The study included a total of 6,138 adolescents and young adults (ages between 15 and 26 years; M=18.5, SD=2.09) of both genders (39.6% males, 60.4% females) from five Spanish provinces (Sevilla, 17.9%; Asturias, 22.9%; Huelva, 57.1%; Coruña, 1.1%; and Pontevedra, 1.4%). Differences on average age for males (M=18.0) and females (M=18.3) were under 1 year (the theoretical measurement error), so they were considered negligible. Only people that had had at least one intimate partner for more than one month in their lifetime took part in the study. All participants were enrolled in secondary (75.8%) or university (24.2%) studies. There was found a significant relationship between sex of respondents and academic level (χ21= 273.24, p<.001, Cramer's V=.211), with a higher proportion on females among university students (78.4%) than in secondary (55.9%) levels. These figures are compatible with demographic information available for Spanish student populations.

DVQ. The original DVQ (Rodríguez-Franco et al., 2010) was included in the study. It measures with 42 items different abuses that could take place in intimate relationships, providing information about perceived frequency of victimization and perceived disturbance using a 5-point Likert scale (from 0-never to 4–all of the time). It measures dating violence across eight domains of abuse: Detachment, Humiliation, Sexual, Coercion, Physical, Gender-based, Emotional punishment, and Instrumental. Although the original questionnaire provides both a measure of perceived victimization (i.e. frequency) and a measure of perceived disturbance, we relied in this study only on the frequency of the victimization scores.

Response bias. The lie-scale of the reduced and revised Eysenck Personality Questionnaire (EPQ-RA;Eysenck& Eysenck, 1985; adapted to Spanish by Ibáñez, Ortet, Moro, Ávila, & Parcet, 1999) was used to evaluate to what extent participants were deliberately attempting to control their responses. The lie-scale is a 6-item scale in a no/yes response format. Responses were summed up and averaged (M=0.51, SD=0.27).

Sampling. Authors invited to take part in the study to all educational centers based on provinces where the research team had some collaborator (A Coruña, Pontevedra, and Principado de Asturias in the north; Huelva and Sevilla in the south of Spain), sending postal and/o electronic mails to valid addresses from official registers. The final sample included data gathered in educative centers that accepted to participate in the study. Participants were explained the main objectives of the study and, also, they were informed that they could drop off from the study at any moment without any penalties.

Provided that IPV represents a main concern in Spanish education, it was already included in academic curricula as a transversal theme to deal with. Anyway, in order to fulfill ethical requirements regarding minor participants (under 18 years old in Spain), schools provided information and asked for explicit consent among parents or legal tutors of potential participants, and assent among adolescents. In the case of people over 18 years old, informed assent was asked before starting the assessment. Anonymity was ensured by developing the evaluation in class groups and delivering analysis only for the complete samples, never for individual cases. Researchers included contact information to give response to any possible discomfort or doubts associated to the study.

Participants were told to select a single dating relationship in order to respond to the DVQ-R. More specifically, instructions asked to “select the most problematic dating relationship you have ever experienced or, if you have had none like this, choose the most important for you”.

Development of short version of DVQ. Although the original questionnaire has shown adequate psychometric characteristics in several countries (so far: Spain, Mexico, and Argentina, Rodríguez-Franco et al., 2010; Italy et al., 2015; and US, López-Cepero et al., 2016), we considered that it could be significantly shortened if several criteria were applied. First, the gender-based scale (including items 3, 11, 17, 19, and 35) measures behaviors directed to the opposite sex (in broad terms) and not necessarily to the couple. Second, the instrumental and emotional punishment scales were also removed due to potentially content overlap with other scales (8, 16, 24, 4, 12, and 28). For instance, item 8 ‘Refuses to have sex with you or give you affection to express his/her anger/annoyance’ might be regarded as a combination of the sexual and coercion scales. Third, some items were removed due to their potential content overlap with other items present in the same scale (7, 17, 18, 22, 29, 31, 34, 36, 37, and 42) (among similar items, those with lower factor loadings in exploratory factor analysis were not retained for further analyses). To obtain a brief, useful but still valid measure, 20 items were finally retained, with four items in each of five the domains of abuse: physical, sexual, humiliation, coercion, and detachment (see Appendix).

