Correspondence concerning this article should be addressed to David A. Cole, Department of Psychology and Human Development, Vanderbilt University, Nashville, TN 37203. Contact: ude.tlibrednav@eloc.divad, phone: 615-343-8712.
The publisher's final edited version of this article is available at Psychol AssessColleges and universities are increasingly concerned about respect for diversity and tolerance of individual differences on their campuses. Nevertheless, no comprehensive measure of peer victimization has been developed and validated for use with college student populations. The Peer Victimization in College Survey (PVIC) is the first such measure. Study 1 (N = 733) reports how PVIC items were empirically derived to ensure construct coverage. Study 2 (N = 100) reports how intuitive PVIC subscales were established to distinguish between subtypes of college peer victimization. Study 3 (N = 520) provides evidence of convergent, discriminant, and construct validity for the PVIC, including its relations to risk factors and to outcomes such as depressive symptoms, anxiety, stress, and college sense of belonging. Study 4 (N = 633) validates several PVIC scaling methods and provides evidence of incremental validity of the measure over current (unvalidated) measures. The PVIC can assess subtypes of peer victimization on college campuses, evaluate the effectiveness of campus intervention efforts, and test hypotheses about the causes and effects of peer victimization.
Keywords: peer victimization, bullying, depression, anxiety, stress, college, sense of belongingThis study introduces the first validated measure of peer-on-peer victimization among college students, the Peer Victimization in College (PVIC) Survey. Derived from college students’ own personal experiences, the study identifies 10 broad types of peer victimization that occur on college campuses. The PVIC can be used both by researchers who study bullying and by college officials who want to understand peer victimization or the effectiveness of interventions on their campuses.
Peer victimization has received increasing and widespread attention, largely due to its potentially severe consequences (Chen & Huang, 2015; Rospenda et al., 2013; Schenk & Fremouw, 2012). Surprisingly, however, a validated and comprehensive measure of peer victimization does not currently exist for use with college students. As the bulk of research on peer victimization has focused on children and adolescents (Hawker & Boulton, 2000; Reijntjes, Kamphuis, Prinzie, & Telch, 2010), most researchers studying peer victimization in college have had to use adolescent peer victimization measures. Such measures are not optimal as we know that both type and frequency of peer victimization behaviors change across age and setting (e.g., Solberg & Olweus, 2003). In the current paper, we report four studies on the construction and validation of the first comprehensive measure of college student peer victimization. In this effort, we defined peer victimization broadly to include being the recipient of any kind of hurtful behavior perpetrated by a fellow student. Such behaviors could be malicious or unintentional, reciprocated or unilateral, verbal or nonverbal, overt or covert, physical or symbolic. We did not require the existence of a power differential between perpetrator and victim (Olweus, 1993). We did require, however, that both parties be students. Victimization by faculty, staff, administrators, or non-college peers were not considered.
Although college peer victimization has been the focus of numerous studies (e.g., Felix et al., 2018; Lento, 2006; Ramsey, DiLalla, & McCrary, 2016), no measure of peer victimization has ever been developed and validated specifically for use with college students. Peer victimization research on college students has relied on unpublished measures, scales developed for use with younger populations, workplace victimization measures adapted for use with college students, and ad hoc questionnaires that have not been empirically validated (e.g., Aquino & Bradfield, 2000; Coyne, Padilla‐Walker, & Holmgren, 2018; Dogruer & Yaratan, 2014; Morales & Crick, 1998; Yu, Branje, Meeus, Koot, Lier & Fazel, 2018). To remedy this, we conducted four studies on the construction and validation of the first comprehensive measure of college student peer victimization, the Peer Victimization in College Survey (PVIC). In this effort, we (1) derived items empirically from actual accounts of peer victimization reported by students across hundreds of colleges and universities, (2) developed subscales representing major types of and reasons for such behaviors, and (3) present evidence of their convergent, discriminant, construct, and incremental validity.
At least three arguments support the need for a valid and comprehensive measure of college peer victimization. First, peer victimization does not disappear upon graduation from high school. Prevalence rate estimates of peer victimization subtypes range from 8 to 25% each; however, no study has assessed the prevalence of all types taken together (verbal, social, physical: Chapell et al., 2004; sexual assault: Fedina, Holmes, & Backes, 2018; cyber- and noncyber victimization: Lund & Ross 2017; online peer victimization: Schenk & Fremouw, 2012). Second, peer victimization in college is different than peer victimization in other settings and at other age levels. From childhood to adolescence to early adulthood, rates of physical aggression drop, and reports of relational and cyber-victimization rise (Scheithauer et al., 2006; Solberg & Olweus, 2003; Varjas, Henrich, & Meyers, 2009). Between high school and college, the number of studies that focus on hazing-related peer victimization increase 10-fold, and the number of studies on sexual assault-related peer victimization increase 18-fold. Peer victimization may even become more insidious, more pernicious, and more institutionalized in college (Franklin, 2008). After college, workplace-related victimization behaviors emerge that are somewhat unusual among peers in college (e.g., unjust evaluations and unnecessary monitoring; Nielsen, Hetland, Matthiesen, & Einarsen, 2012; Ortega et al., 2009). Third, preliminary evidence suggests that the consequences of peer victimization are every bit as severe as they are in younger and older groups: low self-esteem, depressive symptoms, stress and anxiety, and reduced sense of school belonging (Chapell, et.al., 2004; Chen & Huang, 2015; Dilmac, 2009, Kowalski et al., 2012; Kwan et al., 2017; Rospenda et al., 2013; Schenk & Fremouw, 2012; Storch et al., 2004; Tennant et al., 2015).
Our PVIC is unique in that it involved a multi-method approach to item generation. We derived items from (1) first-hand accounts by a diverse sample of college students about their peer victimization experiences, (2) the reports of professionals who work with college victims of peer victimization, and (3) previously developed scales. Such multi-method approaches to scale construction have been used before with considerable success (e.g., Hathaway & McKinley, 1940; Linehan, Goodstein, Nielsen, & Chiles, 1983). This approach generated examples (and items) spanning many domains, including some not always represented in prior measures of peer victimization (e.g., micro-aggressions, hazing, academic victimization, and sexual harassment/assault). In order to comprehend the full array of victimization experiences, we elected to include all such behaviors, despite the fact that some did not always fit particular definitions of peer victimization or bullying.
We also elected to separate the assessment of peer victimization from the assessment of the reasons for the victimization. By treating these as two different parts of the PVIC, we were able to assess broad reasons for victimization without asking multiple questions about each event, which would have made the survey extremely long and unwieldy. For example, an event like “A student called me a mean name” would have required dozens of follow-up questions: “A student called me a mean name because of my race; …because of my sexual orientation; …because of my gender or religion or nationality or accent or clothing or appearance, etc.” Instead, we constructed a second half of the PVIC to assess students’ perceptions of the reasons for any and all victimization experiences they may have had. We recognize that this different than at least one peer victimization survey that created race-related victimization as a separate subscale, somewhat conflating the peer victimization behavior with the presumed reason for the behavior (Desjardins, Yeung Thompson, Sukhawathanakul, Leadbeater, & MacDonald, 2013).
In addition to the construction of the PVIC, we had two other major goals: validation of the measure and the identification of face-valid subscales. Validation involved providing evidence of convergent, discriminant, construct, and incremental validity (Camplbell & Fiske, 1959; Cronbach & Meehl, 1955). As evidence of convergent validity, we expected significant correlations of the PVIC with other measures that have been used with college populations, although they were developed for use with high school or workplace populations. As evidence of discriminant validity, we expected small correlations of the PVIC with a measure of intentional dissembling or lying, reflecting the resistance of the PVIC to response bias. As evidence of construct validity, we expect the PVIC to correlate negatively with measures of school connectedness and positively with measures of depressive symptoms, stress, and anxiety.
