 7.86, SD = 3.50); t(79) = 1.32, p = .19, g = 0.30. However, the result demonstrated a significant effect for grouping in proactive aggression scores, t(79) = 2.32, p = .02, showing incarcerated boys obtained higher scores (M = 5.54, SD = 4.49) compared with the other group (M = 3.58, SD = 3.12). Hedges' g was calculated to assess the effect size due to the different sample sizes of the compared groups (Lakens, 2013). The result represented a medium effect size for the difference in means of proactive aggression between two groups (g = 0.52).
Discussion
This study investigated the factor structure and psychometric properties of reactive and proactive aggression questionnaires across genders. Overall, the result supported the distinction between reactive and proactive aggression among Iranian adolescents by representing a reasonable fit of the 2-factor intercorrelated structure to the data.
While some research indicated different factor structures for the RPQ (Brugman et al., 2017; Colins, 2016; Pang et al., 2013; Smeets et al., 2017), the result of the present study was consistent with those confirming this model's generalizability among various cultures (Cima et al., 2013; Fung et al., 2009; Goodman et al., 1998; Raine et al., 2006). This discrepancy might be justified by the variety of attitudes in various cultures, leading to different expressions of aggression. Besides, since these subtypes of aggression could be more distinguishable in nonclinical samples, the contradictory findings could result from the different samples among various studies (Smeets et al., 2017).
Although the CFA revealed that the fit indices of the 2-factor model reached the acceptable cut-off value, a good-fitting model was not indicated. Similar results in a study by Toro et al. (2020) showed better model fit indicators after executing residual covariances. In this study, similarly, the modification indices suggested covariance of the measurement errors among the items of the same factor; however, no post hoc alteration to the model was done to prevent data-driven changes.
In addition, a second-order factorial model demonstrated a better fit to the current data compared with the 2-intercorrelated factor structure. This result supported the general aggression score, accounting for the association between reactive and proactive aggression. The present result might be rooted in the wording and ambiguity of phrases so that some items can be interpreted as both reactive and proactive. This idea is also confirmed by various studies reporting cross-loading of several items of the RPQ (Brugman et al., 2017; Fossati et al., 2009). The result of an explorative factor analysis conducted among adults revealed that only a subset of items (Items 1, 5, 7, 11, and 14 for reactive and Items 2, 6, 9, 10, 12, and 20 for proactive aggression) could differentiate adequately between these two subscales (Lobbestael et al., 2013).
Moreover, this measure examines aggressive behaviors coupled with aggressive feelings. For example, in the reactive subscale, Items 5, 11, 13, 14, and somehow 22 mainly examine the feeling of anger. It is also suggested that these two forms of aggression would be better distinguished with behavioral observations and the questionnaires categorizing aggression based on the form, function (Polman et al., 2007), and the target of aggression (whether person or object). Future studies should consider the association between such measures and the RPQ factor structure. Furthermore, this result lends support to the dimensional compared to the categorical approach due to the fuzzy distinction between these two types of aggression (Dodge, 1991) and the considerable correlation among them. Besides, proactive aggression could be displayed with a delay from prior provocation (Merk et al., 2005), making it more complicated to distinguish between these subtypes of aggression based on a questionnaire.
In this article, the measurement invariance of the RPQ was examined across genders. The result of CFA revealed that loadings of items had a similar pattern across genders, supporting the distinction between the reactive and proactive aggression in both groups. However, the invariance of the metric model was not confirmed in line with the studies indicating variants of the RPQ across groups in different samples (Rodríguez et al., 2009; Toro et al., 2020). Investigating noninvariant factor loadings demonstrated different items with low factor loadings across groups. Therefore, making obscene phone calls, carrying weapons, and taking things from other students are less related to proactive aggression among girls, while madness and anger after losing a game could not indicate reactive aggression among boys. However, the trait of proactive aggression among Iranian adolescents did not include using force to obtain money, considering the poor factor loading of Item 15 in both groups. Because it has been suggested to stop the invariance and group differences assessment in the case of metric noninvariance (Putnick & Bornstein, 2016; Zager Kocjan et al., 2021), this study did not evaluate the scalar invariance, a necessary prerequisite to compare the mean difference between groups (Putnick & Bornstein, 2016; Zager Kocjan et al., 2021). Therefore, this measure cannot be used to capture the difference in the reactive and proactive aggression scores between Iranian boys and girls.
