Risk factors related to intimate partner violence police recidivism in Spain.

Background/Objective: Some professionals, such as police officers, are required to prevent violent behavior, such as intimate partner violence (IPV). For this task they use actuarial tools designed to estimate the risk of occurrence of further violence after a previous complaint (police recidivism), taking into account risk and protective indicators which they can observe, in spite of they are not behavioral assessment experts. Method: To try to refine the police risk assessments carried out in Spain since 2007 and to improve the two tools available on the Spanish VioGén System, Police Risk Assessment and Risk Evolution (VPR3.1 and VPER3.0), this paper, using an epidemiological design, in a sample of 6,613 new cases of IPV of Spain, studies empirical relationships among 65 indicators (56 risk and 9 protection) and IPV police recidivism up to six months.
Results: It resulted in a recidivism rate of 7.4%, finding statistically significant associations of 46 indicators. Conclusions: Empirical evidence about static indicators and new relevant dynamic indicators in the victims' police protection management is presented. Practical implications for future police risk assessments are discussed.

Violence against women within partner relationships has been awakening tremendous social sensibility, and is now considered a public health issue. Since 2004, Spain has had an Organic Law for the Integral Protection against Gender Violence (L.O. 1/2004), which defines gender violence as a phenomenon carried out by the man against his female partner or ex-partner. However, this definition, while being included in this organic law, has not been accepted by the Real Academia Española (2004). Because of this, this study will use the term intimate partner violence (IPV).

Within this legal frame, institutions heavily involved in IPV (Delegación del Gobierno para la Violencia de Género, 2015b) state that in 2015, 129.193 IPV complaints (1.98% higher than in 2014) were registered. Other data taken from macro surveys about the prevalence of IPV in Spain (Delegación del Gobierno para la Violencia de Género, 2015a) shows that 10.3% of women that are 16 or older have suffered physical violence at the hands of their partners or ex-partners, 8.1% have suffered sexual violence, and 25.4% have suffered psychological violence sometime in their lives. These facts are made aware to the Security Forces (FCS) in 26.8% of cases, and in 1.7% of cases they are directly stated at court, which shows that institutions are only aware of 28.5% of this type of violence. According to the National Statistics Institute, in 2015, women who were victims of IPV and made use of precautionary measures or protection orders were a total of 27.624, 2% more compared to the year before (Instituto Nacional de Estadística, INE, 2015).

In Spain, the annual crime recidivism prevalence rate is between 9 and 17%. It is important to remember that the term “recidivism” is not homogeneous and that each type of recidivism (self-inculpation, penitentiary, judicial, police, penal, etc.) has different rates and characteristics; official records, judicial and penitentiary records to be precise, as well as those that were declared to the police, partner or ex-partner reports and the underlying mechanisms of the aggression being the most commonly used in IPV cases (Babcock et al., 2004, Chereji et al., 2012, Novo et al., 2012).

There are difficulties when comparing IPV recidivism data from different studies, due to the diversity in the methods used, the legal definition of the concept, the type of violence, the size of the sample used and the monitoring periods, which vary according to the objectives of the study, and can range from 3 months long to 10 years long, when studying criminal trajectories (Klein & Tobin, 2008). In global terms, it is estimated that there is a recidivism rate of 20%, even though the values are quite varied (between 8-60%) (Loinaz, 2014, Ortega et al., 2009), and, in the specific case of gender aggressors, recent meta-analyses which were based on data from official records have found a prevalence of 21%, and those that were based on partner reports found a prevalence of 35% (Arias et al., 2013, Babcock et al., 2004). It seems that the critical period or the period of highest recidivism is found within the first six months (Gondolf, 2002, Kingsnorth, 2006). In the spanish Integral Gender Violence Monitoring System, also known as VioGen System (Secretaría de Estado de Seguridad, n.d), administered by the Ministerio del Interior, it is found that out of 378,645 registered cases between 2007 and 2015, 68,430 were recidivism cases (18.07%, and 29.62% of those were cases of multirecidivism). It was also observed that police recidivism was at 6.66% at three months, at 9.04% at six months, at 11.5% at a year, and at 14.65% at two years.

The study of recidivism is important because it allows us to identify factors and their risk and protection indicators that are really associated with the phenomenon, contributing to a greater knowledge about violence, about the risk scenarios and about the variables that intervene. These elements allow more efficient prevention strategies to be designed. A risk or protection indicator can be anything that occurs before a result and which correlates with it, without necessarily implying that the indicator and the result have a cause-effect relationship (Skeem & Monahan, 2011). Violence risk factors are those which increase the probabilities of a behavior taking place (which doesn’t mean they have to be determining), and can be of the historical type (static, difficult to modify) or the dynamic type (can be modified). It seems that dynamic risk factors possess a slight superiority when it comes to predicting violence (Andrews & Bonta, 2010). Protection indicators are those which regulate the impact of the exposition to risk indicators. Because of this, when analyzing violence it will be necessary to take into account both factors.

Therefore, identifying empirically validated risk and protection indicators constitutes the first step in the construction of efficient risk assessment instruments (Rossegger et al., 2013, Skeem and Monahan, 2011), and using those indicators which show a greater relationship with recidivism will help better the precision of the identification of high-risk cases, discarding the ones with low risk and gaining more efficient protection (Kropp, 2008, Stith and McMonigle, 2009). For this empirical testing of the differential characteristics between recidivism and non-recidivism cases, epidemiologic study designs of the prospective and retrospective kind prove to be the most adequate for the empirical testing of the differential characteristics between recidivism and non-recidivism cases (Quinsey et al., 2006, Rodríguez-Díaz et al., 2016, Ureña et al., 2015, Vega-Gea et al., 2016).

