Mapping complexity in realist evaluation: A transferable programme-level theory of change for cross-agency violence reduction

Abstract

This article presents a realist-informed programme-level theory of change developed for a multi-site evaluation of a focused deterrence intervention aimed at reducing serious violence in five UK cities. Focused deterrence, a complex, cross-agency approach, requires theoretical tools that can account for local variation, emergent mechanisms and shifting implementation contexts. Using a five-stage process involving document review, fieldwork, workshops and qualitative interviews, we developed and refined the realist-informed programme-level theory of change to reflect variation in delivery, updated assumptions and context–mechanism–outcome configurations. Our findings reveal divergent delivery models, re-interpretations of core intervention resources and associated mechanisms, non-linear behavioural trajectories and participants’ strategic responses to perceived risks and opportunities. The final model offers a transferable framework for understanding and evaluating how complex interventions unfold across systems. We conclude by outlining lessons for evaluators seeking to develop theory-informed, complexity-aware theories of changes in real-world settings. These are particularly relevant in contexts involving cross-sector coordination, multiple delivery systems and flexible but systematic evaluation designs.

Introduction

A theory of change (ToC) is widely used in evaluation and is critical for helping stakeholders and evaluators understand how complex interventions operate across diverse contexts (see inter alia, Coryn et al., 2011; Ofek, 2017; Vogel, 2012; Weiss, 1995). While ToCs vary in form and flexibility, they often lack a clear articulation of the theoretical linkages and assumptions connecting intervention components (Wilkinson et al., 2021). This limitation is especially acute in multi-component, cross-agency interventions, where causal drivers are dynamic and difficult to trace, as illustrated by violence-reduction approaches such as focused deterrence that sit at the intersection of public health and criminal justice systems.

Developed in the 1990s by David Kennedy and colleagues, focused deterrence is a popular multi-agency strategy for reducing serious violence (Braga and Kennedy, 2021; Kennedy, 1997, 1998). Despite its popularity, the theoretical foundations underpinning its evaluation remain underdeveloped, with their intellectual coherence and theoretical depth varying widely (cf. Broca, 2012; Davies et al., 2016; Kerr et al., 2021; Williams et al., 2014). This inconsistency limits their practical value in high-risk environments where delivery conditions are unstable, and mechanisms unfold contingently. More broadly, complex social problems, for example, youth violence, homelessness and drug-related deaths, are increasingly addressed through coordinated responses across policing, health, education and community sectors. Focused deterrence typifies this wider shift towards shared responsibility across services and provides a timely, transferable case for developing theory-of-change models that reflect the realities of cross-agency implementation.

Focused deterrence, therefore, presents a persuasive use case for exploring complexity in theory of change development. Rather than a standardised programme, it is a framework of interlocking intervention strategies, targeted enforcement, individualised support and community engagement, delivered through coordinated, context-sensitive adaptation across agencies. These strategies operate as resources that shape participants’ reasoning and behaviour, yet their effects depend on how they are interpreted in particular social, institutional and legal contexts, resulting in delivery that varies across sites according to local conditions and capacity (cf. Pawson, 2013). International evaluations have shown wide variation in both implementation and outcomes, underlining the need for evaluative tools that can accommodate this complexity. In the United Kingdom, focused deterrence is gaining policy traction but remains under-theorised. This article, therefore, addresses that gap by presenting a realist-informed programme-level theory of change (FD-PToC) developed for a large multi-site UK implementation of focused deterrence (see Brennan et al., 2023, for the study protocol).

Framing the problem

Youth violence, driven by multiple factors, is one of the most pressing issues related to crime and public health in the twenty-first century (cf. Brennan et al., 2024; Centres for Disease Control and Prevention, 2025). Over the past decade, the United Kingdom has implemented various policies to address youth violence, including both preventive and enforcement strategies. However, identifying interventions that are both effective and transferable remains challenging, as does understanding the conditions under which transfer might succeed (Cartwright and Hardie, 2012). The evidence base for violence prevention is limited, and most validated interventions have been developed and tested outside the United Kingdom, raising concerns about their relevance and applicability in different contexts. Earlier UK applications, including the 2002 Manchester Ceasefire initiative and Operation Shield in London, faced challenges related to mission drift and implementation fidelity (Bullock and Tilley, 2002, 2008), while subsequent scholarship has continued to emphasise the importance of clear deterrence signalling and sustained partnership coordination (Densley and Jones, 2016).

Focused deterrence is a tertiary, evidence-supported approach targeting individuals with a known history of violence. A systematic review of evaluations (cf. Braga et al., 2026) reports that focused deterrence is associated with a moderate, significant 23 per cent reduction in serious violence compared with control conditions. However, nearly all existing evidence is derived from the United States, where the origins and context of violence make it an outlier among developed countries (Garland, 2025). UK implementations have faced both theoretical and operational challenges, and, in most cases, lack rigorous evaluation. As such, there is limited understanding of how focused deterrence might work, or fail to work, in a UK context (Gaffney et al., 2021).

Structure of the article

This article addresses that gap by presenting a structured, realist-informed, programme-level theory of change for focused deterrence in a UK context. The cross-agency structure of focused deterrence amplifies complexity and creates a pressing need for theory-of-change models that can account for variable delivery systems, multiple accountabilities and emergent and unforeseen outcomes. The FD-PToC offers one such approach.

The article’s aims are threefold. First, we outline the focused deterrence intervention within its broader operating context, drawing on existing literature and empirical experience to develop a theoretically grounded implementation framework. Second, we describe the development of the FD-PToC, including its underlying logic, assumptions and the core context–mechanism–outcome (CMO) configurations that shape its epistemic causal claims. Third, we explore how variation in delivery and practitioner adaptation reveal the activation of mechanisms within contextual constraints. A multi-site design was essential to understand how focused deterrence operates under different delivery models, stakeholder environments and local conditions, features common to many violence reduction initiatives. In doing so, we aim to support the design and evaluation of complex cross-agency interventions, offering a replicable framework that integrates realist evaluative thinking with theory-informed practice.

Understanding focused deterrence as a complex intervention

Complex interventions involve multiple interacting components, target diverse behaviours, require varied expertise and operate across different groups, settings or system levels, with complexity arising primarily from interactions with context (Skivington et al., 2021). From a realist perspective, they are best understood not as static packages but as dynamic events within systems shaped by feedback, adaptation and emergence.

