Multidimensional Bullying Victimization Scale: Development and Validation

The prevalence of bullying has increased over the past decade and is a public health concern. Existing measures of bullying lack breadth and/or psychometric support, which may inaccurately represent current rates of bullying. The purpose of this study was to develop and validate a psychometrically sound measure, the Multidimensional Bullying Victimization Scale (MBVS), to evaluate bullying in adolescents. An initial pool of items was generated and rated by 600 adolescents. Exploratory factor analyses yielded three factors—direct bulling, indirect bullying, and evaluative bullying. Confirmatory factor analyses were conducted on independent sample of 623 adolescents, supporting the original three-factor solution. Results indicate that the MBVS demonstrates good internal consistency, construct validity, and concurrent validity. The MBVS scores were significantly and positively correlated with internalizing and externalizing symptoms. The MBVS provides a reliable and valid assessment of adolescents’ experiences with bully victimization.

Keywords: Bullying; bully victimization; adolescents; school; measurement

Approximately 60%-70% of youth witness bullying on a daily basis (Bradshaw, Sawyer, & O'Brennan, 2007; Rivers, Poteat, Noet, & Ashurst, 2009). Research demonstrates that victims of bullying are at increased risk for a range of mental health problems (Arseneault, Bowes, & Shakoor, 2010). Over the past decade, the prevalence of bullying has increased by approximately 25% (National Center for Education Statistics, 2013), with bullying increasing in late elementary school, peaking during middle school, and declining in high school (Sawyer, Bradshaw, & O'Brennan, 2008). Bullying has been regarded as an important public health concern with many schools implementing policies and interventions for reducing bullying; however, existing measures of bullying lack psychometric support or have other limitations. The aim of the current study was to develop and validate a comprehensive measure of bullying victimization in youth.

Bullying has been defined as a specific type of aggressive, interpersonal behavior that involves intent to cause harm, occurs repetitively, and involves an imbalance of power (Olweus, 1978, 1999, 2001). Bullying originally was thought to encompass only physical (e.g., pushing hitting, kicking; Ericson, 2001) or relational acts (e.g., verbal assaults, spreading rumors, social rejection, exclusion; Underwood, 2003). Thus, many of the available instruments assessing bullying largely ignore aspects of bullying we know to be relevant to youth today.

With increased attention to cultural sensitivity and the vast expansion of computer and Internet use, bullying has become multifaceted. Cultural bullying can take many forms including microaggressions (subtle, stereotypical, or insensitive behavior) and overt verbal and physical assaults, with approximately 33% of adolescents reporting experiencing biased-based/cultural bullying (Russell, Sinclair, Poteat, & Koenig, 2012). Additionally, cyberbullying is defined as "an aggressive, intentional act, carried out by a group or individual, using electronic forms of contact, repeatedly and over time against a victim who cannot easily defend themselves" (Thomas, Connor, & Scott, 2015, p. 141), with approximately 20% of high school students reporting monthly, and 5% reporting weekly cyberbullying (National Center for Education Statistics, 2013). Because virtual and electronic communication have become a major component of adolescent social life (Thomas et al., 2015), it is important that cyberbullying be represented in assessment measures.

Recently, researchers have applied Bronfenbrenner's (1977) ecological model of child development to the conceptualization of bullying and victimization. The model conceptualizes the child's environment as a multi-layered set of interrelated systems with varying levels of influence on child development. The closest level, the microsystem, includes demographic variables and individuals who are physically and emotionally nearest to the child (e.g., parents and teachers). Numerous studies have examined bully perpetration and victimization at the microsystem level. For example, children who are overweight or disabled often experience more bullying (Espelage, 2014). Likewise, parental monitoring is associated with decreased bully perpetration and victimization (Li, Fiegelman, & Stanton, 2000) whereas exposure to family conflict is associated with increased bullying (Espelage, Low, & De La Rue, 2012).

The exosystem, the next level, includes a wider ecological net consisting of entities such as politics and neighborhoods. Communities in which violence is frequently modeled or condoned often have increased bullying and victimization, although, the directionality of the relationship is unclear (Swearer & Hymel, 2015). Next is the child's mesosystem, which consists of interactions amongst the microsystems (e.g., parents, peers, and school personnel; Bronfenbrenner, 1977). For example, teacher support of children's need to be connected with their peers has been found to be an effective means for reducing bullying (Lam, Li, Chan, Wong, & Zhang, 2015).

