Abstract
Gun violence remains a pressing public health concern, especially in high-risk urban environments. Community-level violence intervention efforts are being mounted in cities across the United States to prevent and reduce the most severe forms of violence. There is growing evidence to suggest the effectiveness of the Safe Streets Program/Cure Violence as a community-based intervention to reduce homicides and shootings. The mechanism underlying the reductions in community violence is theorized to be linked with changes in attitudes toward violence as well as shifts in social norms related to violence and retaliation, but there are few tools to assess these domains. This preliminary investigation sought to establish the metric properties of the Survey on Attitudes About Guns and Shootings (SAGAS) with the goal of providing an empirical measure of attitudes and community-level norms. Males aged 18 to 24 were surveyed using the SAGAS in two high-violence communities in Baltimore, Maryland, using street intercept methodology. We found acceptable reliability and validity metrics for the SAGAS. Reliability and validity of the SAGAS were assessed using internal consistency and a latent class analysis with violent behavior outcomes. The internal consistency of the total scale was in the extensive range (α = .70-.79) and the internal consistency of the factors was in the exemplary range (α ≥ .80). In addition, latent classes of attitudes were predictive of being arrested or being shot. Future studies will examine if rates of violence decrease in neighborhoods targeted by the Safe Streets Program and the mediating role of attitudes toward gun violence using the SAGAS.
Keywords
Despite recent declines in fatal and non-fatal shootings, gun violence continues to be a major public health problem in many urban areas. According to the Bureau of Justice, fatal shootings decreased 39% from 1993 to 2011; a similar decline was seen with non-fatal shooting. While this trend was also seen in young adults, homicide still remains the third leading cause of death among those aged 10 to 24 years and this age group has the highest rate of homicides in the United States (Centers for Disease Control and Prevention [CDC], 2013). As handguns are illegal for minors to purchase, ownership of handguns among youth suggests a more serious level of involvement in aggression and criminal activity (Shapiro, Dorman, Burkey, & Welker, 1997), and youth who possess handguns tend to endorse more aggressive beliefs about retaliation and gun use (Carter et al., 2013).
Prior studies have shown that violence intervention programs can reduce beliefs supporting aggressive behaviors in urban youth (Chang, Cornwell, Sutton, Yonas, & Allen, 2005). Current community-based intervention violence programs, such as Safe Streets (modeled after the CeaseFire/Cure Violence Intervention), rely primarily on changes in homicides as the metric for the program success. This current investigation employed an existing but unvalidated tool, the Survey on Attitudes About Guns and Shootings (SAGAS), to measure baseline attitudes about the use of guns and violence to resolve disputes. Here, we will assess the preliminary metric properties of the SAGAS in a population of young and emerging adults in high-violence communities.
Method
This investigation focuses on two neighborhoods in Baltimore City. The communities were selected based on similar demographics and rates of fatal and non-fatal shootings. In 2011, the Park Heights community had 1.6 homicides and shootings per 1,000 residents; the rate was 1.8 per 1,000 residents in the Southwest Baltimore community and 1.1 per 1,000 residents in Baltimore City.
Individual blocks within each community were randomly ranked from 1 to the highest number of blocks. Blocks were then visited in the order of their ranking. Blocks were visited up to three occasions or until a maximum of six surveys were completed on that block face. A block face is defined as the even and odd side of the street of the unit block.
Inclusion criteria for the street intercept survey were male youth aged 18 to 24 years and English speaking. Survey staff were trained to approach any youth appearing to be near or in that age range and ask for the respondent’s age. Youth males who appeared to be intoxicated, under the influence of drugs, or otherwise mentally impaired were ineligible to participate. The data collection and data analyses were approved by the institutional review board of the Johns Hopkins University Bloomberg School of Public Health.
Staff Training and Procedures
Interviewers were trained in two in-office orientations that included human subjects training, study protocols, administration of survey to other interviewers, and safety procedures. Each interviewer took training packets in the field with the principal investigator and project manager’s contact information, a summary of the survey being distributed, the study procedures, safety protocols, and participant inclusion requirements. As a safety precaution, all data were collected in teams of two interviewers.
