Refi ning Victims’ Self-Reports on Bullying: Assessing Frequency, Intensity, Power Imbalance, and Goal-Directedness

Bullying can be differentiated from other types of peer aggression by four key characteristics: frequency, intensity, power imbalance, and goal-directedness. Existing instruments, however, usually assess the presence of these characteristics implicitly. Can current self-report instruments be refined using additional questions that assess each characteristic? We examined (1) what proportion of children classified as victims by the commonly used Revised Olweus’ Bully/Victim Questionnaire (BVQ ) also experienced the characteristics of bullying, and (2) the extent to which the presence of the characteristics was associated with emotional (affect, school and classroom well-being), relational (friendship, defending), and social status (popularity, rejection) adjustment correlates among victims. Using data from 1,738 students, including 138 victims according to the BVQ, the results showed that 43.1% of the children who were classified as victims by BVQ experienced all four characteristics of bullying. Frequency ratings of victimization did not capture experiences that involved a power imbalance. Victims who reported all four key characteristics had greater emotional, relational, and social status problems than victims who did not report the four key characteristics. Thus, researchers who focus on victimization for diagnostic and prevention purposes can enrich self-report measurements of bullying victimization by adding questions that assess the characteristics explicitly. This chapter is based on: Kaufman, T. M. L., Huitsing, G., & Veenstra, R. (resubmitted). Refining victims’ selfreports on bullying: Assessing frequency, intensity, power imbalance, and goaldirectedness. Social Development. Binnenwerk Werkbestand Tessa Kaufman.indd 102 29-12-2019 16:07:24 103 Refi ning Victims’ Self-Reports on Bullying Introduc tion Bullying is widely recognized as a unique peer phenomenon (Volk et al., 2017). By defi nition, “a person is being bullied when he or she is exposed, repeatedly and over time, to negative actions on the part of one or more other persons” (Olweus, 1993). More concretely, bullying can be diff erentiated from other types of peer aggression by four key characteristics: frequency, intensity, power imbalance, and goal-directedness (Volk et al., 2014). This distinction between victims of bullying and victims of the broader class of peer aggression is crucial in evaluating anti-bullying interventions. Tackling victimization through bullying requires diff erent interventions than reducing general victimization through aggression (Espelage et al., 2013; Taub, 2002; Van Schoiack-Edstrom et al., 2002). Bullying is embedded in the peer group and thus also requires groupfocused interventions, whereas general aggression may also be resolved by targeting individual skills. Thus, evaluating anti-bullying interventions calls for measures that can diff erentiate between victimization through bullying and through general aggression. Last, the distinction is also important to limit variability in prevalence estimates of bullying across studies (Cook, Williams, Guerra, & Kim, 2010). However, it is unclear whether currently used measures of peer victimization are able to diff erentiate between victimization through bullying versus general aggression (Bauman, 2016; Furlong, Sharkey, Felix, Tanigawa, & Green, 2010; Vivolo-Kantor, Martell, Holland, & Westby, 2014). The Revised Olweus’ Bully/Victim Questionnaire (BVQ; 1996), considered the most widely used instrument (T. Lee & Cornell, 2009), uses a defi nition-fi rst approach, meaning that it fi rst provides children with a defi nition of bullying and subsequently asks about the frequency of aggressive experiences. However, children may not retain this complex, multi-component defi nition in working memory and apply it when answering the questionnaire. They may fall back on their own prior assumptions about the term bullying and, therefore, report their experiences of general peer aggression as bullying (Furlong et al., 2010; Jia & Mikami, 2018). Although the BVQ explicitly measures “frequency”, this does not necessarily guarantee that it captures the other key characteristics of the defi nition, because higher-frequency experiences may not be intense, nor happen in the context of a power imbalance or on purpose (Felix, Sharkey, Green, Furlong, & Tanigawa, 2011). Indeed, previous research has shown that BVQ responses were more likely to detect repeated experiences than power imbalance (Green, Felix, Sharkey, Furlong, & Kras, 2013). Thus, a general concern is that the sole use of frequency ratings in a defi nitionfi rst approach cannot fully discriminate between victims of bullying and victims of other types of peer aggression. 5 Binnenwerk Werkbestand Tessa Kaufman.indd 103 29-12-2019 16:07:24