The next level, macrosystem, consists of indirect influences on child development such as the social structures in the child's immediate environment. Cultural and political, or religious ideologies often influence federal and state laws, as well as school policies, which in turn, impacts children's adjustment (Espelage, 2014). With regard to bullying, decreased bullying and fighting and increased peer intervention is associated with teachers who feel supported by their administration in addressing bullying (Espelage, Polanin, & Low, 2014).

The chronosystem, the outermost level of an individual's social-ecology, includes change over time within the child and their environment (Bronfenbrenner, 1977; Espelage, 2014). This can include familial changes such as divorce and historical events such as war or economic recession. For example, changes in life events (e.g., divorce) could result in psychological changes within the individual (microsystem) and changes in parental involvement/monitoring (macrosystem), which may result in increased peer aggression (Breivik & Olweus, 2006).

The literature consistently demonstrates that being bullied is strongly associated with internalizing symptoms such as depression and anxiety (Benedict, Vivier, & Gjelsvik, 2015; Paul, Smith, & Blumberg, 2013; Waasdorp & Bradshaw, 2015). Victims of bullying often report feeling lonely and pessimistic about their social relationships (Jackson & Cohen, 2012) and having a low self-esteem (Nansel et al., 2001; Waasdorp & Bradshaw, 2015). Being the target of bullying is associated with negative, ruminative thoughts (Nansel et al., 2001), often about physical or social threats and personal failure (Hunt, Peters, & Rapee, 2012), as well as suicidal ideation and behavior (Klomek, Marrocco, Kleinman, Schonfeld, & Gould, 2007).

Although research often focuses on the externalizing symptoms exhibited by bullies, being a victim also is associated with increased externalizing symptoms compared to peers who are not bullied (Kelly et al., 2015). Experiencing bullying is related to increased hostile thoughts (Davidson & Demaray, 2007; Hunt et al., 2012), fighting (Nansel et al., 2001), aggression, and anger (Davidson & Demaray, 2007).

Victimized students often experience increased fear and stress while at school, affecting their ability to learn. This may result in school avoidance, increased absenteeism (Beran & Li, 2007), and difficulty concentrating in class (Buhs & Ladd, 2001). The behavior frequently leads to poorer academic performance (Beran & Li, 2007; Katzer, Fetchenhauer, & Belschack, 2009) and lower academic achievement (Paul et al., 2013).

Bullying primarily has been measured using self-report questionnaires. However, existing measures often lack comprehensive, culturally sensitive items, and psychometric support. The Peer Relations Questionnaire (PRQ; Rigby & Slee, 1993), for example, is a 20-item self-report measure with three subscales: bully, victim, and prosocial behavior. Although the measure assesses a range of bullying and prosocial behavior, the instrument lacks comprehensiveness (i.e., does not include cyberbullying or cultural bullying), and demonstrates minimally adequate psychometrics, with internal consistencies just exceeding.7 for each scale (Rigby & Slee, 1993).

The Olweus Bully/Victim Questionnaire (OBVQ; Olweus, 1996) is a 40-item, self-report measure for students in Grades 3-12. The measure defines bullying and instructs respondents to rate items in terms of frequency (Olweus, 1996). The measure assesses various forms of bullying and classifies respondents as either a nonbully/nonvictim, victim, bully, or bully/victim. Group classification is based on cut points; however, there is minimal support for the use of these cut points as determinants of group classification (Hunt et al., 2012). The OBVQ assesses varying types of bullying; however, the measure does not include cyberbullying (Hunt et al., 2012).

The Personal Experiences Checklist (PECK; Hunt et al., 2012) is a 32-item self-report measure of bullying for children 8 and older. The PECK includes items pertaining to relational, physical, cultural, and technology-based bullying. However, the measure was created and validated with Australian school children (ages 8-16) who were primarily Anglo-Saxon (83.5%; Hunt et al., 2012). Although the PECK provides comprehensive coverage of multiple factors of bullying, many of the cyberbullying items are outdated and do not reflect current social media usage. Further, the generalizability of the measure to diverse American students is unknown.

