Engagement is Not an Option: The Essential Role of Community-Engaged Research for Implementation Science,in Community Violence Studies and Beyond

Abstract

While community-engaged research would seem to have significant overlap with both implementation science and program and policy evaluation, these respective bodies of work have developed largely apart from one another. We argue that disregarding community-engaged epistemologies (or engaging only superficially) when studying how policy or programs “work” significantly increases the likelihood of producing invalid science. Accordingly, we present a case study of our work with a local community-based violence intervention operating in the Northeastern United States to demonstrate how principles of community-engaged research are essential to implementation and evaluation science work.

Introduction

The importance of “communities” has increasingly come into focus in recent decades alongside a variety of issues of concern within criminology and criminal justice (CCJ) (e.g., Gill et al., 2014; Heilbrun et al., 2012; Johnson & Stylianou, 2022; Werth et al., 2020), a “turn” that has seemed only to accelerate. Under guises such as “citizen science,” “participatory research,” and “engaged scholarship,” scholars have also called on their colleagues to more directly engage communities in scientific work, and a rapidly-developing cross-disciplinary body of work has asserted that community-focused, community-reflective, and/or community-involved approaches improve science (e.g., Bracic, 2018; Mikesell et al., 2013; Moczygemba et al., 2023; Ortiz et al., 2020). Arguments supporting non-scientist individuals’ and communities’ involvement in science are even more relevant when researchers seek to proclaim truths about policy arenas critical for community outcomes, such as the nature of crime and justice, or seek to influence the policies and programs meant to directly affect these individuals and/or communities.

Broadly, a policy is a rule, principle, or norm which guides and/or governs present or future actions, often achieved through programs, structured systems – of plans, people, actions, organizational processes, etcetera – through which action is taken to achieve a goal. The fields of policy studies and policy analysis provide the expertise for how best to study the development and design of policies scientifically; these scholars have also begun to address the role of community engagement in doing so (e.g., Welton et al., 2020). Policy enactment must be managed, and scholars in business management, organization studies, public administration, public management, and public affairs have also examined the role of communities in the production of science on policy management (e.g., Schafer, 2019). However, more than simply managed, policies must be implemented, put into effect, carried out, executed. Implementation science is that body of accumulated general truths, obtained through systematic study, about how policies, programs, and other action decisions are best put into effect. Unfortunately, though they have studied community settings and community-based efforts (e.g., Balis et al., 2024; Yapa & Bärnighausen, 2018), to our knowledge, implementation scholars have yet to build a significant body of work on how to best engage communities in their science or on what value, if any, community engagement in science serves (though see McIlduff et al., 2020, for one example).

We hope to assist in placing CCJ scholarship “at the helm” of discourse on the role of community in implementation science. In what follows, we argue that community-engaged research is essential to the scientific endeavor. We briefly describe implementation science, given its relative unfamiliarity among CCJ scholars, then illustrate how community-engaged approaches enhance research by using our work as an example. We find that deeper engagement permits researchers insight into aspects of phenomena otherwise hidden, facilitates more accurate and precise understandings of phenomena than would be likely “from a distance,” and expands the scope, range, and number of opportunities to draw upon the knowledge and wisdom of the researched. We end by discussing how policy- and program-oriented scholars can integrate community engagement and implementation science principles to enhance their work.

Literature Review

The Foundational Role of “Community” in Science

Science is an abstract noun, referring to the accumulation of general truths (what “actually is”) obtained through systematic study (Benjamin, 1949), as well as an action noun, the act of systematically studying in search of truths (White, 1938; see Oakley, 1998, on the origins of “science”). Systematic truth-seeking separates science from, say, conjecture or opinion-based argumentation (Purtill, 1970). Both the content and act of science are characterized by assumptions about the nature of “reality,” or, at least, our experience of it (ontology), how to discern truth about “reality” (epistemology), how to discover truth (methodology), and the value and purpose of truth (axiology). Taken together, one’s stances on these dimensions set the foundation for one’s scientific work.

