Contextual violence and its link to social aggression: a study of community violence in Juárez

Introduction

Methods and Instruments

Results

Table 1 shows the main psychometric and descriptive statistics of each factor identified by the exploratory factor analysis on each instrument. The analysis for the CVCV questionnaire retained 33 of the original 34 items proposed by Gurrola-Peña et al. (2018), eliminating the item “I have seen someone get stabbed in my school” for loading values less than 0.4 for all the factors. The five-factor solution obtained explained 60.92% of the variance of responses, and corresponded to the conceptual content of the original scale. The first factor was called “Non-witness contextual victimization” given that it contained items related to hearing about violent events in areas related to one’s daily life (e.g., “I have heard that someone has been shot in my neighborhood”; “I have heard that in places that I usually go someone has been shot”). The second factor was named “Contextual victimization in the neighborhood” for comprising items that relate to having heard about (e.g., “I heard that someone has been shot in my neighborhood”) or witnessed (e.g., “I have seen drug trafficking in my neighborhood”) threatening situations in areas near the home. The third factor was named “Contextual victimization in recreational areas” because it contained items related to hearing about acts of violence in recreational areas (e.g., “I have heard that in places where I go for fun people have been shot”). The fourth factor was named “Contextual victimization at school” since it contains items related to hearing (e.g., “I have heard that someone has been shot in my school”) or witnessing (e.g., “I have seen how someone has been beaten in my school”) violent acts at school. The fifth factor was called “Witness contextual victimization”, because it is comprised of items that refer to having witnessed violent acts in areas related to one’s daily life (e.g., “I have seen someone being shot in places I usually frequent”; “I have witnessed a non-weapon assault in places where I usually go for fun”). To explore possible differences in contextual violence between men and women, the averages in each indicator were compared using Student’s t-tests. The results only identify statistically significant, although very small, differences in non-witness contextual victimization (t = 2.17, df = 296, p = 0.030, Men M = 2.21, Women M = 1.98) and contextual victimization in school (t = −1.98, df = 283.34, p = 0.048, Men M = 1.27, Women M = 1.39). The other three forms of contextual violence show averages that can be considered equivalent between both groups (t’s < 1.47, p > 0.141).

Analysis of the PTSD checklist showed a structure of two factors plus an indicator that contained only two items, explaining together 59.19% of the variance with the 24 original items. The first factor was labeled “Outcomes” given that it comprises items centered around changes in emotional state and everyday life provoked by traumatic events (e.g., After what happened, I have many difficulties in my relationships with others). The second factor was named “Memories”, containing items related to the recurrence of memories associated with the traumatic event (e.g., “I feel bad when something reminds me of the situation”) and the strategies used to cope with them (e.g., I always avoid thinking or talking about what happened.). The third factor was named “Event” and only has two items, but was kept since both items referred to having recently experienced a violent event (e.g., “Lately I have experienced at least one situation related to death or threats against my life or to other people related to me”) that was considered traumatic (e.g., “because of this situation I have experienced a lot of anguish or excessive fear”), in addition to explaining an adequate amount of variance (see Table 1).

The analysis of the AQ identified four factors consisting of 23 of the original 29 items of the scale, which together explained 49.75% of the variance in responses. Six of the items were discarded because they presented loadings of less than 0.4 on all factors. The first factor was named “Punching”, for referring to the disposition to enter into a conflict and hit other people (e.g., If they provoke me enough, I can hit another person). The second factor was named “Paranoia”, because it refers to a general distrust of other people and the possibility of being hurt by them (e.g., “When people are particularly friendly, I suspect their real intentions”). The third factor was named “Anger”, since its items refer to the tendency to feel angry without a particular reason (e.g., Sometimes I get very angry for no reason). The last factor was called “Arguing”, since its items refer to the willingness to easily enter into discussions with other people (e.g., “When I do not agree with my friends, I argue with them”).

