Design and Validation of the Scale TDV-VP Teen Dating Violence: Victimisation and Perpetration [Violencia en Parejas Adolescentes: Victimización y Perpetración] for Spanish Speakers

Background: This study offers the design and validation of a scale for measuring violence in adolescent couples from the perspective of victimisation and perpetration for young Spanish speakers. Method: Validation study using exploratory and confirmatory factor analysis with online self-selected sampling and the participation of 422 subjects who met the requirements of being between 13 and 21 years old and currently or recently having a partner. Results: A scale of victimisation in adolescent partner relationships was obtained with 25 items and a scale of violence perpetration with 22 items. Both scales presented five factors: psychological violence, verbal violence, control, jealousy, and sexual violence. Significant differences were found between men and women in victimisation and perpetration of sexual violence. Conclusions: The Teen Dating Violence—Victimisation and Perpetration (TDV)—VP complies with the reliability and validity indices, constituting a very useful instrument for the detection and measurement of violence in Spanish-speaking adolescent couples in health-promotion work.


Introduction
Teen dating violence (TDV) refers to different types of assault from one partner to the other, intentionally and during courtship [1]. It reflects both the victimisation or violence suffered and the perpetration of the violence or violence committed. Violence between adolescent couples is a major public health and social justice problem [2]. Nowadays, the gravity of this phenomenon has a strong impact on Spanish society, as this type of violence manifests with a frequency and incidence of between two and three times greater than adult couples [3], although the violence recorded in young couples is usually less intense than that recorded in adult couples [4]. Courtship relationships may begin around adolescence, where the individual continues building his or her own maturity. Adolescents involved in TDV are likely to experience both victimisation and perpetration [5], suggesting that violence within relationships may be mutual. Experiences of victimisation in relationships can have long-term consequences, measurable in longitudinal studies, such as increased use of toxic substances and alcohol in episodic situations, depressive symptomatology, suicidal ideation by women, and increase in antisocial behaviour, suicidal ideation and marijuana use in men [6].
Evaluation of TDV is complex as it manifests in various forms, whether physical, psychological, or sexual violence [1]. Additionally, not all evaluation instruments consider all of these aspects, or they consider them under different categories, making a comparison between results complex and confusing [7]. Moreover, the difficulty of this phenomenon is aggravated by young people who catalogue violence when it manifests in a physical and visible way but allow the psychological and sexual aspects to go unnoticed. This fact may be both perpetration and victimisation. The main objective of this study is to design and validate an instrument for the detection of TDV from the perspective of victimisation and perpetration.

Sample
The research was based on the application of a questionnaire through the Limesurvey platform due to circumstances associated with the health crisis caused by COVID-19. To ensure the inclusion criteria, those who had never had a romantic relationship were asked to refrain from answering. From the 1112 adolescents who answered the questionnaire, the 446 who declared they had had an active relationship in the previous three months were selected. Likewise, those who did not fully complete the TDV-VP questionnaire and those who were outside the age range of 13-21 years were eliminated, as it is from midadolescence that individuals develop a greater self-reflective capacity and a complexity in moral statements.

Transcultural Translation and Dimensions
For the design of the instrument, a bibliographic review of relevant studies was performed. A total of 340 items were collected covering various dimensions: physical violence, psychological violence, sexual violence, harassment/cyberbullying and justification of violence. These items were commented on and analysed by a working group in TDV. A double translation (from English to Spanish and from Spanish to English) was performed, noting that several items matched semantically. It was conducted with the participation of an expert in sexology with English skills and a bilingual person with no knowledge of sexology. Then, to make a first selection in which some repeated items from different instruments were removed, a focus group composed of six experts in sexology and two experts in research methods was made to evaluate the adjustment of the items, leaving a total of 101 items for victimisation and 101 for perpetration of violence.

Validation by Judges
The 202 items were evaluated by 3 psychology experts in sexology and 12 teachers working at TDV, who determined the content validity index (CVI) of each item. A group of judges [26] was selected to analyse each of the test items in the following way: 1 = 'unnecessary'; 2 = 'useful'; 3 = 'essential'. Items with a low CVI were removed, and only those that exceeded the value of 0.51 were selected due to the criterion of having 14 or more experts, according to Lawshe [26]. Additionally, the formulation and wording of each item were evaluated by scoring the clarity from 1 = 'nothing' to 4 = 'many', rejecting items that did not exceed a minimum of 80% agreement and taking into account the experts' suggestions for improvement, leaving a total of 108 items, 54 for victimisation and 54 for perpetration.