Subscale creation is most commonly based upon factor analysis. Factor analysis, however, is based on inter-item correlations and the assumption that these correlations occur because of a common underlying latent cause (Bollen & Lennox, 1991). These assumptions may not pertain to PVIC items, which were judiciously selected to represent discrete events and not multiple aspects of the same event. Consequently, we elected to base subscale construction not on item correlations but on people’s logic and intuition. This involved the collection of additional data on how college students naturally tend to group PVIC items and the application of cluster analysis to these data (See Tryfos, 1997, chapter 15).
Subsidiary goals were to identify characteristics and behaviors that contribute to college peer victimization and to clarify which are related to different subtypes of victimization. We anticipated that one set of contributing factors would consist of prejudicial attitudes towards people’s real or perceived group membership (i.e., bias-based or identity-based bullying; Mulvey et al., 2018). Such factors include bias with regard to race, ethnicity, culture, gender, sexuality, age, or physical characteristics (Kosciw et al., 2012; Neumark-Sztainer et al., 2002; Verkuyten & Thijs, 2002, McGee, 2014; Walton, 2018). Other kinds of contributing factors include behavioral issues, personality, relationship problems, or campus climate (sometimes collectively referred to as non-bias-based factors; Mulvey et al., 2018).
Study 1 involved an empirical, multimethod approach to the selection of items. This process involves three methods: (1) We surveyed students currently attending college; (2) We reviewed extant measures of peer victimization; (3) We consulted with professionals who work with victimized college students.
Participants in the initial item-development phase were students (N = 733), recruited in four ways designed to ensure diversity: (1) from a small liberal arts college (n = 123) via the psychology department subject pool, (2) from a private mid-size university (n = 217) via email recruitments of randomly selected students, (3) from a large state university (n = 181 via the psychology department subject pool, and (4) from a nation-wide sample of college students (n = 212) obtained via Qualtrics Survey Panels. All participants had to be at least 18 years old, fluent in English, and a full-time college student in the United States. The demographics revealed considerable diversity. Regarding age, participants were on average 20.5 years (SD = 3.3, range = 18 to 64). Regarding race/ethnicity, participants were 64.5% White, 13.4% Asian, 10.8% Black, 7.1% Latino/Hispanic, and 4.3% other. Regarding gender, participants were 59.3% female, 39.2% male, and 1.5% transgender. Regarding sexual orientation, participants were 81.9% straight/heterosexual, 3.0% gay/lesbian, 8.6% bisexual, 1.3% pansexual, 1.7% questioning, 0.7% queer, and 2.5% other or preferred not to respond. Regarding ability, 1.6% indicated having some form of physical disability.
Participants in a secondary item-development phase of the study were professional staff at a private mid-size university, employed by the institution to identify and counsel students who had been victimized (n = 18). Although most of their clientele were victims of sexual assault, the staff were very familiar with all kinds of student victimization experiences.
The survey was brief, asking a variety of demographic questions and two peer victimization questions. One was about kinds of peer victimization:
Most interactions among college students are friendly and well intentioned. However, sometimes things go a little bit wrong. Other times, things can go very wrong. Think about a time when a fellow student was hurtful or offensive to you or one of your college peers. We are interested in all kinds of hurtful interactions - both large and small. They could be intentional or unintentional, direct or indirect, in-person or online, physical or emotional or sexual. All kinds. In a couple of sentences, please describe any such hurtful or offensive event, perpetrated by another student. Remember: To help us keep everything anonymous and confidential, please do not use people’s names.
The second was about possible reasons for the victimization: “Some people may be at greater risk than others for being treated like this. In the example you just described, why do you think this happened?”
All procedures were approved by the IRBs at each institution. Participants began by signing online informed consent statements. Participants received a small gift card, course credit, or points toward online purchases. Out of all the student participants, 547 (75%) reported at least one victimization experience and 46 (6%) reported multiple such experiences. Altogether, we collected 582 examples of peer victimization events. Using OptimalSort (2019), a card-sorting program, we collapsed similar events into 50 discrete clusters. We selected or constructed one item to represent each cluster. For example, we created the item, “A student was physically aggressive towards me,” from the following cluster of events: “I got into a physical altercation at the bar,” “My neighbor punched me,” and “Someone purposely spilled their drink on me.” We presented these items to our focus group of college professionals, who recommended rewording some items and adding 12 new items, which largely reflected hazing and sexual assault. We compared these items to those on existing measures peer victimization. All domains of victimization covered in these measures were also captured by our empirically generated items (which included domains not covered by the extant measures). These procedures resulted in 62 distinct peer victimization items.
The survey also generated 536 reasons for peer victimization. Following procedures similar to those above, we reduced these to 66 discrete reasons. To facilitate the readability of this part of the survey, we grouped items under broader headings, such as “Because of the other person’s attitude about…” [followed by a list of demographic characteristics] and “Because the college/university…” [followed by list of campus characteristics].
Using a national sample and samples from small, midsize, and large institutions, Study 1 empirically generated a list of the kinds of peer victimization that actually happen in college. We also generated 66 perceived reasons for the victimization. From these items we generated an initial version of the PVIC. Part one of the PVIC assessed students’ experience of peer victimization. Part two assessed students’ perceptions of the reasons for the victimization events.
Whether the PVIC is used by institutions to assess peer victimization on campuses or by researchers who study peer victimization, being able to separate this highly diverse collection of events into logical, face-valid subgroups is critical. As Study 1 did not assess the structure of the PVIC items, this became the primary goal for Study 2.
The goal of Study 2 was to refine the measure and develop intuitively meaningful peer victimization subscales. Toward this end, we utilized item-level cluster analysis based on college students’ sortings of items into thematically similar groups.
Reasoning that college students could be regarded as experts on college student experiences, we obtained a sample of 100 students from the psychology participant pool at a midsize private university. Mean age was 19.07 years (SD = .92, range = 18 to 22). Regarding race/ethnicity, participants were 65% White, 18% Asian, 11% Black, 1% Middle Eastern, and 1% other (with 3% preferring not to respond). Regarding gender, participants were 60.6% female and 39.4% male. Regarding sexual orientation, participants 93.9% identified as straight/heterosexual, 1% as gay/lesbian, and 5.1% as bisexual.
Participants completed all measures in a lab setting. First, a graduate student research assistant trained participants in the use of OptimalSort (2019). Participants signed consent forms, completed the PVIC about themselves, and then went back through the PVIC items a second time to sort them thematically into clusters using the OptimalSort application on either their personal laptop or a lab computer. Instructions for the card-sorting task were as follows:
Step 1: Take a quick look at the list of items to the left. We’d like you to sort them into groups that make sense to you. There is no right or wrong answer. Just do what comes naturally.
Step 2: Drag an item from the left into this area to create your first group.
Step 3: Click the title to rename your new group.
Step 4: Add more items to this group by dropping them on top of it. Make more groups by dropping them in unused spaces.
Step 5: Please create a minimum of six distinct groups. When you’re done, click “Finished” at the top right.
Upon completion, the research assistant confirmed that participants had sorted all items and created at least of six clusters. Participants received course credit for study participation.
The result of the card sort was a 62 × 62 similarity matrix, representing the number of times participants collectively put pairs of items into the same group. We used this matrix in two ways. First, we identified pairs of items that had an agreement rating of 80% or higher (meaning that at least 80% of our participants grouped these items together). We examined these items to ensure that their co-classification by raters was due to similar thematic content and not item redundancy. Redundancy issues emerged for two pairs. The first pair (“A student was physically aggressive towards me” and “A student hit, shoved, or pushed me”) had 96% overlap with each other. We opted to retain the former, as its content subsumed the latter. The second pair (“A student posted a mean message about me online” and “A student personally and publicly attacked me on social media”) had 90% overlap with each other. We removed the former and reworded the latter to read, “A student publicly posted a mean message about me on social media” to better reflect the intended behavior. See Appendix A for the final version of the PVIC.
