The Problem Behavior Frequency Scale-Adolescent Report form (PBFS-AR; Farrell et al., 2016) is a self-report measure designed to assess adolescents’ frequency of victimization, aggression and other problem behaviors. The PBFS-AR has several advantages over other self-report measures of adolescents’ aggression and victimization. It addresses multiple forms of both in-person and cyber aggression and victimization (i.e., physical, verbal, and relational), substance use, and delinquent behavior. This provides a basis for examining the relations among these constructs and assessing multiple outcomes. It specifies the time frame (i.e., past 30 days) using an operationally-defined 6-point frequency scale (e.g., Never, 1–2 times, 3–5 times), which provides a clear basis for assessing change over time. It also focuses on specific behaviors, which may be more clearly interpreted than items that assess intent (e.g., “When someone hurts me, I end up getting into a fight,” “I am deliberately cruel to others, even if they haven’t done anything to me”; Marsee et al., 2011) or general dispositions (e.g., “I am the kind of person who often fights with others”; Little et al., 2003).

Farrell et al. (2016) evaluated an earlier version of the PBFS-AR that did not include items assessing cyber aggression or victimization using data from 5,532 adolescents from 37 schools in four states. Their confirmatory factor analyses of competing models found support for seven factors. These represented three forms of aggression (physical, verbal, and relational), two forms of victimization (overt and relational), substance use, and other delinquent behavior. They also found evidence of strong measurement invariance (i.e., scalar invariance) such that the structure, factor loadings, and thresholds of items did not differ across sex, geographic locations, or grades. Their analyses of correlations among the seven factors provided clearer evidence for representing physical and relational aggression as distinct factors (r = .74) than for verbal aggression, which was highly correlated with the Physical Aggression and Relational Aggression factors (rs = .89 and .82, respectively).

Farrell et al. (2016) evaluated the concurrent validity of the PBFS-AR based on its correlations with teacher ratings on the Behavioral Assessment System for Children (BASC-2; Reynolds & Kamphaus, 2004) and with adolescent reports on measures of other relevant constructs. The PBFS-AR Physical Aggression, Delinquent Behavior, and Substance Use factors were positively correlated with teachers’ ratings on the BASC Aggression (rs = .20 to .24) and Conduct Problems (rs = .23 to .26) scales and were negatively correlated with the BASC Adaptive Behavior scale (rs = −.26 to −.22). Whereas the PBFS-AR Verbal Aggression factor was also significantly correlated with the BASC Aggression scale (r = .24), it was not as strongly related to the BASC Conduct Problems or Adaptive Behavior scales. These correlations were all smaller for the PBFS-AR Relational Aggression factor (rs = .14 to .16 in absolute value). As predicted, the PBFS-AR Overt Victimization and Relational Victimization factors had small, but significant correlations with the BASC Depression scale (rs = .12 and .17, respectively). Scales on the PBFS-AR also showed the expected pattern of correlations with adolescent reports on measures of related constructs. The PBFS-AR Physical Aggression, Delinquent Behavior, and Substance Use factors were positively correlated with delinquent peer associations (rs = .48 to .50), and with measures of norms and beliefs supporting aggression (rs = .32 to .47). These correlations were generally smaller for the PBFS-AR Relational Aggression (rs = .30 to .41) and Verbal Aggression factors (rs = .32 to .42). Correlations for the two PBFS-AR victimization factors were considerably lower (rs = −.05 to .28).