In this way, systematic reviews (Capaldi et al., 2012, Cattaneo and Goodman, 2005) group factors together, integrating family influences, peer relationships, psychological and behavioral variables, and cognitive factors such as hostile attributions and belief systems. Risk indicators such as age, high levels of stress, and indicators related to general violence and crime are very present in analyzed studies, being considered traditional indicators, and finding more peripheral ones such as the exposition to family violence and childhood abuse, which are moderately associated with IPV. Maintaining antisocial behavior at a young age is shown to be a potent precursor of violence against women during adulthood (Woodward, Fergusson, & Horwood, 2002). Within the field of psychopathology, there is a confirmed presence of antisocial traits, substance abuse and depressive symptoms in aggressors, both men and women. A high degree of conflict within the couple and low satisfaction also appear as robust indicators.

From an evolutionary standpoint, the rise in age shows evidence in favor of protection, as is typical of criminal trajectories (Garrido, Strangeland, & Redondo, 2006). Suicidal ideas are also a good predictor of violence in longitudinal studies (Kerr & Capaldi, 2011). Evidence, however, does not support more proximal indicators such as the mere existence of a denunciation for a violent episode, or its gravity; but it does support indicators such as the existence of an escalation in the violent episode (Capaldi et al., 2012). In this way, they explain that many of the cases brought to the police's attention could be caused by a one-time reaction and that after minimal legal intervention they do not result in similar cases in the future, if they are not associated with other indicators.

In Spain we find works Capdevila, 2009, Capdevila, 2015 that analyze the penitentiary recidivism of IPV aggressors, and they confirm that the indicators shared by the main protection scales, in a national and international context, are very related to the characteristics of the aggressor: previous aggressions towards their partner, previous violence unrelated to the partner (criminal trajectory), having previously broken a sentence, substance abuse, threats, mental disorders, and cognitions that maintain or justify the violence.

Developing strategies to reduce recidivism risk is necessary in a police context, where the prediction of IPV risk stands out in a peculiar way because of the first-hand contact with victims and aggressors (Belfrage et al., 2012). Police records contain data that informs of indicators related to IPV recidivism; however, due to the immediate nature of the assessments it also possesses limitations (Messing, 2008). In this police context, studies show that the indicators more associated with recidivism risk are related to the aggressor's characteristics, especially with alcohol abuse (Jones and Gondolf, 2001, Lin et al., 2009). Even though it is not common to find vulnerability indicators for the victim in police risk assessment instruments, forensic evaluation warns us not only about the high probability of harmful consequences, but also about revictimization (Arce, Fariña, & Vilariño, 2015). The investigation with the Swedish police, led by Belfrage and Strand's (2008) team, contributed acceptable results as it included vulnerability indicators in the IPV risk assessments through the use of the Brief Spousal Assault (B-SAFER; Kropp, Hart, & Belfrage, 2005). In the same manner, investigations undergone by the Australian police highlight that, within the IPV risk assessment instruments, only a limited number of indicators are important for the officers’ decision-making, the victim's perception of their own risk level having to be taken into account at all times (Trujillo & Ross, 2008).

Within the judicial, forensic and especially in the police field, some professionals have the duty to act in order to prevent the repetition of violent behavior. In order to do this, predictions regarding the risk of violence being repeated in the future are carried out. These are used as a base for the decisions concerning the measures that need to be applied to the victim and aggressor in order to avoid violence (Llor-Esteban et al., 2016, Ruiz-Hernández et al., 2015). However, violent phenomena, especially those of the interpersonal kind, are difficult to predict because of their complexity, multi causality and infrequency. Because of this, the probability of identifying aggressors with a higher risk of violence recidivism is low if systematic procedures to evaluate it aren’t used (Andrés Pueyo & Redondo, 2007).

Police assessment of the risk of new IPV in Spain is carried out throughout nearly all of national territory by using two forms: Police Risk Assessment and Police Risk Evolution Assessment (VPR3.1 and VPER3.0), constructed and implemented with this objective in the VioGen System since 2007. They are mechanized instruments of the actuarial kind, part of a Risk Assessment and Management Protocol, possessing 16 and 17 indicators respectively, which present adequate risk prediction parameters (López-Ossorio, González, & Andrés-Pueyo, 2016). This way, in contrast with what occurs with other similar tools across the world, two different, but complementary instruments are used. The initial risk estimate is assessed by the VPR form in order to classify cases and assign police protection measures. After this, police officers use the VPER form for risk monitoring and management, which incorporates risk and protection indicators that are sensitive to the new risk scenarios generated from the denunciation, and has certain similarities with the Level of Service Inventory-Revised (LSI-R) used in the USA (Andrews & Bonta, 1995); with the risk assessment and management instruments used in Latin America (Folino, 2015) and in Spain, with the RisCanvi multiscale risk assessment Protocol (Andrés-Pueyo, Arbach-Lucioni, & Redondo, 2010) and the Self-Appraisal Questionnaire (SAQ) (Andreu-Rodríguez, Peña-Fernández, & Loza, 2016), used in the penal field.

Due to the legislative and social changes in recent years and the growing specialization of police officers in this area, a line of investigation to explore the improvement of this assessment protocol, and thus to explore police IPV risk management with the goal of increasing its performance parameters, was started. This investigation also has the objective of better classifying the more serious cases, differentiating them from the less serious ones.