Introduc tion
Bullying is widely recognized as a unique peer phenomenon (Volk et al., 2017). By defi nition, "a person is being bullied when he or she is exposed, repeatedly and over time, to negative actions on the part of one or more other persons" (Olweus, 1993). More concretely, bullying can be diff erentiated from other types of peer aggression by four key characteristics: frequency, intensity, power imbalance, and goal-directedness (Volk et al., 2014).
This distinction between victims of bullying and victims of the broader class of peer aggression is crucial in evaluating anti-bullying interventions. Tackling victimization through bullying requires diff erent interventions than reducing general victimization through aggression (Espelage et al., 2013;Taub, 2002;Van Schoiack-Edstrom et al., 2002). Bullying is embedded in the peer group and thus also requires groupfocused interventions, whereas general aggression may also be resolved by targeting individual skills. Thus, evaluating anti-bullying interventions calls for measures that can diff erentiate between victimization through bullying and through general aggression. Last, the distinction is also important to limit variability in prevalence estimates of bullying across studies (Cook, Williams, Guerra, & Kim, 2010).
However, it is unclear whether currently used measures of peer victimization are able to diff erentiate between victimization through bullying versus general aggression (Bauman, 2016;Furlong, Sharkey, Felix, Tanigawa, & Green, 2010;Vivolo-Kantor, Martell, Holland, & Westby, 2014). The Revised Olweus' Bully/Victim Questionnaire (BVQ;1996), considered the most widely used instrument (T. Lee & Cornell, 2009), uses a defi nition-fi rst approach, meaning that it fi rst provides children with a defi nition of bullying and subsequently asks about the frequency of aggressive experiences. However, children may not retain this complex, multi-component defi nition in working memory and apply it when answering the questionnaire. They may fall back on their own prior assumptions about the term bullying and, therefore, report their experiences of general peer aggression as bullying (Furlong et al., 2010;Jia & Mikami, 2018). Although the BVQ explicitly measures "frequency", this does not necessarily guarantee that it captures the other key characteristics of the defi nition, because higher-frequency experiences may not be intense, nor happen in the context of a power imbalance or on purpose (Felix, Sharkey, Green, Furlong, & Tanigawa, 2011). Indeed, previous research has shown that BVQ responses were more likely to detect repeated experiences than power imbalance (Green, Felix, Sharkey, Furlong, & Kras, 2013). Thus, a general concern is that the sole use of frequency ratings in a defi nitionfi rst approach cannot fully discriminate between victims of bullying and victims of other types of peer aggression.

104
Chapter 5 A possible way to improve differentiation between victimization groups is to add questions to instruments such as the BVQ that address each key characteristic of the bullying definition explicitly (Bauman, 2016;Furlong et al., 2010;Jia & Mikami, 2018;Volk et al., 2014). However, it is empirically unclear whether the definition characteristics are valid indicators of victimization through bullying: whether experiencing all key characteristics indeed relates to the correlates that conceptually differentiate between victimization through bullying and through general aggression (Jia & Mikami, 2018).
In addressing these concerns, our aim was twofold. First, we aimed to examine the extent to which the method of using a definition-first approach and endorsing frequency ratings only (i.e., the BVQ ) captures experiences that match the definition of bullying. Second, we examined whether adding to the BVQ questions that explicitly assessed the characteristics helped to differentiate victimization through bullying from victimization through general aggression. Did victims who were victimized in line with the bullying definition (compared with victims who were not) show the emotional, relational, and social status adjustment correlates that conceptually relate more strongly to victimization through bullying than to victimization through general aggression?