Focused deterrence, or “pulling levers,” illustrates the dynamics of a complex intervention. Originating in Boston during the mid-1990s in response to gang-related violence, it targets a small number of persistent violent individuals through coordinated law enforcement and partner action (Braga et al., 2001). The strategy operates through three interdependent mechanisms—deterrence, support and community engagement—each of which depends on contextual conditions. Deterrence is implemented through direct communication of sanctions, reinforced by agencies’ visible capacity to act (Kennedy, 1997, 1998). Individuals are informed that ongoing violence will prompt a swift and collective response from the criminal justice system and allied agencies (Braga and Weisburd, 2014). This signalling aims to influence reasoning and behaviour by increasing the perceived costs of violence while simultaneously expanding access to meaningful alternatives.

Alongside sanctions for ongoing involvement in violence, support options range from assistance with exiting violence-involved groups to tailored access to statutory, voluntary and community services. The multi-agency structure is essential, allowing the simultaneous activation of credible threats and credible support as mechanisms for change. Community engagement reinforces this by expressing moral opposition to violence, increasing legitimacy and exerting normative pressure to desist. Focused deterrence is therefore not a standardised package, but a context-sensitive, multi-component intervention shaped by institutional capacity, interagency trust and local political dynamics. It constitutes a complex intervention embedded within dynamic legal, social and organisational systems (Brennan et al., 2024; Simanovic et al., 2024).

Contextual variation of mechanism interaction and implementation

Delivering focused deterrence in different geographies requires substantial adaptation to each site’s specific operating conditions (Engel et al., 2013). This is shaped by local variations in patterns of violence, communities’ sociocultural histories, the organisational capacity of statutory and voluntary partners, and divergent policing strategies and institutional cultures. Populations targeted with focused deterrence include the gang-involved, those engaged in serious interpersonal violence, actors in local drug markets and others identified via police intelligence/records, referrals or risk assessment tools (Braga et al., 2001; Corsaro and Engel, 2015; Saunders et al., 2015; Sierra-Arévalo et al., 2016; Trinkner et al., 2019).

The main elements of the intervention—deterrence messaging, support and community engagement—have also differed in their form and focus during implementation. Some interventions have relied on collective forums (“call-ins”) to deliver deterrence messaging, while others have used one-to-one communication. Similarly, support has ranged from basic service referrals to intensive, personalised casework. Over time, support has evolved from a supplementary feature to a central mechanism for sustained desistance. Community engagement has also varied: in some models, it has been one-off event-based; in others, it has been operationalised through embedded roles such as community “navigators” who maintain ongoing contact with participants (Scott, 2017; Simanovic et al., 2024). This variability illustrates the inherently adaptive nature of focused deterrence as a complex intervention. Its delivery is shaped by a constantly shifting ecology of institutional relationships, resource availability and community trust, which influence both the fidelity and the function of intervention mechanisms.

The theoretical rationale for focused deterrence has historically centred on deterrence theory, particularly the dimensions of certainty, swiftness and severity of punishment. Empirical evidence suggests that certainty and swiftness of sanction are more influential than severity alone (Nagin, 2013), a finding consistent with classical deterrence theory’s emphasis on credible, visible consequences as drivers of rational behavioural calculation (cf. Beccaria, 1986). In focused deterrence, this logic is operationalised through coordinated messaging from enforcement and community actors designed to heighten perceived certainty.

Beyond the deterrent effect, the provision of structured support is theorised to create “turning points” in an individual’s offending trajectory, particularly when it addresses psychological, relational and occupational needs (Sampson and Laub, 1993). Similarly, the expression of anti-violence community norms, whether through symbolic authority or active involvement, is understood as a potential mechanism for change within the focused deterrence framework. While each mechanism may independently influence outcomes, understanding how they interact in context remains an open question—one that realist approaches are well placed to address (Brennan et al., 2024; Simanovic et al., 2024).

Why a UK theory of change is needed

While focused deterrence has shown promise in reducing serious violence, the existing evidence base remains methodologically limited and geographically narrow (Brennan et al., 2024; Simanovic et al., 2024). A 2026 systematic review of 50 evaluations reported overall positive effects (Braga et al., 2026). However, nearly all studies were conducted in the United States, with substantial variation in outcomes, follow-up periods, units of analysis and comparators. Key factors in US violence, such as firearm access, gang dynamics and institutional contexts, are markedly different, which limits the generalisability of these results or their underlying theories of change to the United Kingdom.

Despite growing policy interest in focused deterrence in UK policy, existing evaluations provide limited insight into causal mechanisms or implementation fidelity. To date, only four programmes have been evaluated: the Community Initiative to Reduce Violence (CIRV) in Glasgow (Williams et al., 2014); Operation Shield in London (Davies et al., 2016); Tackling Gangs and Serious Youth Violence in Enfield (Broca, 2012); and CIRV in Northampton (Kerr et al., 2021). The Glasgow and Northampton models leaned towards a public health and support-led approach, while Enfield and Operation Shield were more enforcement-dominant. Results have been mixed and methodologically limited. For instance, the Glasgow evaluation found reductions in weapon possession but no measurable effect on serious violence (Williams et al., 2014). The London project did not proceed to impact evaluation due to implementation failure, and the remaining evaluations did not use designs capable of testing effectiveness.

Despite these limitations, focused deterrence is the only policing intervention classified as “high impact” by the Youth Endowment Fund, and Violence Reduction Units (VRUs) are required to allocate 20 per cent of their budgets accordingly (Home Office, 2023). Yet a recent audit found that of 21 police forces interviewed, only 3 implemented focused deterrence in line with a recognisable framework (OC&C Strategy Consultants and Youth Endowment Fund, 2024). This variation in design and delivery, combined with the absence of robust UK evaluations, reinforces the need for a clearly articulated, realist-informed programme-level theory of change. This gap between policy expectation and delivery consistency emphasises the need for the kind of programme-level theory of change developed in this article.

Challenges of transferring focused deterrence to the UK context

The core theoretical assumptions and delivery mechanisms of focused deterrence were developed in a US context characterised by high firearm availability, elevated homicide rates, militarised policing and extensive prosecutorial leverage, conditions that enable forms of deterrence messaging and enforcement not easily replicated in the United Kingdom (e.g. Braga et al., 2026; Longstaff et al., 2015). UK serious violence is more frequently embedded in knife-related peer conflict involving younger participants and lower legal leverage, altering both the perceived certainty and immediacy of sanction and the social meaning of enforcement encounters. The UK context, therefore, requires not merely adaptation of focused deterrence components, but recalibration of how deterrence, support and legitimacy mechanisms are theorised to operate.

