Introduction

Urban green infrastructures (UGIs), networks of parks, forests, wetlands, and green roofs, offer critical environmental, economic, and social benefits, from carbon sequestration to community well-being1,2. However, their design and management increasingly require multidimensional approaches that integrate diverse perspectives, particularly through participatory governance3,4,5. By shifting from top-down policy development to more inclusive governance structures, participatory governance ensures that urban green infrastructure addresses not only environmental and spatial needs but also the values and aspirations of the communities that use and benefit from these spaces6,7. This approach is vital for addressing social-environmental justice issues, as groups with poor physical health or in racially diverse neighbourhoods often have less access to green spaces8,9.

Co-design, a participatory approach espousing reciprocal knowledge exchange among stakeholders, including communities, professionals, and decision-makers, has emerged as a promising strategy to align UGIs with local needs10,11,12. By emphasising joint decision-making and valuing all expertise, co-design enhances community ownership and trust, delivering long-term social and environmental benefits13,14. Yet, its effectiveness remains debated, with critics citing inefficiencies from diverse perspectives and insufficient expertise, such as challenges in managing conflicting stakeholder priorities or the time-intensive nature of consensus-building, while its processes and impacts lack systematic evaluation15,16,17. These complexities highlight the need for structured tools to analyse co-design dynamics. This complexity, compounded by varying local contexts, hinders comparisons with passive participatory methods and obscures the mechanisms driving its success18,19.

Systems approaches, such as those applied to green and blue space planning, have mapped structural and behavioural dynamics to promote health and social cohesion, yet they often lack procedural guidance for co-design processes20. Similarly, advancements in co-creation, like those from the Health CASCADE project in the health sector, emphasise evidence-based, multi-sectoral collaboration but require structured frameworks to operationalise complex interventions21. We address this gap through a novel application of causal loop diagrams (CLDs), a system-thinking tool that visualises interdependencies and feedback loops in complex systems22,23,24,25,26. CLDs also identify leverage points, high-impact nodes that can drive systemic change, offering insights into effective interventions for significant and desirable changes22,23,27,28. Despite growing use in nature-based solutions28,29,30,31, CLDs have rarely explored co-design dynamics in UGI planning despite research noting their potential31. Here, we aim to:

  1. i.

    Visualise mechanisms and trade-offs in effective co-design;

  2. ii.

    Identify cause-effect chains and leverage points; and

  3. iii.

    Assess empowerment outcomes.

Applying CLDs in a case study in the London Borough of Hammersmith and Fulham (LBHF), UK, we reveal systemic mechanisms, leverage points, and empowerment pathways to advance the effectiveness of co-design approaches (ECDA) for equitable and sustainable UGI design amid global urbanisation and climate pressures. This approach provides a scalable framework to enhance UGI design worldwide, addressing pressing climate and equity challenges.

Results

Identified themes and dynamics of co-design

Analysis of 23 semi-structured interviews and 3 focus groups reveals that ‘effective engagement and participation’ serves as the catalytic core of co-design in LBHF green spaces, orchestrating a network of decision-making influence, social capital, interrelationships, and contextual resilience across six domains (Table 1). These domains align with the theoretical framework’s focus on factors influencing co-design, such as institutional support and stakeholder capabilities17. We identified 43 thematic factors (Supplementary Table 1), with 18 derived from interviews (e.g., respect for diverse experiences, collaboration, empowerment) and all 43 from focus groups (e.g., cultural sensitivity, socio-economic context). Focus group-specific sub-themes, such as ‘capacity of facilitators’ and ‘reflective activities’, were tied to collaborative tasks. The treemap diagram (Fig. 1) illustrates their hierarchical organisation.

Table 1 The six overarching themes identified in the prior systematic review17
Full size table
Fig. 1: Treemap diagram showing the themes and sub-themes identified from the data.
figure 1

The frequency of appearance was calculated using the topic mention length divided by the total conversation length (from interviews and focus groups, n = 38, ~19.5 h) and multiplied by 100. These percentages, detailed in Supplementary Table 1, are derived from thematic coding and indicate topic prominence, with higher percentages reflecting greater centrality (e.g., 37% for cultural sensitivity) and lower percentages indicating less dominance.