Community engagement entails external parties’ efforts to obtain and maintain a community’s interest (e.g., curiosity, desire to know) and involvement (e.g., support, participation) in order to advance a specific goal (O’Mara-Eves et al., 2015; Werth et al., 2020). Manifold goals may motivate community engagement, including persuasion, education, mobilization, even pacification. Community engagement can entail unidirectional action, primarily characterized by activities directed toward either the external party or the community by the other. Engagement may also instead be bidirectional, primarily characterized by a relative balance of action directed by and toward each party. In addition, community engagement can be categorized by whose goals it is intended to serve. It may be undertaken in order to achieve goals determined by and/or benefitting the external actor, goals determined by and/or benefitting the community, or goals collaboratively determined and/or mutually beneficial by both sets of parties. Determination and benefit are, importantly, different; she who decides has power, and she who benefits is made better off from the engagement, but the two need not be the same. Figure 1 illustrates how these dimensions can overlap. In our research, we tend to strive for mutual benefit, a hallmark of feminist methodologies (Cancian, 1992), though this approach may not be necessary, perhaps not even advisable, in all situations and for all researchers.

Figure 1.

Modes of community engagement by direction, action, and benefit

Our understanding of science is informed by inclusive approaches to science, significantly influenced by feminist works. We argue that community engagement intersects with the very foundations of science. For example, positivism and constructivism are the major ontological positions in social science (Aliyu et al., 2014). Positivists assume an objective reality, inclusive of social reality, which can be accurately and consistently measured if only scientists find the appropriate means (Oakley, 1998). Yet, truth-seekers inevitably approach truth-seeking from a specific, subjective standpoint, which limits what they can “see” (Cancian, 1992; Hallen, 1989). Put another way, truth is “bigger” than the “field of view” of any one person; it can only be fully “seen” by combining the views of many individuals standing at differing vantage points (see Illustration 1, first graphic). This observation has been made extensively by feminist philosophers of science, only one among several caveats they highlight regarding the knowledge gained through “traditional” positivism (see Chafetz, 1997, for summary).

Illustration 1.

How different perspectives converge to observe (positivist) or construct (constructivist) reality

Constructivists assume the primacy of a social reality that is largely constructed in the imagination and sensory interpretations of individuals (Aliyu et al., 2014). Of course, this perspective does not require denial of an objective, external reality; only that each individual’s experience is mediated through their own various sensory and processing mechanisms and that this should be the primary object of scientific study. Feminist scholars have been more inclined to adopt this position (Cancian, 1992; Gorelick, 1991; Oakley, 1998; Wylie, 2012). Still, much of their work has noted that, when several individuals have overlapping imaginings that something is true and all act on this assumption in concert, this “truth” has such totalized impact as to have the effect of “objective” truth (see Illustration 1, second graphic). Moreover, the perceived realities of any single individual, while ultimately mediated through that individual’s experience, often originate in forces broader and larger than the individual’s (Gorelick, 1991). One's community is one among several external contexts from which these broader forces emerge.

As communities are social entities reflecting individuals’ shared and/or co-constructed realities and which serve as contextual and moderating factors influencing most realities explained by theoretical or empirical models, an ontology that negates the role of communities is incomplete and inadequate. Accepting that communities are critical components of the realities we study then, discerning the truth of these realities requires epistemologies concerned with communities. Such epistemologies must contend with the notion that scientists simply cannot adequately assess the veracity of community-level ontological truths except by community verification (Koskinen, 2023; Potter & Hepburn, 2005). External onlookers may document observable aspects of stigmatization, including in patterns of others’ behavior (e.g., Blount-Hill et al., in press; Evans & Blount-Hill, 2022; Evans et al., 2025). However, what it feels like to be stigmatized and how it feels to stigmatize others, the motivations for these processes and how they are perceived, understood, and reacted to, all are best described by those directly involved and impacted (e.g., Abulafia et al., 2024; Babu & Blount-Hill, 2025a, 2025).