Theory Key Characteristics of Victimization through Bullying
In the past, bullying was perceived as an impulsive, uncontrolled outburst of aggression toward victims (Olweus, 1978). However, nowadays most people agree that bullying is part of a complex group phenomenon (Salmivalli, 2010) that is characterized by four key characteristics. Bullying (1) repeatedly (2) harms victims in the context of a (3) power imbalance and predominantly involves (4) strategic, goaldirected behavior (Olweus, 1993;Reijntjes et al., 2013;Volk et al., 2014).
These key characteristics have been conceptually and empirically related to correlates of victimization through bullying. First, the repetitive character of bullying refers to frequent experiences of victimization instead of a one-time occurrence (Olweus, 1993;Volk et al., 2014). The frequency of monthly ("two or three times a month") victimization seems a valid lower cutoff point for classifying children as victims, because this distinguishes victims from non-victims in levels of higher psychosocial maladjustment (Solberg & Olweus, 2003). Frequent victimization has been related to lower social support and well-being (e.g., Fullchange & Furlong, 2016;Solberg & Olweus, 2003;Ybarra, Espelage, & Mitchell, 2014).
Second, bullying was proposed to be characterized by intensity, meaning that victims experience bullying as harmful and thus intense, which diff erentiates it from playful teasing or fi ghting (Olweus, 1993;Volk et al., 2014). Perceived intensity may be a powerful predictor of worse adjustment correlates (Volk et al., 2014).
Third, power imbalance means that bullies choose victims who have less physical or social strength (Nelson, Kendall, Burns, Schonert-Reichl, & Kane, 2019;Olweus, 1993) in order to lower the cost of their behaviors. For example, bullies minimize loss of aff ection by choosing victims who are not likely to be defended by signifi cant others (Veenstra, Lindenberg, Munniksma, & Dijkstra, 2010). This imbalance of power diff erentiates bullying from general aggression, in which aggressors can be equal in power, and refl ects bullying's social nature (Volk et al., 2014). Being victimized by someone with greater power may elicit depressive symptoms, because victims feel powerless to change the situation (Hunter, Boyle, & Warden, 2007), and can interfere with victims' relationships with friends and family and their schoolwork (Ybarra et al., 2014).
Finally, bullies' minimization of the loss of aff ection is also refl ected in goaldirectedness. This characteristic refers to the strategic nature of bullying, as opposed to accidental behavior. Whereas the intention of bullying was previously described as wanting to harm another child (Olweus, 1993), recent insights based on evolutionary and sociological theory emphasize that bullying may be aimed at obtaining or maintaining social dominance (Olthof, Goossens, Vermande, Aleva, & Van der Meulen, 2011;Van Der Ploeg, Steglich, & Veenstra, 2020;Volk et al., 2014). This can involve retaliation for previous actions by victims that were aimed at the bullies or their friends (Frey, Pearson, & Cohen, 2015), but does not necessarily need to be reactive.
Can children's experiences with all key characteristics help to discriminate between victimization through bullying and victimization through general aggression? If the BVQ appears not to capture bullying as defi ned, this is only problematic when experiencing all key characteristics would indeed improve the concurrent validity. A way to investigate this is by examining whether experiencing all key characteristics, versus not all, contributes to correlates that conceptually diff er between victimization through bullying and through general aggression (Jia & Mikami, 2018).
Conceptually, victims of bullying are particularly distinct from victims of general aggression in their greater emotional adjustment problems and problems in the peer group, such as compromised social relationships and status (Hunter et al., 2007;Olweus, 1996;Solberg & Olweus, 2003). First, victims of bullying will theoretically 106 Chapter 5 show greater emotional maladjustment because bullying concerns structural social exclusion, which the victim cannot easily escape from, given the power differential. This type of exclusion has a detrimental effect on mental health because it interferes with the fundamental human need to belong to the group (Baumeister & Leary, 1995). Second, the exclusion that characterizes bullying is extra painful because it is personoriented and thus not directed toward a random target (Olweus, 1993;Volk et al., 2014). Third, bullying is a social phenomenon that is embedded in the peer group, and thus relates to victims' social adjustment. Bullies target victims who already have less supportive peer relationships and lower social status in the peer group. Others do no longer want to affiliate with these victims because this could lower their own status and increase the risk of being the next target (Salmivalli, 2010), leaving victims with fewer peers who befriend or defend them. Although bullying also impairs victims' functioning in other domains such as academic functioning, physical health, and parent-child relationships, the majority of those factors may be associated with the emotional and social adjustment problems.
Some evidence already provides support for the suggestion that experiencing multiple key characteristics of the bullying definition relates to greater correlates of victimization through bullying. First, victims of repeated aggression who were less powerful than the bully had poorer mental health than those who only experienced repeated aggression (Ybarra et al., 2014). Second, compared with victims who experienced repeated aggression, victims who also were both less powerful than the bully and experienced that the bully victimized them on purpose had higher levels of depressive symptoms (Hunter et al., 2007;Malecki et al., 2015), anxiety, and lower self-esteem (Malecki et al., 2015).
An important next step in this research on the concurrent validity of the key characteristics of the bullying definition was to also (1) include effects on schoolrelated affect, because bullying is strongly embedded in the school context, and (2) focus on relational and social status adjustment correlates, which particularly characterize this social phenomenon. Lastly (3), previous research has not examined all four key characteristics together and their relative importance; this is central to differentiating victimization through bullying from victimization through general aggression.

Current Study
In this study, we examined whether extending self-reports of victimization with explicit assessment of each key characteristic of the bullying definition can improve the differentiation between victims of bullying and victims of other types of peer aggression. The fi rst research question (RQ1) concerned the extent to which the defi nition-fi rst approach with frequency ratings, which is employed by the most commonly used measure (the Revised BVQ;1996), captures experiences with all key characteristics of the bullying defi nition: frequency, intensity, power imbalance, and goal-directedness (Olweus, 1993;Volk et al., 2014). To this end, we added questions to the BVQ that explicitly assessed the key characteristics, and examined (RQ1a) how many of the children who were classifi ed as victims by the BVQ (victimized at least "monthly"; Solberg & Olweus, 2003) also experienced all key characteristics, and how many did not. Second, we examined which key characteristics may particularly be overlooked when using frequency ratings: (RQ1b) how strongly is frequency associated with experiences of intensity, power imbalance, and goal-directedness? We were interested in both linear and quadratic eff ects, because a very powerful or strategic bully may only need to victimize once to reach their goals.
The second research question (RQ2) investigated whether extending the BVQ with a more narrowband approach, thus posing questions that explicitly address experiences with each key characteristic, improves discrimination between victims of bullying and victims of general aggression. We examined (RQ2a) to what extent victims who experienced the key characteristics, versus those who did not, showed greater emotional (aff ect, school/classroom well-being), relational (friendships, defenders), and social status (popularity, rejection) adjustment correlates that are conceptually stronger among victims of bullying than among victims of more general peer aggression. Last (RQ2b), we examined whether the BVQ was still relevant or could be replaced by the new specifi c questions. We analyzed, within the sample of victims who experienced all key characteristics, diff erences in adjustment correlates between victims who would also have been classifi ed as victims by the BVQ (because they reported systematic, thus monthly, victimization in the BVQ measure) and those who would not have been classifi ed as victims by the BVQ (because they initially reported occasional victimization in the BVQ measure).