In the United States, focused deterrence depends on the credible threat of swift legal sanctions, including pre-trial leverage (e.g. conditional bail) and formal agreements that link compliance to enforcement outcomes. UK legal constraints, especially those connected to youth justice and due process, restrict the use of these tools. For instance, the type of bail-linked call-ins seen in US programmes would raise due process issues under UK and European protections (Graham and Robertson, 2021). The mechanism of “legal leverage” cannot operate as intended when the justice system limits both the immediacy and conditionality of sanctions.

Community engagement is also operationalised differently. While many US interventions embed local faith groups, neighbourhood leaders and community “navigators” into delivery, community involvement in UK settings has typically been less sustained and more symbolic. UK communities are routinely asked to “co-design” interventions and may express concerns about violence, but they are rarely positioned or asked to co-deliver interventions or apply normative pressure on offenders in ways that might activate the legitimacy mechanism (e.g. Braga et al., 2008; Scott, 2017).

Finally, the nature of serious violence itself differs. In the United Kingdom, knife crime and youth-perpetrated group violence dominate, with younger offender profiles than typically seen in US gang contexts (Graham and Robertson, 2021). These distinctions matter because they shape how potential participants interpret threats, offers of support and community messages. If core mechanisms are contingent on context—legal, institutional and cultural—so too must the theory of change be. Understanding how focused deterrence is meant to work in the United Kingdom requires more than adaptation; it requires re-theorisation.

The focused deterrence programme-level theory of change

Epistemological orientation

The development of the FD-PToC is grounded in a pragmatic–realist epistemology. This view suggests that causal mechanisms exist independently of how we perceive them, but our comprehension of these mechanisms is only partial, context-dependent and shaped by social factors. In realist evaluation terms, knowledge about how and why interventions work emerges from the interaction between theory, context and observed outcomes (Pawson and Tilley, 1997). Rather than treating the theory of change as a neutral, technical artefact, we approach it as a tool of continuous evaluative reasoning, shaped by dialogue, contestation and practical engagement with stakeholders.

This orientation follows the realist principle that explanation relies on identifying how mechanisms are triggered under specific contextual conditions to produce outcomes. Following Pawson’s later refinements (cf. 2006; 2013), mechanisms are understood not as programme activities but as the reasoning and behavioural responses of participants when exposed to programme resources within specific contextual constraints. Context is therefore regarded not as background variation but as a set of pre-existing social, institutional, and structural conditions and norms that shape dispositions and limit choices. Rather than viewing programmes as unitary black boxes, the FD-PToC begins by articulating the causal assumptions embedded in delivery and testing how they operate across varied contexts. It treats CMO configurations not as retrospective explanatory tools but as the main building blocks of theory-led intervention design.

The FD-PToC was developed through an iterative, participatory process involving statutory partners, voluntary sector organisations, local delivery teams and community stakeholders. Structured workshops, targeted interviews and facilitated discussions were used to elicit tacit theories embedded in practice, clarify how mechanisms were understood, surface disagreements and explore variations across settings. This co-productive approach reflects the recognition that evaluative knowledge is shaped by institutional standpoint, disciplinary framing and lived experience; by incorporating these influences, we aimed to produce a theory of change that is both analytically robust and grounded in implementation realities. The outcome is therefore presented as a provisional, programme-level account of how focused deterrence is expected to function across diverse UK conditions, open to testing, revision and adaptation.

Intervention overview and study context

The FD-PToC was developed to support the design, implementation and evaluation of seven focused deterrence interventions delivered across five urban sites in England, as part of the Another Chance Fund (ACF1) trial. The sites—Leicester, Manchester, Nottingham, Coventry and Wolverhampton—are mid-sized cities (with populations of approximately 250,000) located in the Midlands and Northwest. All five areas experience high levels of deprivation and serious violence, with knife crime rates exceeding national averages in the years preceding the intervention. Coventry and Wolverhampton, both within the West Midlands region, were identified as having some of the highest rates of knife offences in England and Wales, driven largely by patterns of youth group violence and localised criminal networks (Brennan et al., 2024; Simanovic et al., 2024).

Each intervention adhered to the core principles of focused deterrence but was tailored to the local context and delivery structures. The seven interventions were managed by four multi-agency teams, comprising VRUs, local police forces, and statutory and third-sector partners. Although implementation varied, all interventions targeted individuals aged 14 and over who were either involved in, or at risk of becoming involved in, serious violence and group-based offending. Identification typically drew on police intelligence, crime records, and referrals from statutory or community agencies.

The interventions followed a broadly similar participant pathway, beginning with (1) eligibility screening and pre-engagement planning. This was followed by (2) direct contact, in which participants were informed of the legal and social consequences of continued violence, alongside offers of support. For those who engaged, (3) further assessment and support planning were undertaken, leading to access to services and sustained engagement. For those who continued to offend, (4) targeted enforcement was applied, with (5) efforts made to re-engage the participant.

While this pathway describes the operational sequencing of activities, change was not assumed to follow automatically from this structure. Rather, enforcement, support and community involvement functioned as intervention resources intended to shape participants’ reasoning about risk, identity, legitimacy and future opportunity. Whether desistance was initiated, sustained, resisted or reversed depended on how these resources were interpreted under specific contextual conditions. Regular review of offending and service engagement ensured that the pathway remained dynamic and responsive to changing needs.

Research questions guiding the FD-PToC

The development of the FD-PToC was guided by a series of formative research questions intended to structure inquiry into how the intervention operates across multiple sites and contexts. These questions are not answerable through process or outcome data alone but require theory-led exploration of mechanisms, context and implementation dynamics. Specifically, the FD-PToC was developed to address:

• 1. Who did the intervention work for, and how?

• This question directs attention to variation in participant responses, probing how different mechanisms are activated across populations.

• 2. How are local context and complexity affecting intervention delivery?

• A central concern of the FD-PToC is the interaction between the intervention and the specific social, institutional and structural features of each site.

• 3. Is the intervention being delivered as intended?

• This includes fidelity to the model, as well as intentional adaptations and emergent practices that shape delivery.

• 4. What was learned from how the intervention was delivered?

• Capturing experiential learning from delivery teams and system actors is critical for understanding implementation trajectories and barriers.

• 5. What constitutes business as usual in each site?

• Understanding the counterfactual context is essential to interpreting outcomes and theorising causal contribution.