Validation by Adolescents
A group of ten students participated in the validation of the scale. They were from the third year of secondary education up to the second year of baccalaureate, from a secondary education institute in suburban southern Spain with a large population of immigrants. A semantic validation was made in which terms and expressions were modified, and some items that were considered to be reflected in previous items were deleted, leaving a total of 92 items that were folded into 46 items of victimisation ('my partner did it to me') and 46 items of perpetration of violence ('I did it to my partner'). Items had five answer choices: 0 = 'never'; 1 = 'few times'; 2 = 'sometimes'; 3 = 'frequently'; 4 = 'always').

Field Study
The field study was then conducted with the original version of 46 victimisation items and 46 perpetration items. The survey was applied online by inviting youth associations through Facebook, via email and with the help of teachers from economics, law and education from their courses on the Blackboard platform.
Items with a move away from normal distribution were reviewed. When the outliers were studied through the Mahalanobis test in AMOS, many subjects that needed to be removed appeared, meaning the sample would have reduced by more than 100 subjects. Therefore, we preferred to do a study of normality through a study of the asymmetry and kurtosis of each item, in addition to the multivariate normality also offered by the AMOS program. We decided to remove some items because of the values of asymmetry and kurtosis [27]: V11: 'My partner threw some object at me with the intention of harming me', with a value of 5.7 in asymmetry and 34.3 in kurtosis; V13: 'My partner has physically assaulted me (slap, punch, kick, pulled hair...)', with a value of 3.7 in asymmetry and 18.85 of kurtosis, which helped us approximate the distribution of the data to normal. Additionally, the answers to some items were in their entirety 'Never', making it impossible to execute the factor analysis, as was the case with the V15 variables: 'My partner has used a knife or weapon against me' and V19: 'I lost consciousness from a blow my partner gave me in a fight'. Furthermore, the items of physical violence were not answered affirmatively by any young person, so it could not be included in the final validation. In this way, the final questionnaire (Appendix A) would have the most prevalent dimensions among young people, which are those focused on psychological and emotional aspects.

Statistical Analysis
The measurement of the reliability of the instrument was conducted using Cronbach's alpha test and by constructing validity through exploratory factor analysis (EFA) through the main components, administered for each of the two scales of violence. This technique was applied once discarded using the KMO test (Kaiser-Meyer-Olkin) and Bartlet's sphericity test, that the correlations between the items constitute an identity matrix. The rotation method used was Promax.
A confirmatory factorial analysis (CFA) was subsequently conducted via the maximum likelihood method. Although the sample did not present an absolutely normal multivariate distribution, the authors suggested the execution of this analysis [28,29], in addition to other authors who pointed out that this method is robust to non-compliance with this normal multivariate distribution requirement if the variables have a normal univariable distribution [30]. To evaluate the proposed models for victimisation and perpetration, Chi-Cuadrado ratio divided by the degrees of freedom, (CMIN/DF), Incremental Fit Index (IFI), Comparative Fit Index (CFI), Pratio Parsimonious Normed Fit Index (PNFI), Pratio Parsimonious Comparative Fit Index (PCFI), Root Mean Square Error of Approximation (RMSEA), Tucker-Lewis Index (TLI) and NFI were used [31,32]. Figure 1 shows the instrument validation procedure. In addition, the study is completed by calculating the Composite reliability (CR), the average variance extracted, AVE) and the discriminant validity. strument validation procedure. In addition, the study is completed by calculating the Composite reliability (CR), the average variance extracted, AVE) and the discriminant validity.
Analysis of violence in adolescent relationships was carried out through the Statistical package SPSS 25.0 (IBM, New York, NY, USA) and the AMOS program (IBM, New York, NY, USA).