Interviewers were expected to obtain six surveys per block throughout data collection. Participants were approached in Park Heights and Southwest neighborhoods in Baltimore, Maryland, based on the predetermined blocks. Research assistants approached potential participants and asked if they were in the target age range (18-24 years). Qualifying participants were then informed of the purpose of the study and asked to complete the survey in exchange for a US$10 gift card to a nearby convenience store. Completed surveys were stored in individual envelopes with no personal identifying information other than the location of the block on which it was completed. Surveys were returned to the project manager within 48 hr of collection for data entry.
SAGAS
The 37-item survey included respondent age, exposure to community violence prevention programs, whether they have ever been arrested, whether they had ever been shot or shot at, whether they had seen a vigil, march, or gathering in response to a shooting, and a series of attitudinal items that ask whether the respondents think it is “okay” to shoot someone or threaten him or her with a gun under five common scenarios found in prior research to be “sparks” for shootings involving urban youth. Response options for the attitudinal items were yes, no, or maybe. The survey was anonymous and, to protect confidentiality, self-administered. Respondents completed the survey on clipboards using paper and pencil with “blinders” to conceal responses. Participants were also offered the opportunity to listen to the questions and response options on a portable CD player with headphones to reduce non-response due to literacy.
Statistical Analysis
Descriptive statistics for each of the responses were calculated and are displayed in Table 1. Exploratory factor analysis (EFA) with oblique rotation was used to identify latent factors among the 30-items assessing attitudes about gun violence and resolving conflict. Scree plots as well as fit indices, including root mean square error of approximation (RMSEA) and comparative fit index (CFI), were used to identify the most parsimonious model. Cronbach’s alpha was used to evaluate the internal consistency of the total scale and each of the latent factors identified by EFA. The resultant factors were used for latent class analysis (LCA).
Attitudes Toward Gun Violence (n = 478).
LCA is an example of finite mixture modeling that uses categorical indicator variables. LCA was used to infer class membership based on underlying distributions in each of the scales. Lo–Mendell–Rubin likelihood-ratio tests were used to determine the number of classes; p values less than .05 indicate that the k − 1 class model should be rejected in favor of the k class model. In addition, Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and sample size adjusted BIC were used to evaluate goodness of fit and to identify the solution with the ideal number of classes. Preliminary measures of validity were assessed by including violence (e.g., ever been shot) and previous arrest as outcomes of the LCA model. The model-based approach described by Lanza, Tan, and Bray (2013) was used to estimate the conditional probability of previous arrest and exposure to violence for each latent class. Analyses were conducted in Mplus Version 7.11 and IBM SPSS Statistics Version 22.
Missing data
There were 625 participants who completed the survey. Approximately 77% (n = 478) of participants had complete data for the 30 items about attitudes toward gun violence and conflict resolution. The remaining 23% of participants were missing at least one question. The participants who responded to all 30 items were similar (i.e., p > .05) to the participants with missing data in terms of age, if the participant had ever been shot at, been arrested, seen a vigil, march, or gathering in a neighborhood after shooting. Participants with missing data were more likely to report that they have seen a Safe Streets worker help someone (p < .01).
Results
The mean age of the sample was 20.7 (SD = 2.1). The most common method to resolve conflicts was letting it go or talking with the other person, ranging from 29.1% to 51.5% (see Table 1). Between 14.9% and 30.8% reported that they would or their friend would threaten others with a gun to resolve conflicts. Participants’ endorsing shooting someone was the least common method to resolve conflict; 7.7% to 23.0% of participants would shoot someone or thought their friend would shoot someone to resolve a conflict (e.g., someone owes you money or disrespects you). About a third of the sample reported being shot at in the past and 44% reported ever being arrested.
Reliability Analyses: EFA
The EFA revealed six-factor solution (CFI/Tucker–Lewis fit index [TFI] = .988/.980, RMSEA = .033). One of the factors only contained three items with one indicator that had a factor loading below .400; this factor will not be discussed. The remaining factors were labeled (a) violent attitudes toward personal conflict, (b) non-violent attitudes toward personal conflict, (c) violent attitudes toward conflict involving friends, (d) non-violent attitudes toward conflict involving friends, and (e) gun threats. The factor loadings for each of the indicators are included in Table 2. The loadings ranged from .419 to .813. Internal consistency of the five scales are included in Table 2; Cronbach’s alpha ranged from .848 to .899. The internal consistency for the total scale was .726.
Exploratory Factor Analysis: Attitudes Toward Gun Violence.
p < .05.