Second, we examined the reduced matrix using hierarchical cluster analysis, developed by Ward (1963). Using the Actual Agreement Method (Paea & Baird, 2018) we derived the dendrogram shown in Figure 1 . Dendrograms show hierarchical structure, and can also be used to identify separate clusters at different depths of clustering at the right-hand side of Figure 1 , there are between 9 and 11 distinct clusters, with the most interpretable of these structures contained 10 clusters. Cluster labels and example items were Peer Pressure/Hazing (e.g., “A student pressured me to drink or use drugs more than I wanted”), Sabotage (e.g., “A student tried to sabotage my work for a class”), Belittlement (e.g., “A student laughed at a question I asked”), Broken Trust (e.g., “A student used or took something of mine without asking”), Online (e.g., “A student distributed pictures or videos of me without my consent”), Stereotyping (e.g., “Without getting to know me, a student made false assumptions about me”), Social Exclusion (e.g., “I was excluded from a student group or activity”), Physical (e.g., “A student drugged me without my knowledge”), Verbal Aggression (e.g., “A student called me offensive names”), and Sexual Victimization (e.g., “A student sexually exposed themselves to me without my consent, either online or in-person”). Correspondence of items to subscales appears in Appendix A.
Cluster analysis dendrogram of PVIC items, using the actual agreement method. At different depths within the dendrogram, cluster structure can be identified. At the far left-hand side, each item defines its own cluster. Moving left to right, items that are more similar are combined first, etc. Closer to the right-hand side, there are between 9 and 11 clusters, with a 10-cluster solution providing the most interpretable structure.
Cluster analysis resulted in the distribution of 60 non-redundant peer victimization behaviors into 10 conceptually distinct clusters. These clusters were used to identify 10 face-valid PVIC subscales. Questions about the convergent, discriminant, and construct validity of this measure and its subscales are addressed in Study 3.
In the third study, we utilized a national sample of college students to validate the PVIC. We assessed convergent validity against other measures of peer victimization, discriminant validity against measures of response style, and construct validity vis-à-vis measures of depressive symptoms, anxiety, stress, and college sense of belonging.
We recruited a national sample of college students via the Qualtrics Survey Panels, an online survey administrator. Qualtrics works with various panel agencies that have access to pools of research participants. Potential participants received an email invitation to participate in the study. We required that participants be at least 18 years old, fluent in English, and a full-time college student in the United States. Participants enrolled only in online college courses were excluded. Because atypical students (e.g., older students) may be at particularly high risk for peer victimization, we elected not to impose other sampling restrictions. Through Qualtrics, participants received incentive points (about $5 value) for online purchases for their participation. As participants completed all surveys online, we imposed validity checks recommended by DeSimone and Harms (2018) to screen out participants who were not taking the study seriously or not answering the questions honestly. We excluded respondents who gave incorrect answers to any of three quality control questions: e.g., “For us to check that this online survey is functioning properly, please select ‘4’ as your answer to this question” (n = 13, approximately 2.4% of all recruits). We also examined protocols for speed of responding; however, no respondent completed the survey faster than recommended cutoff speeds (Wood, Harms, Lowman, & DeSimone, 2017).
After screening, our sample consisted of 520 participants. Mean age was 24.47 (SD = 7.60 range = 18 to 60). The sample was highly diverse. (Demographic options were not mutually exclusive, so percentages may not sum to 100.) Regarding gender, 252 were male, 260 were female, 8 were transgender, 7 were gender-variant/non-binary, and 1 indicated “other.” Regarding race/ethnicity, the sample was 63.5% White, 12.9% Black, 12.3% Hispanic or Latino, 8.5% Asian or Asian-American, and 1.9% other. The breakdown of participants by year in school was 21.0% freshmen, 28.3% sophomores, 37.9% juniors, 29.4% seniors, and .1% other. Areas of study included music, fine arts, or arts (15.6%), humanities (11.4%), social sciences (18.1%), science/math (20.6%), business (15.6%), engineering (15.6%), health sciences (15.6%), pre-healthcare (8.3%), undecided (8.3%), and other (12.5%). Regarding sexual orientation, 79.4% identified as straight/heterosexual, 6.5% identified as gay/lesbian, 9.0% identified as bisexual, 3.3% identified as pansexual, 1.2% identified as queer, 1.9% identified as asexual, 1.5% identified as questioning, 1.2% identified as other. Approximately 1.4% (n = 7) of participants failed to complete one or more of the survey questionnaires. People with incomplete data were not significantly different from people with complete data on most measures (ps >.10). The one exception was that people with missing data tended to score lower on our measure of workplace victimization. By using full information maximum likelihood statistical estimation methods, we were able to retain participants with partial data in all analyses.
We administered the PVIC, containing 60 victimization items (presented in random order to minimize the effects of fatigue or order on any particular item) and 66 reasons for victimization items. Additionally, we administered the following convergent, discriminant, and construct validity measures.
As no validated peer victimization measure has been developed for use with college students, we assessed convergent validity with high-school and workplace measures that have been used with this population.
The Dogruer and Yaratan Peer Victimization Scale (PVS; Dogruer & Yaratan, 2014) is a 23-item scale developed out of extant high-school and workplace measures but adapted for use with college students, using 5-point Likert scales ranging from “never” to “always.” A sample item is, “I am ridiculed in front of friends.” In the current study, coefficient alpha was .96.
The Perceived Victimization in the Workplace Scale (PVWS; Aquino & Bradfield, 2000) consists of eight self-report items that measure victimization in the workplace. This scale was developed for use with adults. For the current study, instructions and items were reworded to refer to college peers instead of coworkers. Respondents rate the frequency of their peer victimization experiences using 5-point Likert scales (i.e., “never” to “more than 20 times”). Four items comprise a direct aggression subscale (e.g., “Made an obscene comment or gesture at you”), and four comprise an indirect aggression subscale (“A college peer did something to make you look bad”). In the current study, we combined the two subscales as their correlation was so large (r = .75). Coefficient alpha for the total PVWS was .89.
People may dissemble when answering questions about negative, painful, or embarrassing events, especially when such events contradict one’s self-image (Bagby & Marshall, 2004). Consequently, measures of peer victimization can be confounded by the respondent’s tendency to dissemble or deny such events. To assess the degree to which the PVIC may be subject to this kind of response bias, we compared it to the MMPI-2 Lie Scale (MMPI-Lie), a 15-item measure of the tendency to under-report negative characteristics. Responding “false” to these items, which describe minor personal flaws (e.g., “At times I feel like swearing”), indicates a tendency to “fake good.” In the current study, the KR-20 was .76.
Theory and research link peer victimization to elevated levels of internalizing problems and reduced levels of school belongingness (Hawker & Boulton, 2000; Hodges, Low, Viñas-Racionero, Hollister, & Scalora, 2016). We hypothesized that relations of the PVIC scales to operationalizations of these constructs would be commensurate with these empirically theoretical relations, lending support for the construct validity of the PVIC scales (Cronbach & Meehl, 1955). Consequently, we obtained the following measures of depressive symptoms, stress, anxiety symptoms, and sense of college belonging).
The Beck Depression Inventory – Version 2 (BDI-II; Beck, Steer, & Brown, 1996) is a self-report questionnaire that measures severity of affective, behavioral, and cognitive depressive symptoms in adolescents and adults. It consists of 21 items asking how the respondent has been feeling over the past two weeks. Each item consists of four response options, scored on a 4-point scale (i.e., 0 = “I don’t feel disappointed in myself,” 1 = “I am disappointed in myself,” 2 = “I am disgusted with myself,” 3 = “I hate myself.”) Scores are summed so that higher scores reflect more severe levels of depression. The measure has strong internal consistency, convergent validity, and discriminant validity. It also discriminates well between depressed and nondepressed individuals (Dozois & Covin, 2004). The measure has been validated for use with university populations, showing strong psychometric properties (coefficient alpha = .91; Dozois, Dobson, & Ahnberg, 1998). In the current study, coefficient alpha was .95.
