The CRCEI seeks to bring neighbors together, including youth and adults, build capacity to strengthen neighborhood collective efficacy, and ultimately reduce community violence. The 3-phase intervention brings youth and adults together to facilitate and strengthen trusting relationships, develop restorative and nonviolent intervention and organizing skills, and foster youth leadership and intergenerational partnerships. Consensus organizing strategies are used throughout the intervention to facilitate trusting relationships among neighbors and with community resources, thus building social cohesion, which is a key component of collective efficacy. Consensus organizing engages residents and external and internal resources to address neighborhood problems based on their mutual interests and concerns, which facilitates residents’ willingness to intervene [11,12]. A community-based approach offers opportunities to reach youth outside of school and to build natural mentoring relationships with adult participants and community facilitators.

As indicated in Figure 1, the intervention is implemented across 3 phases. Phase 1 includes conducting a community analysis to gather information on neighborhood history, assets or strengths and challenges, and current resources and efforts to address youth and community violence by meeting with and engaging community residents and organizations, attending community meetings and events, and gathering information on the neighborhood [10]. Residents’ perceptions of collective efficacy and violence exposure are assessed through a community survey. In addition, a community partner and resident facilitators are identified and engaged in this phase, including preparing for the phase II training by assessing facilitation skills, reviewing training content, and adapting it to unique community characteristics and issues, and recruiting residents for the training. Phase II is a 9-week training program that aims to foster collective efficacy among youth and adult residents, including identifying norms, values, and characteristics associated with youth and community violence, safety, and supportive connections in their neighborhoods. Results from the community survey are also shared and discussed. Participants learn and practice strategies for building relationships and safely intervening in neighborhood problems using restorative justice and nonviolent communication principles and neighborhood activism and organizing strategies for addressing structural causes of violence [13]. In phase III, the final phase of the intervention, a community-based prevention project aims to encourage youth and adults to apply newly developed skills to engage the broader community, bringing residents together to foster collective efficacy and reduce violence.

The CRCEI draws on several theoretical frameworks: positive youth development, minority stress and ecosocial theories, and social disorganization theory. Positive youth development (PYD) posits (1) developmental plasticity (ie, young people can change and change their trajectory), (2) developmental assets exist both internally and externally (ie, the psychosocial conditions within which people live significantly influence well-being), and (3) developmental assets and a strength-based orientation are important for understanding and promoting development [14,15,17]. When assets are present and positive adaptation to life events is promoted, PYD holds that young people can thrive. Such thriving is marked by the “6 C’s of PYD,” or competence, confidence, character, connection, caring, and contribution (to self, relationships, and society more broadly).

Within the PYD framework, relationships matter. Adolescent-adult connections serve as important sources of social support that can foster racial-ethnic identity formation and can protect against violence involvement [8,18,19]. Particularly in the context of heightened police surveillance and systemic inequities, adult supports attuned to youths’ lived experiences can serve as “buffers and bridges” critical for envisioning life paths that diverge from violence [20]. Owing to PYD principles and this evidence, providing opportunities for meaningful and skill-building activities fosters adult and youth relationships, which create protective community environments that are a core component of youth violence prevention efforts and the CRCEI [21].

To address racial disproportionality in exposure to community and youth violence, the intervention draws on ecosocial theory and minority stress theory. Ecosocial theory underscores the extent to which structural racism contributes to constrained access to resources [22], while minority stress theory (MST) explains how racism can translate into disparities in health and well-being for individuals [5,23]. MST has been used to explain health disparities, positing that persistent experiences of bias-based discrimination (ie, racism, experiences of sexism, ableism, and homophobia) contribute to social isolation and emotional dysregulation, which, in turn, may contribute to substance use and mental health problems [5]. Chronic, repeated exposure to early trauma and adversity, such as violence and oppression, can result in toxic stress responses that impede children’s healthy development [24].

Findings from prior work in the study sites indicate that exposure to racism and bias-based discrimination is prevalent, with the majority of young people (82%) reporting at least one such instance (eg, “had the feeling that someone was afraid of you”; and “had people assume you’re not smart or intelligent” because of the color of your skin, language, accent, culture, or country of origin). Emerging evidence shows that interpersonal racism and discrimination prevention should include opportunities to promote racial or ethnic identity, increase awareness of unconscious bias (encouraging critical analytic thinking), shift social norms and policies to promote inclusive behaviors, and foster youth-adult connections [24,25]. Additionally, structural racism and discrimination prevention strategies should include adjustments to policies and practices that differentially impact specific groups of youth, as well as training in diversity and trauma-sensitive care. A critical need exists to test violence prevention interventions that directly bolster protective factors while addressing structural racism and discrimination at the community level.