Procedure
First, we developed questions that explicitly assessed the key characteristics of the bullying defi nition. The questions were based on theory and previous research and at the same time were practical to use (e.g., Malecki et al., 2015;Volk et al., 2014;Ybarra et al., 2014). Second, we obtained IRB approval by the Ethics Committee from the Department of Sociology for a pilot study in four schools on the practical use and formulation of the questions, and the duration of fi lling in the questionnaire. After the pilot, we adjusted the questions and administered them together with measures of adjustment correlates to students in a larger sample of schools that took part in the eleventh wave of the ongoing study on the KiVa anti-bullying program (Kärnä et al., 2011) in the Netherlands (Huitsing et al., 2019). Schools were selected based on the criteria that they had distributed informed consent forms to parents and that they administered the bi-annual questionnaire at about the same time.
Information about the study and consent forms were sent to parents prior to assessments. An active consent procedure was used in which parents were asked to indicate whether they wished to allow their child to participate in the research. Students were informed at school about the research and gave oral assent. Students did not participate when they had no permission (parents refused participation or did not return the consent form), when they did not want to participate themselves, or when they were unable to complete the questionnaire. Internet-based questionnaires were completed individually in the classroom during regular school hours with primary teachers present to answer questions and assist students when necessary. The order of questions and instruments used was randomized to avoid systematic effects of question order.

Participants
Of the 2,257 students in the sample of schools in the main study, we used the data of 1,738 (77.0%) students who had permission to participate in the study. Children with and without permission did not differ in gender (as reported by the school) or peer-reported data (nominations received for popularity, rejection, defending, and bullying), p's > .05. The students in the final sample attended 26 schools (196 classrooms; 50.6% boys), in Dutch grades 5 to 8 (US-level grades 3 to 6; M age = 10.6, SD = 1.2). Of these students, 272 (15.7% of the total sample) had been victimized at least once in recent months (victims sample), and thus received questions about the key characteristics. The BVQ-classified systematic victims concerned 138 children (11.1% of the total sample). Last, 1,238 children (71.2% of the total sample) did not report victimization (non-victims). See Figure A5.1 for an overview of the sub-samples.

Measures
Victimization and key characteristics. We measured victimization using the traditional Olweus' (1996) BVQ. The Olweus' BVQ provides children with a definition of bullying (repeatedly harassing another child and the victim has problems defending themselves, with examples that explain that it is intense and happens on purpose); this was presented to children individually in a video that was integrated in the questionnaire. Children responded to one global item ("How often have you been bullied during the past couple of months?") followed by questions about specific forms of bullying. These items distinguished fi ve forms of bullying (7 items in total): physical, verbal (two items), relational (two items), material (taking or breaking others' property), and cyber-victimization (receiving nasty or insulting messages, calls, or pictures). Children answered on a fi ve-point scale how often they experienced each form: 0= not at all, 1= only once or twice, 2= two or three times a month, 3= about once a week, 4= several times per week.
To assess victims' experiences with the key characteristics of the bullying defi nition (referring to whether they were victims of bullying or general aggression), we extended the BVQ with new questions that assessed explicitly whether children experienced each key characteristic. Children were only presented with these questions when they reported being victimized at least "once or twice" on any of the BVQ items (see Table A5.1 for the full questionnaire). Children were fi rst asked to select the names of up to three children that bullied them most often, starting with the person who bullied them the most followed by those who also bullied them. If children were bullied by a group, they could name members of the group separately in this way.
For each bully, we then presented them with seven questions about their victimization experiences (up to 21 items). One item measured the frequency of victimization by a specifi c bully (Olweus, 1996), and one item assessed the extent to which children experienced the bullying as intense (Volk et al., 2014). Power imbalance was assessed using two items that measured whether the victims perceived the bully as being stronger (one item) or more popular (one item) than themselves. The items used were those most often used to assess power imbalance (Felix et al., 2011;Green et al., 2013;Hunter et al., 2007;Malecki et al., 2015;Ybarra et al., 2014) and which have been shown to measure the most relevant types of power imbalance in victimization experiences (Nelson et al., 2019). Goal-directedness has previously been conceptualized as the intention to be mean (Felix et al., 2011;Hunter et al., 2007;Malecki et al., 2015), but we adhered to the theoretical conceptualization of bullying as a strategic behavior aimed at obtaining or maintaining social dominance (Olthof et al., 2011). Therefore, we assessed it using three items that measured to what extent children were sure that the bully bullied them on purpose (one item), and their perception of this purpose: to gain social reputation ("to be cool") and to take revenge (one item each). Children responded to each question using a 5-point scale, representing "not experienced" to "experienced strongly"; from this we created a scale that represented the maximum score across the three items, and not the average, because bullies may have only one goal.