The Depression and Anxiety Stress Scale (DASS; Lovibond & Lovibond, 1995) is a 42-item self-report measure of respondents’ emotional states concerning depression, anxiety, and stress. Respondents rate the extent to which various statements applied to them over the past week on four-point Likert scales (0= “did not apply to me at all” to 3= “applied to me very much, or most of the time”). Sample items include “I felt down-hearted and blue” (depression), “I found it difficult to relax” (anxiety), and “I found myself getting upset by quite trivial things” (stress). Each subscale consists of 14 items. In previous research coefficient alphas ranged from .88 to .94 (Lovibond & Lovibond, 1995). In the current study, coefficient alphas were .97 for the depression subscale, .95 for the anxiety subscale, and .95 for the stress subscale.
The College Sense of Belonging (CSB) is a seven-item scale containing yes/no items from three instruments measuring sense of college connectedness and sense of belonging (Bollen & Hoyle, 1990; Ostrove & Long, 2007; Museus, Yi, & Saelua, 2017). Items were “I feel that I belong on campus,” “I feel a sense of belonging to my college/university,” “I feel I am a member of my university’s community,” “I feel that I fit in well as part of the college environment,” “Overall, to what extent do you feel you belong at your university,” “I see myself as part of the campus community,” and “I feel a strong sense of connection to the campus community.” In the current study, KR-20 was .87.
We computed subscale scores in two ways. First was the number of endorsed subscale items (e.g., a person who endorses 3 out of 6 online victimization items would get a score of 3). Such scores are useful when quantifying one’s amount of victimization. Second was the proportion of subscale items endorsed (e.g., a person who endorses 3 out of 6 items would get a score of .5). Such scores are useful when comparing one subscale to another. Table 1 contains correlations and descriptive statistics regarding PVIC subscales. Correlations ranged from .29 to .70 (median = .46). All were significant (p < .001). Figure 2 depicts number of peer victimization subtypes (out of 10). Only 10.2% of participants reported no peer victimization of any type, revealing that the vast majority (89.8%) had experienced at least one type. Approximately 6.9% experienced some version of all 10 victimization subtypes. All subsequent analyses were based on the proportion (not the number) of endorsed subscale items.
Percent of participants (0–100%) who reported having experienced 0 to 10 subtypes of peer victimization. (Note: anyone who scored > 0 on a particular subscale was regarded as having experienced that subtype of peer victimization; hence the X-axis metric is the number of subscales on which people scored > 0.)
Study 3 Peer Victimization in College (PVIC) Subscale Correlations and Descriptive Statistics
| PVIC subscales | 1. | 2. | 3. | 4. | 5. | 6. | 7. | 8. | 9. | 10. |
|---|---|---|---|---|---|---|---|---|---|---|
| 1. Hazing / Peer pressure (8) a | 1.00 | |||||||||
| 2. Sabotage (4) | 0.60 | 1.00 | ||||||||
| 3. Belittlement (5) | 0.42 | 0.47 | 1.00 | |||||||
| 4. Broken trust (11) | 0.57 | 0.56 | 0.66 | 1.00 | ||||||
| 5. Online (6) | 0.61 | 0.56 | 0.42 | 0.59 | 1.00 | |||||
| 6. Stereotyping (6) | 0.52 | 0.54 | 0.68 | 0.70 | 0.57 | 1.00 | ||||
| 7. Social exclusion (4) | 0.46 | 0.48 | 0.60 | 0.61 | 0.49 | 0.67 | 1.00 | |||
| 8. Physical (4) | 0.59 | 0.47 | 0.29 | 0.45 | 0.54 | 0.41 | 0.41 | 1.00 | ||
| 9. Verbal aggression (6) | 0.53 | 0.53 | 0.60 | 0.70 | 0.53 | 0.69 | 0.64 | 0.50 | 1.00 | |
| 10. Sexual (6) | 0.52 | 0.36 | 0.31 | 0.50 | 0.53 | 0.43 | 0.36 | 0.48 | 0.42 | 1.00 |
| Number of endorsed items per subscale b | ||||||||||
| Mean | 0.88 | 0.49 | 2.23 | 3.57 | 0.88 | 1.64 | 1.22 | 0.31 | 1.66 | 0.86 |
| SD | 1.47 | 0.93 | 1.74 | 3.12 | 1.40 | 1.72 | 1.32 | 0.68 | 1.82 | 1.42 |
| Proportion of endorsed items per subscale | ||||||||||
| Mean | 0.11 | 0.12 | 0.45 | 0.32 | 0.15 | 0.27 | 0.30 | 0.08 | 0.28 | 0.14 |
| SD | 0.18 | 0.23 | 0.35 | 0.28 | 0.23 | 0.29 | 0.33 | 0.17 | 0.30 | 0.24 |
Note. All correlations are significant at p < .001.
a Parenthesized values are the number of items on each subscale. b Maximum values differ by subscale, depending on the number of items per scale.Study 3 Convergent Validity: Regression of PVS and PVWS onto 10 PVIC Subscales
| Predictors (PVIC subscales) | Dependent variable = PVS R 2 = .54 (p < .001) | Dependent variable = PVWS R 2 = .56 (p < .001) | ||||
|---|---|---|---|---|---|---|
| B | SE(B) | β | B | SE(B) | β | |
| Hazing/Peer pressure | 9.96 | 3.78 | 0.12 ** | 0.80 | 1.38 | 0.03 |
| Sabotage | 7.96 | 2.76 | 0.12 ** | 0.95 | 1.01 | 0.04 |
| Belittlement | 0.05 | 1.99 | 0.00 | −0.83 | 0.73 | −0.05 |
| Broken trust | 4.85 | 2.80 | 0.09 | 2.18 | 1.02 | 0.11 * |
| Online | 13.02 | 2.90 | 0.20 *** | 3.88 | 1.06 | 0.16 *** |
| Stereotyping | 4.97 | 2.74 | 0.09 | 3.04 | 1.00 | 0.15 ** |
| Social exclusion | 2.41 | 2.07 | 0.05 | 0.81 | 0.76 | 0.05 |
| Physical | 17.70 | 3.70 | 0.20 *** | 4.69 | 1.35 | 0.14 *** |
| Verbal aggression | 2.70 | 2.44 | 0.05 | 4.80 | 0.89 | 0.26 *** |
| Sexual | 0.83 | 2.48 | 0.01 | 1.58 | 0.91 | 0.07 |
Note. PVS = Peer Victimization Scale, PVWS = Perceived Victimization in the Workplace Scale
Additional evidence of discriminant validity derived from correlations with the MMPI-Lie scale. Total PVIC scores and scores on the Lie scale correlated −.23, which although significant (p < .01) represented only 5% shared variance. Correlations with PVIC subscales were also small, ranging from −.11 to −.29.
We examined construct validity in two ways. First, we tested the relations of our PVIC subscales to factors known to increase risk for peer victimization. Second, we tested PVIC subscales against problems known to result from victimization.
Researchers have divided risk factors for peer victimization into two broad types: bias-based contributing factors and non-bias-based contributing factors (Mulvey et al., 2018). Factor analysis of the 66 contributing factor items supported this typology. Using Mplus, we conducted an exploratory, Geomin, oblique factor analysis using WLSMV estimation. Fit statistics (eigenvalues, RMSEAs, CFIs, TLIs, and SRMRs) suggested that solutions with 10 to 14 factors were defensible. After ruling out solutions with singlets or excessive cross-loadings, a 10-factor solution was the most interpretable (RMSEA = .037, CFI = .943, TLI = .920, SRMR = .045). As shown in Table B1 , four were bias-based (Race/Ethnicity, Speech/Language, Gender/Sexuality, and Health/Disability) and six were non-bias-based (Personality/Popularity, Interpersonal Conflict, Hazing/Initiation, Easy Target, Behavior/Substance Use, and Campus Climate).
