Identifying Victim Types
We conducted a series of latent profile analyses (LPA) using tidyLPA in R (Rosenberg et al., 2018
). LPA compares participants on continuous variables to assign them to mutually exclusive groups. We used four variables: self-reported victimization, and the peer nomination scores for victimization, most popular, and least popular. Fit is determined by comparing models on various statistical information criteria, with the Bayesian information criterion (BIC) considered the most accurate indicator of the number of classes (Nylund et al., 2007
). Examination of the possible classes revealed that the LPA did not identify distinct groups of victims and non-victims; some classes contained both victims and non-victims. Additional information regarding this LPA for the full sample can be requested from the first author.
To address this problem and to ensure identification of distinct groups, we created a binary variable from self-reported victimization to differentiate between victims and non-victims. Participants were categorized as victims if their combined victimization experience across all forms yielded a frequency of approximately 2 to 3 times a month, which is the recommended and widely used threshold to identify victims (e.g., Solberg & Olweus, 2003
). This resulted in approximately 1/3 of the sample (34.8%) being identified as a victim (self-reported victimization = 1) and the remaining 2/3 of the sample being identified as a non-victim (self-reported victimization = 0) which is comparable to prevalence rates from other studies (e.g., Modecki, Minchin, Harbaugh, Guerra, & Runions, 2014
). Following this process, separate latent profile analyses were conducted for victims and non-victims using three variables: the peer nomination scores for victim, most popular, and least popular (see Table 2
for full model comparisons).
Model fit indices for latent profile analyses specifying one to nine latent classes
For participants who self-reported being victimized (n
= 280), model fit continued to improve significantly up until nine classes using a bootstrap likelihood ratio test, which tests the model fit between k
-1 and k
models (Nylund et al., 2007
). In addition to comparison of fit indices, it is recommended to compare the classes that emerge from each model to identify whether they are interpretable and sufficiently large (Nylund-Gibson & Choi, 2018
). On this basis, we chose the three-class model. Overall entropy for the three-class model was 0.99, indicating that the three groups were homogenous.
A similar pattern was found for youth who did not self-report any victimization (n
= 524; see Table 2
). This resulted in a total of six victim type groups: three for participants who self-reported being victimized and three for participants who did not self-report victimization.
presents the means of the six groups on the three clustering variables. Convergent victims (Group 1; n
= 31, 3.9%) self-reported being victimized and scored above average on peer-reported victimization. They scored below average on most popular and above average on least popular. Self-identified victims with high popularity (Group 2, n
= 63, 7.8%) self-reported being victimized but scored below average on peer-reported victimization. They scored above average on most popular and below average on least popular. Self-identified victims with average popularity (Group 3, n
= 186, 23.1%) also scored below average on peer-reported victimization. However, they scored below average on both most and least popular. Peer-identified victims (Group 4, n
= 65, 8.1%) did not self-report being victimized but scored above average on peer-reported victimization. This group scored below average on most popular and above average on least popular. Finally, two types of non-victims were identified with varying levels of popularity. Non-victims with high popularity (n
= 78, 9.7%) did not self-report being victimized and scored below average on peer-reported victimization. They scored above average on most popular and below average on least popular. Non-victims with average popularity (n
= 381, 47.4%) did not self-report being victimized and scored below average on peer-reported victimization. This group scored below average on both most and least popular.
Comparison of independent and dependent variables by victim subgroups identified with LPA
| || || || || || |
| || || || || || |
Loneliness and Self-esteem of Victim Types
To examine group differences in loneliness and self-esteem, we conducted multilevel mixed-effects linear regression with maximum likelihood estimation to account for the nested nature of the data with students nested in classrooms. We ran two unconditional models, one for loneliness and one for self-esteem, to assess the amount of variance between and within classrooms. The intraclass correlation coefficient (ICC) was 0.009 for loneliness and 0.017 for self-esteem, indicating that between 0.9% to 1.7% of the variance in the dependent variables was between classrooms.