By anchoring the FD-PToC in these questions, the aim was to build a generative account of how, why, for whom and under what conditions the intervention produces change. That is, it adopts a realist, theory-led approach to evaluation that engages directly with complexity, variation and practice-based knowledge (Brennan et al., 2024; Simanovic et al., 2024).

Conceptually, the FD-PToC provides a high-level framework for theorising how focused deterrence is expected to reduce serious violence in UK urban settings. As a multi-site initiative involving multiple agencies, it maps the sequence of activities, mechanisms and assumptions through which desired outcomes might be achieved (Brennan et al., 2024). Informed by both the literature and practice-based insight, it highlights the contextual contingencies shaping intervention delivery and effects.

The developed model treats outcomes as the result of mechanisms activated in specific contexts. It integrates evidence on deterrence, desistance and legitimacy through three core CMO configurations. Designed as both an evaluative and planning tool, the FD-PToC enables stakeholders to test and refine underlying assumptions as delivery unfolds.

Five-stage development process

We developed the FD-PToC through a structured five-stage process, adapted from Vogel (2012) and grounded in realist principles. An overview of the five stages is presented in Table 1 to orient the reader before the detailed discussion.

Table 1.

Five-stage process for developing the FD-PToC.

1. Contextual analysis	Identified site-specific delivery conditions, inputs and latent contextual factors shaping mechanism activation
2. Defining long-term impact	Developed a shared, testable vision of desired long-term outcomes for violence reduction with delivery sites
3. Mapping causal pathways	Constructed a sequenced logic model linking inputs, activities, outputs and outcomes
4. Embedding theoretical assumptions	Integrated CMO configurations to capture underlying causal logics informed by realist and criminological theory
5. Modelling the pathway to change	Combined CMO structures and logic model elements into a dynamic, complexity-informed ToC

Stage 1: Contextual analysis

We began by identifying the context within which the intervention would be delivered. We synthesised data from the delivery progress monitoring framework we developed, programme documentation, site-level logic models, stakeholder workshops, site visits and baseline interviews with delivery personnel. Through this process, we identified variations across five implementation sites, related to population dynamics, institutional capacity, perceptions of inter-agency trust, racial equity and community legitimacy. Across sites, contextual variation was further shaped by UK-specific institutional constraints, including youth justice procedural safeguards, limited prosecutorial leverage compared to US models and the embedding of focused deterrence within existing VRU governance structures. These structural features conditioned how enforcement credibility and support coherence could realistically be mobilised, thereby influencing the activation and interaction of core mechanisms and, in turn, the activation of participant reasoning processes. This step revealed latent conditions that shape the activation of mechanisms and informed our contextual framing of the FD-PToC. We therefore treat context not as a simple variation between implementation sites, but as a layered set of pre-existing individual, relational, institutional and structural conditions that constrain choice and shape differential receptivity to intervention resources. In this sense, site-level variation reflects deeper contextual configurations rather than functioning as context itself.

Stage 2: Defining long-term impact

We developed a statement of long-term impact to provide clarity and anchor the ToC. This was not intended as a predictive claim but as a shared vision against which progress could be measured and tested. The resulting statement was as follows: the intervention aims to produce a sustained reduction in serious youth violence in targeted communities by supporting individuals to disengage from violent peer networks, promoting social reintegration and reducing criminogenic opportunity structures. Over time, the programme is expected to generate greater community resilience, institutional legitimacy and perceived safety.

Stage 3: Mapping causal pathways

Using insights from contextual analysis and stakeholder engagement, we developed a focused deterrence logic model to operationalise the sequencing of change from inputs to outcomes. Unlike conventional ToC design, which often works backwards from impact, our approach combined forward and backwards mapping to capture the intervention’s logical flow. The FD-PToC (Figure 1) specifies 7 inputs (in1–in7), 9 core activities (a1–a9), 10 immediate outputs (op1–op10) and 15 intermediate outcomes (ot1–ot15), each linked to assumptions and measurable indicators. This structure reflects both empirical evidence and stakeholder insight, while recognising that real-world change is non-linear, adaptive and context-dependent.

Figure 1.

Linear pathway from inputs to impacts within the FD-PToC, showing how delivery components are sequenced to produce short- and long-term outcomes.

Stage 4: Embedding theoretical assumptions

In line with Pawson (2013), we distinguish between intervention resources, the structured strategies delivered by the programme, and mechanisms, the reasoning and behavioural responses they generate under contextual constraint. To strengthen the explanatory power of the FD-PToC, we embedded theoretical assumptions from deterrence, desistance and legitimacy theories within a realist framework. The three core intervention resources underpinning focused deterrence are outlined below. Their contextual variation and empirical expression are examined in detail in the subsequent modelling section.

Intervention resource 1: Targeted enforcement (r1)

Targeted enforcement refers to the coordinated communication of legal consequences to individuals at risk of serious violence. The associated assumption is that credible and procedurally legitimate sanction signals may prompt recalibration of perceived risk. Whether this resource activates a deterrence mechanism depends on how participants interpret enforcement credibility, sequencing and legitimacy under specific contextual conditions.

Intervention resource 2: Individualised support (r2)

Individualised support encompasses relational engagement and access to tailored services intended to offer viable alternatives to violence. The underlying assumption is that credible, sustained support may enable shifts in identity orientation and perceived future viability. Mechanism activation depends on alignment between participant disposition, relational trust and the continuity of service provision.

Intervention resource 3: Community validation (r3)

Community validation refers to the involvement of credible community actors in reinforcing normative expectations and procedural fairness. The associated assumption is that perceived legitimacy and informal social influence may shape behavioural decision-making. Its activation depends on the perceived authenticity, authority and embeddedness of community voices.

Resources such as deterrence-oriented inputs were refined through expert consultation and stakeholder testing, reflecting the realist emphasis on generative causality. Rather than treating mechanisms as linear triggers, we used the CMO structure to express how complex interventions interact with context to produce contingent, non-linear change.

In revising the FD-PToC, we explicitly distinguished between three analytically distinct elements. First, programme components or intervention resources refer to the structured strategies delivered by the intervention, namely targeted enforcement, individualised support and community validation. Second, assumptions of change refer to the propositional claims about how participants are expected to interpret and respond to these resources, for example, that a credible sanction will prompt risk recalibration or that relational support can enable identity reorientation. Third, mechanisms are understood, following Pawson (2006, 2013), as the reasoning and behavioural responses of participants when exposed to programme resources under specific contextual constraints. This distinction prevents the collapse of delivery activity, causal assumption, and generative mechanism and clarifies that outcomes arise not from programme components alone, but from how participants navigate them within structured conditions.

