Methods
Participants
The study population consisted of a representative sample of Swedish high school seniors in their third and last year at Swedish high school when most were 18 years old. In Sweden, about 91% of all 18-year-old adolescents are enrolled in high school [
17]. The Swedish agency, Statistics Sweden, selected schools that might participate based on information from the Swedish National School Register. Stratification was made on the basis of school size and educational programs (20 programs ranging from those with a vocational profile to those designed to prepare students for entrance into a university) as indicated by data in the National School Register for second year high school student, in the fall term, 2013. One or two study programs were selected from each school.
A total of 13,903 adolescents from 261 of 1215 Swedish high schools were selected and of the 261 schools 238 met the criteria for selection in 2014. An additional sample from Stockholm County was selected using the same selection criteria. The response rate for Stockholm county was lower (48.7%) than for the rest of the country (65.3%). Differences were also seen regarding the size of schools. In Stockholm, fewer of the respondents came from schools with 10–190 pupils (13.9%) compared to the rest of the country (22.1%) and more often came from middle-size schools with 191–360 pupils (51.2%) compared to the rest of the country (41.6%), resulting in a small effect size (Cramer’s V = .10). Few differences were found between the sample from Stockholm and the rest of the country, so answers from Stockholm were used in this study.
Finally, 171 schools with 9773 adolescents agreed to participate in the study and 5873 students in these completed the questionnaire. Thirty-four questionnaires were excluded due to unserious answers or a high amount of missing data, leaving 5839 satisfactory questionnaires. This gave a response rate of 59.7%. The mean age of the participants was 17.97 (SD = .63). An additional 124 questionnaires were excluded since the index question, “Have you gotten to know anyone on the internet during the last 12 months that you had sex with online?” was not answered. The final sample consisted of 5715 adolescents. Participants who answered that they had felt persuaded, pressed or coerced when having sex online (sexually abused online) during the last year, constituted the index group and all other adolescents constituted the reference group.
Procedure
The national agency Statistics Sweden distributed and collected the questionnaires. Information about the study was sent to the principals of the selected schools by mail in August 2014. Questionnaires were answered in digital format by entered answers into computers in 165 schools, where computers were not available, students filled in paper copies of the questionnaire (six schools). A reminder was sent to the schools that had not delivered data by the end of the first month. Information about the study was given to the principals and to the teachers in charge when the questionnaires were to be filled. Students gave their informed consent for participation by answering the questionnaire. All participating students received written information about where to turn for help and support if needed at any time after the day on which they had submitted the completed questionnaire.
Measures
The questionnaire used in the present study was a modified version of a questionnaire used in two previous studies carried out in 2004 and 2009 (Svedin and Priebe [
18,
19]). It comprised 116 main questions. Questions concerned socio-demographic background, experiences of abuse, and risk behaviors. In addition, three standardized instruments measuring relationships with parents and psychosocial health were used.
Socio-demographic background
Demographic questions were drawn up for the purpose of the study (listed in Table
2a). The adolescents self-reported the demographic information.
Abusive experiences
Sexual abuse was measured using the question: “Have you been exposed to any of the following against your will”, followed by six examples (someone flashed in front of you, touched your genitals, you masturbated someone, vaginal, oral, vaginal or anal penetration). The answers were analyzed in two categories, any sexual abuse (all questions) and penetrative abuse (oral, anal or genital penetration), see Table
2b.
Emotional abuse was measured using the question: “Have you prior to the age of 18 been subjected to any of the following by an adult”, with these three examples: been insulted, threatened to be hit, or been isolated from friends, see Table
2b. Participants who answered “yes” to one or more of the questions were considered victims of emotional abuse.
Physical abuse was measured using the same wording used for emotional abuse, but with eight examples of physical abuse (Table
2b). Participants who answered “yes” to one or more of the questions were considered victims of physical abuse.
Relationships with parents
The Parental Bonding Instrument [
20,
21] is an instrument that measures an individual’s perception of parental styles during childhood. The instrument consists of 25 items, where 12 relate to the subscale “care” and 13 relate to the subscale “overprotection”. The response options are presented on a 4-point scale, from “very like” to “very unlike”. The total score for “care” ranges from 0 to 36 and from 0 to 39 for “overprotection”. Items assess perception of maternal and paternal behaviors separately. PBI has been evaluated as an attachment instrument with strong psychometric properties in a review by Ravitz et al. [
22]. Cronbach’s alpha for mother care in the present sample was .87, and for father care .89. Mother and father overprotection were .84, and .78, respectively.
Self-
esteem was measured by the Rosenberg self-esteem scale [
23]. The instrument measures self-esteem using 10 items with four possible answers, ranging from “strongly agree” to “strongly disagree”. The total score varies between 0 and 30, with high scores corresponding to high self-esteem. In the current sample, Cronbach’s alpha for the total scale was .90.
Trauma symptoms were measured using the Trauma Symptom Checklist for Children [TSCC:
24,
25]. The questionnaire includes 54 questions that can be divided into six categories: anxiety, depression, post-traumatic stress, sexual concerns, dissociation and anger. Response options are “never”, “sometimes”, “often” and “almost all of the time”. Cronbach’s alpha in the present sample was .95 for the full instrument and .79–.88 for the six subscales.
Risk behaviors
Health-risk behaviors were measured using questions related to sexual or non-sexual risk-taking. Non-sexual risk-taking was measured with questions about use of alcohol and drugs, see Table
5.