The negative impact of bullying is well established yet measures of bullying are limited. Thus, the current study developed and validated the Multidimensional Bully Victimization Scale (MBVS). The aim of the study was to develop a psychometrically sound, comprehensive measure of bullying for use with adolescents. It was hypothesized that the MBVS would retain items producing four major dimensions of bullying: relational, physical, cultural, and cyberbullying. The scores on each factor were expected to be positively associated with internalizing behavior as assessed by the Youth Internalizing Problems Screener (YIPS; Renshaw & Cook, 2016a) and with externalizing behavior as assessed by the Youth Externalizing Problems Screener (YEPS; Renshaw & Cook, 2016b). The scores on each factor and the composite score were hypothesized to be negatively associated with academic performance. As a measure of content validity, the factor scores of the MBVS were expected to be positively associated with factors scores on the Personal Experiences Checklist (PECK; Hunt et al., 2012).

The participants were 600 youth (54.5% female) recruited from Louisiana schools, youth groups, after-school care programs, and health clinics (see Figure 1). The participants were between 11 and 18 (M = 15.16, SD = 1.72). The students were from Grades 5-8 with the vast majority in high school (16.2% in 9th grade; 19.2% in 10th grade; 31.0% in 11th grade, and 15.0% in 12th grade). The overall sample was predominantly Caucasian (80% Caucasian, 5.3% Black/African American, 3.8% Asian/Pacific Islander, 2.8% Biracial/Multiracial, 2.0% Hispanic/Latino, 1.4% other, and 4.6% declined to answer). The participants primarily attended private schools (84.9%).

PHOTO (COLOR): Figure 1. Stages and samples used in each stage of development of the measure.Note. MBVS = Multidimensional Bullying Victimization scale.

We wrote 74 items based on a review of the literature, existing measures, and the written responses to open-ended prompts from 18 students ages 11-19 about their experiences witnessing or as the victim of bullying. The item pool provided comprehensive coverage of bullying behavior (i.e., 15 items assessing physical bullying, 17 items assessing relational bullying, 23 assessing cyberbullying, and 19 assessing cultural bullying).

After obtaining parent consent and adolescent assent, participants completed a demographic form and the MBVS pilot measure. Participants completed the questionnaires in the classroom at the time requested by the school. Students were informed of the confidentiality of their responses and the consent form was separated from the questionnaires. The researchers were available to answer questions.

Participants completed a short demographic questionnaire that included age, grade, gender, race, school type, parent's marital status, and academic performance.

The 74-item MBVS pilot measure instructed respondents to rate how often they experienced each item on a 4-point scale (never, sometimes, often, very often).

Initial item analyses included an examination of item frequencies, item means, and inter-item correlations. Items that were infrequently endorsed (less than 15% of the time) or that had extreme item means (i.e., do not approach the median value for responses) were considered for initial elimination (DeVellis, 2003). This resulted in 39 items being eliminated due to low frequency. No additional items were eliminated due to low or high means. In addition, items with high interitem correlations (.75 or higher) were examined to determine whether one item in the pair could be eliminated. One item met this criterion and was eliminated. The final item pool consisted of 34 items.

Using the 34 remaining items, principle axis factoring (PAF) exploratory factor analyses were conducted to determine whether underlying dimensions were evident from the data. Analyses were conducted using a direct oblimin oblique rotation. Comrey and Lee (1992) suggested multiple criteria for determining factor solutions, including factor loadings of.40 or greater, eigenvalues of 1.0 or higher, and simple structure (i.e., items load strongly on one factor only). Additionally, results from a Monte Carlo PCA for parallel analysis suggested use of an eigenvalue of 1.40 as the factor cutoff.

The unbound factor analysis suggested a seven-factor solution, as it accounted for the most that variance. However, additional solutions were forced in order to evaluate one-, two-, three-, four-, five-, and six-factor solutions. The one-factor solution was examined to determine whether the items might be best understood as a single variable of bullying. The three-factor solution was examined, as the scree plot of the original unbound factor analysis and the corresponding eigenvalues were suggestive of this solution. Additionally, a four-factor solution was examined to explore the concept of bullying as a four dimensional structure (i.e., physical, relational, cyber, and cultural). The two-, five-, and six-factor solutions were examined to ensure a comprehensive examination of all solutions leading up to the unbound seven-factor solution suggested.