Stage 5: Modelling the pathway to change

Building on the operational sequencing defined in Stage 3 and the mechanism-focused CMO configurations in Stage 4, we formalised these elements into a linear pathway to change (Figure 1), linking inputs through to impacts. In the FD-PToC narrative, this pathway illustrates how context interacts with mechanisms to generate observable outcomes. The model operates as both a planning and evaluative tool, integrating context, mechanisms, proximal outcomes and long-term impacts. It enables stakeholders to make assumptions explicit, monitor progress and adapt delivery in response to emerging conditions.

While this pathway provided a linear representation of the intervention’s flow from inputs to impacts, the next step was to foreground the generative processes hypothesised to drive change. As shown in Figure 2, contextual conditions shape how three core intervention resources, targeted enforcement (r₁), individualised support (r₂) and community validation (r₃), are interpreted and engaged with by participants. The mechanisms of change reside in participants’ reasoning and behavioural responses to these resources, for example, recalibrated risk perception, identity reorientation or shifts in perceived legitimacy. These context-conditioned mechanisms generate proximal outcomes, captured through specific indicators, which may cumulatively contribute to long-term outcomes and impacts.

Figure 2.

Contextual conditions shape how intervention resources are interpreted, triggering mechanisms within participants’ reasoning that generate proximal and long-term outcomes.

Effective implementation of the FD-PToC depends on the coordinated operation of six foundational inputs: (1) recruitment and retention of personnel with the necessary skills and support structures to fulfil their roles; (2) establishment of secure, accurate and timely data systems enabling effective identification and monitoring of the intervention population; (3) creation of clear and efficient referral pathways that connect participants to appropriate support services; (4) strong inter-agency collaboration ensuring that delivery partners across statutory and voluntary sectors are aligned and motivated; (5) well-integrated support services that respond to local need, underpinned by organisational support at both strategic and operational levels; and (6) adequate and stable funding to sustain delivery quality and programme fidelity across multiple sites. Each of these inputs is linked to specific assumptions and indicators, which are used to assess readiness and inform ongoing implementation, monitoring and evaluation.

Building on these inputs, the intervention generates a set of structured activities designed to catalyse change. These activities, such as enforcement call-ins, mentoring sessions, outreach engagement and multi-agency case management, are designed to yield a range of immediate outcomes. These include increased participant engagement, improved compliance with legal and social norms, strengthened partnerships between communities and statutory services, and more consistent feedback and oversight mechanisms. Outputs function as early signals of change and lay the groundwork for a progression of short- to medium-term outcomes.

These outcomes, aligned with their respective CMO configurations, include measurable improvements in community trust and confidence, reduced exposure to and participation in violence, increased uptake of support services, and enhanced inter-agency data sharing. Indicators used to monitor these outcomes are drawn from a range of routine and bespoke data sources, including police records, community surveys, referral logs, exclusion and hospital admission data, and intelligence assessments. Together, these outputs and outcomes contribute to the intervention’s overarching impact goal: achieving a sustained reduction in serious youth violence and building more resilient, cohesive communities.

Surfacing assumptions

The FD-PToC treats assumptions not as background beliefs but as integral to the construction and understanding of causal pathways. Following Wilkinson et al. (2021), assumptions were surfaced iteratively through stakeholder workshops and interviews, focusing on how change was expected to occur and under what conditions mechanisms were seen to activate, falter or backfire.

For targeted enforcement (r1), a core assumption was that individuals respond rationally to credible, proportionate threats. Increased visibility and coordination of enforcement were expected to heighten the perceived certainty and immediacy of sanctions, thereby prompting strategic compliance. However, these assumptions were problematised, particularly for young people affected by trauma, cognitive impairment or substance misuse, where rational appraisal may be impaired.

Individualised support (r2) was premised on the expectation that access to credible alternatives, sustained relational engagement and personalised assistance could activate mechanisms related to identity reorientation, perceived future viability and motivational alignment. Yet views diverged on what counted as “meaningful” support and whether change stemmed from internal motivation or external compliance. This ambiguity informed a broader understanding of support as a mechanism shaped by structural, relational and psychological dimensions.

Community validation (r3) assumed that moral authority and informal social control could influence behaviour by reshaping perceptions of legitimacy and normative alignment. Stakeholder insights, however, revealed a wide variation in trust, cohesion and institutional legitimacy across sites. The credibility of community messaging varied and was frequently contested. Where community messengers were perceived as authentic and procedurally fair, normative pressures were more likely to be recognised. Where credibility was contested, community-oriented resources were less likely to activate legitimacy-based reasoning.

Cross-cutting concerns included the temporal dynamics of change (i.e. how quickly shifts were expected), the availability and quality of support, and the risk of unintended consequences (e.g. enforcement without support undermining legitimacy). Surfacing these assumptions clarified that mechanisms are contingent on context and participant disposition rather than inherent in programme design. Making these propositions explicit enabled targeted empirical testing, identification of constrained choice conditions and adaptive refinement as implementation unfolded.

Engaging with complexity in programme design and evaluation

The development of the FD-PToC exposed several conceptual and practical challenges in capturing complexity within a realist-informed theory of change. As Breuer et al. (2015) note, complex interventions typically involve interdependent components operating across multiple system levels, rendering evaluation design technically and conceptually demanding. One of this article’s key contributions lies in mapping how complexity, expressed through delivery variation, contextual contingency and emergent behaviours, can be embedded into ToC design using realist CMO structures.

Rather than treating complexity as an abstract condition, we drew on a “complexity theory frame of reference” (Byrne, 2011; Walton, 2014) that views interventions as events situated within dynamic systems. Complex systems involve multiple interacting components, such as actors, institutions and histories, that exhibit feedback, emergence and adaptation. Their behaviour cannot be fully understood through linear cause-and-effect relationships.

These systems are nested and open: components (such as statutory agencies) operate within broader systems (e.g. criminal justice, education) and are shaped by external feedback loops (Byrne and Callaghan, 2022; Room, 2011). Interactions within such systems are nonlinear; small contextual differences can generate disproportionately large effects (Anderson et al., 2005; Rickles et al., 2007). Systems often display stable patterns until reaching tipping points, at which they may shift into new “attractor states” within a broader “phase space” of potential outcomes (Capra, 2005; Walton, 2014).