This work, which is part of this line of investigation, has the main goal of identifying and empirically validating risk indicators (historical as well as dynamic) and protection indicators that are more closely associated with IPV recidivism within the police field in Spain, with the restriction that only police officers will record them (not experts in behavioral assessment). Once their impact on recidivism probability has been identified, their incorporation to new versions of police tools for risk assessment and management can be studied. The second objective of this work, which is of a more descriptive nature, is to contribute new information about recidivism within the police field, about an estimate of the characteristics of a recidivist aggressor versus those of a non-recidivist one, and, finally, about the characteristics of the victims of recidivism.

Method

Participants

The sample is formed by 6,613 cases of women who had previously filed a denunciation for IPV in 46 provinces in the Spanish territory, which have been correctly registered in the VioGen System. The victims brought the cases of IPV to the attention of, mainly, the Cuerpo Nacional de Policía and the Guardia Civil. The women have a mean age of 34.5 years (SD = 13.03) and a range of 13 to 68 years. 298 people between 13 and 18 years old (teenage women) who filed a denunciation were registered (4.5%). When it comes to the aggressors, which were all men, their mean age is of 36.51 years old (SD = 14.57) and they have a range of 14 to 68 years. 105 of these men (1.6%) were between 14 and 18 years old (teenage men). Spanish women make up 65.7% of the sample, the rest being distributed between a total of 86 countries. Most of these are from Romania (5.4%), Morocco (4.2%), Ecuador (3.3%) and Colombia (2.3%). On the other hand, 67.9% of the aggressors were Spanish, and the rest of the sample was divided into 89 countries, with similar percentages to those of the women. It is important to highlight that these percentages are not adjusted to the proportion of Spanish and foreign people who live in Spain. It is also essential to point out that within the cases of the Spanish population there are cases of people with foreign origin but that have obtained the Spanish nationality.

The type of relationship that the victim had with her partner or ex-partner was distributed in the following way: girlfriend (22.2%), ex-girlfriend (11.4%), spouse (25.7%), ex-spouse (6.2%), sentimental partner (18.2%), and ex-partner (16.3%). The distinction between girlfriend and sentimental partner is whether they were living together or not.

Instruments

In order to collect information about the new risk indicators, a form version called VPR2015, containing 43 risk indicators (of a historical type), and a VPER2015 version with 22 indicators (of a more dynamic kind), both of risk and of protection (in particular, the I-5, I-6, I-7, I-8, I-9, I-10, I-18, I-19 and I-20), were elaborated in the VioGen System.

When it comes to the themes of the VPR2015 form, its indicators can be grouped into 4 risk factor dimensions: Gravity of the denounced episode, Aggressor-related factors, Victim vulnerability and Aggravating circumstances. These dimensions have a total of 39 indicators to assess. The complementary form, VPER2015, includes indicators of an aggressor's good prognosis, indicators related to the adjustment and disposition of the victim towards protection measures, and the degree of implication in their own protection and as well as their own risk perception.

All indicators had to meet the requirement of being easily perceived by police officers, and by the examination of different sources of information at their disposal: a) denouncer; b) denounced; c) witnesses; d) police reports and other available documents. The risk indicators were marked as present or absent; in the case of not having enough information they were marked as missing data.

Design and methodology

In order to know the strength of the association of each indicator of the VPR2015 form with recidivism, a relational analytical-descriptive study was designed through the use of a prospective and stratified design, estimating the frequency measures of the form's indicators while comparing the control cases with the recidivism ones. This was carried out based on their association with the relative risk coefficient (RR). In order to analyze the VPER2015 indicators, a case-control retrospective design was used, the «cases» being the recidivist assessments and the «controls» being the periodic assessments.

Firstly, a team from the Autonomous University of Madrid (UAM) carried out an exhaustive bibliographic review using databases and documentary sources, both of the primary (original articles) and secondary (systematic reviews and meta-analyses) kind, on IPV risk indicators. Subsequently, other UAM and Universidad de Barcelona (UB) experts were consulted, and they analyzed the suitability of the obtained indicators. Finally, a team of 10 police officers with experience on IPV and risk assessment was assembled, to know their opinion on whether or not the indicators would be suitable in a future police evaluation. In the end, 65 indicators were selected from the ones that had been proposed.

The stratification of the sample was carried out randomly while denunciations were processed in the police dependencies in national territory, urban (65%) or rural (35%), corresponding to 46 out of the 52 Spanish provinces. The greatest percentages were found in Madrid (18.1%), Valencia (7.4%), Alicante (5.9%), Sevilla (5.5%), Málaga (5.2%), Las Palmas (4.9%) and Murcia (4.7%). The denunciations were recorded mainly by the National Police (62.5%) and the Guardia Civil (35%); the rest of the percentage was divided between eight local police forces and the Policía Foral de Navarra.

To ensure that the police officers correctly understood the wording of the indicators, a test with a selection of police units was done during a one-month period. This allowed us to reformulate some indicators for their better understanding by the evaluators. An instruction manual was also elaborated, with explanations about identifying each indicator and minimizing possible interpretation biases.

Afterwards, a two-month period was opened in order for police officers to assess all the new IPV cases registered in the VioGen System (N = 6,613) in two ways: first with the forms that are the focus of this study, completed at the moment of the filing of the denunciation (in the case of the VPR2015) and during the risk management process (for the VPER2015), and immediately after with the forms corresponding to the current police risk assessment protocol (VPR3.1 and VPER3.0).