The seven-factor solution was not chosen, as the eigenvalues did not uphold that suggested by the parallel analysis. Additionally, the pattern of factor loadings was not ideal, with several factors having only few items, and three factors carrying the majority of the item factor loadings. The additional factor solutions (one, two, four, five, and six factors) were not chosen due to poor theoretical coherence, eigenvalues below 1.40, and poor factor loadings. The three-factor solution resulted in the most interpretable factor structure, considering scree plots, eigenvalues, variance accounted for, and interpretability. Additionally, results from the parallel analysis suggested a three-factor solution.

Further item analysis was conducted to determine whether additional items should be eliminated. As a result, six items were eliminated due to poor factor loading (less than.40) and two were eliminated due to loading on more than one factor. Items that resulted in increased reliability (i.e., Cronbach's alpha) when the item is deleted, or items with low total-item correlations (below.20; Floyd & Widaman, 1995) were considered for elimination; however, no items met these criteria and thus no additional items were eliminated.

A factor analysis was conducted using the remaining 26 items. The factor loadings are seen in in Table 1. As seen in Table 1, Factor 1, Direct Bullying, consisted of 11 items representing bullying acts that were direct, and delivered in a face-to-face manner. Factor 2, Indirect Bullying, consisted of eight items that represent experiencing bullying acts that occurred via other people or other mediums (i.e., Internet). Factor 3, labeled Evaluative Bullying, consisted of seven items that assessed bullying that is judgmental or evaluated a person's traits or attributes. Eigenvalues ranged from 9.56-1.44, with the three factors accounting for 51.23% of the variance (36.78%, 8.93%, and 5.52%). The correlation between the three factors ranged from.34-.57.

Exploratory factor analysis pattern of the 26-item MBVS (direct oblimin).

Using Cronbach's alpha coefficient, inter-item reliability estimates were obtained. All three factors demonstrated good internal consistency (α = .89 for Direct Bullying, α = .85 for Indirect Bullying, α = .82 for Evaluative Bullying). The composite, which combined each of the three factors, also demonstrated strong internal consistency (α = .93).

The participants were 652 adolescents (50.9% female) between 11 and 18 years (M = 15.50, SD = 1.41). The sample were from Grades 6-12 with the vast majority being in high school (18.5% in 9th grade; 18.9% in 10th grade; 37.9% in 11th grade; and 15.2% in 12th grade). The sample was predominantly Caucasian (83.3% Caucasian, 6.0% Black/African American, 2.9% Biracial/Multiracial, 4.3% other ethnicities, and 3.6% were missing this information) who predominantly attended private schools (95.3% private schools) in Louisiana.

Participants completed the demographic questionnaire used in Study 1, the 26-item MBVS, the YIP, the YEP, and the PECK. Participants completed the questionnaires in the classroom at a time designated by the administration. The researcher was available to answer questions. The students' consent and assent forms were separated from the questionnaires to ensure confidentiality.

The PECK is a 32-item instrument used to measure youths' personal experiences being bullied (Hunt et al., 2012). The measure yields four factors: Physical, relational-verbal, cyberbullying, and bullying based on culture. Items are rated on a 5-point scale ranging from never to every day, and the total score range is 0-96. The score ranges of the individual subscales were as followed: physical, (0-27), relational-verbal (0-33), cyberbullying (0-24), and bullying based on culture (0-12). The PECK demonstrates adequate internal consistency and adequate test-retest reliability (r = .61-.86; Hunt et al., 2012). This measure was found to demonstrate adequate evidence of construct validity as compared to previously established measures of bullying, particularly for the relational, verbal, and physical bullying domains. Validity estimates for this measure, however, are limited to primarily Caucasian Australian students (Hunt et al., 2012).

The YIPS is a 10-item self-report behavior rating instrument used to screen youth for internalizing problems (Renshaw & Cook, 2016a), including items related to anxiety and depression. This measure yields two classifications of youth: typical and at-risk (>1.5 SD above M). Items are rated on a 4-point scale ranging from almost never to almost always, and the total score range is 10-40. The YIPS has good internal consistency (α = .88) and concurrent validity (Renshaw & Cook, 2016a).