This theoretical lens sharpened our understanding of the challenge: interventions cannot be meaningfully evaluated without attention to the dynamic conditions under which mechanisms are activated. Rather than applying a static input-output model, the FD-PToC was revised to reflect the interdependence of context, delivery structure and mechanism function.

Inputs such as service availability, data quality or organisational trust were initially treated as background conditions. In practice, they emerged as pivotal, variable and often beyond the direct control of delivery teams. Mapping these explicitly allowed us to distinguish between design features and enabling conditions and to highlight which assumptions required active monitoring or adaptation.

Mapping contextual complexity within the FD-PToC

The FD-PToC was designed as an adaptable framework intended to evolve through practical engagement. The development and use of the FD-PToC revealed the conceptual and empirical difficulty of representing complexity in realist evaluation. As with other public health-style interventions, focused deterrence involves multiple interdependent components, each shaped by actors’ intentions, resource flows, institutional constraints and historical contingencies. The linear representations used in early versions of the ToC proved inadequate. As the evaluation progressed, based on qualitative interview data and observations, a clearer picture of the system dynamics emerged, shaped not only by differences between participants but also by deep contextual heterogeneity across sites. As shown in Figure 1, we found that delivery contexts varied significantly along four interlocking dimensions: individual capacity, interpersonal relationships, institutional arrangements and wider system structures. These contextual layers did not simply shape intervention delivery; they influenced how mechanisms were likely to be activated, neutralised or backfire (Brennan et al., 2024).

Following Pawson (2006, 2013), context is treated here as operating across micro and macro levels, from individual histories and network embeddedness to institutional governance structures. These contextual layers constrain available choices and shape differential receptivity to programme resources, thereby contributing to variation in the activation of mechanisms.

A good example of this complexity can be observed in the application of targeted enforcement (CMO1). In Manchester, delivery was coordinated through a decentralised statutory–voluntary partnership model, where community navigators were employed by a charity with strong local ties. Within the FD-PToC, targeted enforcement is conceptualised not as a mechanism, but as an intervention resource intended to trigger recalibration of perceived risk and strategic compliance under conditions of credible sanction. However, the actual deployment of this resource varied significantly from the assumptions embedded in the original theory.

Early versions of the FD-PToC assumed that deterrence messaging would be applied uniformly and with similar credibility across all sites. However, observations and interviews with both practitioners and participants revealed important deviations. Rather than treating these adaptations as implementation shortcomings, we interpret them as strategic responses by practitioners and participants as they navigate competing role expectations, legitimacy concerns and enforcement constraints. Similarly, participants’ movement between compliance, avoidance and re-engagement was interpreted not as evidence of delivery failure, but as strategic decision-making in response to perceived risks, opportunities and constraints within the intervention context. Navigators in Manchester were often reluctant to engage in “hard” deterrence messaging. They avoided directly threatening enforcement consequences, as they perceived such messaging as misaligned with their relational role and likely to undermine trust. Instead, deterrence was communicated more obliquely, often through third parties or softened framing, which weakened its perceived immediacy and certainty.

This adaptation was not simply a fidelity failure, but an emergent behaviour shaped by contextual contingencies, specifically, the moral positioning of navigators, their organisational ethos and the historically strained relationship between the police and communities in the area. These dynamics illustrate the difficulty of assuming that mechanisms like deterrence operate independently of delivery structure and professional identity. Instead, they require careful calibration to context, and their activation is contingent not only on message content but also on the messenger’s credibility and the perceived legitimacy of the institution backing the message (Brennan et al., 2023, 2024; Simanovic et al., 2024). This adaptation altered the conditions under which risk recalibration could plausibly occur, illustrating how professional identity and institutional positioning mediate the mechanism’s activation.

Delivery model typologies and mechanism integration

To represent this complexity further, we moved beyond static accounts of “context” to develop a typology of delivery models: (1) centralised police-led (e.g. Wolverhampton and Coventry); (2) centralised statutory–voluntary partnerships (e.g. Leicester, Nottingham); and (3) decentralised statutory–voluntary partnerships (e.g. Manchester). While all models deployed the three core intervention resources of focused deterrence—targeted enforcement, individualised support and community validation—each delivery model created different conditions under which these resources were interpreted and engaged with by participants. Police-led models tended to confer higher enforcement credibility but often demonstrated weaker embeddedness in community structures. In contrast, voluntary sector-led sites often benefitted from stronger community trust yet faced greater difficulty in signalling credible sanction. These structural differences shaped the likelihood that risk recalibration, identity reorientation or legitimacy-based reasoning would plausibly be activated.

Revised logic model and CMO configurations under complexity

We mapped observed delivery variations onto revised CMO configurations to reflect how layered contextual conditions shape the interpretation and uptake of intervention resources. Rather than treating enforcement, support and validation as stable causal levers, the revised model examines how their activation depended on participants’ reasoning under specific institutional, relational and structural constraints.

In the case of targeted enforcement (r1), deterrence-oriented resources were delivered through both informal warnings and formal sanction pathways. However, their credibility varied across sites depending on policing legitimacy, visible enforcement capacity and the coherence of inter-agency signalling. In the UK setting, deterrence was further constrained by the limited availability of conditional bail leverage and lower perceived sanction severity relative to US jurisdictions. For some younger participants embedded in peer-based knife conflict, perceived reputational risk outweighed formal sanction risk, making compliance a socially costly option. As such, deterrence mechanisms were more dependent on visible inter-agency coordination and procedural legitimacy than on punitive severity alone. When deterrence signals were perceived as consistent and procedurally fair, some participants described reassessing the risks of continued violence and moderating their visible involvement. In other contexts, particularly where enforcement was viewed as lacking credibility, enforcement signals were discounted or strategically navigated. Deterrence, therefore, operated conditionally and was rarely sufficient in isolation.

Individualised support (r2) varied markedly in depth and consistency. In sites with integrated service pathways and sustained relational engagement, participants reported clearer alignment between support offers and personal aspirations, creating space for incremental identity reorientation. Elsewhere, fragmented provision or inconsistent contact limited the plausibility of durable change. In such contexts, engagement sometimes reflected short-term compliance or instrumental use of services rather than substantive transformation. These differences demonstrate how support activation depended not only on service availability but also on relational trust and perceived authenticity.