Accordingly, accurate discernment of the truth of social realities requires, then, that social scientists utilize scientific methods appropriate to access community perspectives, and to enhance the likelihood of the completeness and truthfulness of community input. Community-engaged embeddedness (or embedded community engagement) is one such method. For example, people do not comprehend or notice in great detail several of the social phenomena in which they are involved. A single point-in-time interview can certainly capture an individual’s perception of some or another topic, but these same perceptions may shift in a nonlinear fashion depending on when that interview is conducted. In fact, upon later reflection, a person may even revise their previous understanding and interpretation of their experience. Thus, long-term, ongoing engagement permits researchers access to thoughts and impressions of research participants in timeframes that allow those thoughts to form, settle, and/or evolve. Analyzing a phenomenon across multiple interaction modalities provides better context for communication and external observation, but often requires prolonged contact with research subjects. The resulting familiarity is invaluable, though. Knowing the colloquial language of a community, for example, may lead to a completely different understanding of the meaning and/or affect associated with the exact same set of words. Going still further, and seeking collaborative involvement of community members directly in the research process, facilitates the contribution of their experience and knowledge even more, aiding in the interpretation of data and in understanding the various implications of findings.

Feminist scholars have long argued the need to incorporate impressions of those “closest” to phenomena (Chafetz, 1997). Yet, as Gorelick (1991) puts it, “A subject population does not tell the truth to those in power” (p. 461). This is especially true when the researcher represents a member of the powerful elite and the research “subject” occupies a more marginal social position. So-called “mainstream” theories of intergroup psychology suggest the same (see, e.g., Blount-Hill, 2025, footnote 12). Alternatively, consistent engagement may engender trust and elicit more authentic interactions despite status difference. Overcoming the “trust barrier” creates opportunities for the joining of research and researched perspectives that best approximates “objective” truth. Importantly, feminist scholars have been foremost in identifying how distal processes significantly shape the everyday experience of the “average” person, especially those least in power (Gorelick, 1991). Thus, the mechanisms in which some phenomena originate may be too far to be understood by those they impact. Combining researchers’ often-more-privileged “external” and “outsider” viewpoints with internal, insider access to research subjects and those subjects’ viewpoints over a long enough period of time can reveal truths otherwise imperceptible to either researcher or subject on their own. Restating our basic point, then, the ontological and epistemological bases of “community” necessitate methods involving communities’ engagement.

Axiology concerns the values, ethics and morals that guide the scientist and motivate whether, regarding what, and why the scientist conducts research. Most scholars advocating community engagement do so on the basis of ethical and moral arguments and such grounds have been discussed at length. To note just a few of these arguments, communities should: (1) benefit from the research conducted on them, including the benefit of learning from research findings (Hughes, 2014), about the research process and its associated skills (Roth & Lee, 2004), and how to interpret and utilize research findings themselves (Robinson et al., 2024); (2) be able to have some research efforts directed toward the issues, concerns, or challenges that they prioritize (Koskinen, 2023); and (3) have their voices amplified, especially to those in power, through the legitimizing medium of science (House, 2020). Feminist scholars have been especially prominent in arguing for research methods that foreground the perspectives of communities and place power and control in the hands of those being researched (Gorelick, 1991). We find Cancian’s (1992) work particularly compelling on this point. We would also highlight the assertion, by self-styled inclusive criminologists, that societies have a right to accurate and complete knowledge (Babu & Blount-Hill, in press; Babu & Blount-Hill, 2024; Blount-Hill et al., 2022), which must include knowledge only accessible within communities. As with all other sciences, then, implementation science must account for the essential role of community engagement.

Implementation Science

Implementation science has come to refer to a body of works,

characterized by three aims, namely (1) to describe the process of translating research into practice (process models), (2) to understand what influences implementation outcomes (determinant frameworks, classic theories, implementation theories), and (3) to evaluate the implementation of interventions (evaluation frameworks)

(Braithwaite et al., 2018, p. 6).