The YEPS is a 10-item self-report behavior rating instrument used to screen for externalizing problems (Renshaw & Cook, 2016b), including items related to hyperactivity/impulsivity and conduct problems. This measure yields two classifications of youth: typical and at-risk (>1.5 SD above M). Items are rated on a 4-point scale ranging from almost never to almost always, and the total score range is 10-40. The YEPS demonstrates adequate internal consistency (α = .77) and concurrent validity (Renshaw & Cook, 2016b).

Data were screened for normality, homoscedasticity, and linearity. Multiple participants were multivariate outliers, as their Malhalanobis distance scores were significant (p < .01). However, the participants were retained, as it is expected that some adolescents experience bullying on a frequent basis. Screening for skew and kurtosis showed multiple items that were significantly skewed and kurtotic; however, corrections for skew and kurtosis were not conducted, as bullying behavior is not expected to be normally distributed.

Using Amos version 22, a confirmatory factor analysis (CFA) was conducted to determine whether the three factors from Study 1 were supported. Considering these findings along with Roth's (1994) recommendation for conducting CFA using AMOS with missing data, the listwise deletion method was chosen, as there was less than 5% missing data. Using listwise deletion 29 participants were removed, resulting in 623 total participants examined within this CFA. When estimating adequate sample size, it is commonly accepted in the psychometric literature to use a conservative case to parameter ratio of 10:1 (Bentler & Chou, 1987; Jackson, 2001; Nevitt & Hancock, 2004). The ratio between cases and parameter estimates in in the present study was roughly 24:1 and therefore deemed an appropriate sample size to adequately reduce the risk of Type I error.

To determine the reliability of the factor structure derived from the exploratory factor analysis conducted in Study 1, a confirmatory factor analysis (CFA) using the 26-item MBVS-pilot measure was conducted. Table 2 presents the fit indices associated with the three-factor model tested, specifically the model chi square, Tucker-Lewis Index, the comparative fit index (Bentler, 1990), and the root-mean-square error of approximation (RMSEA; Browne & Cudeck, 1993). As seen in Table 2, the Tucker-Lewis Index (TLI) and comparative fit index (CFI) values between.90-.95 and RMSEA values (with 90% confidence interval) between.05 and.08 indicated adequate data-model fit (Kenny, 2014). Additionally, factor loadings, λ > .50 were considered strong, as they accounted for over 25% of variance. Thus, items with factor loadings below.50 were considered for elimination. Latent construct reliability was considered desirable if H ≥ .70, indicated a strong intrafactor correlation (Mueller & Hancock, 2008).

Confirmatory factor analysis of the MBVS.

The CFA Model 1 tested the 26-item MBVS three-factor structure. Using the aforementioned model validity standards, Model 1 did not yield an adequate data-model fit (Χ2 = 1693.17, df = 296, p < .001, CFI = .818, TLI = .800, RMSEA [90% CI] = .087 [.083,.091]). Model 1 was characterized by a wide range of factor loadings for each construct (λ range = .42-.81, p < .001), and adequate maximal reliability for all factors (H range = .86-.90).

Due to low factor loading on Model 1, two items were considered for elimination in CFA Model 2. One item ("Screenshot Snapchats that I send and post on the Internet") was eliminated due to poor factor loading and the resulting fit indices were examined. Model 2 yielded a marginally stronger, yet still inadequate data-model fit (Χ2 = 1471.60, df = 272, p < .001, CFI = .837, TLI = .820, RMSEA [90% CI] = .084 [.080,.088]). Model 2 was characterized by a wide range of factor loadings for each construct (λ range = .40-.81, p < .001), and adequate maximal reliability for all factors (H range = .86-.90).

Due to low factor loading on Model 2, one item ("Post embarrassing videos of me on Snapchat, YouTube, Facebook) was eliminated in Model 3. Model 3 yielded a marginally stronger, yet still inadequate data-model fit (Χ2 = 1399.80, df = 249, p < .001, CFI = .841, TLI = .823, RMSEA [90% CI] = .086 [.082,.091]). Model 3 was characterized by robust factor loadings for each construct (λ range = .52-81, p < .001), and adequate maximal reliability for all factors (H range = .86-.90).