The present study evaluated a revised version of the PBFS-AR (Version 2) administered to a large, predominantly African American sample of middle school students. Revisions included adding some additional items to assess in-person verbal aggression, in-person physical victimization, and in-person verbal victimization, and substance use. A major revision was expanding the measure to include items representing cyber forms of both aggression and victimization. This study had several objectives. The first was to compare competing models of the structure of victimization, aggression, and other problem behaviors based on this expanded pool of items. We hypothesized that support would be found for factors representing specific forms of aggression and victimization, and factors representing substance use and delinquent behavior. Based on the Farrell et al. (2016) study we expected to find support for factors representing physical, verbal, and relational forms of in-person aggression. Although Farrell et al. found support for two victimization factors (overt and relational), their study included only one item representing verbal victimization. We anticipated that our inclusion of additional verbal victimization items would result in three distinct in-person victimization factors (physical, verbal, and relational). We further hypothesized that cyber and in-person forms of both aggression and victimization would represent distinct factors based on previous studies and on the qualitative differences between interactions via electronic communication technologies and in person interactions (Landoll et al., 2015; Mehari et al., 2014; Mehari & Farrell, 2018). However, we were less certain whether cyber aggression and victimization items would be best represented by single aggression and victimization factors, or by factors that paralleled their in-person counterparts (i.e., physical, verbal and relational forms of cyber aggression and victimization).

A second objective was to test for measurement invariance. There has been increasing recognition of the need to establish measurement invariance, or the degree to which measurement properties are consistent across individuals, contexts, and over time (Rosen et al., 2013). Despite its importance, few attempts have been made to evaluate the measurement invariance of measures assessing constructs such as aggression, victimization and problem behaviors (e.g., Farrell et al., 2016; Marsee et al., 2011; Queirós & Vagos, 2016; Rosen et al., 2013). Based on analyses of the earlier version of the PBFS-AR by Farrell et al. (2016), we hypothesized that there would be strong measurement invariance across sex and grade.

Finally, we evaluated the concurrent validity of the PBFS-AR by examining its relation to school office discipline referrals. Prior work has supported the use of school office discipline referrals to validate teacher and parent report measures of student behavior (for a review, see McIntosh, Fisher, Kennedy, Craft, & Morrison, 2012). We hypothesized that students who had office discipline referrals for physical aggression would have higher scores on the PBFS-AR Physical Aggression factor than those without such referrals, and that students with office discipline referrals for disruptive behavior and those with out-of-school suspensions would have higher scores on each of the problem behavior factors compared with those that did not. We did not expect to find relations between office discipline referrals and suspensions and victimization factors.

The primary objectives of this study were to compare competing models of the structure of victimization and adolescent problem behaviors, including aggression, substance use, and other delinquent behavior, and to evaluate the validity of the PBFS-AR based on its relations to school office discipline referrals. A key question was the extent to which there are separate factors representing physical, verbal, and relational forms of aggression and victimization, and whether these distinctions are evident in both in-person and cyber forms of aggression and victimization. The results of this study were generally consistent with both theory and the findings of other studies that have made a case for differentiating between physical and relational aggression (for a review see Voulgaridou & Kokkinos, 2015). Specifically, we found support for in-person physical and relational acts as distinct forms of aggression. However, the evidence was less clear for verbal aggression. This lack of clarity is consistent with the results of Farrell et al.’s (2016) analyses of the structure of in-person aggression. They found that correlations between verbal aggression and some constructs (e.g., teacher ratings of aggression, norms for aggression) were similar to those found for physical aggression. However, correlations between verbal aggression and other constructs (e.g., teacher ratings of adaptive behavior, delinquent peer associations) were more similar to those found for relational aggression.

Whereas the literature has been clear in differentiating between physical and relational aggression, it is less clear where verbal aggression fits within this framework. This lack of differentiation may reflect a lack of conceptual clarity regarding the definition of verbal aggression. Physical and relational aggression are defined based on intent. Physical aggression involves the intent to cause physical harm, whereas relational aggression involves the intent to damage social relationships (Voulgaridou & Kokkinos, 2015). Items are defined as verbal aggression based solely on the mode of delivery, which may not reflect the intent or context. For example, insulting someone or calling them names may represent a direct attack on the individual’s sense of self or worth, a precursor to a physical confrontation, or an attempt to damage the individual’s social standing in a public setting. Verbal aggression may encompass a range of behaviors, some of which are more closely related to physical aggression, some of which are more closely related to relational aggression, and some of which sit squarely in the middle. For example, in this study, the items “Yelled at someone or called them mean names” and “Put someone down to their face,” both of which could be construed as an attack designed to provoke, loaded solely on the In-Person Physical Aggression factor. In contrast, the items related to making fun of someone to make others laugh and saying something disrespectful about someone’s family, behaviors which are likely intended to damage an individual’s image or reputation, loaded more strongly onto the Relational Aggression factor. It is possible that previous studies that found an overt (or direct) aggression factor included items assessing verbal attacks that may have been precursors to physical aggression. Future studies may need to parse out verbal aggression items based on whether the intent is to provoke, harass, or taunt an individual, or to damage the individual’s image, reputation, or relationships.

