Online sexual harassment and negative mood in Croatian female adolescents

In an era of heightened concerns about adolescent exposure to OSH, we sought to address a gap in understanding possible association between OSH and negative mood in the period between middle and late adolescence, and the role of resilience. The following two research questions were explored: (RQ1) does baseline frequency of OSH predict negative mood levels 2 years later and (RQ2) does resilience buffer the association between the two constructs? Following the literature and previous analyses that used the same dataset [17], we focused on female adolescents because of their higher risk of exposure [5], greater likelihood of distress [3], and higher vulnerability to negative mood [18].

A panel sample of high-school sophomores, aged 15.8 (SD = 0.48) years at baseline, was recruited from 14 secondary schools in Rijeka, the third largest city in Croatia,¹ as a part of a longitudinal study of the role of sexually explicit material in adolescent sexual socialization and sexual health (for more details about the study see [17]). For logistical reasons, only larger schools (with 63% of the city’s second-grade high-school student population) were selected in the sample. The initial survey (n = 1287) took place in December 2015 (Time 1) and was repeated five times in 5–6-month intervals. The final wave (T6; n = 892) was carried out in March 2018. The notable reduction in panel size was mostly due to the fact that students who enrolled in a 3-year vocational school program completed their education and left school before T5. For the current study, we only used data from female students who participated in both T1 and T6 (n = 477).

Attrition rate reflected school absenteeism (due to illness, educational field trips, obligatory out-of-school practical courses, and truancy) as well as occasional difficulties in questionnaire linking (approx. 1.4 questionnaires per class per wave). We carried out multivariate logistic regression analysis to investigate possible bias introduced by attrition. Independent variables were age, mother’s and father’s education, academic achievement, resilience, and reported frequency of OSH and negative mood symptoms at baseline. Two significant differences emerged between female students who participated in both T1 and T6 (coded 1) and those who did not (coded 0). Compared to the latter, the former group was characterized by higher odds of reporting better grades (AOR = 1.70, p = 0.000) and somewhat more educated mothers (AOR = 1.36, p = 0.035).

Data were collected by a self-administered paper and pencil survey administered in classrooms. To maximize confidentiality, large screens were placed between participants. In addition to being printed on the first questionnaire page, information required for students’ consent was also delivered by a research assistant. Contact information for a youth psychological health center was provided at the end of the questionnaire, which took about 20–25 min to complete. Prior to the study launch, all parents/carers were sent a leaflet with basic information about the study. All study procedures were approved by the Ethical Research Committee of the Faculty of Humanities and Social Sciences, University of Zagreb.

Symptoms of depression and anxiety were also assessed at T1 and T6 with the Patient Health Questionnaire for Depression and Anxiety (PHQ-4 [19]), a brief 4-item measure which was used to assess symptoms of depression (two items, e.g., Little interest or pleasure in doing things) and anxiety (two items, e.g., Feeling nervous, anxious or on edge) experienced in the 2 weeks preceding the survey. A 4-point scale ranging from 1 = not at all to 4 = nearly every day was used to record the frequency of symptoms. The construct was characterized by good reliability (Cronbach’s α was 0.86 at both T1 and T6), and had acceptable stability over time (r = 0.43).

Resilience was measured at baseline, T4, T5, and T6. It was assessed by a shortened version of the Connor-Davidson Resilience Scale [20]. This brief 5-item measure (e.g., I can deal with whatever comes, I can stay focused under pressure, and I think of myself as strong person) was pre-tested in a preparatory pilot study carried out in early 2015. A 5-point Likert-type scale, ranging from 1 = it never relates to me to 5 = it always relates to me, was employed to record answers. Following the notion that resilience is not a static individual trait [14], but a dynamic process [21], a composite variable was built by averaging the five items across data collection waves. Over time, the indicator had satisfactory reliability (Cronbach’s α ranged from 0.75 at baseline to 0.79 at T6) and stability (r = 0.73–0.53). To reflect middle to late adolescents’ resilience, the four composites were then averaged into a single one. For multi-group testing, this global resilience indicator was categorized (i.e., tertilized) into three roughly equally sized groups: (1) low resilience, (2) moderate resilience, and (3) high resilience.

Cross-lagged path analysis was used to explore associations between OSH and negative mood over the period of 26 months. The analytical procedure consisted of several steps. As the precondition for exploring structural associations, the key constructs’ measurement models were tested for configural, weak factorial, and strong factorial invariance across time by applying progressively more restrictive equality constraints [22]. The standard test of Chi-square difference (Δχ²) was used for comparisons of nested models. After confirming strong factorial invariance of depression and anxiety symptoms and partial strong factorial invariance of OSH, we explored the two-wave cross-lagged path analytic model presented in Fig. 1. Next, the model was controlled for participants’ resilience at baseline; the composite indicator was added to the model and connected to all other latent constructs (OSH at T1 and T6 and negative mood at T1 and T6). Finally, as recommended by Fergus and Zimmerman [14], a multi-group analysis was employed to address the protective role of resilience by directly comparing paths between OSH and negative mood among female adolescents characterized by low, moderate, and high resilience. Model fit was evaluated using χ², CFI, and RMSEA statistics. Following standard guidelines for longitudinal structural equation assessment, TLI and CFI values ≥ 0.95 and RMSEA values ≤ 0.05 were taken to represent excellent fit [22]. Due to multivariate non-normality (the distribution of OSH was particularly skewed), path analytic models were bootstrapped with 5000 resamples to produce 95% confidence intervals around the relevant estimates [23].