Analysis of the AVS yielded a three-factor structure containing 12 of the original 14 items which together explained 53.86% of the total variance in responses. The first factor was named “Acceptance of violence against children”, as it contained items such as “A child who is habitually disobedient must be punished physically”. The second factor was named “Acceptance of couple violence”, as it contained items such as “It is a right of the couple to hit the other if they are insulted or ridiculed”. In the original proposal, this factor contained items about violence against children and was named “Family violence”, but in this analysis it only contained items related to violence in couples. The third factor was named “Acceptance of military violence” containing items such as “Our country should be aggressive internationally using military force”.

Once the factors were identified, the data were analyzed using the SEM of maximum likelihood in AMOS 22 (Arbuckle, 2013). This initial model, presented in Fig. 1, shows significant factorial loads of all observed variables on their respective latent variables, with CR values among 6.36 and 15.78 with p < 0.001., confirming that the structure of the instruments identified by the exploratory factor analysis was adequate for this sample. The model confirms a positive and significant effect of contextual violence on PTSD (B = 0.13, CR = 3.88, p < 0.001), explaining 6% of its variance. Contextual violence (B = 0.08, CR = 2.07, p = 0.038) PTSD (B = 0.32, CR = 4.25, p < 0.001) had positive direct effects on disposition to aggression, which added to an indirect effect of contextual violence on disposition to agresion (B = 0.04, z = 2.86, p = 0.004) (Sobel, 1986), explain together 13% of the variance. However, the fit of this model was not adequate (Chi² = 140.18, gl = 51, p < 0.001, RMR = 0.036, GFI = 0.92, AGFI = 88, CFI = 0.94, RMSEA = 0.077 90% CI [0.062–0.092]). An analysis of the residual covariances of the model made it possible to identify that these discrepancies derive mainly from the contextual violence observed variables, which are all part of its latent variable, but only “at school” and “in the neighborhood” variables have a predictive effect on PTSD. This would seem to indicate that the complete construct of contextual violence is not necessary for the prediction of PTSD. The model is more parsimonious if we take only the indicators of contextual violence at school and in the neighborhood, which could also be considered the most common locations for the sample in this study. It is noteworthy that these two factors have a more important effect than variables related to direct exposure to violence such as the “witnessed violence” factor.

With this model as a backdrop, a new model was tested (see Fig. 2) which preserved only the factors of contextual violence at school and neighborhood for their significant weights on the PTSD. These variables were preserved allowing free covariation between their measurement errors as part of the same construct, but contextual violence was removed as a latent variable, while three of its five original observed variables were eliminated as well. The errors of the AQ variables “punching” and “arguing”, as well as the errors of the variables of the PTSD dimension “event” and “consequences”, were also allowed to freely correlate. It is interesting to note that the covariation between the “event” and “consequences” variables is negative (B = −0.10, CR = −3.69, p < 0.001), which could indicate a sustained effort to overcome the traumatic event by continuing with everyday life (i.e., those most exposed to traumatic events would be more aware of the importance of continuing their routine activities).

In this new model, contextual violence at school (B = 0.13, CR = 2.55, p < 0.011) and in the neighborhood (B = 0.13, CR = 5.08, p < 0.001) together explained the 13% of the variance in responses to the PTSD questionnaire. In turn, PTSD (B = 0.23, CR = 3.82, p < 0.001), contextual violence at school (B = 0.06, CR = 2.53, p = 0.011) and in the neighborhood (B = 0.06, CR = 1.26, p = 0.205) had positive effects on disposition to violence, although this last one is not statistically significant. These direct effects added to indirect effects of contextual violence at the school (B = 0.03, z = 2.13, p = 0.036) and in the neighborhood (B = 0.03, z = 3.07, p = 0.002) (Sobel, 1986), explain 17% of the variance of responses to the AQ. This indicates that 17% of the explained variance of the AQ is given by a direct effect of PTSD, but part of this variance is explained by contextual violence at school and in the neighborhood. This model shows better goodness of fit (RMR = 0.020, GFI = 0.97, AGFI = 0.93, CFI = 0.97), although the Chi-square indicator and RMSEA keep signaling significant discrepancies between the relationships established in the model and those observed in the data (χ² = 41.63, df = 21, p = 0.005), and an acceptable possibility to generalize this model to the population from which the sample was extracted (RMSEA = 0.058 90% CI [0.031–0.083]) (Kline, 2011). While Browne & Cudeck (1993) propose that maximum values of 0.08 for RMSEA indicate a reasonable goodness of fit, more demanding criteria such as that of Hu & Bentler (1999) propose that levels above 0.06 indicate that the model could not be generalized to the population from which the sample was extracted.