Full size image

Across all domains, the sub-themes with the highest frequency were related to inclusivity and diversity such as inclusive activities (20/23 interviews, 3/3 focus groups), diverse participation (15/23 interviews, 3/3 focus groups), and open and inclusive atmosphere (19/23 interviews, 3/3 focus groups). These sub-themes enriched discussions, prompting further exploration of related concepts. For example, the ‘diverse participation’ sub-theme sparked discussions on respecting others’ voices, while ‘clarity of purposes and processes’ highlighted the need for deliberative engagement. Participants noted, “It’s important to have a deliberative approach to engagement” (Participant 1), and “Public consultations should facilitate meaningful conversations” (Participant 2).

Themes concerning interrelationships also received attention, with participants valuing ‘collaboration’, ‘resource exchange’, and ‘equity of community participation’. One stated, “We must work together to align designs with community needs and local context,” while another noted, “Ensuring fair participation secures public support.” Resource sharing was also key, as reflected by, “I appreciate all the inputs from other participants. Feel like I’m learning a lot.” Conversely, the ‘power dynamics’ sub-theme (14/23 interviews, 3/3 focus groups) had a negative impact, with participants noting that authority imbalances hindered participation. One remarked, “There’s no hierarchy. It’s bothersome if someone is more superior,” while another said, “I feel a bit daunted if someone directs me.”

Supplementary Table 1 shows the detailed themes and sub-themes constructed from the data, together with their description, frequency of appearance, and sample excerpts.

Causal loop diagrams: Influential factors, cause-effect chains, and leverage points

The CLDs were developed to illustrate the factors identified to affect co-design practices and the outcomes arising from the interactions among these. We created three separate causal loop diagrams, one focusing on interview insights (Supplementary Fig. 1), one from focus groups (Supplementary Fig. 2), and a synthetic CLD (Fig. 2) integrating all datasets and the theoretical framework17. The interview-based CLD maps 18 factors with five reinforcing loops and three balancing loops. The focus group-based CLD, denser with all 43 factors across six domains (Table 1), includes six reinforcing loops and five balancing loops, emphasising trade-offs and empowerment. Building on these single-method analyses, the synthetic CLD (Fig. 2) offers a holistic perspective on the dynamic interactions among all 43 factors across the six identified domains.

Fig. 2: Unified causal loop diagram combining the insights across interview and focus group datasets and the framework on co-design from17.
figure 2

Arrows indicate causality; ‘+’ denotes an increase in the independent variable raises the dependent variable (∂Y/∂X > 0), ‘-’ denotes a decrease (∂Y/∂X < 0). Loop labels ‘R’ indicate reinforcing feedback, ‘B’ balancing feedback. Black represents variables, with forest green marking outcomes.

Full size image

The synthetic CLD reveals that co-design relationships are dynamic, with effectiveness varying in response to external factors, including socio-economic and cultural contexts, often beyond direct control, and internal factors, including personal values and organisational structures. This underscores the need for adaptability in co-design practices to navigate these influences effectively. Synthesis across datasets highlights significant overlap, notably the roles of inclusive organisational arrangements, local context sensitivity, and internal perceptions like perceived trust (29%, meaning nearly a third of the conversation focused on trust) in enhancing ECDA, thus strengthening social capital (e.g., “cultural sensitivity” at 37%, over a third of the discussion) and community empowerment (30%). For instance, after collage-making activities, participants reported converging on a shared idea, which strengthened trust. One participant reflected, “The trust we built allowed us to share ideas freely,” while another noted, “Seeing my ideas valued by people from different backgrounds made me feel empowered”.