110
Chapter 5 Emotional adjustment correlates. We assessed positive and negative affect using the PANAS-C (Watson, Clark, & Tellegen, 1988), which consisted of ten adjectives that described emotions. Children indicated how often they had experienced each feeling during the previous two weeks. We computed the mean of the five items for both scales. We assessed well-being at school using the mean of five items concerning perceptions of the classroom and school (Kärnä et al., 2011). Students responded to items such as "I feel accepted as I am at school" (1 = never, 4 = always), α = .82. We measured classroom comfort using four items from the Classroom Peer Context Questionnaire that focused on comfort (CPCQ; Boor-Klip, Segers, Hendrickx, & Cillessen, 2016), for example, "In this class, I belong to the group" (1 = never, 4 = always), α = .84. Latent Factor Analysis (LFA) showed that the four indicators represented one overarching construct (CFI = .98, TLI = .99): they all showed significant and acceptable to high Geomin Rotated factor loadings (positive affect λ = 0.68, negative affect λ = -0.45, school wellbeing λ = 0.90 and classroom comfort λ = 0.87). To minimize the number of analyses we used this latent factor in the analyses of RQ2.
Relational and social status adjustment correlates. To assess defending, children were first asked to read a piece explaining that some children help children who are bullied by supporting, comforting, or otherwise helping them. We then asked them to nominate the classmates who defended them: "Which classmates defend you when you are victimized?". In addition, they nominated the classmates they perceived as their best friends ("Which classmates are your best friends?", friendship), as most popular ("Whom are the most popular students in your class?", popularity), and whom they disliked ("Which classmates do you dislike?", rejection). For each student, nominations received (for defending: outgoing) for each variable were summed and divided by the number of participating classmates, resulting in proportion scores (0-1).

Analyses
We conducted the analyses in Mplus 7 and used Maximum Likelihood Estimation (MLR), which is robust to violations of non-normality. We computed intra-class correlations (ICC) of the manifest variables of victimization and adjustment correlates at the classroom level. The level of explained variance at the classroom level varied between ICC = .03 (negative affect) and ICC = .32 (friendships). We thus used a multilevel structure with the cluster command to take into account the dependent structure of the data. We controlled for children's gender and age. RQ1: Analysis steps 1 and 2. First (RQ1a), we examined to what extent the BVQ classifi es children as victims when they have experiences that are in line with the defi nition of bullying. Within the sample of children who would be classifi ed by the BVQ as victims, we examined how many of these victims did, and how many did not, (1) experience each key characteristic, and (2) experienced all characteristics together, thus in line with the defi nition. The cut-off criteria to determine that a characteristic was experienced (versus not experienced) were that, at least with one of the three bullies, victimization had happened (1) at least monthly and the victim perceived: (2) it as at least a little bit intense; (3) the bully as at least a little bit stronger or more popular; (4) it as on purpose.
Second (RQ1b), we examined which key characteristics are especially missed using measures that only assess frequency explicitly, such as the BVQ. Using regression analyses we examined to what extent frequency was associated (1) with experiencing the other key characteristics (intensity, power imbalance, and goal-directedness) all together, and (2) with the extent to which each individual key characteristic was experienced, using the original, continuous, key characteristic measures (0 = not experienced to 4 = experienced strongly).
RQ2: Analysis steps 3 and 4. We examined the concurrent validity of the addition of the key characteristics to the BVQ. First (RQ2a), did victims who experienced the key characteristics (victims of bullying) and those who did not (victims of general aggression) indeed diff er on correlates that are conceptually greater among victims of bullying than victims of general aggression? We focused on emotional, relational (friendship, defending), and social status (popularity, rejection) adjustment correlates. Next, we examined the contribution of each individual characteristic by regressing the eff ects of each key characteristic on the adjustment correlates. We allowed intercorrelations across all outcomes and key characteristics.
Last (RQ2b), we examined whether the BVQ was still relevant or could be replaced by the new specifi c questions. We analyzed, within the sample of victims who experienced all key characteristics, diff erences in adjustment correlates between victims who would also have been classifi ed as victims by the BVQ (because they reported systematic victimization in the BVQ measure) and those who would not have been classifi ed as victims by the BVQ (because they initially reported occasional victimization in the BVQ measure).
To reduce the risk for false discovery rates (FDR) (regression analyses with fi ve outcomes), we used an FDR controlling procedure when determining statistical 112 Chapter 5 significance (Benjamini & Hochberg, 1995). Thus, p values across the five outcomes were first ordered from smallest to largest, ranking them i = 1 to i = 5. A threshold of significance (critical value) was established according to the formula: critical value (p i ) = Q (m = number of tests, Q = percentage of false discoveries 5% = .05). This procedure resulted in the following critical values: p (1) ≤ .01, p (2) ≤ .02, p (3) ≤ .03, p (4) ≤ .04, p (5) ≤ .05. Each ranked p value was then compared with its corresponding critical value, starting with i = 5. The critical value that was used was that of the highest ranking p value that was below its corresponding critical value. Thus, the lowest p-value was compared to p (1) ≤ .01, the second lowest to p (2) ≤ .02, and so on. We applied this method to all analyses except the RQ2b analyses, because the small sample size in those analyses (N = 82) would increase the risk for Type 2 error; instead, we used the p < .05 threshold.