In contrast to in-person aggression, we did not find support for distinct physical or relational forms of cyber aggression, which is consistent with previous work (Mehari & Farrell, 2018). This may be because items representing cyber forms of physical aggression involve threats of physical aggression rather than actual physical aggression. Although we were not able to differentiate among different forms within cyber aggression, we did find support for cyber aggression as distinct from in-person aggression. This supports the notion that perpetrating aggression via electronic communication technologies has unique aspects that may influence which adolescents are more likely to engage in this form of aggression. Interestingly, its patterns of relations with office disciplinary referrals was more similar to the pattern for physical aggression than to the pattern found for relational aggression. That is, perpetrating cyber aggression was associated with a higher likelihood of office referrals for disruptive behavior and suspensions than was perpetrating relational aggression. It is possible that perpetrating cyber aggression is indicative of more nonconforming or deviant behaviors, which is supported in part by the lower prevalence of cyber aggression compared to in-person aggression (e.g., Wang et al., 2009).

Our evaluation of the PBFS-AR found support for strong measurement invariance across sex and grade. Although measurement invariance is a prerequisite to making meaningful comparisons across groups (Widaman & Reise, 1997), few prior studies have evaluated the measurement invariance of aggression and victimization (e.g., Marsee et al., 2011; Marsh et al., 2011; Rosen et al., 2013). The finding of measurement invariance across sex and grade supports the use of the PBFS-AR for examining differences across these groups. We also found mean differences across both grade and sex. Cyber aggression had similar rates of prevalence across sixth, seventh, and eighth grade, whereas in-person aggression was highest in sixth grade, consistent with other trends suggesting that in-person aggression declines throughout middle school (Farrell, Goncy, Sullivan, & Thompson, 2017). The stability of cyber aggression throughout the course of middle school suggests that it continues to be reinforced, or at least not effectively addressed through informal social control, as adolescents grow. Interestingly, trends over the past ten years suggest that in-person aggression is decreasing (Waasdorp, Pas, Zablotsky, & Bradshaw, 2017), whereas cyber aggression appears to remain stable (Smith, 2012). Taken together, these findings suggest a need to incorporate and evaluate elements of cyber aggression prevention into violence prevention programming, and to target online norms to increase effective bystander intervention in cyber aggression situations.

Although we did not find sex differences in the mean frequency of physical or relational aggression, we found that boys reported lower frequencies of cyber aggression and victimization, and in-person victimization. Differences were small for cyber aggression and in-person victimization, with medium sized effects for cyber victimization. Most prior studies have found a higher frequency of direct aggression (i.e., physical and verbal) among boys compared with girls, and negligible sex differences in relational aggression (Card et al. 2008). Our study’s lack of significant sex differences in physical aggression may reflect our focus on a predominantly African American sample. This is consistent with Card et al.’s (2008) meta-analysis, which revealed smaller sex differences in the frequency of direct aggression in samples with higher percentages of minority youth, and smaller sex differences for self-report measures. Our finding of higher frequencies of cyber aggression and victimization among girls than among boys underscores the need for further research to replicate these findings and identify factors that are responsible for these differences, particularly the larger differences for cyber victimization.