Feedback loops and leverage points

The synthetic CLD (Fig. 2) comprises seven reinforcing loops (R1-R7) and five balancing loops (B1-B5) that shape design dynamics, with all variables interrelated in a multi-directional way. R1, centres on thoughtful engagement, where “deliberativeness of public engagement and participation” (20%) and “clarity of purposes and processes” (32%) amplify the proven value of co-design. A focus group participant highlighted this dynamic, stating, “A deliberative approach with clear structures boosts our growth and sustainability” (Focus Group 1). R2 accentuates community understanding, with “community understanding of their role and contribution” (18%) and “willingness of community to participate” (10%), strengthening ECDA. An interviewee noted, “When communities feel their input is valued, engagement becomes empowering, not performative” (Interviewee 14).

R3 links social capital, interrelationships, and perceptions, where “social capital” (e.g., “cultural sensitivity” at 37%, “trust” at 29%) builds “interrelationships” (e.g., “open and inclusive atmosphere” at 45%), shaping “perceptions” (e.g., “degree of influence on decision-making” at 34%), leading to outcomes like “community empowerment” (30%) and “co-generated solutions.” A focus group participant emphasised the role of cultural sensitivity, stating, “Cultural respect deepens our bonds” (Focus Group 3), while another interviewee highlighted the outcome, noting, “Successful solutions fuel action and build momentum for change” (Interviewee 17). R4 emphasises inclusive engagement, with “community engagement involvement” and “equity of community participation” (28%) driving “continuous sharing and learning” (44%) and a “shared vision.” A focus group member reflected, “Diverse voices align our vision” (Focus Group 2). R5 highlights reflective practices, where “reflective activities” (12%) and “evaluation processes” (14%) enhance “workshops,” reinforcing ECDA.

R6 connects “degree of influence on decision-making processes” (34%) and “willingness of community to participate” (10%) with ECDA, promoting “equity of community participation” (28%) and “interrelationships” (45%), ultimately leading to “co-generated solutions,” “informed decision-making,” “community needs addressed,” and “community empowerment” (30%). R7 involves “awareness of the needs and expectations of the public” and “respect for diverse experience and knowledge bases,” enhancing “willingness to engage in deliberative public engagement” (10%) and “organisational arrangements.” A focus group participant noted, “Understanding needs with respect drives meaningful engagement” (Focus Group 4), underscoring the role of awareness and respect in co-design.

Among the reinforcing loops, R3 exerts the most significant impact due to the high frequency of its codes, identifying three key leverage points: social capital, interrelationships, and perceptions. These highly connected nodes serve as strategic intervention points, directly yielding desirable outcomes. For instance, social capital, encompassing “trust,” “cultural sensitivity,” and “social support”, initiates a positive feedback loop that amplifies effects across perceptions and interrelationships, resulting in outcomes like co-generated solutions, community empowerment, and addressed community needs. A targeted shift in these leverage points can produce sustained, transformative impacts within the co-design system.

Balancing loops regulate the system by addressing challenges. B1 tackles transparency issues, where “lack of transparency” (15%) limits “diverse participation” (29%), balanced by ECDA, as participants emphasised, “The process needs to be transparent - what is our role?” B2 mitigates mismatched expectations (23%), reducing “perceived trust” and stabilising through ECDA, with a participant noting, “The outcome doesn’t reflect what I want.” B3 counters power dynamics (32%), where imbalances lower “sense of fairness” (8%) and “leadership” (5%), balanced by the ECDA as participants stressed, “Everyone should play an equal role.” An interviewee further highlighted this challenge, stating, “Persistent authority gaps erode trust and limit engagement’s reach” (Interviewee 9). B4 addresses socio-economic barriers, where “socio-economic context” (15%) and “cultural context” (13%) constrain “diverse participation,” balanced by ECDA, with a focus group member noting, “Economic gaps block agreement” (Focus Group 2). B5 manages resource constraints, where “socio-economic context” limits “structured systems,” balancing ECDA (section 4.1.2). B3 exhibits the greatest influence among balancing loops due to the high frequency of “power dynamics” (32%), underscoring leadership and power imbalance challenges.