Community validation (r3) was similarly contingent. In locations where credible community actors or lived-experience navigators were embedded within delivery structures, normative messaging was more likely to be interpreted as legitimate and locally grounded. Unlike many US models, where faith leaders play a central role in deterrence, UK sites exhibited more fragmented community authority structures. Consequently, the activation of legitimacy mechanisms relied less on symbolic moral condemnation and more on the relational credibility of individual navigators operating within local trust networks. In more fragmented environments, community involvement was symbolic or contested, limiting its influence on participant reasoning. The activation of legitimacy-based mechanisms, therefore, depended on the perceived authority, coherence and embeddedness of community voices within specific contextual settings.

Across all three domains, outcomes were non-linear. Participants cycled between engagement, desistance, relapse and re-engagement, reflecting strategic navigation of intervention resources under constrained conditions. These trajectories were documented through qualitative interviews, site observations and routine process data, and varied according to disposition, network embeddedness and structural constraint. The revised CMO configurations thus emphasise that mechanisms reside in participant reasoning rather than in programme components, and that contextual alignment determines whether reasoning processes are plausibly activated. Because these reasoning processes unfolded contingently and interacted over time, we present the configurations in narrative form. Narrative exposition enables tracing how mechanisms emerge, combine, weaken or reconfigure across delivery models without implying fixed causal chains or uniform programme effects.

Modelling mechanism activation across contexts

To examine how context influences mechanism activation, we analysed variation in the operationalisation of enforcement, support and validation across the five trial sites. These elements were treated as intervention resources embedded within dynamic CMO configurations, with mechanisms understood as participants’ reasoning and behavioural responses under specific contextual constraints. Outcomes were therefore conceptualised as emergent rather than deterministic. Outcome statements were derived from triangulation of qualitative interview data with participants and practitioners, site-level observations and routine process indicators, enabling us to link reported behavioural shifts to specific configurations of resources and contextual conditions.

Consistent with Pawson’s (2006) argument that programmes operate under conditions of constrained choice and produce differentiated benefits, the FD-PToC recognises that focused deterrence does not generate uniform effects. Participants with weaker peer embeddedness and stronger attachments to alternative institutions were more likely to interpret support-oriented resources as credible opportunities for change, activating mechanisms related to identity reorientation or future orientation. In contrast, those deeply embedded in violent networks or facing acute structural marginalisation were more likely to discount deterrence-oriented resources, particularly where enforcement credibility was limited or reputational costs outweighed formal sanction risks. Mechanisms are therefore located in participants’ decisions about how to engage with, resist or strategically navigate intervention resources, ranging from compliance and cautious experimentation to overt resistance or instrumental engagement designed to manage enforcement risk. Variation in outcome reflects differences in disposition, network position and structural constraints, with context shaping which individuals were well-disposed to engage and which were less receptive, rather than variation in delivery alone.

Targeted enforcement (r1): Conditional credibility and non-linear deterrence

Targeted enforcement operated differently across delivery models. Local resource availability, delivery structure and trust in policing shaped how deterrence-oriented resources were enacted and perceived. Where police coordination and visible enforcement capacity were strong, the threat of sanction appeared more credible. In contrast, low trust or delivery via non-enforcement partners diluted the perceived certainty of consequences. In contexts characterised by strong group cohesion and reputational pressures, continued violence could also be interpreted as a rational survival strategy, where immediate peer standing or personal safety outweighed the perceived risk of formal sanction.

Rather than constituting a mechanism, targeted enforcement functioned as an intervention resource intended to trigger risk recalibration and strategic compliance. It was expressed through both soft (e.g. informal warnings, verbal cautions) and hard (e.g. tagging, control orders) approaches. Its effectiveness was rarely standalone; activation of risk-based reasoning was more likely when enforcement signals were accompanied by credible support and community validation, and less likely when these were absent or inconsistent. Deterrence was not uniformly applied, even within sites, and often reflected the moral or professional stance of individual practitioners.

Outcomes were rarely linear. Some participants modified their behaviour, others tested boundaries, relapsed or moved incrementally towards compliance. These behavioural trajectories reflected how participants interpreted enforcement signals, considering peer dynamics, delivery sequencing and perceptions of legitimacy. On its own, deterrence-oriented resources seldom produced lasting change, but in combination with other intervention resources, they contributed to shifts in behaviour where risk recalibration was plausibly activated. For example, in some sites, participants responded to credible and procedurally legitimate enforcement signals by reducing visible involvement in violence and engaging with support services, suggesting that risk recalibration combined with emerging identity reorientation could initiate behavioural adjustment. In contrast, where enforcement was perceived as inconsistent or lacking legitimacy, some participants disengaged from support and adopted avoidance strategies, limiting sustained change. In such instances, delivery teams adapted sequencing, strengthening relational engagement before reintroducing deterrence messaging to restore coherence and credibility.

Individualised support (r2): Credibility, alignment and discontinuity

Individualised support was equally sensitive to contextual variation. The depth and coordination of services varied considerably, from well-integrated support ecosystems (e.g. mental health, education, housing) to fragmented, under-resourced provision. These structural differences influenced both the credibility and sustainability of support offers.

In some areas, navigators with lived experience of offending played a dual role, delivering personalised assistance while also modelling what desistance could look like in practice. Their credibility and proximity allowed them to function as vicarious role models, offering not just practical support but also motivational pull. In this sense, support operated as an intervention resource capable of triggering mechanisms related to identity reorientation, future orientation and perceived viability of non-violent pathways. For some participants, this pull towards a different possible self, anchored in trust, shared experience and relational continuity, was as powerful as the threat of sanction. These insights revealed the importance of personal connection and identity work in activating these reasoning processes, particularly for individuals navigating unstable or traumatic environments.

Support engagement was varied. Participants might initiate but not sustain contact; engage with one service while avoiding others; or cycle through moments of desistance, relapse and recommitment. These patterns reflect how participants interpreted and navigated support-oriented resources under specific contextual constraints, highlighting the non-linear nature of change.

Community validation (r3): Trust, voice and normative alignment

Community validation was shaped by trust in statutory actors and the presence of credible community voices. Some sites benefitted from embedded navigators and active local networks; others lacked identifiable community representation. Variability in legitimacy and capacity influenced the conditions under which legitimacy-based reasoning could plausibly be activated.

Community messaging ranged from symbolic endorsement to sustained involvement. Where credible and procedurally fair, it generated normative pressure through family, peers or respected figures. In this sense, community validation functioned as an intervention resource that could trigger shifts in perceived legitimacy and normative alignment. However, these reasoning processes were fragile and easily disrupted by weak enforcement signals or inconsistent support, which undermined the perceived coherence of the intervention.