Results
For the scale of victimisation, Bartlett's sphericity test was conducted. It was found that the level of significance (p ≤ 0.000) was less than 0.05, with a Chi-square value of 7907.37 (gl = 300) and 0.948 the sample rate of KMO. According to these results, the exploratory factor analysis was performed. Initially, seven factors appeared: one with a single item and another factor with two items. The factorial weights of the items were analysed and studied, regrouped into five factors after eliminating the corresponding ones. Thus, the exploratory factor analysis confirmed the existence of five factors that explain 71.07% of the variance through 25 items. The factors and reliabilities are presented in Table 1. It is suggested that an internal consistency value of 0.6 can be considered acceptable for scales with less than 10 items [33]. For the scale of perpetration of violence, Bartlett's sphericity test (p ≤ 0.000) was conducted, resulting in a Chi-square value of 3570.750 (df-231) and 0.896 the KMO index, so the Exploratory Factor Analysis (EFA) was made with the 22-item scale. Five factors were extracted that explained 57.925% of the variance.

Model Fit: Confirmatory Factor Analysis
The Chi-Cuadrado ratio divided by the degrees of freedom, CMIN/DF, was between values 2 and 5; Incremental Fit Indices (IFIs) and Comparative Fit Indices (CFIs) had a value greater than 0.92; Pratio PNFI and Pratio PCFI had a value greater than 0.7; RMSEA had a value less than 0.07 and TLI and NFI were greater than 0.92 [31].
The results for the victimisation scale obtained a good goodness of fit between the proposed model and the observed data, always complying with the recommendations of the author previously mentioned.
The results for the perpetration scale obtained an acceptable goodness of fit between the proposed model and the observed data. The IFI and CFI incremental adjustment values exceed the value 0.9, and the root of the average approximation quadratic residue (0.066) was even lower than that of the victimisation scale, which is favourable. The rest of the values were good, following the recommendations [31]. This provided two models with five factors whose goodness adjustment rates are presented in Table 2.  Figure 2 shows the confirmatory factor analysis path diagram on five proposed dimensions for victimisation and perpetration. Table 3 shows the Standardized Regression Weights (R) and the Square Multiple Correlations (R2) for the two models, victimisation and perpetration. The Square Multiple Correlations (R2) measure the percentage of common variance between the observable variable and its latent variable, and it is advisable that the values are greater than 0.50.  Table 4 shows the average variance extracted (AVE) of each of the victimisation scales studied was greater than 0.5, a critical value [34]. For this reason, it showed a good value in the five latent variables when used as an indicator of convergent validity, so that the constructs explained more than half of the variance of the respective indicators. It did not occur with these factors on the perpetration scale since the only value greater than 0.5 was obtained on the control scale. In the comparison between the square root of AVE (in bold) and the correlation between the scales, only in the sexual variable, both in victimisation and in perpetration, it was higher than the value of the correlation between them, which indicated that the models (victimisation and perpetration) did not show good discriminant validity.   The Square Multiple Correlations in the victimisation scale ranged from the value 0.321 to 0.814. The latent variable "Psychological violence" had two observable variables: "Turning friends against the partner" and "Removing comments, photos or videos of the partner from social networks because they made them jealous", with lower values but very close to 0.50, for what they were considered adjusted to this latent variable. The observable variables "Insisting on touches that are not pleasant to the partner or that the partner did not want" and "Treating the partner as a sexual object" had a percentage of common variance with the latent variable "Sexual violence" with values of 0.346 and 0.331, respectively. The rest of the observable variables showed a shared common variance with their respective latent variables greater than 0.5.

Composite Reliability, Average Variance Extracted, Convergent and Discriminant Validity
Regarding the scale of perpetration, the Square Multiple Correlations ranged between 0.191 and 0.678, with seven correlations greater than 0.5. There were three observable variables on the "Psychological violence" scale that shared between 0.2 and 0.3 of the variances with it. In "Verbal violence", the variable "Feeling that you can't talk to the partner because she's almost always mad at me" was the only one that presented the least common variance with its latent variable (0.191). The three observable variables of sexual violence presented low values of common variance with its latent variable. The victimisation scale was the one with the most common variance between the observable variables and their respective latent variables. However, the perpetration scale presented less common variance between the observable variables and the latent variables that compose it. Table 4 shows the average variance extracted (AVE) of each of the victimisation scales studied was greater than 0.5, a critical value [34]. For this reason, it showed a good value in the five latent variables when used as an indicator of convergent validity, so that the constructs explained more than half of the variance of the respective indicators. It did not occur with these factors on the perpetration scale since the only value greater than 0.5 was obtained on the control scale. In the comparison between the square root of AVE (in bold) and the correlation between the scales, only in the sexual variable, both in victimisation and in perpetration, it was higher than the value of the correlation between them, which indicated that the models (victimisation and perpetration) did not show good discriminant validity.