Supplementary Analyses Using Different Operationalizations
We conducted two types of supplementary analyses. First, in our main analyses, we used a dichotomized instead of continuous approach to compute the victimization groups (i.e., victims of bullying vs. victims of general aggression), to ensure that the victims of bullying experienced all, and not only some, key characteristics. The definition of bullying proposes that all four characteristics need to be present in order to classify experiences as bullying; if two people fight regularly to achieve social dominance but they are equally strong, it is not bullying (Olweus, 1993). This required a dichotomous approach because if we had used the mean across all characteristics (the continuous approach), victims could also receive a high score on "being bullied" when one characteristic was highly present (e.g., frequency), but another characteristic was absent (e.g., there was no power imbalance). However, we conducted a sensitivity analysis to determine whether the results (of RQ2a) were consistent when the effects of a continuous measure of victimization through bullying on adjustment correlates were analyzed. In this analysis, victimization through bullying represented the average score across all key characteristics.
Last, in our main analyses, the score on each individual key characteristic represented the highest (maximum) score across the three bullies, because key characteristics only need to be experienced with one of the bullies in order to have an impact. However, computing each key characteristic as the average across three bullies also seems informative to indicate the severity of victims' problems. For example, if victims experience a power imbalance with three bullies (high average for power imbalance), this may signal that they have extreme difficulty in defending themselves. Therefore, we replicated the analyses in RQ2a by computing the key characteristics as the average across three bullies. i m

Results
Step 1: Experiences of Key Characteristics of the Defi nition We fi rst examined the extent to which victims in the BVQ reported the key characteristics of bullying. In total, there were 138 systematic victims who reported the name of at least one bully: 135 (97.8% of the victims) selected the name of one bully, 91 children (65.9%) selected a second bully, and 55 children (39.9%) selected a third bully. Table 5.1 shows how many self-reported victims reported the key characteristics: it shows whether frequency (row 1), intensity (row 2), power imbalance (rows 3-5), goal-directedness (rows 6-8), and all characteristics together (row 9) were experienced. The columns show how often the characteristics were experienced on average across three bullies (column 1), and, most centrally, among at least one of the three bullies (column 2).
Children mentioned all characteristics on average in 29.8% of the victimization experiences (column 1). Victimization was repetitive in, on average, 59.6% of the cases and was experienced as intense in 79.4% of the cases. On average, 59.8% of the victimization experiences were characterized by a power imbalance; the majority were characterized by a diff erence in strength (68.7) or popularity (82.6%). Last, on average 70.7% of the children experienced victimization as being goal-directed, and within this group, a majority perceived the goal to be to increase one's status, and a minority perceived the goal to be to take revenge. Thus, overall, each key characteristic was present in most victimization cases, but not in all.
Victims were somewhat more likely to experience the key characteristics in any bully (column 2), compared with the averages across all bullies in column 1. About twofi fths (43.1%) were victimized by any bully in line with the defi nition, and threefi fths were not. Further, the majority of the children experienced each individual characteristic in at least one bully, ranging from 71.0% in which a power imbalance was present in at least one bully, to 79.7% in which victimization was goal-directed for at least one bully, and 87.6% in which victimization was experienced as intense.

114
Chapter 5 Note. 1 Refers to how often children experienced the key characteristic, on average across the three Bullies. 2 Refers to experiencing the key characteristic in any of the three Bullies (thus, in at least one Bully).

Step 2: Regressions of Intensity, Power Imbalance, and Goal-Directedness on Frequency
Children who experienced all key aspects in any bully were victimized more frequently than those who were victimized but did not experience at least one key aspect, β = 0.25, SE = 0.05, p < .001, R 2 = .07. Further, intensity (β = 0.39, SE = 0.06, p < .001) and goal-directedness (β = 0.19, SE = 0.05, p = .001) but not power imbalance (β = 0.01, SE = 0.05, p = .846) were linearly associated with higher frequency, R 2 = .23. Age and gender were unrelated to frequency, and quadratic effects of the predictors on frequency were not significant (p's > .05). Thus, high frequency ratings were also associated with intensity and, to a lesser extent, goal-directed experiences, but frequent victimization was not necessarily characterized by a power imbalance.