Implementation science is cross-disciplinary, focused not just on whether an intervention (e.g., policy, program) is effective, but also on how and under what circumstances the intervention is effective, and what strategies promote its utilization (Taxman, 2025; Wilson & Kislov, 2022). Implementation science has a long history in the medical field, dating back to the nineteenth century work of Florence Nightingale (Wilson & Kislov, 2022). Still, implementation science is not commonly drawn upon in CCJ (Taxman, 2025). Without working more closely with practitioner communities to understand their implementation – especially what factors hinder implementation with fidelity – there will continue to be a large gap in translation from practice to research to practice, and on. The community of scholars who take an explicitly feminist approach to science are among the most committed to a science that translates to action, particularly to social change (Cancian, 1992).

While policy implementation is a subject of study for several CCJ scholars (even if not under the label “implementation science”), evaluation studies constitute perhaps the primary way that CCJ researchers engage in the study of program implementation. Generally, evaluations are studies of whether and how a specific cause initiated to resolve a specific problem brings about a predicted effect within a specific field of action (Deniston et al., 1968; Weiss, 1999). The primary difference between evaluation and more general hypothesis testing is that the latter is a “basic science” function, designed to produce generalizable knowledge, while the former is “applied science,” meant to produce knowledge specific to the subject studied (Niiniluoto, 1993). Whether a cause brings about (or is associated with) its intended effect is the subject of an outcome evaluation. This entails applying methods of scientific explanation to testing the following hypothesis: [Something] causes (or is correlated with) [some intended effect/outcome]. In CCJ, the most common “causes” to be evaluated are policies or programs. Often enough, we also evaluate models, idealized frameworks for action.

How a cause engenders one or more outcomes is the subject of a process evaluation, and typically entails applying methods of scientific exploration or description to present a continuous or stepwise causal process. Process evaluation may be approached in different ways. The evaluator may describe or explore the process by which observed outcomes are brought about by policies or programmatic activities (i.e., exploratory causal process evaluation). Alternatively, the evaluator may conduct a hypothesis test: [Program] causes [outcome] by [prescribed process] (i.e., explanatory causal process evaluation). Model evaluations often are designed with this approach in mind, to test not only how activities bring about outcomes, but whether these outcomes are brought about through the specific process presumed by a given model. Still, some programs may be initiated with a certain model in mind, but fail to ever implement that model as intended. In such cases, evaluations of whether a program maintained fidelity to its espoused model constitute another category of process evaluation (i.e., fidelity evaluation; Miller & Miller, 2015). Unfortunately, many CCJ process evaluators are often unclear regarding which of these process evaluation types they are engaged in (e.g., fidelity versus causal process).

Regarding the “whether” question (i.e., whether the cause is associated with its intended effect), the vast majority of evaluations in CCJ are based on assessments of correlation, rather than causation, a weaker basis on which to make strong, generalizable claims. We would argue that the “how” question is often much more important and insightful, and may help fortify the limitations of correlation-based outcome evaluations. While a researcher might determine statistical association between the program (independent variable) and an outcome (dependent variable), answering the “how” question reveals the explanation for the relationships observed (see Corley & Gioia, 2011, defining theory). Epistemologically, not only can theory support hypothesizing about, and designing statistical study of, causal relationships (in fact, these activities must be based on theory), the inability to answer this question suggests, at best, incomplete explanatory models and, worse, the possibility of spurious results. Methodologically, though all research questions should be addressed through a variety of methods, the “how” question tends to make this imperative much clearer. Axiologically, policymakers, the public, and specific communities tend to be much more interested in the various pathways through which something might be achieved than merely whether it was achieved or not (see, e.g., Blount-Hill & Szkola, 2025, on practitioners’ desire for socially-just violence prevention). Knowing how something works allows one to apply value judgments to assess whether alternative pathways might be preferred, as well as economic judgments about whether modifications to a causal pathway might increase relevant input-output ratios.

If an evaluation is of a policy or program intended to effect, influence, or benefit a community, or be implemented within or by a community, or work through or because of a community, etc., then the necessity of community engagement is evident. This truth is magnified for any policy or program described as “community-based.” Thus, community-engaged research on community-based interventions provides an excellent resource to illustrate how a community-engaged approach enhances the scientific endeavor. We present our work with a community-based violence intervention as an illustration.