In 2.640 cases (39.9% of the sample), the VPER2015 form wasn’t completed due to the suspension of the case after judicial dismissal. In the rest of the cases (n = 3,973) various evaluations per case were registered, and in different moments (as is frequent in these cases). Because of this, for the study of this form's indicators it was decided to use only the first VPER2015 of each case, independently of the date in which it was registered. In this way, the evolutionary assessments could be motivated by a new denunciation (n = 240) or by a new periodic revision of the case (n = 3,727).

Finally, throughout another six months after the initial assessment, the cases in which a new denunciation against the same aggressor was filed (n = 490), which are the ones considered as recidivism within the police field in this study, were identified.

In addition to the VPR2015 and VPER2015 forms, the police reports of 368 recidivism cases were also gathered, to carry out an exhaustive examination through an analysis of the content of all the available information from each case, studying both the seriousness of the new violence (mild, serious or very serious) and its type (physical, psychological, sexual and harassment).

Statistical analyses

The study's independent variables are the sample's descriptors and the forms’ indicators. The dependent variable is recidivism after six months. Univariate analyses were used to describe frequencies and averages. Through the use of bivariate or relational analyses, the relative risk values (RR) for the VPR2015 indicators were obtained, as well as the odds ratio (OR) for the VPER2015 indicators. The analyses were carried out through the use of the IMB SPSS Statistics program (version 20), the bivariate analysis being then used as statistical adjustment procedures and measures of effect, using Pearson's Chi-square, and the RR and OR coefficients. The missing data category was considered a missing value.

Results

Analysis of recidivism, the characteristics of a recidivist aggressor versus a non-recidivist one and the victim's characteristics

Within the 6,613 studied cases, 490 new denunciations arose. Due to this, the IPV recidivism rate in a three month interval in this sample was 4.6%, and 7.4% in a six month interval. Some form of judicial protection measure was present in 39% or recidivism cases (191 cases). The analysis of the denunciations’ content showed 63% mild violence and 37% serious or very serious violence. When distributed according to the type of violence, the following results were found: 31.8% of violence was of the physical kind, with 2.7% of it being serious; 56.2% was psychological violence, 29.4% being serious; 2.2% was sexual violence, being serious in all of the cases; and 2.5% were harassment cases, with 10.1% being considered serious. The sum of all the percentages surpasses 100% due to the overlapping that exists between different kinds of violence.

When it comes to the characteristics of the recidivist aggressor, statistically significant increases in percentages stand out in the following categorical variables: percentage overrepresentation in comparison to the global sample was observed in indicators such as: presence of a criminal record χ2 (1, N = 6,310) = 31.57, p˂.0001; exaggerated jealousy χ2 (1, N = 6,003) = 21.94, p˂.0001; presence of any type of addiction or substance abuse χ2 (1, N = 5,730) = 12.18, p˂.0001; presence of a gender or domestic violence history within their family χ2 (1, N = 4,248) = 11.62, p˂.001; presence of physical/sexual aggression records χ2 (1, N = 6,173) = 18.36, p˂.0001; presence of suicidal ideas χ2 (1, N = 5,917) = 17.55, p˂.0001; and presence of problems in their life χ2 (1, N = 5,130) = 13.17; p˂.0001. In terms of age, 8.6% of young men (between the ages of 14 and 18) (n = 105) were recidivists. This 8.6% formed 1.8% of the total of recidivism cases (n = 490). The percentage of women between the ages of 13 and 18 who had been victims of a new aggression was found to be 8.1%, higher than the global sample (7.4%).

The cases of multi recidivism (n = 49) –more than three denunciations-, show different sociodemographic data. Victims have a similar mean age (32.8), but aggressors have an mean age of 34.82, two years younger in comparison to the sample (SD = 16.89), and a range of 15 to 63 years. 83.7% of women were Spanish, compared to the 65.7% of the initial sample. On the other hand, 69.4% of multi recidivist men were Spanish, in comparison to the 67.9% from the initial sample. The most common type of sentimental relationship is that of the sentimental expartner (42.9%), compared to the 16.3% in the first denunciation. 61.2% of these cases had a Judicial Protection Order.

VPR2015 form's indicators

Table 1 shows the result of the analysis of the association between the denounced episode's indicators, gathered in the VPR2015 form, and recidivism at six months. In the first group of indicators (and sub-indicators), aimed to explore the denounced violence episode, significant results were found for psychological violence, the use of weapons and the presence of serious threats. Particularly, the use of cold weapons and suicide threats appear as the best predictors, even with a low occurrence rate. According to sexual violence, the results lack sufficient power because of a shortage of contingencies (1-β < 0.80).

Table 1

Reported episode indicators associated with recidivism after 6 months included in the VPR-2015 form.