Steps 3 and 4: Regressions of the Key Characteristics on Adjustment Correlates
Children who were victimized but did not experience all key characteristics ( well-being and classroom comfort. Victims of bullying had also the fewest friendships (β = -.14, SE = .06, p = .019, R² = 0.10), the fewest defenders (β = -.11, SE = .05, p = .028, R² = 0.03), were the least popular (β = -.22, SE = .04, p < .001, R² = 0.08), and the most rejected (β = .11, SE = .05, p = .042, R² = 0.04). These diff erences between victims of bullying and general victims were found to be similar in an additional analysis in which victimization through bullying represented the continuous mean score across all key characteristics, with negligible diff erences in the sizes of the beta coeffi cients (diff erences < .02; Table A5.2, row 8). Thus, experiences of all key characteristics were indicators of emotional, relational, and social status adjustment correlates that are conceptually related to victimization of bullying.
Further regression analyses showed that each separate key characteristic, except for intensity, was associated with adjustment correlates when the other key characteristics were controlled for (Table 5.2). Being frequently victimized and a power imbalance were related to emotional adjustment and having fewer friendships (only power imbalance). Moreover, goal-directedness of victimization additionally explained children's emotional adjustment over and above frequency and power imbalance. Only being less powerful than the bully, but not the other key characteristics, was associated with being less popular. Last, only greater frequency, but not the other key characteristics, was related to greater rejection by classmates. All results were similar in supplementary analyses in which each key characteristic represented the mean score (instead of the highest score) across all bullies, with small diff erences in the sizes of the beta coeffi cients (diff erences < .05; Table A5.2, rows 1-4) and with the exception of the now marginal instead of signifi cant eff ect of frequency on rejection.
Overall, all key characteristics except for intensity were thus in some way related to victims' adjustment correlates, particularly to school-and classroom-related wellbeing.

116
Chapter 5 classification partly helped to identify victims of bullying who had worse adjustment problems that are conceptually related to bullying.

Discussion
This research addressed the question of to what extent the widely used BVQ instrument for the assessment of bullying can be extended to make a more stringent differentiation between victims of bullying (according to its definition) and other types of peer aggression. This distinction is considered essential to evaluate anti-bullying interventions and for prevalence assessments of this unique group phenomenon (Jia & Mikami, 2018). Our research showed that less than half of the children who reported being victimized according to the BVQ also experienced all key characteristics of the bullying definition (frequency, intensity, power imbalance, and goal-directedness), despite being provided with a definition. More than half of the self-reported victims did not experience all characteristics of bullying, and, therefore, seemed to be victims of another type of peer aggression. Particularly power imbalance, and to a lesser extent goal-directedness, were experienced least often by children who were victimized according to the BVQ. Frequency ratings only do not seem to capture experiences that match the full definition of bullying.
Does adding explicit questions about these key characteristics help to diff erentiate victimization through bullying from victimization through general aggression (concurrent validity)? In support of earlier research focusing on one or some of the key characteristics of bullying (Green et al., 2013;Hunter et al., 2007;Malecki et al., 2015), children who mentioned the key characteristics had greater emotional, relational, and status adjustment problems that are conceptually related to victimization; not only as compared with non-victimized children but also as compared with children who experienced more general peer victimization. Overall, our fi ndings imply that measures that use the defi nition-fi rst approach, such as the BVQ , could gain more precision in identifying victims of bullying specifi cally according to their defi nition. This could be done by adding to the measure a relatively minor number of questions that explicitly assess experiences with each key characteristic.
Children who did not stick to the defi nition of bullying when answering the BVQ may have relied on their own defi nitions instead of adhering to the defi nition provided. Children's defi nitions often deviate from the formal conceptualizations of bullying (Vaillancourt et al., 2008), which can be problematic because their experiences require diff erent interventions.
Zooming in on the key characteristics, many children who reported victimization did not experience a power imbalance (29.0%) or goal-directedness (20.3%). This is in line with previous research that focused exclusively on power imbalance, and not goaldirectedness; suggesting that the BVQ was less sensitive to power imbalances (Green et al., 2013). Our results make sense because a higher frequency of victimization, the only characteristic explicitly assessed in the BVQ , was not or to a small extent related to power imbalance and goal-directedness. Previous research has shown that many children did not spontaneously mention power imbalance (74.0%) or goal-directedness ("intentionality"; 98.3%) as essential characteristics of bullying (Vaillancourt et al., 2008), which might explain why not all of them included it in their answers, despite being given a defi nition of bullying. Nevertheless, a power diff erential is acknowledged as a central characteristic of bullying because it embodies its social nature (Olweus, 1993;Volk et al., 2014). In our study, victims who reported a power imbalance also had poorer emotional, relational, and social status adjustment. Considering goal-directedness is relevant because this characteristic defi nes the strategic nature of bullying (Olthof et al., 2011;Veenstra et al., 2007;Volk et al., 2014) and was also associated with emotional maladjustment at school in our study.
Notably, the intensity associated with victimization did not predict any adjustment correlates. Perhaps the measure had too little variation: the vast majority of the children (87.6%) experienced intensity. Alternatively, the children's recollections of intenseness may not be valid, because they may have suppressed or positively reappraised the experiences to decrease their emotional impact (Gross, 1998). Indeed, victimization was associated with poorer adjustment regardless of whether children reported it as intense, and omitting the intensity item did not affect the results. That most children experienced victimization as intense is a sign that intense experiences are incorporated in children's definition of bullying.
Our findings highlight the relevance of the use of the definition to discriminate between victimization through bullying and through general aggression, by showing that experiences that were in line with the definition were not only associated with mental health (Hunter et al., 2007;Malecki et al., 2015), but also with social adjustment. Compared with victims who did not report all key characteristics of bullying, bullied children experienced greater emotional problems in general and related to school, had fewer friends and defenders, and were less popular and more rejected. Power imbalance played an important role in the associations with social adjustment; this characteristic was most strongly related to fewer friendships and lower popularity. Frequency was the only correlate of rejection which could, although tentatively, support the expectation that bullies aim to send a strong signal of dominance, but do so in a way that minimizes the costs to other aspects of their reputation and pick targets who are already rejected. Overall, the findings support conceptualizations of bullying as a unique social phenomenon that interferes with functioning in the peer group (Salmivalli, 2010).