An Example: Community-Engaged Scholarship on Community-Based Violence Intervention

The Community Violence Prevention Project (CVPP) is a collaborative effort of the two authors to assist community-based violence interventions (CBVIs) and agencies and organizations that support these interventions. Much of our work relates to how CBVIs are implemented. Approximately three years ago, a CBVI program requested that we conduct an outcome evaluation of two replications of the Cure Violence (CV) gun violence intervention model in a large city and a smaller neighboring city in the northeastern United States. The CV model is built on the assertion that violence is “transmitted” between individuals like a contagious disease (Slutkin, 2013). CV programs rely on three strategies to address violence: (1) direct interruption of potentially-violent conflicts through conflict mediation, to stop the spread of violence; (2) long-term mentoring and case management for those most at risk of involvement in violence, treating and “curing” those already “carrying” the violence “disease”; and (3) public messaging that violence is unacceptable to a given neighborhood community, thus changing behavioral and/or cultural norms that, if left unchanged, would catalyze and spread violence in a neighborhood (Butts et al., 2015). Importantly for this program model, services are delivered by “credible messengers,” usually local residents whose turn from violence enhances their ability to inspire intended program participants’ belief in a message against violence (Roman et al., 2026; Szkola & Blount-Hill, 2025). The program that contacted us was hosted at a local university and funded by state government. Its leadership had implemented the model years earlier, with different staff, but suspended implementation due to lack of funding. The program was now being restarted.

When the program contacted us, its service provision had not yet begun. Concerned that an outcome evaluation of the program would not be feasible within its current funding period, we instead offered to provide technical assistance; conduct ongoing evaluability and fidelity assessments; and collect survey data on service area residents’ attitudes towards violence as a baseline for later evaluators’ comparative purposes.

Embeddedness as Method

By being embedded, we mean consistent and frequent contact with program personnel, relatively unobstructed access to observe daily operations, participating across a wide gamut of program activities and operational functions, and developing a somewhat “insider” status as, in a sense, “a part of the team.” Being embedded provided us with a way of building trust with program staff and knowledge of program activities “in real time,” which also permitted us opportunities to provide feedback at many points throughout program implementation. Because frequent contact provided the opportunity to make relationships, we could observe day-to-day operations without staff feeling as if they needed to alter their behavior as people often do in the presence of strange guests.

We conducted weekly phone meetings with the program’s executive director and conducted over 30 visits to the site. Weekly phone meetings included updates, discussion of challenges, and collaborative problem-solving. Site visits consisted of informal “check-ins” and discussions with staff, both as a group and one-on-one. We also attended scheduled and impromptu meetings with staff and with external stakeholders and partners, including the funding agency. Throughout, we took field notes and debriefed with each other often.

Our embedded approach differs from typical evaluation. More often, researchers begin after a program has been established and operating for several years. Analyses of the implementation process is usually based on retrospective qualitative response methods, like interviews and focus groups, though sometimes accompanied by descriptive analyses of program administrative data, assuming relevant and reliable data are available. This “non-embedded” approach offers the advantage of “neutral” disinterest and its presumed assurance against bias. It is also adequate, in our view, to determine whether a program “works” or not, if “works” means empirical correlation with a stated outcome. Nonetheless, in our experience, these approaches are often inadequate to determinatively assess how or why a program works.

Embeddedness requires the researcher to obtain and maintain a community’s interest and involvement in order to advance a specific goal – that is, to pursue community engagement, in this case, with the program community. Only through sustained and in-depth exposure to the program did we realize the scope and depth of several major implementation challenges. In what follows, we present overall “lessons learned” from taking this community-engaged embedded approach, based on our field notes, debrief materials, research outputs, and reflections.

Lessons Learned

Reviewing and reflecting on our notes, research materials, and memories, numerous factors affected the program’s operation, implementation, and future evaluability, which we learned only through our embedded approach. Had we not used this approach, we might have carried on our work misunderstanding the program’s operations and misrepresenting its impact.