	Coefficient		Frequency (N = 6,613)
Indicators	RR [95% CI]	aχ2	Present	Absent	bN/S	% Valid
I-1. Presence of any type of violence carried out by the aggressor	1.10 [0.89-1.37]	0.81	5,139	1,404	70	98.9
I-2. Psychological Violence	1.19 [0.98-1.48]	3.37	4,115	2,204	294	95.6
Serious psychological violence	1.37 [1.11-1.68]**	8.81	1,089	5,230	294	95.6
I-3. Physical violence	0.99 [0.82-1.19]	0.00	4,369	2,150	94	98.6
Physical violence with injuries	1.09 [0.91-1.30]	0.91	2,299	4,220	94	98.6
Serious physical violence	1.46 [0.92-2.32]	2.60	154	4,295	2,164	67.3
I-4. Sexual violence	1.14 [0.80-1.62]	0.52	360	5942	311	95.3
Sexual violence with injuries	-	2.15	27	6,275	311	95.3
Serious sexual violence	-	0.47	6	5,963	644	90.3
I-5. Victim's defensive reaction to physical aggression	1.11 [0.92-1.35]	1.234	1,718	4,305	590	91.1
I-6. Use of weapons/objects against partner	1.46 [1.15-1.86]**	9.51	673	5,813	127	98.1
Use of a cold weapon	1.84 [1.33-2.54]***	13.481	267	5,813	533	91.9
Use of a firearm	1.40 [0.47-4.13]	0.37	30	5,813	770	88.4
Use of objects	1.25 [0.89-1.73]	1.743	394	5,813	406	93.9
I-7. The aggressor has access to firearms	1.13 [0.78-1.63]	0.43	347	5484	782	88.2
I-8. The aggressor is an expert in combat techniques	1.37 [0.85-2.21]	1.71	164	5,351	1,098	83.4
I-9. Presence of threats/plans aimed to cause the victim harm	1.07 [0.89-1.28]	0.53	3,954	2,274	385	94.2
Serious threats from the aggressor	1.54 [1.28-1.85]***	21.61	1,606	4,622	385	94.2
Suicide threats from the aggressor	1.68 [1.24-2.28]***	11.35	413	2,274	3,926	40.6
Economic threats from the aggressor	0.95 [0.69-1.30]	0.10	688	2,274	3,651	44.8
Death threats from the aggressor	1.19 [0.96-1.47]	2.75	1,953	2,274	2,386	63.9
Threats to social reputation	1.14 [0.87-1.50]	0.97	868	2,274	3,471	47.5
Threats to children's integrity/custody	1.03 [0.77-1.38]	0.06	807	2,274	3,532	46.6
I-10. An escalation in aggression or threats in the last 6 months	1.32 [1.11-1.58]**	9.70	2,640	3,398	575	91.3

Note. The relative risk coefficient (RR) indicates protection when values are inferior to 1 and risk when they are superior to 1, as long as the range does not contain the unit. Significant Value (r) for risk

p˂.01.

p˂.001.

IC 95%.

The N/S Category (missing data) groups the cases where this option was marked and the lost cases together.

The analysis of the relationship between the aggressor's characteristics and recidivism within the police field shows significant results in all indicators, except for mental disorder (Table 2). The indicators of jealousy, controlling behaviors and suicidal ideas make up the more clinical aspect of the aggressor's characteristics in cases of IPV, showing a relation with recidivism. The more antisocial aspect of the aggressor's characteristics is formed by indicators that are strongly associated with repeated violence and non-compliance towards formal measures of social control, like breaking judicial measures, and shows good result sets. The presence of problems in the aggressor's life, acting as an indirect indicator of maintained stress, is also a good predictor.

Table 2

Indicators related to the aggressor associated with recidivism after 6 months included in the VPR-2015 form.

	Coefficient		Frequency (n = 6613)
Indicators	RR [95% CI]	aχ2	Present	Absent	bN/S	% Valid
I-11. They have shown exaggerated jealousy or have suspected their partner of infidelity in the last 6 months	1.54 [1.28-1.85]***	21.94	2,748	3,255	610	90.8
I-12. They have shown controlling behaviors in the last 6 months	1.80 [1.49-2.18]***	38.36	2,979	2,991	643	90.3
Physically controlling behaviors	2.01 [1.58-2.56]***	33.28	974	2,991	2,648	60.0
Psychologically controlling behaviors	1.80 [1.47-2.20]***	33.50	2,177	2,991	1,445	78.1
Controlling behaviors towards academic/labor aspects	1.95 [1.28-2.96]**	9.81	229	2,991	3,393	48.7
Economically controlling behaviors	1.59 [1.16-2.18]**	8.27	561	2,991	3,061	53.7
Cybernetically controlling behaviors	1.88 [1.46-2.44]***	23.86	843	2,991	2,779	58.0
I-13. In the last 6 months they have shown harassing behaviors	1.38 [1.15-1.66]***	11.90	2,149	3,765	699	89.4
I-14. In the last year they have done material damage	1.30 [1.07-1.57]**	7.54	1,774	4,290	549	91.7
I-15. In the last year they have disrespected an authority figure	1.58 [1.22-2.04]***	12.00	572	5,353	688	89.6
I-16. In the last year they have committed an aggression against third parties and/or animals.	1.31 [1.01-1.71]	4.18	687	5,143	783	88.2
I-17. In the last year they have made threats and slights towards third parties.	1.26 [1.03-1.56]	5.05	1,344	4,385	884	86.6
I-18. Presence of problems in their life in the last 6 months.	1.45 [1.18-1.72]***	13.17	1,482	3,648	1,483	77.6
Economic or work-related problems	1.46 [1.14-1.86]**	9.31	835	3,648	2,130	67.8
Problems with the justice system	1.93 [1.31-2.84]***	10.90	200	3,648	2,765	58.2
I-19. Presence of a criminal record	1.64 [1.38-1.95]***	31.57	2,581	3,729	303	95.4
I-20. Presence of past breakings of sentence conditions	1.85 [1.34-2.56]***	13.73	267	5,901	445	93.3
I-21. Presence of physical/sexual aggression records	1.58 [1.28-1.96]***	18.36	934	5,239	440	93.3
I-22. Presence of gender violence records against other partners	1.44 [1.15-1.80]***	10.29	861	5,322	430	93.5
I-23. Presence of a mental/psychiatric disorder	1.27 [0.89-1.80]	1.81	343	5,328	942	85.8
I-24. Presence of suicidal ideas or attempts	1.67 [1.32-2.13]***	17.55	625	5,292	696	89.5
I-25. Presence of any type of addiction or substance abuse (alcohol and drugs)	1.39 [1.15-1.67]***	12.18	2,499	3,231	883	86.6
I-26. Presence of a gender or domestic violence history within their family	1.63 [1.23-2.16]***	11.62	547	3,701	2,365	64.2

Note. The relative risk coefficient (RR) indicates protection when values are inferior to 1 and risk when they are superior to 1, as long as the range does not contain the unit. Significant Value (r) for risk

p˂.01.

p˂.001.