Within both in-person and cyber media, we found little support for differentiating among the forms of victimization experienced by adolescents. Our findings differed from the findings of Farrell et al.’s (2016) study of the previous version of the PBFS-AR, which found support for distinct in-person relational and physical victimization factors. However, their study included a limited pool of victimization items (e.g., only one verbal victimization item). Our conclusions also differ from those of several studies of the Revised Peer Experiences Questionnaire (RPEQ), which was designed to assess multiple forms of both aggression and victimization. Prinstein, Boergers, and Vernbert (2001) found support for distinct factors representing overt and relational victimization. In a subsequent study of an expanded pool of items, De Los Reyes and Prinstein (2004) found support for separate factors representing overt, relational, and reputational victimization. In both studies, the factor structure was determined by exploratory rather than confirmatory analyses. Queirós and Vagos (2016) found support for the same three victimization factors as De Los Reyes and Prinstein in their analysis of a Portuguese translation of the RPEQ using a confirmatory factor analysis. Although the factors identified in these three studies of the RPEQ fit the data well, none of these studies evaluated competing models of its structure. It is therefore not clear how well the models they identified would improve upon the fit of models specifying a smaller number of factors. McLaughlin, Hatzenbuehler, and Hilt (2009), for example, found support for a measurement model of the RPEQ that included a single victimization factor that was based on parcels, where each parcel included items representing overt, relational, and reputational victimization items.

The current study suggests that whereas individual adolescents may differ in their propensity to engage in physical versus relational aggression, those who are victimized are likely to report multiple forms of victimization, though they may experience different forms from different people. There may be underlying commonalities of victimization experiences (e.g., threats to security, belongingness, and acceptance at the peer and/or school level) that supersede form and intended method of hurt or harm (i.e., physical versus relational). This is consistent with research that suggests victimization during adolescence can serve as a mechanism to marginalize or exclude youth from peer groups (Brown, 2004). This does not imply that different forms of victimization all have the same impact on adjustment. By definition, physical victimization causes physical harm, and all forms of victimization carry a risk for psychosocial adjustment difficulties. Prinstein et al (2001), for example, found that relational victimization was related to high school students’ concurrent levels of loneliness and low self-esteem even after controlling for overt forms of victimization and both overt and relational forms of aggression. Further work is needed to evaluate the value added by differentiating among specific forms of victimization, in particular the extent to which they differ in their impact on adjustment.

Although we did not find support for differentiating among forms of victimization within media, we did find that in-person and cyber forms of victimization appear somewhat distinct (but highly correlated, at r = .85). With few exceptions (e.g., Landoll et al., 2015), most prior research has not tested whether in-person and cyber victimization are separate constructs. Instead, researchers have generally based research questions, study design, and empirical analyses on the untested assumption that they are distinct phenomena. This study supports that commonly held assumption. Many researchers posit that cyber victimization may cause greater harm due to circumstances surrounding the perpetration of cyber aggression that may magnify its effect, including the possibility of an unlimited audience and revictimization through views and shares (or the victim choosing to review the aggressive behavior; e.g., Tokunaga, 2010). This underscores the need for longitudinal research to explore whether there are differences in severity or types of outcomes, or interaction effects (Salmivalli et al., 2013).

Our analysis of the relations between scores on the PBFS-AR and office discipline referrals and outcomes provided support for its criterion validity. Of particular note was the finding that students with office referrals for physical aggression differed more on the PBFS-AR Physical Aggression factor than on the other measures of aggression and problem behavior. This is despite the fact that these factors, particularly those representing different forms of aggression, were highly correlated. This provides some evidence of discriminant validity for this measure. Students with office referrals for disruptive behavior and those receiving out-of-school suspensions showed more consistent differences in their frequency of the three forms of aggression, delinquent behavior, and substance use. This is consistent with the notion that students who engage in each of these types of problem behaviors are also likely to engage in other behaviors viewed as disruptive and be subject to suspensions. Office referrals were also related to victimization. Adolescents who are victimized may behave in disruptive or deviant ways, which elicits discipline from their teachers and aggression from their peers. This is consistent with research that has identified a subset of adolescents described as aggressive-victims (Bettencourt & Farrell. 2013). Such youth are more likely to have poor emotion regulation, which may increase their likelihood of engaging in disruptive classroom behavior. It is also possible that adolescents who are victimized become more disruptive, perhaps due to feeling rejected, isolated, or unsafe.