Social disorganization theory posits that neighborhood-level characteristics (ie, the preponderance of social and economic disadvantage, residential instability, and family disruption) contribute to violence and other social and health problems, while collective efficacy (ie, the preponderance of social cohesion and social control) can protect against these problems [25,26]. Collective efficacy is a neighborhood-level construct comprised of social cohesion (mutual trust) and informal social control, wherein neighbors support prosocial behavior and intervene in problems [16,26]. Engaging youth and adults to intervene in community problems enhances social connections [27], increasing residents’ ability to influence and control their environment [28,29]. Intentional collaborations among youth and adults can increase collective efficacy and promote community well-being [28].

Theoretically, collective efficacy buffers the impact of structural inequality on violence and its sequelae. In neighborhoods where collective efficacy is high, unstructured socializing, antisocial behavior, delinquency, and violence tend to be low, as are violence-related mental health problems (eg, depression and anxiety), even after accounting for disadvantage and disorder [10,16,26,30]. Thus, interventions to increase collective efficacy have the potential to facilitate a variety of positive individual and community outcomes, but the links between collective efficacy and these outcomes are, to date, primarily correlational.

Together, PYD, ecosocial, and MST frameworks suggest that strengths-based, community-level strategies that incorporate young people and caring adults are essential for addressing structural racism, bias-based discrimination, and violence. Meanwhile, social disorganization theory and research indicate that collective efficacy is linked to lower levels of community violence and greater well-being. For these reasons, the CRCEI seeks to reduce community and youth violence by fostering adult-youth connections, raising awareness of structural factors and racism that influence community well-being, and building collective efficacy.

As discussed earlier, the CRCEI takes place across 3 phases. In phase I (approximately 3 mo), community mobilization and engagement takes place. Researchers identify and partner with a community-based organization in each neighborhood whose mission, goals, and programs align with the intervention. The community partner and other community organizations help to identify 3 residents as community facilitators. This phase includes conducting a community analysis, a baseline community survey, and community engagement activities, attending community events and engaging with and getting feedback on the intervention from community residents and organizations. It also includes preparing the community partner and facilitators for the training, including reviewing and adapting training facilitation guides and case scenarios for role plays and other activities for each neighborhood, and recruiting youth and adult residents to attend the training.

Phase II is the collective efficacy training, which includes 9 training sessions over approximately 10 weeks (including a 1-wk break). The training location, dates, and times are discussed and decided with community partners and facilitators. The training sessions are held in accessible community locations, and dinner is provided at the beginning of each session. The youth present and get feedback on their community project ideas in the last session, and community resources are invited to share information and listen to the youth’s ideas. A community project planning committee is also formed in the last session.

In phase III (approximately 4 to 6 mo), the community partner and facilitators work with the community project planning committee to further develop and implement a community project. Each community receives a US $5000 grant to implement the project, which focuses on increasing collective efficacy and social connections, using intervening skills to take action on issues identified by participants in the training, providing alternatives to violence, particularly for young people, and increasing community resources. The project, selected by the community members, may include youth-centered community events and tangible products (eg, rehabilitation of recreational spaces) for young people and the wider community.

The comparator for this intervention follows a similar cadence of interactions with community members, that is, an attention control. A combination of asset mapping and existing partnerships helps to identify a community organization in each comparison neighborhood to host psychoeducational sessions. With each neighborhood having its own unique character, shaped by history, demographics, and local culture, a list of health and wellness topics is developed based on the needs and wants of community members. This list of community needs and wants is generated through community surveying, discussions with the hosting community organization, and feedback from each wellness session. Topics for the health and wellness sessions include physical health practices and resources; mental health practices and resources; child and adolescent nutrition; public transportation resources; Alzheimer disease signs and treatment; and concussion prevention, signs, and treatment. A typical health and wellness session consists of a presentation as well as a question-and-answer session. Each session is led by a speaker from the university, health care delivery system, county, and community organizations (nonuniversity affiliated speakers receive an honorarium for participating), with expertise relevant to the topics identified. The location for the sessions is at a facility that is considered safe and trustworthy by community partners.

For this attention control design, both intervention and comparison sites involve approximately 18 hours of in-person time, generally spread out over 6- to 9-week periods. The program is delivered with some variation in schedules to meet the needs of community partners. Such configurations include, but are not limited, to, 2 health education sessions offered on the same day. We had different participation structures in the intervention and comparison sites to balance pragmatic concerns and resource constraints. The intervention was purposefully designed in prior pilot studies to be implemented with 3 structured phases in this study, therefore requiring more resources than the health sessions in the comparison sites.