IC 95%.

The N/S Category (missing data) groups the cases where this option was marked and the lost cases together.

Table 3 shows the results obtained on the total of vulnerability indicators of the victim exposed to situations of violence, especially when the exposition is somewhat chronic. The only indicator to obtain a significant result is the foreign origin of the victim, which is more of a protection factor rather than a risk one (RR = 0.73). Table 4 illustrates six indicators which record aggravating circumstances, in terms of the victim, in the form of behaviors, experiences and beliefs. Having suffered IPV by the hands of other aggressors in the past, having expressed their desire to end the relationship, and the fear of suffering a very serious aggression (being attacked in a violent way or even being killed) are the indicators that show significant values when it comes to risk.

Table 3

Indicators of victim vulnerability associated with recidivism after 6 months included in the de VPR-2015 form.

	Coefficient		Frequency (N = 6,613)
Indicators	RR [95% CI]	aχ2	Present	Absent	bN/S	% Valid
I-27. Presence of any type of disability	1.30 [0.77-2.20]	0.97	136	6,310	167	97.5
I-28. Victim is in gestation period	1.06 [0.61-1.84]	0.05	154	6,255	204	96.9
I-29. Victim suffers from a serious illness	1.00 [0.55-1.84]	0.00	137	6,185	291	95.6
I-30. Victim is a foreigner	0.73 [0.58-0.92]**	6.98	1,698	4,862	53	99.2
I-31. Victim has no social/family support	0.98 [0.76-1.26]	0.01	903	5,104	606	90.8
I-32. Victim has a mental/psychiatric disorder	0.82 [0.49-1.38]	0.53	230	5,963	420	93.6
I-33. Victim has suicidal ideas/attempts	1.34 [0.87-2.05]	1.81	206	6,036	371	94.4
I-34. Victim has any kind of addiction or engages in substance abuse (alcohol and drugs)	1.23 [0.82-1.84]	1.02	260	5,753	600	90.9
I-35. There is a history of gender or domestic violence within their family	1.40 [0.91-2.15]	2.39	205	5,504	904	86.3
I-36. Victim is economically dependent on the aggressor	0.86 [0.69-1.07]	1.69	1,446	4,537	630	90.5
I-37. The victim has minors or family members under their care	1.05 [0.89-1.25]	0.40	3,661	2,952	0	100

Note. The relative risk coefficient (RR) indicates protection when values are inferior to 1 and risk when they are superior to 1, as long as the range does not contain the unit. Significant Value (r) for risk

p˂.01. *** p˂.001.

IC 95%.

The N/S Category (missing data) groups the cases where this option was marked and the lost cases together.

Table 4

Indicators of aggravating circumstances associated with recidivism after 6 months included in the VPR-2015 form.

	Coefficient		Frequency (N = 6,613)
Indicators	RR [95% CI]	aχ2	Present	Absent	bN/S	% Valid
I-38. The victim has withdrawn complaints in the past	1.13 [0.78-1.62]	0.43	352	5,953	308	95.3
I-39. The victim has suffered gender violence by the hands of other aggressors in the past	1.39 [1.07-1.81]	6.29	619	5,504	490	92.6
I-40. The victim has expressed their desire to end the relationship to the aggressor at least 6 months ago	1.48 [1.22-1.79]***	16.42	3,373	2,610	630	90.5
I-41. There have been reports of mutual violence	1.01 [0.77-1.32]	0.01	834	5,004	775	88.3
I-42. The victim fears for the physical integrity of the minors or family members under their care	1.20 [0.97-1.42]	2.98	1,186	5,427	0	100
I-43. The victim thinks that the aggressor is capable of attacking them in a violent way or even of killing them	1.33 [1.10-1.61]**	9.18	2,020	3,510	1,083	83.6

Note. The relative risk coefficient (RR) indicates protection when values are inferior to 1 and risk when they are superior to 1, as long as the range does not contain the unit. Significant Value (r) for risk

p˂.01.

p˂.001.

IC 95%.

The N/S Category (missing data) groups the cases where this option was marked and the lost cases together.

VPER2015 form's Indicators

The results from this second form show significant values for the majority of its indicators. As seen in Table 5, the risk indicators with higher odds ratio values are the fact that the aggressor has gotten closer to the victim in any way, except through the use of third parties, the fact that the victim has resumed living with the aggressor –with or without the presence of restraining measures-, the victim engaging in activities that threaten their own security, and the perception of high risk by the victim. Protection indicators reach significant values in all cases, except for when the aggressor establishes a new sentimental relationship (I-18). The perception of nonexistent risk by the victim also proves to be a protection factor, with a value of 70.7% (OR = 0.29).

Table 5

Indicators associated with recidivism cases included in the VPER-2015 form.