The findings of the current study are comparable with studies indicating measurement invariance of the RPQ after modification of the model. For example, a study by Baker et al. (2008) reported the invariance of the RPQ across genders after implementing certain modifications (e.g., allowing for correlation of error variance). Structural invariance of the RPQ was also confirmed among Portuguese youth after excluding Item 21 (Pechorro et al., 2018).
The impact of gender on the expression of aggression alongside cultural issues should be considered to interpret these findings. Additionally, because few studies assessed the measurement invariance of the RPQ across genders, further studies should be conducted to support these findings and ascertain the variant of this scale. Keeping in mind that there were different ages in each group in this research, further research on the invariance assessment of this scale would benefit from more homogeneous samples.
Another purpose of this study was to examine the reliability of the RPQ across genders. Omega and alpha coefficients indicated sufficient internal consistency of the RPQ (Deng & Chan, 2017; Nunnally, 1994), except for the proactive subscale in females. Because heterogeneity in this subscale is not the case regarding the inter-relatedness of the items (Tavakol & Dennick, 2011), further investigation is needed to examine whether a low omega coefficient might result from the reduction in the items of the proactive subscale (Tavakol & Dennick, 2011) or the inability of this questionnaire to measure proactive aggression in girls (Brugman et al., 2017).
The total score of the RPQ, like in previous studies, was correlated to subscales of CBCL and SDQ in both groups, supporting the convergent validity of the RPQ (Bartels et al., 2018). In line with previous studies (Fossati et al., 2009; Raine et al., 2006), residualized scores were used due to a moderate correlation between both subscales to take into account the shared variance. The score of rule-breaking behaviors was significantly associated with purely proactive aggression among girls when the residualized scores were used. This result is consistent with the notion that this kind of aggression is more instrumental and planned (Fossati et al., 2009). In contrast, the rule-breaking score was correlated with a purely reactive score among boys, suggesting that boys who demonstrate aggression followed by provocation tend to engage in delinquency and violate rules.
Furthermore, the aggressive behavior score showed a moderate correlation with purely reactive aggression in both groups, implying reactive aggression is a predominant form of aggression (Thomson & Centifanti, 2018). However, the association between aggressive behaviors and purely proactive aggression was significant only among girls, suggesting distinctive patterns of association across genders (Connor et al., 2003). This result implied that girls with high aggressive behaviors display both profiles of aggression, which is consistent with studies indicating the mixture of both kinds of aggression among the more aggressive individuals (Colins, 2016; Euler et al., 2017; Fossati et al., 2009; Merk et al., 2005).
Since residualized proactive aggression did not show any significant correlation among boys, further investigation is needed to find out the reason behind these results. Using the residualized score to control for the effect of the other dimension may contribute to these findings due to increasing error variance (Raine et al., 2006). Keeping in mind that there are common correlates for reactive and proactive aggression, subsequent studies should notice more specific variables to support more evidence for the differential correlates of each subscale. Overall, our hypothesis regarding the convergent validity of the RPQ scale across genders was supported.