To assess the relation of these 10 Contributing Factors to our 10 Peer Victimization Factors, we used canonical correlation analysis. Roots 1 to 5 were significant (F36,2141.33 = 1.904, Wilks’ lambda = .87): r1 = .81, r2 = .55, r3 = .47, r4 = .33, and r5 = .28. Examination of the canonical loadings (see Table 3 ) revealed five relations. First, people with more contributing factors of any sort tended to experience more victimization of all sorts (root 1). Second, contributing factors such as hazing, initiation, and substance use were associated with physical aggression and peer pressure (root 2, a non-biased-based factor). Third, contributing factors such as having a disability or health-related problems were associated with social exclusion (root 3: bias-based). Fourth, contributing factors such as being a racial or ethnic minority were associated with greater stereotyping (root 4: bias-based). Fifth, contributing factors pertaining to one’s gender or sexuality were associated with more sexual victimization (root 5: bias-based).
Study 3 Canonical Loadings for Roots 1 to 5, Relating Contributing Factors to PVIC Subscales
| Subscale | Root 1 | Root 2 | Root 3 | Root 4 | Root 5 | . |
|---|---|---|---|---|---|---|
| Set 1: Contributing Factors | ||||||
| Race/ethnicity | 0.54 | 0.17 | 0.32 | 0.53 | 0.03 | . |
| Speech/language | 0.64 | 0.31 | 0.38 | 0.28 | −0.21 | . |
| Behavior/substance | 0.60 | 0.49 | −0.26 | 0.02 | 0.20 | . |
| Gender/sexuality | 0.41 | 0.19 | 0.28 | −0.04 | 0.52 | . |
| Personality/popularity | 0.76 | 0.00 | 0.37 | −0.12 | −0.34 | . |
| Interpersonal/conflict | 0.95 | −0.22 | −0.19 | 0.04 | 0.04 | . |
| Hazing/initiation | 0.45 | 0.66 | −0.20 | −0.10 | −0.31 | . |
| Health/disability | 0.57 | 0.29 | 0.53 | 0.06 | −0.03 | . |
| Easy/target | 0.55 | 0.02 | 0.29 | −0.39 | −0.06 | . |
| Campus/climate | 0.65 | 0.25 | 0.32 | −0.14 | 0.22 | . |
| Set 2: Peer Victimization | ||||||
| Hazing/Peer pressure | 0.68 | 0.63 | −0.16 | −0.06 | −0.16 | . |
| Sabotage | 0.65 | 0.35 | 0.16 | 0.15 | −0.40 | . |
| Belittlement | 0.86 | −0.33 | −0.15 | −0.04 | −0.03 | . |
| Broken trust | 0.88 | 0.03 | −0.22 | 0.07 | 0.08 | . |
| Online | 0.68 | 0.38 | 0.25 | −0.04 | 0.14 | . |
| Stereotyping | 0.85 | −0.09 | 0.21 | 0.43 | −0.01 | . |
| Social exclusion | 0.79 | −0.11 | 0.40 | −0.25 | −0.22 | . |
| Physical | 0.50 | 0.55 | 0.21 | 0.13 | 0.22 | . |
| Verbal aggression | 0.76 | 0.05 | −0.07 | 0.14 | −0.11 | . |
| Sexual | 0.57 | 0.33 | 0.14 | −0.18 | 0.55 | . |
Note. Loadings > .40 appear in boldface.
We also tested the bias-based hypothesis by comparing majority to minority students on peer victimization subscales, anticipating that belonging to a racial/ethnic, gender, sexuality, or age minority group would lead to higher instances of peer victimization. As shown in Table 4 , being a racial/ethnic minority was associated with greater stereotyping and online victimization. Transgender students also reported more stereotyping. Students who were not straight were more likely to report stereotyping, social exclusion, and verbal aggression. Students who were older (> 23 years old) scored higher on the peer pressure, sabotage, online, social exclusion, physical, and verbal aggression subscales. No significant effects were in the unexpected direction.
Study 3 Means (and Standard Deviations) and t-tests for Majority-Minority Comparisons on PVIC Subscales
| Subscale | Race/ethnicity | Gender | Sexuality | Age | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| White n = 330 | Not White n = 186 | t515 | Cis n = 506 | Trans n = 12 | t517 | Straight n = 407 | Not straight n = 108 | t514 | 18 – 23 yrs old n = 342 | > 23 yrs old n = 176 | t517 | |
| Hazing/Peer pressure | 0.10 (0.18) | 0.12 | −1.01 | 0.11 | 0.11 | −0.13 | 0.11 | 0.11 | 0.30 | 0.09 | 0.14 | −2.73 ** |
| 0.18 | 0.18 | 0.17 | 0.18 | 0.18 | 0.15 | 0.22 | ||||||
| Sabotage | 0.11 (0.22) | 0.14 | −1.07 | 0.12 | 0.10 | 0.25 | 0.12 | 0.12 | −0.11 | 0.10 | 0.17 | −3.22 *** |
| 0.25 | 0.23 | 0.23 | 0.23 | 0.23 | 0.21 | 0.26 | ||||||
| Belittlement | 0.44 (0.34) | 0.45 | −0.16 | 0.44 | 0.47 | −0.22 | 0.49 | 0.43 | 1.77 | 0.44 | 0.45 | −0.17 |
| 0.37 | 0.35 | 0.38 | 0.36 | 0.34 | 0.35 | 0.35 | ||||||
| Broken trust | 0.32 (0.28) | 0.33 | −0.51 | 0.32 | 0.30 | 0.33 | 0.36 | 0.31 | 1.89 | 0.32 | 0.33 | −0.37 |
| 0.29 | 0.28 | 0.29 | 0.30 | 0.27 | 0.28 | 0.29 | ||||||
| Online | 0.13 (0.22) | 0.18 | −2.49 * | 0.14 | 0.21 | −0.98 | 0.18 | 0.13 | 2.08 * | 0.13 | 0.18 | −2.70 ** |
| 0.24 | 0.23 | 0.20 | 0.25 | 0.22 | 0.22 | 0.25 | ||||||
| Stereotyping | 0.25 (0.27) | 0.31 | −2.26 * | 0.26 | 0.55 | −3.46 | 0.35 *** | 0.25 | 3.42 *** | 0.26 | 0.30 | −1.60 |
| 0.30 | 0.28 | 0.27 | 0.29 | 0.28 | 0.28 | 0.30 | ||||||
| Social exclusion | 0.31 (0.32) | 0.30 | 0.29 | 0.30 | 0.49 | −1.95 | 0.39 | 0.28 | 3.00 ** | 0.28 | 0.34 | −1.81 |
| 0.34 | 0.33 | 0.36 | 0.35 | 0.32 | 0.32 | 0.34 | ||||||
| Physical | 0.07 (0.16) | 0.09 | −1.02 | 0.07 | 0.15 | −1.52 | 0.09 | 0.07 | 0.92 | 0.06 | 0.10 | −2.16 * |
| 0.17 | 0.16 | 0.25 | 0.18 | 0.16 | 0.15 | 0.19 | ||||||
| Verbal aggression | 0.27 (0.29) | 0.29 | −0.68 | 0.27 | 0.40 | −1.48 | 0.34 | 0.26 | 2.51 * | 0.26 | 0.32 | −2.15 * |
| 0.32 | 0.30 | 0.33 | 0.34 | 0.29 | 0.30 | 0.31 | ||||||
| Sexual | 0.14 (0.23) | 0.14 | −0.17 | 0.14 | 0.25 | −1.62 | 0.18 | 0.13 | 1.69 | 0.14 | 0.15 | −0.32 |
| 0.24 | 0.23 | 0.24 | 0.26 | 0.23 | 0.23 | 0.25 | ||||||
Note. Degrees of freedom vary slightly because of missingness.
