Data for this study were drawn from the Quebec Youths’ Romantic Relationships Survey. The primary goals were to document the prevalence of dating violence, explore associated risk factors and mental health outcomes in high school adolescents ages 14 to 18. The survey was completed through a one-stage stratified cluster sampling of high schools in 2011-12. Schools were randomly selected from an eligible pool from the Quebec Ministry of Education. To obtain a representative sample of students in grades 10 through 12, schools were first classified into strata according to metropolitan geographical area, status of schools (public or private schools), teaching language (French or English) and social economic deprivation index.

Data were collected in 34 schools. Depending on the teaching language of the school, the survey was completed by students either in English or in French. A double-translation technique was used to translate English measures into French. Class response rates and the overall student response rate were determined as the ratio between the number of students that accepted to participate and the number of solicited students. Response rate was 100% for the majority (320/329) of classes; while for the remaining, the response rate ranged from 90% to 98%. The rate of partial nonresponse was less than 3.5% for the majority of variables and no additional adjustment was made for nonresponse as bias and loss of power are likely to be inconsequential (Tabachnick & Fidell, 2013). In addition, analyses of the partial nonresponse did not reveal any specific pattern. A total of 8,230 teenagers (Grades 10, 11 and 12) responded to the self-reported questionnaire. After data verification, 36 participants were excluded for invalid or completely missing data for a final sample of 8,194 participants. Participants were given a sample weight to correct biases in the non-proportionality of the schools sample compared to the target population. The weight was defined as the inverse of the probability of selecting the given grade in the respondent's stratum in the sample multiplied by the probability of selecting the same grade in the same stratum in the population. A weighted sample of 6,531 youths resulted and is used in the further analyses.

The questionnaire administered in class required approximately 40minutes to complete. Two research assistants explained the aims of the survey and assisted participants if needed. Students agreed to participate on a voluntary basis and signed a written consent form. Students were provided with a list of resources (phone help lines, websites, community organizations, health services, etc.). The institutional review board of the Université du Québec à Montréal approved this study.

Sociodemographic information was collected on gender, age, family structure (living with two parents under the same household, living with two parents in different households (shared custody), living with one parent, other family structure arrangements), main language spoken at home (French, English or other), education level and ethnicity of parents.

Dating violence experienced in the past 12 months was assessed among youth who reported a current or past-year dating relationship. Psychological TDV (3 items), physical TDV (3 items) and threatening behavior (2 items) were assessed with items from an adapted version of the Conflict in Adolescent Dating Relationships Inventory (CADRI). Six (6) of these items were derived from a short version of the CADRI (Wekerle et al., 2009) while the two (2) remaining items were from the long version of the CADRI (Wolfe et al., 2001). Sexual TDV was measured by 9 items adapted from the Sexual Experiences Survey (SES) (Koss et al., 2007). Only acts occurring specifically during a conflict or argument with the dating partner in the last 12 months were surveyed. Response options were: Never, 1 to 2 times, 3 to 5 times, and 6 times or more.

PTSD symptoms and emotional distress following TDV were assessed by asking participants to refer to the most difficult situation they had ever experienced in the past 12 months. PTSD following TDV was measured using 6 items adapted from the Children's Impact of Traumatic Events Scale II (CITES-II) (Wolfe, 2002). Items reflected core dimensions of PTSD including reexperimentation (“I have dreams or nightmares about what happened”), avoidance (“When I’m reminded of what happened I try to think of something else”), hypervigilance (“Loud or sudden noises make me jump or feel afraid”) and distanciation (“I feel distant from other and I do not feel as interested in things I used to enjoy”). Participants indicated whether statements were “Not true”, “Sometimes true”, or “Very/Often true”. A total PTSD score ranging from 0 to 12 was computed. Internal consistency was satisfactory (α=.70).

Two (2) items were used to assess emotional distress following the most difficult situation experienced among the listed behaviors: “I was upset by what had happened to me” and “I was scared”. The first item originates from a study by Sears and Byers (2010), while the second item was designed for the present study. Responses of both items were on a three-point scale: Not true, Sometimes true and Very/Often true. Given the correlation between the two items was low (r=.40), we considered them separately in the analysis. A dichotomized score for each item was computed according to whether the participant reported the statement was sometimes or very true (1) or not (0).

Data analyses were conducted using STATA (StataCorp, 2011) complex survey command and Mplus (Muthén & Muthén, 2015). Prevalence rates with 95% confidence intervals (CI) and TDV chronicity in the past 12 months were computed. Prevalence rates were obtained by dichotomizing items and scale scores (0 vs. at least one occurrence). Second, we computed chronicity scores for both items and scale scores to describe the patterns of TDV recurrence. Chronicity scores indicate how often the set of acts measured by each item and scale occurred. They were obtained by computing the mean of items included in each form of TDV. As suggested by Straus, Hamby, and Warren (2003), chronicity scores were computed for those who reported an episode of victimization.

