Methods
Sample
Study procedures were approved by the institutional review board of The Ohio State University, including procedures to ensure the confidentiality and anonymity of subjects' responses (e.g., data were immediately stripped of identifying information). The analytic sample comprised 585 subjects ages 18 to 21 enrolled at The Ohio State University, recruited in two data collection efforts. Subjects completed a one-time only online survey to assess current health and retrospective dating violence histories from age 13 to 19 (described below). The recruitment procedures were as follows:
Study 1, conducted from March, 2011 to April, 2011, involved randomly sampling 730 students from the registrar's office, from a total of 32,716 students age 18 to 21 enrolled at Ohio State University. Using students' university email account, we sent a recruitment email, which included the study description and link to the online survey. Two follow-up reminders were sent by email, three and seven days after the initial email. The cumulative response rate at each recruitment email was as follows: initial email (31.6%, 231/730); second email (41.0%, 300/730); final email (46.7%, 341/730). Subjects who completed the survey were credited $20 to their university account. Of the 341 subjects who completed the survey, 44 were excluded because they were older than age 21 (n = 7) or because they never had a dating partner from age 13 to 19 (n = 37). After these exclusions, the eligible analytic sample comprised 297 subjects who had a dating partner from age 13 to 19 (n = 190 females; n = 107 males).
Study 2, conducted from May, 2011 to March, 2012, involved recruiting students ages 18 to 21 enrolled in four undergraduate Human Development and Family Science courses at The Ohio State University. Through an introductory email which included the study description and survey link, instructors offered completing the survey as extra credit coinciding with the coverage of relationship-related topics in class. A total of 311 students completed the survey; according to the class rosters, this represented 98% of eligible students. Of the 311 students, the eligible analytic sample comprised 288 subjects who had a dating partner from age 13 to 19 (n = 255 females and 33 males). The over-distribution of females in the sample is consistent with the gender distribution in the Human Development and Family Science undergraduate program.
Our current analysis combined the two samples, totaling 585 subjects (n = 297 from the first study plus 288 from the second study). The over-distribution of females in Study 2 (recruited from undergraduate classes) compared to Study 1 (recruited through the university registrar) represented the main difference between the two samples, with otherwise similar characteristics between the two samples due to our narrow eligibility criteria (e.g., subjects had to be between the ages of 18 and 21).
Survey
Health and Health Behaviors. To reduce response bias, subjects were first asked about health before they were asked about dating violence victimization. Asking subjects details about dating violence first, which could be a traumatic experience, could potentially cause bias in their responses to the health items; specifically, subjects might provide lower health ratings if the experience of completing the dating violence questions was traumatic.
Depressive symptoms over the last two weeks were assessed using two questions from the nine-item Patient Health Questionnaire, which have a sensitivity of 74% for detecting depressive symptoms among adolescents relative to the Diagnostic and Statistical Manual of Mental Disorders. The two questions included:
Having little interest in doing things;
Feeling down/hopeless.
The two depression questions were scored as separate items, not as a composite measure; the response options for each question ranged from 0 (not at all) to 3 (nearly every day). In the analysis, the item responses were collapsed into a binary category (0 = not at all; 1 = experienced symptoms on any day).
Unhealthy/disordered eating behaviors were assessed using three questions from the Youth Risk Behavior Surveillance System, including whether subjects ever:
Fasted, vomited, or took diet aids to lose weight (each its own separate question).
Response options to the questions were binary (yes/no).
Binge drinking and smoking were assessed using questions from the Youth Risk Behavior Surveillance System, including:
Ever smoked daily for 30 or more days;
Had 5 or more drinks on six or more days over last month.
Response options to the questions were binary (yes/no).
Sexual behavior was assessed by asking about subjects' ever engagement (yes/no) in vaginal/penile intercourse, oral sex, and anal sex, and the number of partners they engaged in the sexual activity with (subjects reported the number of partners). The cut point for frequent oral and vaginal sex was defined as having five or more partners. Anal sex was less common, so the definition included having had any anal sex. While these measures are likely to indicate sexual health risk from sexual transmitted infections and unwanted pregnancy, they may not constitute equivalent risk.