Considering that women have been especially affected by the violence in Ciudad Juárez, a gender-based comparison of the model was conducted. The indicators show an adequate adjustment of the model without restrictions for CFI and RMSEA (CFI = 0.96, RMSEA = 0.050 90% CI [0.030–0.069]), but not for Chi-squared indicator (χ² = 72.90, df = 42, p = 0.002).

The prediction effects of the model are significant for the women’s model, with p < 0.004 and CR values between 2.92 and 4.43, except for direct effects of contextual violence at school (CR = 0.80, p = 0.419) and in the neighborhood (CR = 1.73, p = 0.082) on AQ. Indirect effects of violence at school (B = 0.06, z = 2.18, p = 0.028) and in the neighborhood (B = 0.06, z = 2.65, p = 0.008) remain statistically significant, and added to direct effects in the model explain 20% of variance of PTSD and 23% of AQ (Fig. 3). The fact that the significant indirect effects of contextual and neighborhood violence on aggression, and that the direct effects are not significant, point to the PTSD as a mediating variable of the relationship between these forms of contextual violence and aggression. That is, contextual violence in the neighborhood and school increases the predisposition to aggression only when they are high enough to increase PTSD.

In contrast, the only statistically significant paths for men’s model are contextual violence in the neighborhood on PTSD (CR = 2.78, p = 0.005), PTSD on AQ (CR = 2.59, p = 0.010) and contextual violence in the neighborhood on AQ (CR = 2.13, p = 0.032). Indirect effects of violence at school (B = 0.02, z = 1.23, p = 0.21) and in the neighborhood (B = 0.02, z = 1.90, p = 0.056) were not statistically significant (Fig. 4). This model results clearly more adequate for women than for men, considering that explained variance decreased to 9% for PTSD and 14% for AQ, almost half than in the women’s model.

These differences between both models would explain the significant discrepancies observed in the χ² indicator of the unrestricted model. As the equivalence between both models was not guaranteed in this first level, no comparison was performed in the following levels of restrictions.

In order to test the hypothesis of the effects of contextual violence and PTSD on the acceptance of violence, the AQ was replaced with the Acceptance of Violence measurement in the previous model. The resulting model did not confirm our hypothesis. There was adequate adjustment of the model (χ² = 13.68, df = 17, p = 0.68, RMR = 0.011, GFI = 0.98, AGFI = 0.97, CFI = 1.00, RMSEA < 0.001 90% CI [0.00–0.04]), but an explained variance of 0%.

A possible alternative explanation of the results is that the perception of contextual violence may be affected by participant individual differences, in that more sensitive people would tend to report higher levels of perceived violence. To explore this possibility, high and low groups were formed in the measurement of emotional stability of the TIPI; Gosling, Rentfrow & Swann (2003) based on the median value (4.5). Comparisons were made between both groups using the Student’s t test in the indicators of contextual violence at school and in the neighborhood. The results do not identify statistically significant differences in any of the two indicators of violence, with values of t < 1.57 and p > 0.119 for both comparisons, indicating that reports of contextual violence were not affected by emotional stability.

Discussion

The current study examined whether degree of contextual violence could determine levels of PTSD and disposition to aggression. The SEM analysis only partially confirmed the hypothesis that contextual violence was predictive of the emotional affectations associated with PTSD. This is because only the indicators of violence at school and in the neighborhood were able to predict any variance in responses to the PTSD instrument. In addition, our analysis showed that these variables were only predictive of PTSD in women, and that the only predictor for men was violence in the neighborhood. As the responses came from a sample of students, it makes sense that the main predictors of PTSD would be violence in the neighborhood and at school, as these are likely to be their most frequented locations. However, the result that violence at school is predictive for women but not men may indicate that school is perceived as a safe space by men only. It is also notable that the factors of contextual violence at school and in the neighborhood, composed of items that mainly relate to hearing that violent acts have occurred, have a greater predictive capacity than factors such as witnessed violence. This may be due to the fact that signs of violence in highly-frequented locations have a greater impact than those in less-frequented areas or in non-specific locations, even if it was personally witnessed.