	Coefficient		Frequency (N = 3,973)
Indicators	OR [95% CI]	aχ2	Present	Absent	bN/S	%Valid
I-1. The aggressor has contacted the victim through technological means	6.69 [4.43-10.10]***	106.82	155	3,499	319	92.0
I-2. The aggressor has contacted the victim through third parties	1.60 [0.91-2.83]	2.75	188	3,463	322	91.9
I-3. The aggressor has physically gotten closer to the victim	11.99 [8.47-16.96]***	295.67	208	3,461	304	92.3
I-4. The aggressor is an escaped convict or is in an unknown location	1.82 [1.06-3.11]	4.98	153	3,820	0	100
I-5. The aggressor has distanced themselves from the victim	0.06 [0.05-0.08]***	551.82	3,399	574	0	100
I-6. The aggressor shows a peaceful attitude	0.09 [0.06-0.11]***	417.63	3,368	605	0	100
I-7. The aggressor exteriorizes a respectful attitude towards the law	0.18 [0.14-0.24]***	166.58	3,523	450	0	100
I-8. The aggressor shows regret	0.28 [0.20-0.41]***	52.28	1,199	767	2,007	49.5
I-9. The aggressor joins Help programs	0.33 [0.19-0.57]***	17.20	500	1,318	2,155	45.8
I-10. The aggressor follows the requirements of the charges and of the family separation	0.26 [0.17-0.39]***	48.00	2,009	534	1,430	64.0
I-11. The victim has resumed living with the aggressor	4.36 [2.67-7.12]***	41.16	104	3,869	0	100
I-12. The victim doesn’t report any episodes to file a complaint about, and/or turns down the offer of protection	2.85 [2.09-3.88]***	48.18	447	3,526	0	100
I-13. The victim engages in activities that threaten their own security	6.15 [4.37-8.64]***	138.08	217	3,756	0	100
I-14. The victim is in a situation of economical dependency	1.26 [0.90-1.77]	1.94	604	3,369	0	100
I-15. The victim has minors or family member under their care	1.22 [0.94-1.58]	2.27	2,125	1,848	0	100
I-16. Judicial proceedings to carry out a separation/divorce unwanted by the aggressor	1.52 [1.11-2.08]**	6.95	635	3,338	0	100
I-17. The victim establishes a new sentimental relationship, which is not approved by the aggressor	1.41 [0.85-2.33]	1.86	215	3,758	0	100
I-18. The aggressor establishes a new sentimental relationship	1.41 [0.90-2.21]	2.29	268	2,383	1,322	66.7
I-19. The aggressor has a stable economic and labor situation	0.63 [0.47-0.85]**	9.15	1,823	1,327	0	79.3
I-20. The aggressor has favorable family and social support	0.58 [0.42-0.82]**	9.87	2,070	690	1,213	69.5
I-21. Presence of a conflict about their responsibilities in the children's care	2.74 [2.00-3.75]***	42.99	522	2,980	471	88.1
I-22. The victim feels that the risk level is nonexistent	0.29 [0.19-0.44]***	36.38	990	2,512	471	88.1
The victim feels that the risk level is low	1.01 [0.76-1.33]	0.00	2,179	1,323	471	88.1
The victim feels that the risk level is high	4.00 [2.91-5.50]***	83.54	333	3,169	471	88.1

Note. The odds ratio coefficient (OR) indicates protection when values are inferior to 1 and risk when they are superior to 1, as long as the range does not contain the unit. Significant value (r) for risk or protection

p˂.01.

p˂.001.

IC 95%.

The N/S Category (missing data) groups the cases where this option was marked and the lost cases together.

Discussion and conclusions

The recidivism rate during the six-month risk period is consistent with the prevalence data in other IPV studies. However, it is two pints below the global VioGen data. Recidivism does not behave in a lineal or progressive way; it shows itself more prominently during the first three months. A possible explanation has to do with the time spent by the denounced men adapting to the new conditions after the denunciation, and the emotional attachment to their partner.

In this investigation, the weight of each indicator is recorded through the odds ratio, based on the investigation's design and its temporal direction. These empirical weights are very useful to analyze the association between risk factors and recidivism, and are also very intuitive since they indicate the number of times that the probability of occurrence of the criterion associated with the predictor rises.

The RR values found in the VPR2015 form's indicators confirm that in contexts where protection of the victim is demanded based on the risk, the majority of risk indicators can be considered static or dynamic elements to be assessed by the police with the information available to them. The construction of prediction tools must, because of this, bear this empirical evidence in mind, and must also consist of these elements (Canales et al., 2013, Folino, 2015).

The results obtained with the indicators that explore the denounced violence episode don’t reach good predictor results, since they do not properly discriminate cases based on recidivism probability; partly, because of its prevalence, whether it is due to excess or deficit, as it happens with sexual violence. The victim's defensive reaction, recorded in 26% of cases, does not show positive data either. This evidence confirms the results of previous revisions (Capaldi et al., 2012), which explain that these types of predictors, while very related to a punctual episode, do not usually work as good recidivism indicators. Another possible explanation of these results is derived from the presentation of the indicators, in the sense that when they are all grouped up, they illustrate that serious psychological violence, the use of weapons –in a Spanish context, cold weapons-, the presence of threats and the escalation of aggressions obtain significant RR values, like the revisions show (Capaldi et al., 2012). This, more defined, group of indicators is usually present in IPV risk assessment instruments -B-SAFER, RVD-BCN, ODARA, SARA-, which also add serious psychological violence as a significant predictor.