One initial challenge was bridging program staffs’ and researchers’ understandings of the underlying theory guiding the program’s professed model. Embeddedness granted us the opportunity to help enhance the program’s fidelity to its model, allowing greater confidence in attributing its outcomes to the model versus idiosyncrasies in implementation (Miller & Miller, 2015). For example, initially, the program’s service area was to include two police districts, a larger service area than is typical of CV programs. In fact, at least one previous evaluation of a CV implementation suggested that the model was not appropriate for such large service areas (Fox et al., 2015). Nonetheless, our program partner had promised to serve these districts while seeking the support of the local politicians allocating funding for the program. At first, program administrators did not understand this as a deviation from the CV model. Building from our familiarity with CV program guidelines, we were able to persuade them to reduce the size of the program’s service area. Using geospatial analytics of shooting patterns in its service area, we helped them designate areas of more manageable size where gun violence was concentrated. Without being embedded, though, we would not have discovered this difference in our conceptualizations of the CV model until much later and we would not have been able to correct it. Still more, our proximity sensitized us to the more-frequent-than-anticipated political and similar pressures impinging on model implementation.

During one weekly check-in call, we were informed that, because shootings had declined over previous months, the program would be modifying its service areas to new violence “hot spots” to show that it was (a) responsive to real time crises and (b) effective, namely by showing declining numbers of shootings in “hot” service areas as opposed to maintaining low numbers in areas that had gone “cold.” While some violence interventions may be designed for this kind of mobility, the CV model assumes a workforce with historical social ties to specific socio-geographic areas; these individuals often lack similar connection to other places, even in the same city. If the program relocated, there was a significant chance that its staff would lack the necessary “credibility” to carry out the model in the new service areas. Notably, we were told that administrators had adjusted program service areas during their previous implementation of CV, with unclear impact on the validity of prior-reported evaluation results. This time, embeddedness allowed us to intervene. Drawing on CV guidelines, we were able to convince program leadership to maintain relative permanence in its service areas, with only infrequent adjustments over longer timeframes. Less involvement with the program would likely have meant only being alerted to service area changes several months or years after, if at all.

The credibility of CV staff is usually related to their previous histories involved in “street life,” including violent or criminal networks, a commonality they share with the individuals they now seek to persuade to renounce violence. However, from its outset, the program had significant difficulty persuading its host university to hire individuals with criminal records. Individuals, though, with this history who, nonetheless, have reformed their lives and identities, possess the interpersonal skill needed to work with “hard to reach” people in sometimes tense circumstances, and have the passion and moral commitment to persist in this work despite its challenges, are few and far between. Not just anyone with a criminal record will have these skills and characteristics. We became privy to a repeating process of the program identifying one of these few uniquely-qualified candidates, having them languish in the hiring process, trying to maintain candidates’ interest, watching candidates take jobs elsewhere rather than wait, all the while placing pressure on existing staff to maintain positive outcomes. The team was chronically understaffed in its larger city; approximately nine months after the program’s documented start date, it had only two staff members, with its executive director functioning as interim program manager. No team was in place in its smaller city until the second year of its funding period, with the program, up until that point, existing only “on paper.”

Most CCJ outcome evaluations are paired with a fidelity evaluation, and thus the lack of staffing would likely have been discovered by most evaluators. We have heard about the challenge of hiring CV staff in single-point interviews often enough, across jurisdictions, so we do not claim once-in-time interviews would not have revealed this hiring challenge. However, through long-term engagement with the program, initiated almost at its start, we were able to better appreciate the scope of this challenge. By year three of its implementation, the program was still not fully staffed in either site, making an evaluation inappropriate and almost certainly misleading and unethical. Not only was the challenge of hiring of a scope which previous CV literature did not prepare us for, but it was also underemphasized in program administrators’ representations to external partners, in which they more often aimed to portray the program in its most positive light. Our closeness engendered the kind of frank engagement that enabled us to document the true significance of this challenge as an external threat to the ability of the program to maintain fidelity with its intended model.