Sexual risk-
taking behaviors were measured using questions about age of onset for sexual debut and having had more than six sexual partners, see Table
5.
Internet behavior was measured with questions about time spent on the internet and seven questions mainly about sexual behavior on the internet during the last year, see Table
5.
Pornography consumption was measured by two questions, see Table
5.
Data analyses/statistics
Bivariate statistical analyses were performed using Pearson’s Chi square statistics on categorical variables. Kolmogorov–Smirnoff test was performed to examine whether the PBI, Rosenberg, and TSCC scales (totals and subscales) could be assumed to be normally distributed. As these tests indicated that they were not normally distributed, bivariate analyses on these variables were performed using Mann–Whitney’s U test.
Furthermore, as there were too many variables to be included in a multiple logistic regression model, the number of variables to be included in a “final model” was reduced by performing stepwise multiple logistic regression analyses for each main table separately (each table identifies different group of factors that could be associated with sexual abuse on the internet, Table
4 excluded), Table
6.
All analyses were performed using SPSS, version 22.0 (IBM Inc., Armonk, NY). A p value < .05 (two-sided) was considered statistically significant.
Ethics
The study was approved by the Regional Ethical Review Board of Linköping (Dnr, 131–31).
Discussion
To our knowledge, this study is the first to study adolescents with experiences of online sexual abuse by a person they had met online and where they had felt persuaded, pressed or coerced. The results of the study can be summarized in four main findings.
First, the study showed that most sexual contacts online were positive experiences with persons of about the same age or only slightly older. However, previous studies have shown that having a sexual relationship with a person met online can be viewed as a risk behavior since this kind of contact increases the risk of facing negative consequences later, for example receiving unwanted sexual approaches [
12]. Similar reasoning has been put forward by Baumgartner et al. [
14,
26] in defining online sexual risk behaviors as the exchange of intimate sexually insinuating information and material with someone only known online. In the current study, 5.8% of the adolescents had had sexual experiences online with a person they had only met online, and of those, 9.7% reported that they had been persuaded, pressed or coerced meaning that they, by definition, had been sexually abused online. Girls were more often the victims and for girls, the perpetrators were generally older.
Second, there were no significant differences in socio-demographic background between the index group and the reference group. This result can be compared to studies on children victims of online grooming [
13] or adolescents sending nude images [
5] were it was also found that the socio-demographic background did not differ from children without these experiences.
Third, the adolescent victims of online sexual abuse had backgrounds with significantly more numerous and/or varied experiences of different forms of abuse including physical, psychological as well as sexual abuse, especially penetrative sexual abuse than those who had not been victims of online sexual abuse. Earlier findings indicate that the more severe the form of sexual abuse the more serious the subsequent associated health issues will be, with penetrating child sexual abuse at the upper end of the scale of severity [
27]. This study underlines these earlier findings but also adds to our knowledge that online abuse per se is also associated with poor health, low self esteem and a poorer relationship between parent and child. As concerns health, as measured by TSCC, online sexual abuse only was associated with poorer health, at least on the same level as offline sexual abuse only, with those students who had been sexually abused both online and offline scoring highest, supporting the polyvictimization model [
28].
These results are also supported by earlier studies [
15,
16,
29‐
31] stating that online sexual victimization, also including cyberbullying, are associated with adverse emotional and psychological consequences. In the current study, the final multiple logistic regression model showed that online sexual abuse was strongly associated with depression. This is in line with the results from studies focusing on youth who had sent sexual pictures (sexted), where both Van Ouystel et al. [
32] and Dake et al. [
33] found an association between sexting and depression. In the study by Temple et al. [
34] associations were also found between sexting and depression in their unadjusted models, but not when prior sexual behavior, age, gender, race, ethnicity, and parental education had been adjusted for. It is, however, important to bear in mind that the studies referred to above do not examine if the motivation factor for sending the images was, for example, sending the image just for fun and with no negative consequences afterwards or if it was because of coercion leading to the taking and sending of the image.
Fourth, adolescents abused online also had more online risk behaviors such as sharing personal information significantly more often, looking for someone online to talk sex with, or posting nude pictures on a community site. These behaviors might increase the risks of later being a victim of online sexual abuse [
17].
The results in the study should be read in light of the following limitations. The response rate was rather low at 59.7%. Part of this can be explained by the fact that on a typical day 10% of students of this age are absent from school. An assumption is that the absent group probably would have added some individuals to the index group and thereby affected the results slightly, since people dropping out from research more often come from families with poorer support and are more often burdened with psychosocial health issues and lower motivation to participate in school surveys [
35]. On the other hand, other studies that have found little evidence for substantial bias as a result of nonparticipation [
36]. Recall bias is always a limitation in questionnaire-based studies, as is the question of whether the answers are trustworthy. All answers were reviewed before the analyses and 34 questionnaires were excluded due to unserious answers. Another limitation is the small size of the index group which may cause low statistical power. The main concern regarding study power arises when the index group is separated into two groups. When comparing these two groups to the reference group, statistical significance is detected, even though the power is well below 80%. However, in all but one comparison between the two subgroups (SA internet, SA offline and internet) no statistical difference was detected. Having a larger power would probably result in more statistically significant findings. The implication of the low power is that we underestimate rather than overestimate the presence of actual differences between the groups.
Finally, the index question did not contribute to any additional probing to determine what online sexual activities or sexual abusive behaviors respondents might be referring to when they endorsed these items, nor did it allow them to describe the behavior further. It would have been conceptually interesting to have a fuller description and examples from respondents.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.