The results of the present study show that variables of the aggressor, as the main person responsible for the violent behavior, are very associated with recidivism. The indicators are grouped into three main blocks: IPV-specific behaviors, violent or generally antinormative behavior, and clinical indicators. These groups integrate valid predictors and a variable that measures stress, in terms of problems in the aggressor's life. Data that shows a decrease in violent behavior recidivism with age was found, observing a clear decrease after 51 years are reached. The data obtained also shows that recidivism tends to increase in the cases of a sentimental ex-partner relationship, in which the partners do not live together and have shared children (Richards, Jennings, Tomsich, & Gover, 2014).

Out of the three clinical variables –suicidal ideas, substance abuse and mental disorder-, the first two obtain a significant RR value. Certainly, the mental disorder indicator groups up all mental illnesses, when only a certain group of disorders (e.g. Disorders that have to do with delusional jealousy, or Borderline Personality Disorder) are associated with violent behavior, without the police officers being able to correctly assess them. However, this indicator usually has a greater relationship with the seriousness of the violence than with mere recidivism. The present study's results also confirm previous studies about the limited security that protection orders offer victims, since many disobediences of judicial orders are found (Richards et al., 2014).

The indicator which evaluates how the victim perceives the risk of their own situation proves to be a good predictor. The victim's vulnerability indicators, however, do not reach a significant association. These types of variables, which are especially related to the dynamic characteristics of the victims, are very interesting when it comes to clinical or forensic exploration, however, studies that reach good empirical data for these types of indicators are scarce (Belfrage & Strand, 2008). On the other hand, it is possible that elements of this nature must be included in risk assessment tools, since they allow for a more precise knowledge about the victim's needs and facilitate their referral to specific resources. The foreign nationality of the victim, instead of presenting itself as a risk variable, is actually associated with recidivism in terms of protection, in contrast with what other studies show. This being said, it seems beneficial to continue investigating about the role of vulnerability indicators in IPV.

The risk and protection indicators contained in the VPER2015 form show a good enough association with recidivism to be integrated in a police risk evolution assessment form, used for the management of IPV protection resources. The performance of the indicators included in this form is dual: On one hand, they prove to be sensible to the changes produced during new risk scenarios, when police and judicial resources have began mobilizing. On the other, they are cemented as good recidivism predictors.

The inclusion of VPER2015 protection indicators is another one of the innovations of a future assessment tool, as the combination of risk and protection factors in instruments of this nature, which are used in police contexts, is rare. It's frequent for the protection management to be monitored through the same instruments that were used in the first assessment, because risk reduction is understood as the absence of these types of indicators (Folino, 2015). Here, however, the VPER2015 indicators are different to the ones in the VPR2015. VPER2015's indicators, combining both the risk and protection ones, prove to be very useful during the process of risk management in IPV. This empirically demonstrates that risk is not only reduced through the absence of risk indicators, but also through the presence of suitable protection indicators.

This new group of indicators, in contrast with what occurs with the VPR2015 form, are assessed by officers that know these cases with higher precision due to the periodically established contact with the victims, which aids with the better weighing of the indicators. All of this brings about results that are well related with both risk and protection aspects, supplying evidence about indicators that allows for assessments which compliment these two aspects, to build more dynamic risk assessment instruments through the use of semi-actuarial evaluations in a police context (Messing et al., 2013, Sabri et al., 2014). The aggressor's attitude towards the victim, translated into behaviors that allow the observation of whether or not they accept and respect the new legal scenario, is translated into risk or protection indicators of which there was no empirical evidence.

The main goal of this study consisted on empirically relating IPV indicators with police recidivism, being the first study with such a numerous sample of Spanish population, which allows for the generalization of the results to the general population of women residing within national territory who have denounced being victims of this kind of violence. As it can be seen, this objective has been reached, obtaining evidence on risk and protection indicators that can be included in police IPV risk assessment instruments.

Studies about risk factors related to any kind of human behavior, like violence, do not give way for going any further than probabilistic associations. However, the relationship and strength of the association constitute the main substance on which cause-effect relations are based.

The present study is not free of limitations. Firstly, the recidivism measure does not record cases of violence that was not denounced, leaving out a missing «dark figure». On the other hand, there are probably cases of denounced episodes that are not genuine. It could also be argued that the information was gathered by police officers and not by the investigators, which can create a bias in the quality of the information. However, the officers’ experience in risk assessment, the training they received through the manuals that were given to them, and the fact that they worked on real cases, creating reports for the judges, has probably contributed to guaranteeing the quality of the data recorded, reducing this limitation and increasing the ecological validity. The application of police victim protection measures after the denunciations, based on their risk level, constitutes an inherent circumstance of these studies, especially at the highest risk levels (where the protection is greater). Its influence over recidivism is still missing data and presents an engaging scenario for future investigation. As the investigation on police evaluations (Trujillo & Ross, 2008) points out, the data allows for optimism when it comes to the possibilities that these kinds of indicators open up in risk assessment forms that are specifically designed for the management of IPV victims’ security. The data presented in this work is bound to a period of six months, as was explained previously. However, it would be very interesting to carry out studies with wider time periods from now on. Because of this, the cases will remain under observation in order to be able to present new information that enables the continued addition of data in the future. This will update and improve police risk assessment instruments.

To summarize, the obtained data is coherent with previous investigations on IPV risk indicators, and new protection indicators have been found with empirical evidence for the police management of violence risk. Nonetheless, it is still necessary to continue advancing in the perfection of risk assessment through studies on risk and protection indicators that enable the improvement of IPV victims’ security.