The CRCEI was initially funded by the Substance Abuse and Mental Health Services Administration as part of a broader effort across the county to promote community mental health and increase resiliency. Subsequently, the team also received research funding from the Centers for Disease Control and Prevention to conduct a rigorous evaluation of this intervention (a cluster randomized controlled trial across similar neighborhoods is ongoing). As CRCEI was already being implemented in several neighborhoods to assess impact, a quasi-experimental matched comparison group design was used, using participatory propensity score matching to identify comparison neighborhoods with similar baseline characteristics as the sites already implementing CRCEI. Four neighborhoods receiving the CRCEI intervention are compared with 4 propensity-matched neighborhoods receiving the comparison health sessions. The study uses a community-based participatory and partnered approach. For this reason, we offered an alternative intervention in comparison communities rather than “care as usual” or “no intervention” to promote reciprocity. Furthermore, our aim was to match the demographics and characteristics of the communities and use the same data collection methods and measures in the intervention and comparison communities to analyze the aims for this study.

As illustrated in Figure 2, survey data were collected from participants in the intervention and comparison groups at 3 time points: baseline (T1; prior to exposure to the training or health sessions), post training or health sessions (T2; immediately following the 9-wk CRCEI training or last health session), and follow-up (T3; approximately 6 mo after the T2 to coincide with the completion of phase III in the intervention arm). To address secondary aims and inform neighborhood adaptation of the CRCEI, community survey data are collected at baseline (T1) and follow-up (T3), and secondary data on community and youth violence incidence are measured annually. In addition, an ethnographic process evaluation, including implementation science frameworks, is conducted to elucidate how CRCEI is implemented across intervention neighborhoods [31,32]. The evaluation focuses on dimensions of fidelity, reach, dosage, responsiveness, differentiation or characteristics, and adaptation and draws on a mix of observation, feedback questionnaires, and attendance, interview, and secondary data.

This study includes 8 neighborhood sites in the greater Pittsburgh, Pennsylvania, region. Four neighborhoods were initially chosen as intervention sites based on having high levels of gun and community violence incidents and desiring or engaged in efforts to increase community safety and identified as priority neighborhoods by the Allegheny County Department of Human Services, Office of Behavioral Health, interested community partners, and the project Steering Committee. The Steering Committee guides and informs the project and includes community partners, resident facilitators, interdisciplinary and countywide partners, and research team members.

To identify 4 comparison sites, the team used a process we refer to as, “participatory propensity score matching.” Building on the work of the abovementioned stakeholder groups, the research team constructed a list of 29 additional neighborhoods that were similarly high in violence. Steering Committee members then rated the identified neighborhoods in terms of appropriateness and need, recommending the removal of 5 neighborhoods that were least comparable. The research team then calculated propensity scores for the remaining 24 neighborhoods.

The propensity scores were calculated based on measures used to construct the Social Vulnerability Index, which uses 15 social factors grouped into 4 related themes (socioeconomic status, household composition, race, ethnicity, or language, and housing or transportation) [33]. On the basis of local norms and steering committee input, we modified these slightly and included the following variables: population, percent below poverty level, percent age 65 years or older, percent age 17 years or younger, percent older than age 5 years with a disability, percent speaking English “less than well,” percent Hispanic, percent multiunit structures, percent mobile homes, percent crowding, percent on vehicle, and group quarters. Propensity scores were estimated for each neighborhood using a 15-covariate linear regression model, which included the census variables described earlier. Propensity scores were estimated using logistic regression in SPSS, and matching was performed with the FUZZY extension (version 2.0.1) using a nearest neighbor greedy algorithm with a caliper of 0.18. For each intervention neighborhood, the research team noted the top 4 comparison neighborhood matches. The research team then reviewed the recommended matching neighborhoods and data on those neighborhoods with the steering committee and identified the best match (based on score and steering committee knowledge of other factors) for each of the 4 neighborhoods already selected for intervention.

These 8 intervention and comparison neighborhoods struggle with poverty, school “push-out” (disciplinary actions that push youth, especially Black boys, out of the regular school system), and have among the highest rates of gun violence in the county (see Table 1 for neighborhood characteristics).