Data on this study support the idea that proactive aggression is related to delinquency and antisocial behaviors (Brendgen et al., 2001; Deng & Chan, 2017), as incarcerated boys showed a significantly higher level of proactive aggression but not reactive form. This result contrasts with the study showing the association between delinquency and the severity of aggression but not the types (Stickle et al., 2012). However, this result supports the association of severe aggression, psychopathic and callous–unemotional traits with proactive but not reactive aggression (Brugman et al., 2017; Cima et al., 2013; Cima & Raine, 2009). This result signifies differential trajectories for these two kinds of aggression (Poulin & Boivin, 2000; Raine et al., 2006). The nonsignificant mean difference in reactive aggression score between nonincarcerated and incarcerated groups is consistent with the fact that a moderate level of reactive aggression is normative (Brugman et al., 2017; Fossati et al., 2009; Raine et al., 2006), suggesting reactive aggression as a relatively common type (Cima et al., 2013; Fung et al., 2009). However, the impact of the small sample size should be considered when interpreting this finding since the effect size was small.
Furthermore, this study can provide evidence for the ability of this scale to screen adolescents with proactive aggression, despite a disagreement on the existence of a solely proactive aggressive population (Merk et al., 2005; Smeets et al., 2017) or the ability of the RPQ in recognizing such group (Brugman et al., 2017). However, this result contrasts with research reporting significantly higher reactive and proactive aggression scores in violent offenders compared with nonoffenders (Cima et al., 2013). This discrepancy could be explained by the disparate types of crimes in our study, including both violent and nonviolent crimes.
The current study confronted some limitations that should be considered. This study was the first to examine the psychometric properties of the RPQ while comparing the RPQ factor structure across two genders in Iranian population. Since there is strong evidence for validation of the theoretical factor structure of the RPQ across different cultures, the result of the current study should be interpreted cautiously. More research is needed to examine whether this result is affected by methodological artifacts or actual cultural context differences. This cross-sectional study could not present the differential pathways of these two types of aggression and the temporal stability of the RPQ. A future longitudinal study is encouraged to measure the invariance of factor structure across time. Since this study used one source of information for all assessments, future examination for the evaluation of criterion validity of the RPQ should consider observational assessment, computational instruments, and multiple informants such as friends, parents, and teachers. In this study, the psychometric properties of the RPQ were evaluated in a nonclinical sample. We suggest replicating this study with a higher risk sample to evaluate the usefulness of this brief and easy answer questionnaire for the clinical population, considering the co-occurrence of these two types of aggression among highly aggressive individuals (Merk et al., 2005). Therefore, regarding the moderate to a strong correlation between these subtypes of aggression (Fossati et al., 2009; Hubbard et al., 2010; Poulin & Boivin, 2000; Raine et al., 2006) and concerns about the usefulness of this distinction, repetition of this study in cases with the extreme scores may be of value.
Since there was no female incarcerated, the generalizability of the result in this regard is limited. Regarding a severity model of reactive and proactive aggression against a typology model (Stickle et al., 2012), future studies should take into account the severity of aggression and psychiatric diagnosis in the relationship between these types of aggression and delinquency in a larger sample, including both genders (Cima et al., 2013). Such studies would provide the possibility to compare the exclusive role of reactive and proactive aggression in various types of crimes across genders.
Altogether, this article provides more evidence for reactive and proactive aggression independence in line with the previous research (Connor, 2004). However, data in this study did not support the equivalence of factor loadings across genders, implying that boys and girls express reactive and proactive aggression differently. However, the 2-factor structure represented the data as a reasonable well-fitting model with adequate reliability and validity across genders. This study signified future investigation of the distinctive correlation between these two forms of aggression and emotional and cognitive deficits across both genders and ages. In this regard, scales dissociating different facets of aggression alongside functions are recommended for a better configuration of the distinctive correlates of reactive and proactive aggression. Such studies would provide support for the unique etiological and consequential pathways and shed light on producing more efficient and specific intervention and prevention strategies through attention to specific underlying mechanisms.
References
. (1991). Integrative guide for the 1991 CBCL/4-18, YSR, and TRF profiles. Department of Psychiatry, University of Vermont.
(2001). Manual for the ASEBA school-age forms & profiles: An integrated system of multi-informant assessment. University of Vermont, Research Center for Children, Youth & Families.