Recall, the program operated two sites in close, but independent, cities. Its executive director had longstanding, historical relationships with stakeholders in the larger city. In contrast, the director had significantly fewer connections in the smaller city. This likely explained the comparative difficulty in recruiting and hiring staff there; only one staff member was hired in the smaller city after two years of program implementation. More significantly, the implementation of CV actually had a negative history in the smaller city, where previous promises of a fully-funded implementation had not been kept and those qualified to be CV staff members did not accord this latest attempt requisite credibility. Still more, an influential community activist proactively campaigned against the program after his negative experience with the previous implementation attempt. Programs may not always highlight challenging histories associated with their brand or personnel for fear of losing the confidence of funding authorities. This information can likely only be gained through expanded engagement, building staff members’ trust that their disclosures will be handled in a way that protects their reputations and fully explains their points of view, which proved absolutely critical for understanding the program’s small city implementation challenges.

Of course, much has been said about the need for better data collection and management by practitioners, especially nongovernment service organizations, and especially community-based, “grassroots” organizations. Concerning our partner program, we found it difficult to reconcile anecdotal accounts of programmatic activity, our in-field observations, and the program’s records. Four factors seemed most salient in this challenge: (1) similar to other “nontraditional” employees with incarceration histories and “disadvantaged” backgrounds, the program’s staff lacked digital literacy and technological skill; (2) staff were unfamiliar with professional data collection and management and struggled to prioritize learning these skills and taking time away from service delivery; (3) program leaders lacked expertise and training in evidence-based decision-making, resulting in the underutilization of data in operations and strategic planning; and (4) the data system provided to the program was not well designed with this type of user in mind.

Drawing from strategies detailed by other scholars in implementation science (Powell et al., 2015), we implemented an audit-feedback plan, providing initial training and acculturation related to data, then conducting several impromptu audits of data input practices, and provided additional training when necessary. Key to this plan was staff’s view of us as “internal” trusted associates who wanted to help them mitigate challenges, not just document them. We developed a rapport that allowed staff to feel comfortable admitting their struggles with data entry and data management in a relational context in which they believed they would not be judged and it would not be used against them. Through continued contact with staff, we could take advantage of opportunities to reinforce previous lessons and messaging related to data, and could determine when modifications to our plan were needed. Nevertheless, we ultimately abandoned the effort after several attempts were unsuccessful. Other implementation challenges came to dominate the attention of the program staff. Being embedded is not a panacea for common implementation and research-related challenges, and evaluation of community-based programming will be hard under even more favorable circumstances. But our embeddedness also permitted a much clearer view of the oft-unaccounted-for heterogenous moderators influencing apparent empirical relationships reported throughout CV and CBVI evaluation literature.

Discussion and Conclusion

In our view, community engagement is critical for scientists to obtain the most comprehensive and accurate understanding of virtually any social phenomenon. We find that it is impossible to examine either “objective” reality or reality forged through shared construction without including and accounting for the perspective of communities whose standpoints are uniquely held and otherwise inaccessible. For several phenomena, researchers can hardly know if their findings are valid without verification by the communities who experience the phenomena studied, and, often enough, this verification process is best facilitated using multiple, overlapping methods over time. Of course, many others have noted compelling ethical and moral reasons for community engagement, including, among others, the right of communities to participate and have access to knowledge (Mathiesen, 2015) and the right of societies to accurate and complete knowledge (Blount-Hill et al., 2022).

While communities play an essential role in all of science, we have noted their particular relevance in the study of implementation. First, formal and systematic approaches to the study of implementation may be even more neglected than sustained community engagement as a feature of CCJ scholarship. Accordingly, CCJ scholars must more explicitly accept the value of studying implementation. CCJ’s embrace of implementation science also provides an opportunity. Scholars of implementation have not focused as much as some in other fields on the necessity of community engagement for their work. CCJ scholars might be positioned to lead in addressing this relative absence.