Study 3 Construct Validity: Four Outcome Variables onto 10 PVIC Subscales
| Predictors | B | SE(B) | β | t(508) | r |
|---|---|---|---|---|---|
| Dependent variable = Depressive composite, R = .40 | |||||
| Hazing/Peer pressure | −0.14 | 0.08 | −0.10 | −1.69 | 0.20 *** |
| Sabotage | 0.08 | 0.12 | 0.04 | 0.66 | 0.24 *** |
| Belittlement | 0.04 | 0.07 | 0.04 | 0.62 | 0.24 *** |
| Broken trust | −0.04 | 0.04 | −0.06 | −0.91 | 0.25 *** |
| Online | 0.11 | 0.08 | 0.08 | 1.26 | 0.28 *** |
| Stereotyping | 0.03 | 0.08 | 0.02 | 0.35 | 0.28 *** |
| Social exclusion | 0.41 | 0.09 | 0.28 | 4.58 *** | 0.35 *** |
| Physical | 0.24 | 0.16 | 0.08 | 1.51 | 0.25 *** |
| Verbal aggression | −0.04 | 0.07 | −0.04 | −0.60 | 0.25 *** |
| Sexual | 0.21 | 0.07 | 0.15 | 2.90 ** | 0.28 *** |
| Dependent variable = Anxiety symptoms (DASS), R = .44 | |||||
| Hazing/Peer pressure | 0.05 | 0.40 | 0.01 | 0.13 | 0.31 *** |
| Sabotage | 0.70 | 0.58 | 0.07 | 1.20 | 0.31 *** |
| Belittlement | 0.02 | 0.34 | 0.00 | 0.07 | 0.24 *** |
| Broken trust | −0.31 | 0.22 | −0.10 | −1.46 | 0.27 *** |
| Online | 0.93 | 0.41 | 0.13 | 2.27 * | 0.36 *** |
| Stereotyping | 0.16 | 0.39 | 0.03 | 0.41 | 0.31 *** |
| Social exclusion | 1.81 | 0.44 | 0.25 | 4.13 *** | 0.36 *** |
| Physical | 1.70 | 0.78 | 0.12 | 2.18 * | 0.33 *** |
| Verbal aggression | −0.20 | 0.34 | −0.04 | −0.57 | 0.29 *** |
| Sexual | 0.56 | 0.35 | 0.08 | 1.61 | 0.29 *** |
| Dependent variable = Stress (DASS), R = .42 | |||||
| Hazing/Peer pressure | −0.59 | 0.43 | −0.08 | −1.35 | 0.23 *** |
| Sabotage | 0.41 | 0.64 | 0.04 | 0.65 | 0.27 *** |
| Belittlement | 0.01 | 0.37 | 0.00 | 0.02 | 0.25 *** |
| Broken trust | −0.09 | 0.23 | −0.03 | −0.37 | 0.28 *** |
| Online | 0.68 | 0.45 | 0.09 | 1.53 | 0.31 *** |
| Stereotyping | 0.44 | 0.42 | 0.07 | 1.04 | 0.33 *** |
| Social exclusion | 2.22 | 0.47 | 0.28 | 4.68 *** | 0.38 *** |
| Physical | 1.28 | 0.85 | 0.08 | 1.50 | 0.27 *** |
| Verbal aggression | −0.23 | 0.37 | −0.04 | −0.61 | 0.28 *** |
| Sexual | 0.64 | 0.38 | 0.09 | 1.69 | 0.26 *** |
| Dependent variable = Sense of belonging (CSB), R = .30 | |||||
| Hazing/Peer pressure | 7.41 | 2.02 | 0.23 | 3.66 *** | 0.03 |
| Sabotage | 0.26 | 1.48 | 0.01 | 0.18 | −0.08 |
| Belittlement | −0.82 | 1.06 | −0.05 | −0.77 | −0.17 *** |
| Broken trust | −0.73 | 1.50 | −0.04 | −0.49 | −0.13 ** |
| Online | −1.33 | 1.55 | −0.05 | −0.86 | −0.10 * |
| Stereotyping | −1.64 | 1.46 | −0.08 | −1.12 | −0.18 *** |
| Social exclusion | −4.15 | 1.10 | −0.24 | −3.76 *** | −0.24 *** |
| Physical | −1.23 | 2.00 | −0.03 | −0.62 | −0.05 |
| Verbal aggression | 1.50 | 1.32 | 0.08 | 1.14 | −0.11 ** |
| Sexual | −0.68 | 1.33 | −0.03 | −0.51 | −0.07 |
Note. DASS = Depression Anxiety and Stress Scale, CSB = College Sense of Belonging scale
Study 3 began with the empirically generated items from Study 1 along with the subscales identified in Study 2 and provided evidence of convergent, discriminant, and construct validity for the PVIC. Results also revealed that peer victimization is highly prevalent in college populations, with 90% of students reporting at least one kind of victimization. Furthermore, analyses exposed gaps in coverage of the peer victimization construct by extant measures (e.g., scores on the PVS were not predicted by peer pressure, sabotage, belittlement, social exclusion, or sexual victimization, and scores on the PVWS were not predicted by belittlement, broken trust, stereotyping, or social exclusion. Finally, the study documents the association of specific peer victimization subtypes with key risk factors and important psychological/educational outcomes. Not assessed in Study 3, however, were issues of scaling and incremental validity, the primary goals of Study 4.
In the fourth study, we utilized a second national sample of college students (a) to examine the relative utility of different scaling systems for the PVIC items and (b) to test the incremental utility of the PVIC over and above extant measures of peer victimization. Regarding scaling, we compared the original binary scoring system (i.e., events did or did not happen) to three alternative scaling methods: event frequency, effect duration, or perceived impact. Regarding incremental validity, we tested the predictive utility of the 10 PVIC subscales in relation to depressive symptoms, anxiety symptoms, perceived stress, and college belongingness, while statistically controlling for the PVS and PVWS measures of peer victimization.
As in study 3, we recruited an independent national sample of college students via the Qualtrics Survey Panels. We required that participants be at least 18 years old, fluent in English, and a full-time college student in the United States. Incentives, validity checks, and exclusionary criteria were the same as in study 3. After all screening procedures, our sample consisted of 633 participants. Mean age was 21.24 (SD = 3.45 range = 18 to 46). The sample was highly diverse. Regarding gender, 41.5% were male, 55.9% were female, .5% were transgender, and 2% were gender-variant/non-binary. Regarding race/ethnicity, the sample was 64.6% White, 12.5% Black, 6.1% Hispanic or Latino, 12.7% Asian or Asian-American, 1.9% Middle Eastern and 2.4% other. The breakdown of participants by year in school was 19.9% freshmen, 27.9% sophomores, 22.2% juniors, 24.6% seniors, 2.8% other, and 2.5% prefer not to respond. Areas of study were diverse: 13.5% in music, fine arts, or arts, 13.1% in humanities, 16.3% in social sciences, 13.8% in science/math, 7.8% in business, 6.6% in engineering, 12.1% reported health sciences, 4.2% reported pre-healthcare, 1.1% were undecided, 10.9% reported other, AND .5% prefer not to respond. Regarding sexual orientation, 75.0% identified as straight/heterosexual, 4.6% identified as gay/lesbian, 11.1% identified as bisexual, 2.7% identified as pansexual, 1.3% identified as queer, 1.9% identified as asexual, 3.3% identified as questioning, .2% identified as other. Approximately 1.4% (n = 7) of participants failed to complete one or more of the surveys.