(1993). Distinguishing instrumental and hostile aggression: Does it make a difference? Journal of Abnormal Child Psychology, 21(4), 355–365. 10.1007/bf01261598
(2008). Differential genetic and environmental influences on reactive and proactive aggression in children. Journal of Abnormal Child Psychology, 36(8), 1265–1278. 10.1007/s10802-008-9249-1
(1973). Aggression: A social learning analysis. Prentice-Hall.
(2018). Childhood aggression and the co-occurrence of behavioural and emotional problems: Results across ages 3–16 years from multiple raters in six cohorts in the EU-ACTION project. European Child & Adolescent Psychiatry, 27(9), 1105–1121. 10.1007/s00787-018-1169-1
(2012). Proactive and reactive aggression are associated with different physiological and personality profiles. Journal of Social and Clinical Psychology, 31(5), 458–487. 10.1521/jscp.2012.31.5.458
(2001). Reactive and proactive aggression: Predictions to physical violence in different contexts and moderating effects of parental monitoring and caregiving behavior. Journal of Abnormal Child Psychology, 29(4), 293–304. 10.1023/a:1010305828208
(2015). Confirmatory factor analysis for applied research. Guilford Publications.
(2017). Examining the reactive proactive questionnaire in adults in forensic and non-forensic settings: A variable-and person-based approach. Aggressive Behavior, 43(2), 155–162. 10.1002/ab.21671
(2001). Is it time to pull the plug on hostile versus instrumental aggression dichotomy? Psychological Review, 108(1), 273–279. 10.1037/0033-295x.108.1.273
(2006). Proactive and reactive aggression in childhood and adolescence: A meta-analysis of differential relations with psychosocial adjustment. International Journal of Behavioral Development, 30(5), 466–480. 10.1177/0165025406071904
(2016). Validity and reliability of the reactive-proactive aggression questionnaire in Turkish adolescents. Educational Research and Reviews, 11(20), 1931–1943. 10.5897/ERR2016.2937
(2007). Sensitivity of goodness of fit indexes to lack of measurement invariance. Structural Equation Modeling: A Multidisciplinary Journal, 14(3), 464–504. 10.1080/10705510701301834.
(2002). Evaluating goodness-of-fit indexes for testing measurement invariance. Structural Equation Modeling: A Multidisciplinary Journal, 9(2), 233–255. 10.1207/s15328007sem0902_5
(2009). Distinct characteristics of psychopathy relate to different subtypes of aggression. Personality and Individual Differences, 47(8), 835–840. 10.1016/j.paid.2009.06.031
(2013). Validation of the Dutch Reactive Proactive Questionnaire (RPQ): Differential correlates of reactive and proactive aggression from childhood to adulthood. Aggressive Behavior, 39(2), 99–113. 10.1002/ab.21458
(1995). Constructing validity: Basic issues in objective scale development. Psychological Assessment, 7(3), 309–319. 10.1037/1040-3590.7.3.309
(1988). Statistical power analysis for the behavioral sciences (2nd ed.). Lawrence Erlbaum Associates. 10.4324/9780203771587
(2016). Assessing reactive and proactive aggression in detained adolescents outside of a research context. Child Psychiatry & Human Development, 47(1), 159–172. 10.1007/s10578-015-0553-z
(2006). The correlates of proactive and reactive aggression in early childhood [Unpublished doctoral dissertation]. The University of North Carolina at Greensboro, USA. https://libres.uncg.edu/ir/uncg/f/umi-uncg-1221.pdf.
(2004). Aggression and antisocial behavior in children and adolescents: Research and treatment. Guilford Press.