A zero-inflated regression was used to predict both the expected zero values and the counts of reported dating violence experiences. Counted scores were computed by summing the midpoint for each category, as proposed by Straus et al. (2003). Items were scored 0 (never), 1.5 (1-2 times), 4 (3-5 times) and 10 (6 times or more). Zero-inflated models, deemed more accurate in modeling violence frequency data with a preponderance of zeros (Swartout, Thompson, Koss, & Su, 2014), assume two distinct populations, that is participants who are likely to report TDV and those who are not. A logistic portion of the model aims at identifying the variables that distinguish those reporting TDV than the true-zero TDV group. The count part of the model aims at identifying variables significantly associated with TDV frequency using a Poisson or Negative Binomial regression. Choosing the best model between a zero-inflated Poisson (ZIP) and a zero-inflated Negative Binomial (ZINB) depends on the dispersion of the count data. Teens reported 0 to 30 events of psychological and physical TDV (psychological TDV: Mean=1.46, Variance=11.80; physical TDV: Mean=0.77, Variance=7.33). Threatening behaviors values ranged from 0 to 20 (Mean=0.26, Variance=2.24) and sexual TDV values ranged between 0 to 90 (Mean=1.11, Variance=25.67). Because overdispersion was observed in our data, as evidenced by variance estimates greater than the means in the TDV counted scores, we performed a ZINB model (Preisser, Stamm, Wong, & Kincade, 2012). Lower values of AIC for ZIBN compared to ZIP approaches confirmed that ZINB models provided better fit to the data. A regression model was performed for each form of TDV, including gender (being a girl coded as 1) as the independent variable and the following sociodemographics as controls: mother and father level of education, language spoken at home, family structure, parents’ ethnicity, partner's gender, age and the square of age.

Statistical analyses exploring possible gender differences in PTSD and emotional distress following TDV were also performed. When analyses were conducted on dichotomous variables (dichotomized TDV items and each form of TDV), a design-based test with Fisher statistics was used to examine differences between included variables. For continuous variables (PTSD score), linear regressions were used to examine differences between girls and boys as t-tests are not supported with the complex design command in STATA. Indeed, when considering two variables in which the dependent variable is continuous, linear regressions can be used as an alternative to t-tests.

Sociodemographic characteristics of participants are presented in Table 1. The majority (76.4%) reported French as the main language spoken at home and close to half (58.6%) lived with both parents, while 38.6% lived either in single-parent families or in shared custody and 2.8% described another living arrangement. A total of 2,094 (55.4%) girls and 1,344 (48.8%) of boys reported having a dating relationship in the last 12 months. Due to missing values, sum of categories may not reach the total of participants.

Prevalence and 95% CI of the four different forms of TDV and endorsement for each item for teenage girls and boys are reported in Table 2. Results showed statistically significant differences in prevalence rates according to gender for each form of TDV. Girls were more likely than boys to report experiencing psychological TDV (56.4% vs. 45.8%, F(1, 26)=43.01, p < .001), physical TDV (15.7% vs. 12.8%, F(1, 26) =15.88, p < .001) and threatening behaviors (6.8% vs. 3.9%, F(1, 26)=11.05, p = .003). Gender disparity was especially salient for sexual TDV, with girls nearly 3 times more likely to report being victimized by a dating partner in the last 12 months (20.2% vs. 5.7%, F(1, 26)=116.28, p < .001). The prevalence rate of girls experiencing at least one episode of any form of TDV was also significantly higher (62.7% vs. 49.5%, F(1, 26)=66.68, p < .001).Chronicity scores presented in Table 2 show that girls were more likely to report greater frequency of all forms of TDV experiences, with the only exception being that boys were more likely than girls to report 6 or more events of physical TDV. Table 2 also provides information on specific TDV items contributing to explain the gender patterns in all forms of TDV.

Table 3 presents the regression models estimates controlling for sociodemographic variables. Teen girls were less likely to belong to the true-zero TDV group for psychological TDV (β=-0.35, p<.001), physical TDV (β=-0.36, p<.001) sexual TDV (β=-1.45, p<.001), as well as threatening behaviors (β=-0.86, p<.001). The TDV frequency increased among girls for psychological TDV (β=0.17, p=.003) and sexual TDV (β=0.48, p=.013), while gender did not discriminate frequency of physical TDV (β=-0.01, p=.947) and threatening behavior (β=0.06, p=.700).

Results on PTSD symptoms and emotional distress associated with the most difficult episode of TDV experienced in the past 12 months are presented in Table 4. Analyses revealed a significant effect for gender, t(26)=2.7, p = .01, with girls (M=2.5, SE=0.1) reporting higher scores of PTSD symptoms associated with TDV than boys (M=2.1, SE=0.16). The percentages of teenagers endorsing “being upset” and “being scared” were respectively 33.8% and 23.5% among participants, with girls endorsing significantly higher prevalence for both items of emotional distress following the most difficult episode of TDV.