Due to high interitem correlation, resulting in high modification indices, one set of items was correlated within CFA Model 4. "Make fun of my weight" and "Make fun of my size" were correlated in this model, due to similarity in content. Model 4 yield a marginally stronger, yet still inadequate data-model fit (Χ2 = 1281.20, df = 248, p < .001, CFI = .857, TLI = .841, RMSEA [90% CI] = .082 [.077,.086]). Model 4 was characterized by robust factor loadings for each construct (λ range = .51-82, p < .001), and adequate maximal reliability for all factors (H range = .86-.90).

Due to additional high interitem correlations, as per the modification indices, four additional pairs were correlated in CFA Model 5, due to similar in content ("Make fun of me" with "Tease me," "Punch or hit me" with "Bump into me on purpose," "Spread rumors about me in text message" with "Spread rumors about me," and "Make fun of me for being smart" with "Make fun of me for my grades"). Model 5 yield an adequate data-model fit (Χ2 = 985.22, df = 244, p < .001, CFI = .912, TLI = .900, RMSEA [90% CI] = .070 [.065,.074]). Model 5 was characterized by robust factor loadings for each construct (λ range = .50-84, p < .001), and adequate maximal reliability for all factors (H range = .86-.90). Given that Model 5 was more psychometrically and statistically sound and theoretically coherent, it was selected as the preferred measurement structure for the MBVS (see Figure 2). The final 24-item version of the MBVS includes 11 items within the Direct Bullying scale, 6 items within the Indirect Bullying scale, and 7 items within the Evaluative Bullying scale.

PHOTO (COLOR): Figure 2. Preferred confirmatory factor analysis measurement model for the MBVS.Note. MBVS = Multidimensional Bullying Victimization Scale; H = latent construct reliability coefficient.*Standardized factor loadings (λ) significant at the p < .001 level.

Further analysis of Model 5, the preferred measurement model, included reliability estimates by conducting internal consistency analyses using Cronbach's alpha and coefficient omega to assess reliability for each factor and the composite scale. In the psychometric literature, omega is recommended as an acceptable alternative to Cronbach's alpha, due to concerns about the limitations and inappropriate use Cronbach's alpha (Cronbach & Shavelson, 2004; Dunn, Baguley, & Brunsden, 2014; Kelley & Pornprasertmanit, 2016; Sijtsma, K., 2009, Yang & Green, 2011). All three factors demonstrated good internal consistency estimates with alphas ranging from.84-.89 (α = .89 for Direct Bullying, α = .85 for Indirect Bullying, α = .84 for Evaluative Bullying) and omegas ranging from.88-.94 (omega [95% CI] = .939 [.934,.946] for Direct Bullying, omega [95% CI] = .881 [.879,.881] for Indirect Bullying and omega [95% CI] = .886 [.884,.886] for Evaluative Bullying). The composite, which combines each of the three factors, demonstrated strong internal consistency (α = .93; omega [95% CI] = .928 [.928,.929]).

Construct validity was assessed through correlation analyses between the MBVS factors and composite and the factors of the PECK (Hunt et al., 2012; Physical, Relational-Verbal, Cyberbullying, and Bullying Based on Culture), academic performance, and composite scores of the YIPS and YEPS. A Bonferroni correction was applied with a conservative p-value of.003 due to the number of correlations being conducted. Descriptive statistics of validity measures are presented in Table 3. Construct validity information is presented in Tables 4 and 5.

Descriptive statistics of YIPS, YEPS, PECK, and academic performance.

Validity correlations of MBVS, YIPS, YEPS, and academic performance.

Validity correlations of MBVS and PECK.

It was hypothesized that MBVS factors and composite scores would be positively correlated to the YIPS and the YEPS. The hypothesis was supported. The MBVS Total, MBVS Direct Bullying, MBVS Indirect Bullying, and MBVS Evaluative Bullying were all positively related to the YIPS (see Table 3). Similarly, the MBVS Total, MBVS Direct Bullying, MBVS Indirect Bullying, and MBVS Evaluative Bullying were all positively related to the YEPS (see Table 4).

The hypothesis, that the MBVS factors would be positively correlated with PECK factors could not be fully examined, as the MBVS factor structure produced a three-factor model, rather than the four-factor model as predicted. Thus, the relationships between the resulting three MBVS factors, the MBVS composite, the PECK factors, and the PECK composite were examined. Table 5 presents the correlations amongst the scores. As seen in Table 5, all MBVS scale scores and the composite were positively related to the PECK total score, Relational-Verbal Bullying subscale, Physical Bullying subscale, Cyberbullying subscale and the Bullying Based on Culture subscale.