(2003). Gender differences in reactive and proactive aggression. Child Psychiatry and Human Development, 33(4), 279–294. 10.1023/a:1023084112561
(2017). Testing the difference between reliability coefficients alpha and omega. Educational and Psychological Measurement, 77(2), 185–203. 10.1177/0013164416658325
(1987). Social-information-processing factors in reactive and proactive aggression in children's peer groups. Journal of Personality and Social Psychology, 53(6), 1146–1158. 10.1037/0022-3514.53.6.1146
(1991). The structure and function of reactive and proactive aggression. In D. Pepler & K. Rubin (Eds.), The development and treatment of childhood aggression (pp. 201–218). Lawrence Erlbaum Associates, Inc.
(2017). Distinct profiles of reactive and proactive aggression in adolescents: Associations with cognitive and affective empathy. Child and Adolescent Psychiatry and Mental Health, 11(1), 1–14. 10.1186/s13034-016-0141-4
(2009). A cross-cultural study of the psychometric properties of the Reactive–Proactive Aggression Questionnaire among Italian nonclinical adolescents. Psychological Assessment, 21(1), 131–135. 10.1037/a0014743
(2009). Cross-cultural generalizability of the Reactive–Proactive Aggression Questionnaire (RPQ). Journal of Personality Assessment, 91(5), 473–479. 10.1080/00223890903088420
(1998). The Strengths and Difficulties Questionnaire: A pilot study on the validity of the self-report version. European Child & Adolescent Psychiatry, 7(3), 125–130. 10.1007/s007870050057
(2009). Multivariate data analysis (7th ed.). Prentice Hall.
(2018). Exploring the differential associations between components of executive functioning and reactive and proactive aggression. Journal of Clinical and Experimental Neuropsychology, 40(1), 62–74. 10.1080/13803395.2017.1314450
(1995). Structural equation modeling: Concepts, issues, and applications. Sage.
(1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6(1), 1–55. 10.1080/10705519909540118
(2010). Reactive and proactive aggression in childhood and adolescence: Precursors, outcomes, processes, experiences, and measurement. Journal of Personality, 78(1), 95–118. 10.1111/j.1467-6494.2009.00610.x
(2015). Factor validity of a proactive and reactive aggression rating scale. Journal of Child and Family Studies, 24(9), 2734–2744. 10.1007/s10826-014-0075-5.
(2005). Reactive and proactive aggression in children A review of theory, findings and the relevance for child and adolescent psychiatry. European Child & Adolescent Psychiatry, 14(1), 11–19. 10.1007/s00787-005-0432-4
(2006). A parent questionnaire for distinguishing between reactive and proactive aggression in children. European Child & Adolescent Psychiatry, 15(1), 38–45. 10.1007/s00787-006-0502-2
(2015). Principles and practice of structural equation modeling. Guilford Publications.
(2013). Calculating and reporting effect sizes to facilitate cumulative science: A practical primer for t-tests and ANOVAs. Frontiers in Psychology, 4,
Article 863 . 10.3389/fpsyg.2013.00863(2018). The neural basis of reactive aggression and its development in adolescence. Psychology, Crime & Law, 24(3), 313–333. 10.1080/1068316x.2017.1420187
(2013). The relationship between adult reactive and proactive aggression, hostile interpretation bias, and antisocial personality disorder. Journal of Personality Disorders, 27(1), 53–66. 10.1521/pedi.2013.27.1.53
(2005). The distinction between reactive and proactive aggression: Utility for theory, diagnosis and treatment? European Journal of Developmental Psychology, 2(2), 197–220. 10.1080/17405620444000300
(2006). Adaptation and standardization of child behavior checklist, youth self-report, and teacher’s report forms. Journal of Exceptional Children, 6(1), 529–558.
(2010). Testing measurement invariance across groups: Applications in cross-cultural research. International Journal of Psychological Research, 3(1), 111–130.
(1994). Psychometric theory 3E. Tata McGraw-Hill Education.