Outcomes included incremental shifts in perceptions of legitimacy, changes in peer influence and informal compliance. As with other intervention resources, effects were rarely binary; rather, varied inputs contributed to gradual behavioural adjustment through contingent and context-dependent social dynamics.

From linear chains to complex configurations

Across all three domains, outcomes were non-linear. Participants did not simply “change” in response to intervention resources; they relapsed, resisted or re-engaged in unpredictable ways depending on how these resources were interpreted under specific contextual constraints. These variations were not treated as implementation instability, but as expressions of mechanism activation, revealing how participants strategically navigated intervention resources under constrained conditions.

This prompted a shift from static “if-then” CMOs to dynamic configurations, where timing, sequencing and system state altered how mechanisms, understood as participants’ reasoning processes, were activated. This shift away from linear “if–then” chains reflects Pawson’s critique of programme theories that collapse resources and mechanisms and overlook how participants navigate interventions through varying degrees of commitment, resistance or strategic compliance (2013).

Drawing on complexity theory, we treated these intervention pathways as subject to system feedback, shifting attractor states (Byrne and Callaghan, 2022; Capra, 2005) and sensitive dependence on initial conditions (Room, 2011). As a result, even minor contextual variations could produce significant differences in behavioural outcomes.

We also applied Mackie’s (1974) INUS condition logic, treating each mechanism as an insufficient but necessary part of an unnecessary but sufficient combination. For example, deterrence-oriented resources may be necessary to initiate risk recalibration, but without credible support or validation, sustained behavioural change is unlikely. Failure may result not from the absence of a mechanism, but from its misalignment with context or the weakness of complementary system elements. The UK FD-PToC, therefore, differs from US-derived models not in the strategic architecture of focused deterrence, but in the contextual recalibration of how intervention resources are interpreted and how mechanisms are likely to be activated under specific legal, cultural and institutional constraints.

Reflections on practice: What is distinctively UK about the FD-PToC

These reflections are shaped by the specific legal, institutional and governance conditions of the UK context, including youth justice safeguards, VRU structures and variable community–police legitimacy. For those seeking to develop realist-informed theories of change in complex, real-world environments, this section shares practical learning from our experience. Rather than offering a universal template, we reflect on what worked, what we adapted and what we would do differently in future. These insights may assist evaluators, implementation teams and researchers designing multi-site evaluations of adaptive interventions.

Treat the ToC as a scaffold, not a product

The instability observed in mechanism activation, understood as shifts in participant reasoning across delivery typologies, demonstrated that the ToC could not function as a fixed blueprint. Its value lay in making assumptions explicit and revisable as contextual conditions changed. Structured logic models and CMOs helped elicit assumptions, test interpretations and clarify differences between sites. We recommend using ToC and logic model materials as prompts for engagement, not as fixed artefacts.

Accept iteration and loosen sequencing

While we followed a five-stage process (see Table 1), development was mainly nonlinear. We repeatedly moved between logic models, stakeholder insights and CMO formulation. In future, we would introduce partial CMO sketches earlier to stimulate shared theorising, even before implementation begins.

Use typologies to clarify variation

Developing a delivery model typology helped us make sense of how the same intervention logic played out differently across sites. If we were to do this again, we would incorporate typology development into the early stages of ToC construction as a tool for anticipating divergence and structuring comparison.

Co-production is essential—but requires pacing

Our approach involved workshops, interviews and embedded engagement. While valuable, these were resource-intensive. Qualitative interviews also included a substantial number of children and young people participating in the intervention across sites; their accounts of enforcement encounters, support relationships and perceived legitimacy directly informed revisions to CMO assumptions and our understanding of mechanism activation. In the future, we plan to hold shorter, more frequent check-ins to sustain learning without overwhelming delivery teams or evaluators.

Track ToC versioning and assumption shifts

The transition from version 1 to version 2 of the logic model was a turning point. Explicitly documenting how assumptions changed in response to delivery insights strengthened the model’s credibility and evaluability. We recommend maintaining a version history with accompanying narrative justification for revisions.

In short, we found that designing a theory of change under conditions of complexity requires not just technical skill but also openness to emergence, divergence and adaptation. The ToC becomes most powerful when used not to fix logic but to make programme assumptions and participant reasoning visible, contestable and revisable under changing contextual conditions. In practice, this implies that commissioning arrangements should allow for adaptive refinement rather than fixed contractual logics, and that implementation fidelity should focus on preserving core intervention resources while permitting contextual adaptation.

Conclusions and next steps

The FD-PToC presented here is not a direct import of US-focused deterrence logic; rather, it is a recalibrated model that accounts for UK legal constraints, youth justice governance, patterns of violence dominated by knife crime and the evolving public health infrastructure. Its transferability rests on its analytical structure rather than on the assumption that contexts are equivalent. The cycling between compliance, relapse and re-engagement observed across sites reinforced the need for evaluative designs capable of tracing participant reasoning over time rather than assuming linear behavioural response. In this respect, the FD-PToC demonstrates that a realist-informed theory of change can retain explanatory clarity while directly engaging the complexity, variation and unpredictability characteristic of real-world violence reduction.

In realist terms, the FD-PToC therefore clarifies for whom focused deterrence is more likely to work, namely those with viable alternative attachments and credible exposure to coherent resources; in what circumstances, specifically where enforcement legitimacy, support capacity and community trust align; and in what respects, primarily through shifts in risk perception, identity orientation and legitimacy recognition rather than uniform compliance effects.

The FD-PToC contributes to evaluation theory and practice by demonstrating how ToCs can operate as iterative, theory-building tools rather than static implementation maps. It extends realist evaluation by embedding CMO configurations directly into the theory development process and by explicitly modelling assumptions for testing and adaptation. The model supports not only more robust evaluation of focused deterrence but also the advancement of theory-led evaluation practice in dynamic, multi-sector policy environments. These methodological recommendations stem from observing how contextual variation shaped mechanism activation and generated differentiated participant trajectories across sites. A detailed outline of the recommended realist mixed-methods evaluation framework aligned with the FD-PToC is provided in the Supplementary Material.

Supplemental Material

sj-docx-1-evi-10.1177_13563890261444323 – Supplemental material for Mapping complexity in realist evaluation: A transferable programme-level theory of change for cross-agency violence reduction

Supplemental material, sj-docx-1-evi-10.1177_13563890261444323 for Mapping complexity in realist evaluation: A transferable programme-level theory of change for cross-agency violence reduction by Paul McFarlane, Alex Sutherland, Tia McPherson, Iain Brennan and William Graham in Evaluation