The hypothesis that the MBVS factor and composite scores would be negatively related to academic performance was not supported, as the correlations were not statistically significant when using a Bonferroni correction and significance level of.003.

The purpose of the current study was to develop a multidimensional measure of bully victimization using a sample of adolescent youth. Bullying was defined as a specific type of aggressive, interpersonal behavior that involves intent to cause harm, occurs repetitively, and involves an imbalance of power (Olweus, 1978, 1999, 2001). The results of the study produced the MBVS. Further the study provided validity data on the association of MBVS scores with an existing measure of bullying and measures of internalizing and externalizing symptoms.

Study 1 was designed to generate a list of potential items for inclusion in a pilot measure of bully victimization. Items were generated by reviewing relevant literature and existing measures, written student responses to open-ended prompts, as well as from reviewing previous measures and bullying literature. A large sample of adolescent participants (N = 600) rated the frequency in which they experienced each item on a 4-point scale (i.e., never, sometimes, often, very often). After eliminating low frequency and highly correlated items, exploratory factor analyses were conducted and yielded a 26-item scale with three factors, direct bullying, indirect, and evaluative bullying. Factor 1, Direct Bullying, consisted of 11 items that assessed experiencing bullying in a personal, direct, and face-to-face manner. Factor 2, Indirect Bullying, consisted of eight items that assess experiences of bullying through other people or through other mediums (e.g., Internet). Factor 3, Evaluative Bullying, consisted of seven items assessing experiences of bullying that are judgmental or negatively evaluating a person's traits or attributes.

Confirmatory factor analysis, conducted in Study 2, supported the nonnormal distribution of the pilot items, factors, and the composite obtained in the initial factor analysis. Additionally, the three-factor latent structure and construct reliability of the MBVS were confirmed, with some refinement of item inclusion. Two items were eliminated, as they did not demonstrate strong factor loadings. Thus, the final measure consists of 24 items measuring the three factors, Indirect Bullying, Direct Bullying, and Evaluative Bullying. Two items from the pilot version of the MBVS were removed from the Indirect Bullying factor, resulting in a six-item factor. The Direct Bullying and Evaluative Bullying factor items remained consistent with Study 1 item retention. The final measure of the MBVS consists of 24 total items.

The three factors and composite of the MBVS demonstrated good internal consistency with all factors having moderate-to-strong correlations with one another and the composite score. The composite score, therefore, may be used as a general index of overall bully victimization, or the three factors may be used to assess experiences of specific types of bullying. Additionally, the readability analysis suggested that the measure is appropriate for youth who read at the third-grade level or higher.

An important purpose of the study was to examine the construct and concurrent validity of the MBVS. Construct validity was assessed in several ways. The MBVS factors and composite scores were strongly correlated, suggesting that the factors are significantly related to the broader construct of bully victimization. Additionally, the three factors were moderately correlated, suggesting that each factor measures a unique subset of victimization experiences. These findings are consistent with previous research that indicates that bullying behavior is often correlated. Specifically, multiple studies have found that one third of cyberbullying victims also are victims of traditional bullying (Erdur-Baker, 2010; Li, 2005).

Internalizing (Benedict et al., 2015; Hunt et al., 2012; Nansel et al., 2001; Paul et al., 2013; Waasdorp & Bradshaw, 2015) and externalizing symptoms (Bradshaw, Waasdorp, & Johnson, 2015; Davidson & Demaray, 2007; Hunt et al., 2012), and poor academic performance (Beran & Li, 2007; Katzer et al., 2009) are consistently found to be strongly correlated with experiences of bully victimization. Thus, it was hypothesized that higher scores on the MBVS factors would be associated with higher scores on the YIPS, YEPS, and lower scores in academic performance. This hypothesis was partially supported. Similar to previous research, the MBVS factors and composite scores were positively associated with internalizing (i.e., YIPS; Renshaw & Cook, 2016a) and externalizing (i.e., YEPS; Renshaw & Cook, 2016b) symptoms. However, academic performance was not associated with any factor or with the composite score. It is possible that the MBVS factors and composite scores were not related to academic performance as expected, due to the homogeneity of the sample with most participants reporting above-average grades.