All Study 4 measures are described in Study 3. In the current study, coefficient alphas were .93 for the PVS, .86 for the PVWS, .94 for the BDI-2, and .97 for the DASS. For the CBS, KR-20 = .91. Additionally, participants completed the 60-item PVIC (in random order) with revised response options. Respondents indicated whether or not an event had happened to them (0/1 scoring), the frequency with which the event occurred (0 = “never,” 1 = “one time,” 2 = “2 or 3 times,” 3 = “4 to 10 times,” and 4 = “more than 10 times”), the size of the impact (0 = “did not affect me at all,” 1 = “mild effect,” 2 = “moderate effect,” 3 = “big effect,” and 4 = “very big effect”), and the duration of the impact (1 = “one day or less,” 2 = “several days,” 3 = “a week,” 4 = “several weeks,” and 5 = “more than a month”). Checklist scoring consisted of the sum of the binary responses. Frequency scoring was the sum of the item frequencies. Duration scoring was the sum of the event durations. Impact scoring was the sum of the event impact ratings.
Table 6 contains means, standard deviations, and correlations for all study measures. Table 7 reports subscale correlations with emotional and educational outcome variables.
Study 4 Means, Standard Deviations, and Correlations
| Measure | 1. | 2. | 3. | 4. | Mean | SD |
|---|---|---|---|---|---|---|
| 1. PVIC checklist (0/1 scoring) | 1.00 | 18.98 | 13.60 | |||
| 2. PVIC frequency (0–4 scoring) | 0.91 | 1.00 | 39.28 | 34.58 | ||
| 3. PVIC duration (1–5 scoring) | 0.87 | 0.88 | 1.00 | 45.22 | 41.13 | |
| 4. PVIC impact (0–4 scoring) | 0.86 | 0.86 | 0.95 | 1.00 | 32.12 | 30.39 |
| 5. Depression (BDI-2, DASS) | 0.34 a | 0.37 b | 0.39 b | 0.38 b | 0.00 | 0.96 |
| 6. Anxiety symptoms (DASS) | 0.42 a | 0.45 b | 0.46 b | 0.46 b | 11.61 | 9.79 |
| 7. Perceived stress (DASS) | 0.38 a | 0.41 b | 0.43 b | 0.43 b | 15.49 | 10.69 |
| 8. College Belonging Composite | −0.11 a | −0.15 b | −0.17 b | −0.17 b | 0.00 | 0.81 |
| 9. Peer Victimization Scale (PVS) | 0.61 | 0.65 | 0.58 | 0.61 | 9.89 | 11.78 |
| 10. Perceived Victimization in the Workplace Scale (PVWS) | 0.59 | 0.65 | 0.59 | 0.57 | 4.28 | 5.07 |
Study 4 Pearson Correlations between PVIC Subscales and Emotional/Educational Outcome Variables
| Depression composite | Anxiety (DASS) | Stress (DASS) | College Belonging | |
|---|---|---|---|---|
| Hazing/Peer pressure | 0.22 *** | 0.31 *** | 0.23 *** | −0.05 |
| Sabotage | 0.24 *** | 0.33 *** | 0.26 *** | −0.07 |
| Belittlement | 0.32 *** | 0.39 *** | 0.42 *** | −0.15 *** |
| Broken trust | 0.34 *** | 0.42 *** | 0.39 *** | −0.10 ** |
| Online | 0.29 *** | 0.35 *** | 0.30 *** | −0.11 ** |
| Stereotyping | 0.35 *** | 0.41 *** | 0.37 *** | −0.22 *** |
| Social exclusion | 0.41 *** | 0.42 *** | 0.40 *** | −0.29 *** |
| Physical | 0.17 *** | 0.25 *** | 0.17 *** | −0.05 |
| Verbal aggression | 0.28 *** | 0.39 *** | 0.31 *** | −0.11 ** |
| Sexual | 0.24 *** | 0.28 *** | 0.24 *** | −0.04 |
We treated our outcome variables (depressive symptoms, anxiety symptoms, stress, and college belonging) as the criteria by which to judge the four methods for scaling the PVIC. To assess the relative merits of each scaling approach, we began with Pearson correlations (see Table 8 ). Although the differences were small, statistical comparisons (Steiger, 1980) revealed that correlations between the checklist scaling of the PVIC and the four criteria were significantly smaller than were the comparable correlations involving the frequency, duration, and impact scalings of the PVIC, suggesting that moving beyond binary item scaling (i.e., whether or not each event happened) may increase the predictive utility of the PVIC.
Study 4 Partial Correlations of Some Peer Victimization in College (PVIC) Scalings Controlling for Others in the Prediction of Emotional and Educational Outcomes
| Criterion variable | Controlling for Checklist | Controlling for Frequency | Controlling for Impact | Controlling for Duration | |||
|---|---|---|---|---|---|---|---|
| Frequency of peer victimization | Duration of effect | Impact of effect | Duration of effect | Impact of effect | Duration of effect | Impact of effect | |
| Depressive symptoms (BDI & DASS) | 0.15 *** | 0.21 *** | 0.19 *** | 0.16 *** | 0.14 *** | 0.10 * | 0.04 |
| Anxiety symptoms (DASS) | 0.18 *** | 0.20 *** | 0.21 *** | 0.14 *** | 0.16 *** | 0.08 | 0.09 |
| Perceived stress (DASS) | 0.15 *** | 0.22 *** | 0.22 *** | 0.17 *** | 0.17 *** | 0.08 | 0.07 |
| College belonging composite (CBS) | −0.14 *** | −0.15 *** | −0.16 *** | −0.08 | −0.09 * | −0.01 | −0.04 |
Note. BDI = Beck Depression Inventory-2, DASS = Depression, Anxiety, Stress Scale, CBS = College Belonging Scale.
Next, we tested the incremental validity of the PVIC over and above the PVS and PVWS. For this test, we opted to use the frequency-scale version of the PVIC, despite the fact that the duration and impact scales proved to be slightly better predictors in the last set of analyses. With the goal to make this as fair a comparison as possible, we based this decision on two factors. First, both other measures also use frequency scales (PVS: “never” to “always”; PVWS: “never” to “more than 20 times”). Second, as the criterion variables in the current study focus on the consequences of peer victimization, using duration and impact scales that also focus on the consequences of peer victimization consequences would not be fair. Although duration and impact scaling would be appropriate for other research questions, we regarded frequency scaling as the most appropriate method for the current study.
We conducted four hierarchical regressions to test our incremental validity hypotheses. Specifically, we focused on the ability of the PVIC scales to predict our four criterion variables (taken separately) over and above the PVS, a high school measure sometimes used in college populations, and the PVWS, a workplace victimization sometimes used with college students. In each analysis, we entered the PVS and PVWS in step 1 and the 10 PVIC subscales in step 2. The statistic of primary interest was the change in R-squared from step 1 to step 2. As shown in Table 9 ), the PVIC scales collectively accounted for significant variance over and above the control variables in every analysis, supporting the incremental validity of the PVIC.
Study 4 Hierarchical Regressions: Incremental Validity of PVIC Subscales over Two Extant Measures of Sometimes Used with College Students
| Dependent variable | Step: Predictors | R | R 2 | ΔR 2 | ΔF (10, 616) | p |
|---|---|---|---|---|---|---|
| Depressive symptoms | 1: PVS & PVWS | .384 | .147 | |||
| 2: PVS, PVWS, & 10 PVIC subscales | .490 | .240 | .093 | 7.56 | .001 | |
| Anxiety symptoms | 1: PVS & PVWS | .486 | .236 | |||
| 2: PVS, PVWS, & 10 PVIC subscales | .530 | .281 | .045 | 3.851 | .001 | |
| Perceived stress | 1: PVS & PVWS | .403 | .163 | |||
| 2: PVS, PVWS, & 10 PVIC subscales | .508 | .258 | .096 | 7.931 | .001 | |
| College belonging | 1: PVS & PVWS | .173 | .030 | |||
| 2: PVS, PVWS, & 10 PVIC subscales | .357 | .127 | .097 | 6.852 | .001 |
Note. PVS = Peer Victimization Scale, PVWS = Peer Victimization in the Workplace Scale
Two broad findings emerged from Study 4. First, empirical support emerged for scoring the PVIC items with respect to either (1) the frequency of their occurrence or (2) the perceived duration or impact of their consequences. Such scaling methods were superior to binary item scaling inherent in a checklist approach. When using the PVIC to predict adverse personal outcomes, however, we caution that duration and impact scaling could slightly conflate the assessment of victimization with the assessment of its consequences. Second, empirical support also emerged for the incremental validity of the PVIC over and above the PVS and PVWS. The superiority of the PVIC is not surprising, given that it was empirically developed for, and specifically validated with, college populations. Furthermore, it assesses a broader array of peer victimization subtypes than does either of the other two measures.