Within each neighborhood, a different organization works with the research team on the project. In the intervention neighborhoods, community partners whose mission, goals, and programs align with the project are engaged through a formal agreement and scope of work and receive funds to pay for identified staff to work with the research team on the project. In the comparison sites, a community organization is identified that can host and inform the health sessions and is given a stipend to assist with covering the time and effort for hosting the sessions at their facility. Community partners in the intervention neighborhoods include community development organizations, grassroots organizations working to promote safety and provide support for victims of violence and their families, or other neighborhood-based organizations working to improve community outcomes. In addition, 3 residents are hired as community facilitators in the intervention neighborhoods based on their engagement with youth and/or adult residents and in local safety and community organizing and improvement efforts. Community organizations hosting health sessions in the comparison communities include faith-based organizations, community centers, neighborhood associations, and libraries. Asset maps are created for each intervention and comparison neighborhood with the goal of identifying community champions and relevant resources to support the project. The research team relies on partnerships with key constituents in each of these sites (including site coordinators, facilitators, partners, and community members) to facilitate recruitment and retention as described below.

Table 2 contains a description of the primary and secondary outcome measures (Table 2). The primary outcome is self-reported collective efficacy at the individual participant level across 3 points in time (ie, T1, T2, and T3). Collective efficacy is measured with the 10-item Collective Efficacy Scale, which has been previously validated and widely used, including among low-income, African American, and youth samples [28,34,35]. The measure includes a 5-item “social cohesion and trust” subscale and a 5-item “informal social control” subscale. Each item is assigned values based on a 5-point Likert-style scale, and final scores are calculated by averaging responses across items (range 1‐5).

The evaluation has 2 groups of secondary outcomes, those at the participant level and those at the neighborhood level. Intermediary, individual-level beliefs and behaviors related to collective efficacy are measured, including participants’ willingness to intervene, social connections, and prointervening norms and values. Willingness to intervene is a modified version of the Collective Efficacy social control subscale, with questions focusing on whether the participant is likely to intervene in social problems (ie, in lieu of whether they perceive their neighbors as being willing to intervene) [36]. Social connections are measured via 3 questions asking how many neighbors (youth and adults) that participants know by sight, name, and talking to on a regular basis (range 0 to >10). Community norms, values and attitudes about intervening is a scale asking participants the degree to which they agree with 10 statements related to norms and values pertaining to intervening in social problems [10]. Exposure to violence [37] is measured using 17 items with binary yes or no response options, including 9 items capturing direct exposure to violence and 8 items capturing indirect exposure to violence (ie, witnessing violence). This exposure to violence measure has been previously validated [37]. Exposure to violence values are calculated by the proportion of respondents endorsing each item (ie, responding “yes”). These data are collected from residents in the intervention and comparison neighborhoods via the community survey and at 2 time points, baseline and follow-up.

At the neighborhood level, we measure collective efficacy using the same measure of collective efficacy described earlier [26], which is aggregated to a mean value based on responses among all participants within a given neighborhood on the community survey. Prior research uses self-reported survey measures to measure collective efficacy at the individual and neighborhood levels. We measured collective efficacy at the individual level by surveying participants in the intervention and comparison sessions. We also measured collective efficacy at the neighborhood level through our community survey, and similar to Sampson et al, the responses are aggregated to the neighborhood level.

Finally, violent crime data for City of Pittsburgh neighborhoods are obtained from the Pittsburgh Bureau of Police open data portal, while data for municipalities outside the city (eg, Wilkinsburg, Duquesne, Turtle Creek, and Braddock) are drawn from the Pennsylvania Uniform Crime Reporting system and Allegheny County sources, including police blotter data and the county’s integrated data warehouse.

Eligibility for participation in the community survey requires that participants reside in the specified neighborhood, meaning that they hold an address within the preselected community or communities, and speak English. Initially, only individuals aged ≥18 years were allowed to complete the survey, but this enrollment criterion was later amended, beginning in Homewood in November 2023, to include individuals aged ≥13 years due to the importance of understanding youth experiences and perspectives.

Youth aged ≥13 years and adults who speak English and live within the identified neighborhoods are eligible to participate. Individuals can participate in the CRCEI or Health Sessions without participating in the research procedures.

Individual participants in the intervention training and the comparison health sessions are sampled by availability. Comparison and intervention sites use the network of community partners identified in the asset maps created for each neighborhood. This includes site leaders, program facilitators, recruiters with strong connections to their community, prevention specialists embedded in schools, and tabling at community outreach events. In the intervention sites, the research team also works collaboratively with community partners and facilitators, as well as community residents and organizations engaged in phase I of the intervention to recruit youth and adult residents to participate in the training. In addition, each community partner staff and resident community facilitator personally identifies and recruits residents in their personal and professional networks. Participant recruitment started in February 2023 and continued until the eighth neighborhood completed enrollment, through November 2024.