Differences in Sex
By performing the Kolmogorov-Smirnov normality test of the items that made up the scales of victimisation and perpetration (p = 0.000), the execution of Mann-Whitney U nonparametric tests for comparisons based on sex justified was performed. As for the comparison by reason of sex, significant differences were found between men and women only in terms of sexual violence. On the one hand, women recognised greater suffering from this violence, while men recognised greater perpetration of violent behaviours within the sexual framework of the relationship of the couple, as demonstrated in Table 5. On the other hand, these differences were considered small due to the value of r de Rosenthal [35] interpreted similarly to Cohen's d [36], so an effect size between 0.2 and 0.5 would be understood as small; between 0.5 and 0.8 would be moderate, and a value greater than 0.8 would be large [37].

Discussion
The objective of this study was to design and validate a scale for the detection and measurement of TDV-VP. Due to the results, it is stated that the victimisation scale meets good reliability and validity criteria for application in the Spanish-speaking youth population. In the case of the perpetration scale, the composite reliability had better scores than Cronbach's Alpha. The advantages of this phenomenon are that the composite reliability is a more complete criterion since it is not influenced by the number of items of the latent variable [34]. However, the perpetration scale obtained worse results than the victimisation scale that may be explained by social desirability [38], so it is important to include this phenomenon in future studies to control its influence. The TDV-VP provides a useful tool for the recognition of victimisation and the perpetration of violence, measuring all five aspects: psychological violence, verbal violence, control, jealousy and sexual violence. The population used in the validation of the instrument did not provide enough informa-tion to contemplate a sixth aspect in relation to physical violence, as this is contemplated by instruments such as CADRI (12) or CUVINO [7].
However, the TDV-VP examines sexual violence, which is considered one of the strengths of the scale in considering this type of violence as an indispensable factor in the measurement of TDV. Moreover, this instrument has focused on the violent aspects that are more present in teen dating relationships (psychological, control, jealousy and verbal violence). These types of behaviours are those that manifest themselves in a more masked and subtle way than physical violence and, therefore, they can be more difficult to detect by the adolescent and young population.
Neither of the two scales had good discriminant validity, and it can be explained by the conceptual closeness between the constructs of psychological violence, control, jealousy and verbal violence. These four dimensions presented high correlations with each other, but not with the sexual violence factor. This discriminant validity must be improved in future studies by analysing each dimension in detail.
Sexual violence has been the most controversial dimension according to the results for reasons of differences between the sexes. The results of our study have been directed in the same direction according to reasons of sex [11,39]. Little research has achieved similar results in the character of sexual violence within the TDV [40], with girls showing lower rates of perpetration than boys (3% vs. 10%) and higher rates of victimisation than boys (14% vs. 8%). One factor studied in relation to the prevalence of sexual violence is the consumption of pornography. Exposure to violent pornography was associated with all types of TDV, although some patterns differed by gender [14,41]. Similarly, men who scored higher in what they called 'male hostility' and were frequent consumers of pornography were more likely to report sexual coercion [42]. The importance of contemplating sexual violence is indisputable. There is a strong need to address the consequences of shortcomings in the educational system regarding comprehensive sex education, especially from the perspective of equality and sexual diversity.
The absence of significant relationships according to reasons of sex with control, verbal violence, jealousy and psychological violence was considered to be remarkable. Neither men nor women saw a pronounced difference in the experience of victimisation and perpetration of those types of violence. This could be explained by the normalisation and naturalisation of these behaviours within the relationships of adolescents. Verbalemotional violence [39] is the most likely kind of violence experienced by young people between the ages of 14 and 20. In fact, one of the most frequent types of aggression was insulting or ridiculing in networks, which coincides with the most prevalent behaviour of verbal behaviour [39,43]. The relationship between victimisation and perpetration sheds light on the experience of mutual violence in relationships of couples involved in TDV [5], which can also be explained by the intergenerational transmission theory of violence. It would be interesting to analyse violence factors, such as relationships between adolescents and their parents, from this perspective. Two-way violence could also be explained as self-defence, among other reasons [1]. This bidirectionality of violence would justify the high and moderate correlations of these four dimensions in both victimisation and perpetration. However, in the case of sexual violence, a different phenomenon occurs, and it is related to an imbalance of power between the aggressor and the victim [11] that derives from sexist and patriarchal beliefs and a lack of sexual education.