Validity Analyses: LCA
The Violent Attitudes Toward Personal Conflict scale was used for the LCA. Table 3 displays the model fit indices for the latent class solutions. The Lo–Mendell–Rubin Likelihood Ratio test and the BIC suggested the three-class solution had the best fit, however, AIC and sample size adjusted BIC suggested a four-class solution. We decided to go with the more parsimonious model. The mean maximum posterior probabilities were .98, .92, and .96 for Classes 1 to 3, respectively. Class 1, Non-Violent Approaches to Personal Conflict accounted for 46.2% of participants. This class had the highest probability of responding no to solving personal conflicts by threatening or shooting others with guns. Class 2, Undecided Participants accounted for 15.7% of the sample. This class included the highest probability of maybe responses compared with the other two classes; there was a low probability of responding yes to the items. Class 3, Threats of Violence included the remaining 38.1% of the participants. The class had the highest probability of responding yes to the indicators. The participants in this class had higher probabilities for threats of using violence versus actual violence. The item probability scales are included in Table 4.
Latent Class Analysis Fit Indices (n = 614).
Note. AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion. The boldfaced numbers indicates the selected model for the paper. There are three class model based on the fit indices presented in this table.
Results of Latent Class Analysis of Violent Attitudes Toward Personal Conflict Factor.
Note. LCA = latent class analysis.
p < .05.
The probability of being arrested was the lowest in the Non-Violent Approaches to Personal Conflict class, namely, .328, which was significantly lower than the probability of being arrested in the Undecided Participants class (probability = .626, χ2 = 19.69, p < .001) as well as the probability of being arrested in the Threats of Violence class (probability = .521, χ2 = 12.03, p = .001). There were no differences in the probability of being arrested between the Undecided Participants and Threats of Violence classes (χ2 = 2.417, p = .120). Similarly, the probability of being shot in the past was the lowest in the Non-violent Approaches to Personal Conflict class (probability = .219). The difference in the probability of being shot in the past in the Non-violent Approaches to Personal Conflict class was significantly lower than the probability of being shot in the Undecided Participants class (probability = .424, χ2 = 10.02, p = .002) as well as the probability of being shot in the Threats of Violence class (probability = .433, χ2 = 22.84, p < .001). Again, there was no difference between being shot in the past between the Undecided Participants and Threats of Violence classes (χ2 = .015, p = .904).
Discussion
This preliminary investigation sought to establish the metric properties of the SAGAS. The investigation found acceptable reliability and validity metrics for this survey on attitudes about gun violence. The internal consistency of the total scale was in the extensive range (α = .70-.79) and the internal consistency of the factors was in the exemplary range (α ≥ .80). To assess validity, we used the latent classes to predict two outcomes: being arrested or being shot. The non-violent class was less likely to have been arrested or shot in the past as compared with the other classes. The study also found that up to 17% of participants in high-violence communities would use a gun to resolve a personal conflict.
While this investigation focuses on an often-overlooked population of young adults to better understand attitudes toward gun violence, there are several limitations that should be discussed. First, several sociodemographic characteristics were not assessed (e.g., highest level of education or income). These characteristics could be used to identify subpopulations that use gun violence to solve conflicts. These items were added to follow-up assessments and will be reported in future reports. Also, given the goal of this investigation was to understand violence in two relatively small geographic areas, the study design limits the generalizability of the findings, but nonetheless offers promising directions for future studies of community attitudes and norms around gun violence in urban centers. Similarly, these two communities were racially homogeneous (majority African American) and the investigation was limited to males as they account for the vast majority of perpetrators and victims of gun violence. Future investigations should examine the utility of this instrument in more diverse populations and geographic areas. The study also relies on self-report measures although strategies were in place to ensure confidentiality and truthful responses (e.g., anonymous survey with blinders).
Future studies will examine how the attitudes toward gun violence change with the introduction of targeted interventions. The Safe Streets Program is being implemented in several neighborhoods in Baltimore (Webster, Whitehill, Vernick, & Curriero, 2013; Whitehill, Vernick, & Parker, 2012). We suspect that changes in attitudes about violence and subsequent shifts in community norms around violence will account for overall reductions in shootings and homicides. Metrics validation of the SAGAS improves our capacity in future studies to draw such conclusions.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the Centers for Disease Control and Prevention (CDC) 1U01CE001954-01A1.