While interventions related to fields like medicine and public health may be more often inspired by science at the outset, CCJ-related interventions are often designed and implemented without the direct input of scientists. More often, CCJ researchers are called upon, if at all, after a CCJ policy or program has been implemented to provide evidence of its efficacy. Understanding that evaluation will be the most likely entry point for community-engaged implementation science, we give particular attention to this subject. CCJ scholars have generally tended not to prioritize process evaluation, or not to integrate process findings into impact assessments with much rigor, which, incidentally, is where both community engagement and implementation science are most clearly required. Process evaluation is a crucial variant of implementation science and likely a primary way for CCJ researchers to enrich the validity of our evaluations while demonstrating the usefulness of science to our implementing partners.

We hope we have made a compelling case, then, for why community engagement is “not an option” for accurate and comprehensive evaluation or implementation science, or, for that matter, more basic, theoretical science. Referring to our own work with a community-based violence intervention (CBVI), engagement with practitioners permitted us to understand its implementation process with significantly more clarity than would otherwise have been possible, with implications for eventual assessments of outcomes as well. Perhaps most importantly, we observed how significant the institutional and political context of the program was for its effectiveness in achieving its intended outcomes. External bureaucratic processes from its hosting agency, felt pressure to respond to the requests or desires of politicians responsible for its funding, conflicting motivations regarding adherence to its purported model, and the legacies of its previous implementation and institutional affiliation all were significant, identifiable factors in its ability to hire and maintain appropriate staff, obtain neighborhood community support, and operate as necessary given its aims (see Lynn et al., 2000; Sampson et al., 2013). These became far more salient than we initially anticipated, as important for the program’s outcomes as its stated model. Moreover, we recount several times when we cast aside neutral and distant observation to point out to our study subjects when these non-model factors seemed to push them to act in contradiction to their intended model. Again, we believe this greatly enhanced the degree to which the program maintained fidelity to its intended model, crucial for generalizing any of its outcomes to similar implementations.

The implementation challenges we found can be understood better by tracing them to specific political, institutional, social, and historical contexts. Observing these contextual factors and how they influenced decisions made by the program team allowed us to better account for the heterogenous moderators that often go unobserved but nonetheless influence the implementation of a program that is situated within a community. For example, staffing challenges related to hiring those with a criminal record arose from the institutional context of being hosted at a university, and hiring challenges also arose from the historical context of mistrust and failed implementation in the program’s smaller city site. The strong influence of political context was directly observed in the program’s early decision-making about where to work or whether to retain its original service areas. In both circumstances, decisions on implementation were not motivated primarily by reducing gun violence, but instead by responding to external political pressures to maintain support for the program, though this responsiveness would violate the program model’s theoretical assumptions. In instances like these, we learned that being embedded in the program community provided us with a more nuanced understanding of the numerous factors shaping its operating environment and how these affected the implementation of the program.

Observing these challenges provides valuable information for others who will implement similarly-situated programs. Indeed, we might have better anticipated these issues if they were more accurately and fully represented in extant literature on CV. If these issues had been anticipated, practitioners may have made better attempts to mitigate them at the outset, before they became the larger challenges that they evolved to. As demonstrated here, understanding these contexts are also important for evaluating the effectiveness of interventions. Such understanding will help researchers more accurately interpret results from evaluation analyses, especially important given that evaluations may be too often viewed as definitive determinations of the effectiveness of a given program model and not the results of idiosyncratic implementation of just the one program. A program’s underlying logic and services could be effective, but local conditions may still prevent that program from being implemented as intended. Without this knowledge, researchers may conclude that the program model is ineffective, as opposed to program implementation, producing invalid science.

We are not arguing that our approach necessarily, and in all cases, produces evidence superior or more truthful than others, nor that some sources of data or information necessarily outweigh others. The evaluation of information veracity and usefulness is always particular to a specific situation. We are arguing that community-engaged embeddedness expands the avenues for data collection, quality control, and validation, and greatly enhances the place of communities (of practice, interest, residence, or otherwise) as sources of information. Communities provide us, if not superior then, at least, another source of truth, and a critical one at that. Continued and in-depth engagement through embeddedness allowed us to combine knowledge learned from a community with our own scientific knowledge in order to form a clearer “picture” of how a program was put into effect, carried out, executed, and, thereby, a truer picture of the program’s outcomes and impact.