Six major results emerged from the current set of studies. First and foremost is the introduction of the first validated, comprehensive measure of peer victimization for use with students in college. Second, as we developed PVIC items empirically, their clusters not only serve as subscales for scoring purposes but also characterize the kinds of peer victimization that actually occur on college campuses. Third, we found evidence of convergent and incremental validity for the PVIC vis-à-vis other measures sometimes used to assess peer victimization in college. Fourth, evidence indicated that frequency scores, impact ratings, and duration estimates were superior to the use of the PVIC as a simple peer victimization checklist. Fifth, correlations with demographic and behavioral risk factors were commensurate with identity-based and non-biased-based hypotheses about causes of peer victimization, providing evidence of construct validity. Sixth, the PVIC subscales scales correlated with important clinical and education-related outcomes. Collectively, these findings suggest that the PVIC is well suited for both basic and applied research on bullying and victimization in college.
First and most importantly, the current studies introduce and validate the first comprehensive measure of college peer victimization. The absence of such a measure in the literature is quite surprising, especially given that (1) college students represent the population most commonly researched by psychologists, (2) the current data clearly show that peer victimization remains a serious problem in college with 89.2% reporting at least one type of peer victimization and the vast majority reporting multiple types, and (3) the consequences of college peer victimization can be quite severe, even life-threatening (Klomek, Sourander, & Gould, 2010). Our creation of such a measure is important given certain trends in post-secondary education. For over well over 100 years, college and university student populations have grown in diversity. As important as this process is, it can be accompanied by increased rates of intolerant and prejudicial behavior that college officials must seek to understand. Furthermore, as colleges launch campus-based efforts to promote diversity tolerance, they must undertake efforts to assess program effectiveness. The PVIC is a tool that can help meet both of these needs.
The second contribution was our identification of PVIC subscales, which correspond to real subtypes of peer victimization. We based the PVIC on peer victimization experiences actually reported by college students around the country. Consequently, the PVIC subscales represent an empirically driven characterization of the kinds of peer victimization that occur in U.S. institutions of higher learning. In this effort, we chose to cast our net broadly and include all kinds of peer-on-peer aggressive behaviors, even some that might not meet certain previous definitions of peer victimization. For example, most high school measures of peer victimization do not include sexual misconduct or assault (e.g., Desjardins et al., 2013), although stand-alone measures of sexual victimization do exist (e.g., Krebs, 2014). We included such events in order to be comprehensive. Likewise, most measures of peer victimization do not include hazing, although stand-alone measures of hazing do exist (Campo, Poulos, & Sipple, 2005), perhaps because hazing typically serves as a prelude to social inclusion (not exclusion) and because people essentially agree to such treatment a priori. We included hazing behaviors, however, because such behaviors sometimes go further than what was agreed upon and because they can have negative psychological and physical consequences. A third example is micro-aggression, possibly excluded because such behaviors are often unwitting or unintentional (although separate measures do exist; Torres-Harding, Andrade, & Romero Diaz, 2012). We included micro-aggressions because intentionality is difficult to assess and the consequences may be serious irrespective of the perpetrator’s intent. In the PVIC, these types of peer victimization constitute separate subscales; consequently, investigators can elect to exclude them if they do not represent their construct of interest or if they are measured by other means.
The third finding was evidence not only of convergent validity with extant measures of college peer victimization, but of incremental validity of the PVIC scales over and above such measures. The unique relation of the PVIC to mental health and educational outcomes is not surprising, given that pre-existent tools to measure peer victimization in college (a) were developed for use with younger or older populations, (b) were never validated for use with college students, and (c) do not assess the full range of peer victimization behaviors that occur on college campuses. Clearly, the PVIC fills a need.
Fourth, our findings support the use of several possible scaling methods. Likert-type scaling options that assess the frequency of each peer victimization event, the perceived impact of such events, and the duration of such impact were all superior to the use of the PVIC as a simple binary checklist. Although duration and impact scaling were highly redundant of each other, both yielded information not captured by frequency scaling. Care should be taken, however, when using the PVIC to predict emotional or mental health outcomes, as the duration and impact scales literally quantify the consequences of peer victimization (not the peer victimization events per se), which could conflate victimization with its consequences. To avoid such confounds, we recommend the use of event frequency scaling.
Our fifth major set of findings were correlations between PVIC subscales and various risk factors. Demographic differences largely reflected the bias-based bullying hypothesis that being a minority of almost any type increases risk for peer victimization (Aboujaoude, Savage, Starcevic, & Salame, 2015; Farrington & Baldry, 2010). Specifically, greater victimization scores were associated with one’s demographic minority status regarding race/ethnicity, gender, sexuality, and age. This conclusion was further supported by students’ own suppositions about the reasons for their victimization, which included their race, ethnicity, gender, sexuality, health, and disability (Bucchianeri, Gower, McMorris, & Eisenberg, 2016; McGee, 2014; Mulvey, Hoffman et al., 2018; Walton, 2018). Empirical support also emerged for non-bias-based risk factors, such as hazing/initiation rituals, substance use, and other risk-inducing behaviors (Bucchianeri et al., 2016; McGee, 2014; Mulvey et al., 2018; Walton, 2018). These results identify the specific kinds of victimization for which particular high-risk subgroups are at increased risk.
The sixth major set of findings consisted of correlations between the PVIC and known consequences of victimization, such as depressive symptoms, anxiety symptoms, stress, and sense of belonging. These correlations help to clarify the specific subtypes of peer victimization that are especially related to these outcomes. Depressive symptoms and elevated levels of stress were more often reported by students who experienced social exclusion. Anxiety symptoms were correlated with social exclusion, online, and physical victimization. College sense of belonging was lower among students who experienced greater peer pressure and social exclusion. Despite the relative paucity of research on peer-victimized college students, these outcomes are similar to those reported for more often studied samples of children and adolescents (Hawker & Boulton, 2000; Reijntjes et al., 2010).
Certain shortcomings and limitations of the current research suggest important avenues for future research. First, the design of our validation study was cross-sectional. Knowing whether depressive symptoms, anxiety, stress, and college sense of belonging are truly consequences or merely correlates of peer victimization will require longitudinal research. Second, our measures of depression and anxiety symptoms were based exclusively on self-report methods. Knowing whether peer victimization relates to major depression or anxiety disorders will require clinical interview methods. Third, we suspect that the PVIC (like almost all measures of stressful life events) will be subject to the memory-biasing effects of the respondent’s emotional status (Harkness & Monroe, 2016; Monroe & Simons, 1991), a potential confound that can be circumvented through the use of interview-based assessment methods (e.g., Brown & Harris, 1989; Hammen, 2006). Fourth, it is possible that a single complex episode of victimization (in which several kinds of victimization co-occur) could result in the endorsement of more than one PVIC item. This possibility (which also exists for most life event checklists) should be taken into consideration when interpreting raw PVIC scores. Fifth, we demonstrated a degree of discriminant validity in that PVIC scores were not seriously affected by intentional dissemblance; however, other threats to discriminant validity also exist that should be explored.
In sum, the current studies describe the development and validation of the first empirically derived, comprehensive measure of college-student peer victimization. Hopefully, our introduction of PVIC will enhance peer victimization research in this important population by both college officials and researchers. 3
This research was supported in part by Patricia and Rodes Hart. Contributions by Elizabeth Nick were supported by NIMH T32MH018921-26.