Measurement Model. First, we tested the measurement model of the 20 items of the DVQ-R through a series of confirmatory factor analysis. The measurement model was first estimated in a random independent sample of 3,069 participants and next cross-validated in the remaining 3,069 participants. The initial model was a 5 correlated factors model. In this model, all items loaded in their corresponding factors, all factors were correlated and no factor cross loadings were allowed for any of the 20 items. Model fit was far from adequate: S-B χ2160=551.60, p< .001, robust TLI=.81, robust CFI=.84, robust RMSEA=.028, 95% C.I.=.026, .031. Inspection of the Lagrange Multiplier Test suggested that releasing four fixed parameters should significantly improve model fit. All of these fixed parameters were initially set to zero and corresponded to correlation among item error terms. After a closer inspection of these correlated error terms we found that all of them corresponded to within-factor item error terms. These covariations were among items 1-9, 3-15, 20-21, and 40-41 of the original scale (see Appendix). Freely estimating these parameters significantly improved model fit: S-B χ2156=305.69, p< .001, robust TLI=.94, robust CFI=.95, robust RMSEA=.018, 95% C.I.=.015, .021. This final model was also estimated using data from the random independent sample (N= 3,069) and model fit was adequate: S-B χ2156=223.15, p< .001, robust TLI=.94, robust CFI=.95, robust RMSEA=.017, 95% C.I.=.011, .021. To obtain more accurate parameter estimates, the final model was estimated using the complete sample (N=6,138) (see Figure 1): S-B χ2156=462.12, p< .001, robust TLI=.94, robust CFI=.95, robust RMSEA=.018, 95% C.I.=.016, .020.

Figure 1.

Standardized parameter estimates for the Measurement Model of the 20 items of the DVQ-R (N=6,138). All p‘s<.001.

(0.24MB).

Figure 1 presents standardized parameter estimates for final model. The internal consistency of the five scales was between .64 and .74. Breakdown of descriptive information regarding DVQ-R scales is included in Table 1. As for the frequency of victimization, it was found that detachment was the most common, followed by coercion, while the scales of sexual and physical violence obtained the lowest scores.

To test for the factorial invariance of the instrument we conducted a series of multigroup analyses for males and females. First, a highly constrained model was initially estimated simultaneously for the two groups. This model imposed that all factor loadings, all factor correlations and all error term covariations were equal across groups. This model showed an adequate fit to the data: S-B χ2339=694.29, p< .001, robust CFI=.93, robust RMSEA=.018, 95% C.I.=.017, .020. Inspection of the Lagrange Multiplier Test results for releasing constraints across groups showed, however, that four parameters were significantly different for male and female participants and should be released. The final model with the released cross-group constraints showed a better fit to the data: S-B χ2335= 637.40, p< .001, robust CFI=.94, robust RMSEA=.017, 95% C.I.=.015, .019. The Likelihood Ratio Test (LRT) between these two models showed that the model with the four released constraints fitted significantly better to the data: ΔS-B χ2=56.89, Δ d.f.=4, p< .001. LRT was not significant when releasing additional constraints. The released cross-group constraints were the covariation of one error term, two item loadings, and one covariation between factors. The covariation of errors for the items has beaten you and has slapped your face, pushed or shaken you was greater for men (.072, p< .001) than for women (.022, p<.01). The unstandardized relationship between the item forces you to undress even if you don’t want to and the sexual factor was greater for men (1.148, p< .001) than for women (.780, p< .001). The unstandardized relationship between the item ridicules or insults you for the ideas you uphold and the humiliation factor was lower for men (.960, p< .001) than for women (1.112, p< .001). Finally, the covariation between coercion and physical factors was greater for men (.058, p< .001) than form women (.049, p< .001). Despite these trivial differences in models for men and women, all the estimated parameters were statistically significant and factorial invariance for men and women might be tenable. Overall, results from the multigroup analyses showed that the 20 items of the DVQ-R reflected an invariant structure across sex, thus adding generalizability to the general model.