To test hypotheses regarding group differences, we ran a series of models, changing which victim group served as the reference group. To account for these multiple comparisons, we used the Holm-Bonferroni p
-value adjustment to assess whether group differences were significant (Abdi, 2010
; Holm, 1979
). Analyses controlled for gender (0 = boys, 1 = girls) given significant gender differences in victim group membership in the current sample, χ2
= 804) = 12.233, p
= 0.032, with more boys being popular non-victims and more girls being non-victims with average popularity than expected. Table 3
presents the adjusted means post model estimation, accounting for gender.
As expected, convergent victims reported significantly more loneliness than all other groups, ps < 0.001. Self-identified victims with average popularity had the second highest levels, significantly higher than the remaining groups (ps < 0.001). Self-identified victims with high popularity reported significantly less loneliness than self-identified victims with average popularity, p < 0.001. Popular self-identified victims reported significantly more loneliness than popular non-victims (p = 0.004) but did not differ from average popular non-victims (p = 0.059). Peer-identified victims were significantly lonelier than both non-victim groups (ps < 0.001). As expected, there was no significant difference in loneliness between the non-victim groups with high and average popularity (p = 0.061).
Convergent victims and self-identified victims with average popularity had similar levels of self-esteem (p = 0.73), which were significantly lower than peer-identified victims and both non-victim groups (ps < 0.002). Popular self-identified victims had lower self-esteem than non-victims with average popularity, but the difference failed to reach the adjusted p-value threshold of 0.0056 for significance (p = 0.008). In fact, popular self-identified victims did not significantly differ from any other group in self esteem (0.017 < ps < 0.073). There were no significant differences in self-esteem between popular non-victims, average popular non-victims, and peer-identified victims (ps < 0.958).
In the primary analyses, self-reported victimization was treated as a binary variable because the full-sample LPA was unable to identify distinct groups of victims and non-victims with the continuous self-reported victimization variable. Accordingly, the primary analyses did not differentiate the extent to which youth self-reported victimization. As an exploratory analysis, we also identified groups of victims and non-victims using cut-off scores (> 0.5 SD
above the mean) for self- and peer-reported victimization. Consistent with past research (e.g., Dawes et al., 2017
; Scholte et al., 2013
), we identified convergent victims (n
= 43), self-identified victims (n
= 154), peer-identified victims (n
= 57), and non-victims (n
= 550). There was more variance in popularity for self-identified victims and non-victims (SDs
= 1.07 and 1.01, respectively) than for convergent and peer-identified victims (SD
s = 0.22). Of the 154 self-identified victims, 44 were high in popularity (>M
+ 0.5 SD
). Of the 550 non-victims, 134 were high in popularity. This resulted in six victim types (see Table 4
): convergent (5.3%), self-identified with high popularity (5.5%), self-identified with average popularity (13.7%), peer-identified (7.1%), popular non-victims (16.7%), and average popular non-victims (51.7%).
Comparison of independent and dependent variables by victim subgroups identified with cut-off scores
| || || || || || |
Peer nominated victimization
| || || || || || |
We again conducted two separate multilevel mixed-effects linear regression for loneliness and self-esteem. We still controlled for gender, even though the results indicated no significant gender differences in group membership, χ2
= 804) = 8.472, p
= 0.132. We accounted for multiple comparisons with the Holm-Bonferroni p
-value adjustment (Abdi, 2010
; Holm, 1979
). Table 4
presents the adjusted means post model estimation. The results were very similar to the primary analyses.
Convergent victims had significantly higher levels of loneliness than all other groups (ps < 0.005). Self-identified victims with average popularity status reported the second highest levels of loneliness, significantly more than the other victim and non-victim groups (ps < 0.001). Peer-identified victims had the third highest level, more than the non-victim groups (ps < 0.001). There was no significant difference in loneliness between peer-identified and self-identified victims with high popularity (p = 0.879). The non-victim groups with high and average popularity were less lonely than popular self-identified victims (ps < 0.002) and did not differ from one another (p = 0.025; Holm-Bonferroni adjusted p-value for comparison p < 0.025).
For self-esteem, convergent victims, popular self-identified victims, and self-identified victims with average popularity had significantly lower levels of self-esteem than the other groups (ps < 0.002) but did not significantly differ from one another (ps > 0.25). The two non-victim groups and peer-identified victim group did not differ significantly in self-esteem (ps > 0.627).