Considering that < 2% of information was missing on the key variables, full information maximum likelihood (FIML) estimation was used to deal with missing data [24]. Data nestedness in school classes was disregarded based on the finding that only 3% of variance in female adolescents’ OSH and < 7% of variance in negative mood was explained at the level of class.

All statistical analyses were carried out using AMOS 24 and IBM SPSS 24 software packages.

The majority of participants (80.3%) reported living with both parents at baseline. The majority of participants reported that their mother and/or father were secondary educated (56.9% and 60.5%, respectively). A college-educated parent was reported by over a third of participants. A substantial minority (28.9%) of students reported attending religious services at least once a month (16.2% reported no religiosity). At baseline, 18.0% of students were sexually active.

Over a third of participants (37.5%) reported (lifetime) OSH at baseline. At T6, OSH in the past 6 months was reported by 15.7% female adolescents. Overall, 42.6% of participants experienced OSH at some point. On average, exposure to OSH in this panel sample was low at both data collection points (M_T1 = 4.33, SD = 2.37 and M_T6 = 3.54, SD = 1.75). Importantly, the average levels of depression and anxiety symptoms decreased between T1 (M = 9.51, SD = 3.37) and T6 (M = 8.52, SD = 2.98).

Table 1 shows bivariate associations between the key constructs. Of the four possible associations between OSH and negative mood, only the one between the two constructs at T6 was statistically significant, albeit small (r = 0.18). There were no associations between adolescent resilience and OSH, but resilience was consistently and negatively related to depression and anxiety symptoms reporting (r = − 0.25 and r = − 0.26). The effects were small to moderate in size.

A multivariate, path analytic assessment of links between OSH and negative mood is shown in Fig. 1. In the model, which was characterized by adequate fit to the data (χ²(73) = 187.29, TLI = 0.954, CFI = 0.968, RMSEA = 0.057 [90% CI 0.047–0.068]), the key constructs were significantly associated at both baseline (r = 0.16, p = 0.003) and T6 (r = 0.17, p = 0.001). However, cross-lagged paths were not significant, indicating that baseline levels of OSH were not predictive of reported depression and anxiety symptoms 2 years later and vice versa. Bootstrapping the model with 5000 resamples confirmed this pattern of (non) significant structural links.

With resilience controlled for in the path analytic model (χ²(83) = 206.79, TLI = 0.951, CFI = 0.966, RMSEA = 0.056 [90% CI 0.046–0.066]), the associations between T1 and T6 levels of OSH and negative mood remain significant and almost identical in size. Corroborating bivariate findings, adolescents’ resilience was significantly related to depression and anxiety symptoms at T1 (β = − 0.30, p = 0.000) and T6 (β = − 0.15, p = 0.004), but not to OSH.

To address the question of whether resilience can buffer the association between the key constructs, we carried out a multi-group path analytic assessment, in which groups represented participants with different resilience levels (Table 2). At bivariate level, the three groups differed significantly in the reported OSH at T6 (F = 3.38, p = 0.035) and depression and anxiety symptoms at both T1 (F = 14.78, p = 0.000) and T6 (F = 28.68, p = 0.000), with higher levels of OSH and negative mood observed in the low compared to the high resilience group. The multi-group model, which was characterized by acceptable fit (χ²(108) = 481.24, TLI = 0.903, CFI = 0.914, RMSEA = 0.052 [90% CI 0.045–0.059]), pointed to substantial differences in the associations between OSH and negative mood among the three groups. In the low resilience group, associations between OSH and negative mood were significant both at T1 (r = 0.22, p = 0.034) and T6 (r = 0.24, p = 0.027)—unlike in the high resilience group, where neither reached statistical significance. The moderately resilient participants were characterized by a non-significant association at T1 and a significant relation at T6 (r = 0.32, p = 0.003).

The current study has several strengths, including repeated measurements over a developmentally important 26-month period and the use of robust and reasonably powered analytical approach. In addition, there was no evidence of relevant attrition bias beyond educational attainment. However, there are also several limitations. Apart from self-reported nature of our data, it should be noted that the measure of OSH differed in its timeframe; at baseline we asked for lifetime experience, while at T6 we inquired about the past 6 months. Our study also did not distinguish between distressing and non-distressing OSH, although the measure did focus on ‘unwanted’ experiences, which the literature specifies as problematic. Future studies should address this issue and how it may reflect on the role of resilience. Finally, compared to lengthier indicators, the brief measure of depression/anxiety symptoms used in the current study is characterized by a more limited variability. Although this may have affected the estimation of key associations, the marginal effect sizes observed in the current study suggest that the limitation is minor.

Future research could also explore a few other plausible moderators of the observed cross-sectional association between OSH and negative mood. Of particular relevance might be peer support, which can be assessed in detail using ego-centered network analysis [31], illicit drug and alcohol use [25], and the parallel experience of (cyber)bullying [32]—which may amplify emotional reactions to OSH.