(2013). Patterns of reactive and proactive aggression in young adolescents in Singapore. Social Development, 22(4), 794–812. 10.1111/sode.12024
(2018). The Reactive–Proactive Aggression Questionnaire: Measurement invariance and reliability among a school sample of Portuguese youths. Child Psychiatry & Human Development, 49(4), 523–533. 10.1007/s10578-017-0772-6
(2015). Adaptation of the Basic Empathy Scale among a Portuguese sample of incarcerated juvenile offenders. Psychology, Crime & Law, 21(7), 699–714. 10.1080/1068316x.2015.1028546
(2007). A meta-analysis of the distinction between reactive and proactive aggression in children and adolescents. Journal of Abnormal Child Psychology, 35(4), 522–535. 10.1007/s10802-007-9109-4
(2000). Reactive and proactive aggression: Evidence of a two-factor model. Psychological Assessment, 12(2), 115–122. 10.1037/1040-3590.12.2.115
(2016). Measurement invariance conventions and reporting: The state of the art and future directions for psychological research. Developmental Review, 41, 71–90. 10.1016/j.dr.2016.06.004
(2006). The reactive–proactive aggression questionnaire: Differential correlates of reactive and proactive aggression in adolescent boys. Aggressive Behavior: Official Journal of the International Society for Research on Aggression, 32(2), 159–171. 10.1002/ab.20115
(2009). Cuestionario de agresión reactiva y proactiva: Un instrumento de medida de la agresión en adolescentes [Reactive and proactive aggression questionnaire: An instrument for measuring aggression in adolescents]. Revista de psicopatología y psicología clínica, 14(1), 37–49. 10.1002/ab.20115
(2004). A beginner's guide to structural equation modeling. Psychology Press.
(2008). Differential correlates of reactive and proactive aggression in Asian adolescents: Relations to narcissism, anxiety, schizotypal traits, and peer relations. Aggressive Behavior: Official Journal of the International Society for Research on Aggression, 34(5), 553–562. 10.1002/ab.20269
(2017). Are proactive and reactive aggression meaningful distinctions in adolescents? A variable-and person-based approach. Journal of Abnormal Child Psychology, 45(1), 1–14. 10.1007/s10802-016-0149-5
(1999). Juvenile offenders and victims: 1999 national report. Office of juvenile justice and delinquency prevention. https://ojjdp.ojp.gov/library/publications/juvenile-offenders-and-victims-1999-national-report
(2012). Gender differences in psychopathic traits, types, and correlates of aggression among adjudicated youth. Journal of Abnormal Child Psychology, 40(4), 513–525. 10.1007/s10802-011-9588-1
(2011). Making sense of Cronbach's alpha. International Journal of Medical Education, 2, 53–55. 10.5116/ijme.4dfb.8dfd
(2007). Validity of farsi version of strengths and difficulties questionnaire (SDQ). Advances in Cognitive Science, 8(4), 33–39. http://icssjournal.ir/article-1-266-en.html
(2018). Proactive and reactive aggression subgroups in typically developing children: The role of executive functioning, psychophysiology, and psychopathy. Child Psychiatry & Human Development, 49(2), 197–208. 10.1007/s10578-017-0741-0
(2020). Factorial analysis and invariance testing for age and gender of the Reactive-Proactive Aggression Questionnaire (RPQ). International Journal of Psychological Research, 13(1), 62–70. 10.21500/20112084.4190
(2017). Use of multi-group confirmatory factor analysis in examining measurement invariance in counseling psychology research. The European Journal of Counselling Psychology, 6(1), 75–82. 10.23668/psycharchives.2041
(2006). Subtypes of aggressive behaviors: A developmental perspective. International Journal of Behavioral Development, 30(1), 12–19. 10.1177/0165025406059968
(1998). Reactive and proactive aggression differentially predict later conduct problems. The Journal of Child Psychology and Psychiatry and Allied Disciplines, 39(3), 377–385. 10.1017/s0021963097002102
(2021). Measurement invariance of the subjective happiness scale across countries, gender, age, and time. Assessment, 29(4), 826–841. 10.1177/1073191121993558