In a first step, univariate ANOVA was carried out in order to compare victimization frequencies reported by males and females. Physical, sexual and coercion scales showed statistically significant differences, although only two of them (physical violence and coercion) reached an appreciable (although small) effect size.

In a second step, relationships between victimization and sex, age and educational attainment were analyzed, including the bivariate relationships among age and educational attainment and the DVQ-R scale scores. First, the five scale scores of the DVQ-R were submitted to a Multivariate Analysis of Variance (MANOVA) to ascertain if dimensions of the DVQ-R were different across sex: Wilk's λ=.963, F (5, 6132)=47.26, p< .001, η2=⋅037)⋅ Male and female participants scored significantly different on the scales of the DVQ-R. Univariate tests showed that men scored significantly higher on the physical (F=160.68, p< .001, men=.16; women=.06), coercion (F=86.03, p< .001, men=.55; women=.42), and sexual (F=26.69, p< .001, men=.19; women=.14) scales. Bivariate Pearson's correlations showed that age was positively related to humiliation (r=.04, p< .001) and negatively related to sexual (r=-.04, p< .001) and coercion (r=-.05, p< .001), although these relationships were almost trivial. Finally, given the significant relationship between sex and age (r=.06, p< .001) we re-estimated the MANOVA of sex on DVQ-R scales now including age as another factor. To do so, we clustered the age variable into two groups, as suggested by two-step cluster analysis results. The first group (15-18 years-old) included 65.9% of participants and the second group (older than 18 years-old) included 34.1% of participants. Results from MANOVA indicated a significant interaction effect sex*age groups on DVQ-R scale scores: Wilk's λ=.997, F (5, 6130)=47.26, p=.002, η2=⋅003)⋅ Thus, although we previously found that men scored higher on several DVQ-R scale scores, we should be cautious about this result given that; a) female participants were older than males; and, b) older participants scored lower on various DVQ-R scales.

We submitted the lie-scale scores of the EPQ to a two-step cluster analyses to estimate the number of potential clusters that better classify participants. Results showed that three clusters better described the distribution of lie-scores across participants. A first group of potentially heavy liars ranged their scores from 0.60 to 1.00 (41.3%); a second group of potentially moderate liars ranged their scores from 0.33 to 0.50 (39.4%); and a third group of non-liars ranged their scores from 0.00 to 0.17 (19.3%). Next, we analyzed if sex, age and educational background were statistically related to the belonging to the three groups. Chi-square analyses revealed that there were more men in the potentially heavy liars group than expected (1200 counted and 995 expected, p< .001). The reverse tendency was found for women: there were more women in the non-liars group than expected (776 counted and 716 expected, p< .001). Univariate Analyses of Variance (ANOVA) revealed that participants in the group of potentially heavy liars were significantly younger than non-liars (17.99<18.27, p< .001). Also, potentially heavy liars had lower educational attainment (M=3.76) than moderate liars (M=3.86, p=.014) and non-liars (M=4.05, p< .001). Finally, we performed a Multivariate Analysis of Variance to estimate differences in the DVQ-R scale scores across groups of liars. Results showed significant differences in the DVQ-R scores across groups of liars: Wilk's λ=.996, F (10, 11412)=2.49, p=.005, η2=⋅002⋅ Given the previously found relationship among sex, age groups and groups of liars, we re-run the analysis now including sex, age group, and group of liars as factors. This inclusion led to a non-significance of the differences previously found of the DVQ-R scores across groups of liars: Wilk'sλ=.997, F (10, 11402)=1.58, p=.10, η2=⋅001⋅ Overall, DVQ-R scale scores seemed to be free of reporting bias once age and sex were taken into account.