Individual youth and adult participants in the community survey are sampled randomly and by availability. In the first 2 intervention neighborhoods, availability sampling that primarily relied on facilitator recruitment did not yield adequate response rates and raised concerns about sampling bias. The team then adapted the data collection strategy and moved toward random sampling, adapting the Centers for Disease Control and Prevention’s CASPER model [38]. A dataset of residential addresses from county parcel, address, and property assessment data was assembled, and 30 residential addresses were randomly selected [39], their related US Census blocks were extracted, and those blocks were used to survey each of the 8 participating neighborhoods. The research team canvasses door to door, inviting participation in the community survey from each residence within each selected block. The number of blocks across the 8 neighborhoods ranges from 12 to 28 blocks. The number of addresses per neighborhood ranges from 524 to 2403. Given time limitations to the team’s ability to visit addresses multiple times, inability to reach some residents at home, and underrecruitment of community survey participants, random sampling is supplemented with recruitment of participants in public areas and community events (ie, availability sampling).

Descriptive statistics will be used to summarize the sample with regard to baseline characteristics of interest. Means and SDs will be presented for continuous variables, while sample proportions will be provided for categorical variables. Ninety-five percent CIs will accompany all sample statistics.

Analyses are tailored to the quasi-experimental design (with considerations for clustered data by neighborhood) and aligned with the study aims. The primary independent variable of interest is treatment assignment at the neighborhood level (intervention vs comparison). Dependent variables include both individual-level outcomes (eg, collective efficacy and exposure to community violence) and neighborhood-level outcomes (eg, collective efficacy aggregated to the neighborhood, police-reported crime, and violence rates). Models will incorporate both fixed effects (eg, treatment, baseline values, and covariates) and random intercepts to account for clustering where appropriate. Inferences will rely on cluster-robust SEs with small-sample corrections (df=6). Effect sizes and 95% CIs will be reported throughout, with emphasis on estimation and interpretation of magnitude rather than binary hypothesis testing.

The individual is the unit of analysis, but clustering within neighborhoods is explicitly modeled. The independent variable of interest is neighborhood-level treatment assignment, with covariates including baseline individual scores, demographic characteristics (eg, age, sex, race or ethnicity, and education), and relevant neighborhood-level indicators. Continuous dependent variables will be analyzed using linear mixed effects models of the form:

where YijY_{ij}Yij is the outcome for individual iii in neighborhood jjj, XijX_{ij}Xij are individual covariates, uju_juj is a random intercept for neighborhood, and εij\varepsilon_{ij}εij is the individual-level error term. The random intercept uju_juj captures unobserved contextual influences shared by individuals in the same neighborhood (eg, social environment, policing, or community resources). By including uju_juj, the model allows each neighborhood to have its own baseline level of the outcome, and treatment effects are estimated net of this shared clustering. For binary dependent variables (eg, presence or absence of exposure to violence), generalized linear mixed models with a logit link will be used, also including a neighborhood random intercept.

The intercept term represents the expected outcome value when predictors are set to 0 (or to reference categories). In the context of these analyses, a random intercept for neighborhood allows the model to adjust for systematic differences between neighborhoods not explained by observed covariates. This ensures that individuals are not treated as statistically independent when they belong to the same neighborhood and that SEs correctly reflect the clustered design. Fixed intercepts anchor the regression line, providing the baseline level against which covariate and treatment effects are interpreted.

The neighborhood is the unit of analysis. The dependent variables are neighborhood-aggregated outcomes (eg, mean collective efficacy scores and rates of crime or violence). The independent variable is treatment assignment. The primary analysis will use an analysis of covariance specification, where the neighborhood-level outcome at follow-up is regressed on treatment assignment and the baseline value of the same outcome. This model provides an adjusted mean difference in follow-up outcomes between intervention and comparison neighborhoods, controlling for baseline heterogeneity. As only 8 clusters are available, models will not include random effects but will instead rely on cluster-robust variance estimators with small-sample corrections. Adjustment for additional neighborhood-level covariates (eg, baseline socioeconomic disadvantage, racial composition, and baseline crime levels) will be included cautiously in sensitivity analyses, given the limited degrees of freedom. These covariates will be added one at a time to assess the robustness of the primary results. Sensitivity analyses will also include difference-in-differences models, where the dependent variable is the change from baseline to follow-up. Results will be expressed as standardized effect sizes for comparability across outcomes and to provide planning parameters for future trials.

Missing data will be addressed using full information maximum likelihood in the mixed effects models, which provides unbiased estimates under a missing-at-random assumption. As a sensitivity analysis, multiple imputation by chained equations will also be conducted for key outcomes and covariates. Both approaches preserve variability and prevent the downward bias in SEs that can arise with complete-case analysis. The extent and patterns of missing data will be reported, and the results will be compared across approaches to confirm robustness.