With the rise in new technologies, new areas where violent acts are committed have emerged. Nowadays, control of one partner over the other has shifted to digital screens. Surveillance and control is one of the most common forms of violence in adolescent relationships, and one of the variables that contribute to the explanation of cyberbullying in dating relationships is jealousy [43], which, under the myth of romantic love, justifies this type of behaviour [39].
Setting the point of view on control across networks depending on sex, it was found that just over a quarter of young people in a current or recent relationship having experienced some form of victimisation of network abuse in the previous year [44]. In addition, women suffer more victimisation than men, especially sexual abuse. One in ten young people acknowledged that they had perpetrated cyber-dating abuse. In contrast, young males were significantly more likely to report cyber-dating sexual abuse. Victims of cyber-dating sexual abuse were seven times more likely to have suffered sexual coercion.
The main limitations of this study lie in the length of the scale for both the validation by judges and for the collection of data of the participants. These may have led to tiredness or exhaustion, which may have influenced the responses. For example, there was a question whose answers were distractors, and it was answered. A possible way to overcome this would be a second validation by judges and a reduction of items. Additionally, the data collection was originally going to be in schools in the province of Almería.
Another limitation of the present study was that the sample was generally female. Because the sample was collected online, the explanation may be given by greater maintenance of privacy by men and greater openness and predisposition to share personal experiences and feelings by women.
The absence of affirmative responses in the items on physical violence, and their disappearance in the validated TDV-VP scale may respond to different explanatory factors. On the one hand, it may be related to a change in the forms of violence experienced in adolescence. It may, in turn, be influenced by the moralisation and social stigmatisation of the issue of intimate partner violence that leads young people to deny the practice of violence (social desirability). On the other hand, it can be greatly affected by the characteristics of the sample participating in the study, which is made up of a group mainly of women, middle class, indigenous, who voluntarily participated in research with a sample through social networks (volunteer bias).
However, due to COVID-19, it was adapted into an online survey, and some 21-yearolds were kept in the sample due to this low participation situation. In this way, despite the fact that the sample of sexual diversity is not wide, it is necessary to study the phenomenon of differences by sex and sexual orientation in future studies. For this reason, a search for a larger homosexual and bisexual sample will be considered in future investigations to address aspects such as intragender violence. With regard to other research, a multicultural sample should be available to analyze the perspectives and beliefs around intimate partner violence and assess the limitations of this questionnaire in a cross-cultural way.

Conclusions
The study of violence in couple relationships must be adapted to new platforms where violent behaviours occur, such as social media. This investigation has highlighted aspects of violence that occur through physical contact, such as sexual violence, as well as behaviours that are manifested in novel ways, such as control and cyberbullying.
While gender violence is an issue that requires research, education and prevention, the study of domestic violence-which encompasses the experiences of couples outside of relational hegemony-can not be overlooked.  Institutional Review Board Statement: The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Institutional Bioethics Committee of University of Almeria (UALBIO2020/003, 21/02/2020).

Informed Consent Statement:
Informed consent was obtained from all subjects involved in the study.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author. 19. Mi pareja siente que no puede hablar conmigo porque casi siempre estoy enfadado/a con él/ella. (My partner feels that he/she cannot talk to me because I am almost always angry with him/her.) 20. He forzado a mi pareja a practicar alguna actividad sexual cuando él/ella no quería.
(I have forced my partner to practice some sexual activity when he/she did not want to.) 21. He insistido en tocamientos que no le son agradables o que mi pareja no quería. (I have insisted on touching that is not pleasant or that my partner did not want.) 22. He tratado a mi pareja como un objeto sexual. (I have treated my partner as a sexual object.)