Indirect effects of contextual violence on AQ were also identified, implying that the PTSD is mediating this relationship. However, the comparison analysis between groups confirmed this result only in the sample of women. This difference may be since the levels of variance explained, both of PTSD and AQ, are markedly higher in women than in men. In this regard, it should be considered that the level of contextual violence in the school reported by women is barely significantly higher than that of men, and their averages of contextual violence in the neighborhood are equivalent. That is, with clearly close levels of contextual violence, the harmful effects on emotional health (PTSD) and its willingness to aggression (AQ) are more important in women. In this regard, it should be recognized that these scales do not evaluate some forms of violence that only women typically suffer such as sexual harassment, whether in work or educational environments, which represents a solid research objective for new projects in this same line.

It would be interesting to explore whether the level of education affected the results. Considering that higher education is associated with rational thinking, it is interesting to consider whether university students who are over-exposed to violence would not accept or justify the use of violence (conscious or rational thinking), even if they tend to rely on violent behaviors (unconscious or reactive acting) in threatening situations. We also propose that community violence studies would benefit from moving beyond common sampling criteria such as age, sex, and education level in order to shed further light on the patterns of social-cognitive functioning among individuals exposed to violence. Due to resource and time restraints, this rational could not be followed in the current study, and as such constitutes a serious limitation. Shahinfar, Kupersmidt & Matza (2001) suggest two ways to accomplish this. Firstly, it is important to carefully consider the modality of exposure to violence of the participants (e.g., if studying effects of indirect exposure, filter those who have had direct exposure to violence). Secondly, the severity of the exposure is a factor that must be emphasized and included in analysis.

The current study confirmed the hypothesis on the role of emotional affectations (PTSD) as potential generators of aggression or at least a tendency to attack others. The direct effects of contextual violence on disposition to aggression show that contextual violence predisposes one to aggression against others in general, both by itself and through its emotional affectations. Several notable differences were identified between men and women models. Although in absolute terms 23% explained variance in the model with women may seem like a small amount, it should be considered that in relative terms, it is twice the amount of variance explained in the model with men. This difference may derive from the increased sense of vulnerability among women from exposure to rape, kidnaping, domestic violence or human trafficking; all these types of crimes with notably high rates in Juárez City.

The elements uncovered in the models presented here could potentially relate to the spread of violence, a phenomenon whose dynamics have been studied among criminal groups in Mexico (Osorio, 2015) but not among general populations. Previous studies have shown that increases in a city’s violence show patterns similar to those of an epidemic (Berkowitz & Macaulay, 1971), where exposure to especially violent events transforms existing norms and the meaning, for example, of “being violent” from an undesirable attribute to a characteristic that provides respect and security. One interpretation of this phenomenon is that as new norms spread among the population to the extent, those endorsing them are more likely to “adjust” (in evolutionary terms). As a consequence, the adopters of these norms can gain access to better resources and develop better solutions for common problems such as exposure to violence (Fagan, Wilkinson & Davies, 2007).

Although no effects on norms related to acceptance of violence were identified in this study, the instruments used did not cover aspects such as acceptance of the possession of weapons, the legitimacy of revenge, the exercise of vigilante justice, or other similar norms which would imply a greater willingness of citizens to join the spiral of violence in their city, that is, what Eisenhammer (2014) has called “bare life”. These elements remain a vein for future research to explore.

Conclusions

While the model in this investigation is not a definitive model for contextual victimization, we do consider it as a starting point for the study of contextual violence consequences, given that it has been centered mainly on its measure and not on its effects. Future research could also improve by measuring whether the participants have or have not been exposed to direct or indirect violence, in order to limit possible correlates on contextual victimization only. Finally, though this line of research centers on understanding the underlying mechanisms of violence exposure, it must be stressed that the implicit goal of studies in this area should be to improve interventions to reduce aggressive behavior in individuals exposed to violence (Calvete & Orue, 2011), and reduce community violence as a result.

Supplemental Information