Relationship and Dating Violence Histories. We used a method similar to the timeline follow-back interview to assess dating violence histories retrospectively from age 13 to 19. We previously used this method to document domestic violence and child abuse histories in more than 4,000 women and men. While retrospective dating violence assessment may result in mis-estimation of abuse due to recall bias, retrospective assessment is the field's standard for capturing adolescent dating violence experiences and our assessment method used memory prompts to facilitate recall. The timeline follow-back interview method has been used extensively to capture other risky health behaviors, such as drug and alcohol use. First, to establish relationship histories, subjects were asked whether they had a dating, romantic or sexual partner between age 13 and 19; this could include a boyfriend/girlfriend, someone the subject liked romantically or was involved with sexually but did not consider to be a boyfriend or girlfriend, or someone the subjected "hooked up with". They were then asked specific details about their three most recent partners, starting with their most recent partner, including the partner's gender, the age the relationship began and ended, and the partnership type (e.g., boyfriend/girlfriend). We used memory prompts, such as asking the subject to remember the year they were in high school, to facilitate recall of the age that a relationship began and ended. For operational practicality, we asked details about subjects' three most recent partners. After we asked subjects detailed questions about their three most recent partners, we asked about the total number of partners subjects had beyond those three from age 13 to 19.
After information about subjects' relationship history was gathered, dating violence victimization was assessed retrospectively using eight questions covering the three core conceptual areas of intimate (including dating) violence (physical, sexual, and non-physical) outlined by the Centers for Disease Control and Prevention (Table 1):
Our eight questions were adapted from the CDC's Youth Risk Behavior Surveillance System, Foshee and Swahn's studies, and Coker's dating violence survey currently being administered in a CDC-funded intervention study (unpublished data, personal communication with Dr. Coker). Additionally, our questions included newer forms of dating violence/abuse, including harassment/stalking through text messaging and email. Initial validity data from the eight dating violence questions we used in the present study were presented at the Women's Health Congress in Washington, D.C. in March 2013, and the validation manuscript is undergoing peer review; in brief, a confirmatory factor analysis of the eight dating violence questions showed that the questions loaded onto the hypothesized conceptual abuse factors (physical, sexual and non-physical abuse).
For each question in Table 1, subjects were asked whether they ever experienced dating violence between age 13 and 19. Subjects who responded with "yes" were considered exposed to that abuse type. We created the following exposure groups based on prior studies that have conceptually and empirically examined physical and sexual violence within a single category, and psychological abuse only in a separate category.
Physical and/or sexual dating violence exposure included subjects who reported experiencing physical and/or sexual types of dating violence. Subjects in this group could also have exposure to non-physical abuse. We included exposure to sexual pressure involving either (or both) verbal and physical coercion, as verbally coerced sexual acts have been shown to have lasting trauma for victims.
Non-physical only dating violence exposure included subjects who were not exposed to physical/sexual dating violence, but who reported experiencing any of the non-physical dating violence types listed in Table 1. For ease in reporting, this group is referred to as the "non-physical TDV" group.
Non-exposed subjects included those who reported never experiencing physical, sexual, or non-physical types of dating violence from age 13 to 19.
Other (Non-dating) Abuse Exposures. Using three questions from the Centers for Disease Control, we asked about 1) whether subjects had ever been bullied between ages 13 and 19 (1 question); and 2) whether subjects experienced other types of abuse before age 18, including being punched, kicked, choked, or receiving a more serious physical punishment from a parent or other adult guardian (1 question) and being touched in a sexual place or being forced to touch another person when they did not want to (1 question).
Analysis
All analyses were gender stratified. Chi-square tests were used to compare health indicators for subjects who reported any dating violence victimization with those who reported no victimization. Generalized linear models with a log link and robust sandwich variance estimators were used to obtain prevalence ratios (PRs) for each dichotomous health indicator for exposed compared to unexposed subjects, using a modified Possion regression approach. Logistic regression models were not used because the health outcomes were not rare, and the odds ratios from these models would not closely approximate relative risks (or equivalently, prevalence ratios). The analyses investigated the effects of the type of dating violence experienced (physical/sexual versus non-physical only) on health indicators; specifically, regression coefficients compared health indicators in subjects with physical/sexual dating violence exposure compared to never-exposed subjects, and compared those with non-physical only dating violence compared to never-exposed subjects. We fit unadjusted and adjusted models; the adjusted models were adjusted for age, bullying victimization from age 13 to 19, and other non-dating physical and sexual abuse before age 18 (see Methods section and footnote in Table 2 for definitions)—all of which are theoretically linked and empirically associated with health impairments. Analyses were completed using Stata statistical software, version 12.0.