As hypothesized the MBVS factors and composite scores were positively related to the PECK scores, with higher scores on one resulting in higher scores on the other. This hypothesis was fully supported, with moderate-to-strong associations between all MBVS and PECK factors and composites. The strong correlations between this previously established measure of bullying and the MBVS demonstrate the presence of convergent validity, suggesting that the MBVS is a valid measure of bully victimization.

Taken together, the results of this study provide initial evidence that the MBVS is a reliable, stable, and structurally valid assessment measure of adolescent bully victimization. While only found to be moderately psychometrically sound, the hypothesis that the MBVS would demonstrate a promising, multidimensional structure was supported. This suggests that the MBVS is a technically adequate instrument for use in adolescents as an assessment of bullying victimization. As adolescents demonstrate a unique social hierarchical pattern compared to children (Cairns & Cairns, 1991; Krappman, 1999; Schafer, Korn, Brodbeck, Wolke, & Schulz, 2005), the measure was developed specifically for use with adolescents ages 11-18. In addition, efforts were made to ensure that items were developmentally appropriate for youth in wording and content. Finally, the reading level was appropriate for all ages of which this measure is postured to assess, providing further evidence that the MBVS is appropriate for youth samples. Because of its multidimensional nature, the MBVS is useful for assessing bully victimization and its specific dimensions.

While many different types of schools were invited to participate in this study, the sample consisted primarily of students from private school settings (85% in Study 1 and 95% in Study 2). Therefore, the homogeneity of the current sample may have affected the frequency and variability of items endorsed. Further validation studies of the MBVS including a more heterogeneous sample of students should be conducted. Additionally, future research may benefit from exploring the differences in types of bullying victimization experienced in various school settings, especially public school. This is a notable limitation of the present study given that research has found children who attend public schools endorse more bullying, victimization, and fighting than those attending private schools (Shujja, Atta, & Jawwad Muhammad, 2014). Presently, there is no empirical evidence supporting the generalization of bullying research in private school settings to public school setting and vice versa. Further validation studies of the MBVS are warranted to improve its generalizability and broaden the scope of its utility.

The lack of racial/ethnic diversity in the participants may have resulted in low levels of endorsement of items related to culture-based bullying, which may have ultimately resulted in the lack of support for this hypothesized factor. Additionally, with ethnicity and bullying involvement being dependent on the ethnic composition of the classroom (Vervoort, Scholte, & Overbeek, 2010), the lack of diversity within this study may have affected the endorsement of victimization experiences in general, which may affect the generalizability of the results. Future studies aiming to refine the development of the MBVS should seek to obtain reliability and validity data using more racially, ethnically, and culturally diverse samples. Such studies may benefit from the inclusion of items that were deleted due to low frequency, as the current sample was predominantly Caucasian and it is possible that these items would be rated more frequently by a more diverse sample.

Additionally, although the sample collected included participants with ages ranging from 11-18 and Grades 5-12, this study's sample was heavily loaded in the high school years. This may affect generalizability of the results to younger adolescents, as some research suggests higher rates of bullying in middle school compared to grades 9 through 12 (Nansel et al., 2001). Future research would benefit from the inclusion of a more balanced sample of participants within both age and grade levels.

Clinical implications of the MBVS include its usability within schools and mental health clinics. Administration of the MBVS to adolescents, within the school context, may provide school officials with important information about the presence of bullying within their student population. Further, the type of bullying being experienced would be available, thereby facilitating a more nuanced, school-wide bullying intervention, tailored specifically to the school's demonstrated needs.

In conclusion, it appears that the MBVS is a promising assessment of adolescent's experiences with bullying victimization within the school setting. The MBVS is a moderately psychometrically sound measure with an adequate model fit, demonstrating evidence of strong internal consistency, and initial evidence of construct and concurrent validity. This measure will be useful for when there is a need for a comprehensive understanding of bullying behaviors. The MBVS overcomes problems associated with outdated terminology, subjective thresholds, and single-item assessments of bullying. Subsequently, the MBVS has the potential to inform practice within schools, clinics, and theory within bullying research.