Implications and Suggestions for Future Research
Our findings provide empirical support for suggestions (Bauman, 2016;Furlong et al., 2010;Jia & Mikami, 2018;Volk et al., 2014) that assessment of victimization through bullying may require an extension of current instruments, by explicitly addressing experiences with the key characteristics of the bullying definition.
More specificity in the assessment of bullying seems especially relevant for prevalence estimation and intervention evaluation.  (Ferguson, Miguel, Kilburn, & Sanchez, 2007). Therefore, tackling bullying calls for unique strategies. For example, the power imbalance of bullying calls for teaching children to defend themselves by fi nding support, and the social status goals that often characterize bullying might be addressed by providing bullies with alternative prosocial strategies to achieve these goals (Ellis, Volk, Gonzalez, & Embry, 2015).
Refi ning the assessment of victimization through bullying for such research goals may be simple: the addition of seven additional questions to the BVQ per bully achieved 56.9% more precision in the diff erentiation between victims of bullying and victims of general aggression. In developing the questions, we balanced accuracy with practical use. For example, asking for all possible types of power imbalance involved in bullying would lead to a lengthy questionnaire, while only asking a general question about "whether the other was more powerful" would be too abstract. We therefore focused on two types of power imbalance (in strength and popularity) that had been used previously (Felix et al., 2011;Malecki et al., 2015), had the highest loadings on physical and social factors distinguishing power imbalance (Nelson et al., 2019), and were often experienced by our participants. However, the additional questions should not replace the BVQ but should be used to complement it. We showed that the BVQ functioned as a "gatekeeper", by helping to identify victims with worse adjustment problems that are conceptually related to victimization.
More generally, our fi ndings highlight the importance for bullying researchers in general of diff erentiating clearly between bullying and more general aggression in their communication with other researchers and practitioners, and of explicating that bullying is a separate and more harmful form of aggression that has specifi c characteristics.
Future research with even larger sample sizes is needed to test (1) whether children's self-reported experiences of intensity can be associated with maladjustment correlates, (2) whether latent profi les can be distinguished based on experiences of key characteristics, and whether the fi ndings diff er (3) across age or grade groups, (4) between offl ine and online victimization (Corcoran, Guckin, & Prentice, 2015), and (5) depending on the number of bullies. Moreover, future research can examine the predictive validity of assessing each key characteristic.

Limitations and Strengths
Our results need to be interpreted with some limitations in mind. First, the participating children were in schools that participated in an anti-bullying program, and thus were familiar with the differences between bullying and other forms of aggression: this is part of the curriculum. Our findings might have overestimated the proportion of children who adhered to the bullying definition. Second, we used the victims' perspective to assess the definition's key characteristics, but victims might not always be the most reliable informants. They can misinterpret bullies' goals, adjust their recollections of victimization to suppress negative emotions (Gross, 1998), or underreport their problems because of social desirability. Although victims' perceptions determine the impact of victimization (Olweus, 2013), we recognize that a multi-informant perspective might also be valuable. Last, we could not examine the temporal order of associations between the key characteristics and adjustment. However, we were interested in correlates of victimization through bullying (concurrent validity), regardless of whether those are precursors or consequences of victimization.
This study was the first to assess to what extent the most commonly used method to assess victimization can discriminate between victimization through bullying and other types of peer aggression, in accordance with its definition. We relied on information from a large sample of 1,738 children, including 138 systematic victims of bullying; we focused on their experiences with specific bullies to prevent them from reporting the presence of different key characteristics across bullying occasions. Using these data, we showed that less than half of the children who reported victimization in the widely used BVQ also experienced all key characteristics of the bullying definition, and that victims who endorsed all key characteristics had poorer emotional, relational, and social status adjustment, which conceptually related to victimization through bullying as compared with victimization through general aggression. Providing children with a multi-component definition, and hoping that they will take this into account when responding to questions about how frequently this happened to them (ignoring the power imbalance, intensity, and goaldirectedness) may not be the most valid approach to differentiating victimization through bullying from victimization through general aggression. Figure A5.1 Flow Chart of Sample Selection.

Appendices
Note. Dotted squares represent samples used in the analyses.