Outcomes and impacts

As a result of the co-design practices, outcomes and impacts such as informed decision-making and co-generated solutions are identified. Across datasets and the literature framework, community empowerment, co-generated solutions/ works, improved power and capacity of the community, and sustainable community growth are the outcomes and impacts. These are not endpoints but inputs perpetuating the co-design cycle, with short-term gains (e.g., “shared visions,” enhanced community knowledge) and long-term impacts (e.g., evidence-based decision-making, sustainable growth). Participants reflected, “Seeing the results; it’s the best reward,” and “It wasn’t just talk; I learned stuff,” highlighting the tangible benefits of co-design. An interviewee further emphasised the value of collaboration, stating, “Meaningful collaboration produces tangible outcomes that wouldn’t exist otherwise” (Interviewee 10), reinforcing the significance of co-generated solutions.

Discussion

Co-design has gained prominence in public realm design worldwide10,11,12. This is particularly so with green space design due to perceived benefits in ensuring that spaces are designed in a locally specific context that meets the needs of users. Despite its widespread adoption, there remains a lack of rigorous evaluation, hindering the identification of best practices17,32. The assumption that stakeholder engagement is inherently beneficial often overshadows co-design’s core values, including collaboration, equity, and local expertise, complicating efforts to assess its impact33. The lack of discourse on co-design methodologies and the factors influencing its effectiveness further poses a significant barrier to achieving meaningful engagement, especially with diverse stakeholders17,33,34,35.

To address these gaps, we employed CLDs to visualise the factors, cause-effect chains, trade-offs, challenges, and leverage points shaping co-design effectiveness, drawing on interviews, focus groups, and the theoretical framework from17). Our findings demonstrate that CLDs are a powerful tool, not only capturing immediate outcomes but also illuminating the long-term dynamics of inclusive and sustainable solutions over time. The CLD approach uniquely elucidates the temporal evolution of feedback loops22,36,37, enabling the evaluation of complex interdependencies such as trust-building, equity, and power asymmetries.

This systems perspective offers a novel methodology for evaluating co-design in urban planning and is broadly transferable to other domains such as public health and community development, where similarly complex and context-dependent systems require inclusive and adaptive design. In doing so, it equips policymakers with actionable insights to optimise co-design practices across diverse settings and populations. Table 2 maps each feedback loop to policy recommendations and measurable indicators, supporting practical application for policymakers and practitioners in achieving equitable, sustainable outcomes.

Table 2 Co-design system loops and associated policy recommendations and indicators
Full size table

Building on these insights, systemic constraints, such as entrenched power dynamics and funding shortages, significantly hinder co-design effectiveness. Our CLDs reveal that imbalances in leadership and authority can marginalise participants and undermine trust. Decision-making processes often rely on ‘experts’, typically referred to as those with professional experience and knowledge of the field, and the communication with lay people has been characterised by a deficient top-down, one-way model that restricts shared learning and innovation38,39. Rather than advocating for a complete shift to a bottom-up model, we propose a symmetric and inclusive communication framework that bolsters trust, respect, and mutual understanding between experts and the public, ensuring both contribute to shaping outcomes.

Urban green space design has increasingly embraced public participation, yet the focus must shift toward facilitating equitable dialogues that bridge knowledge gaps and organisational barriers40. While not explicitly modeled in the CLD, political structures significantly influence power dynamics, particularly through constraints in collaboration, capacity, and funding41,42. In the UK, for instance, government commitments to “level up” through green space improvements, backed by £39 million in the 2022 white paper43, are undermined by a decade of austerity, with park funding cut by £330 million annually since 20109,44,45. This has led to reduced staff and deteriorating infrastructure, creating a dissonance between policy goals and implementation means. This funding shortfall exacerbates inequities in green space access, particularly for marginalised communities, as evidenced by the socio-economic barriers identified in B4 and B5. Addressing these systemic constraints requires strategic interventions, such as reallocating resources to support diverse stakeholder participation and creating windows of opportunity (e.g., aligned local and central government priorities) to enhance public engagement in green space design. Such interventions can also promote environmental justice by ensuring equitable access to green spaces.

Although resource-intensive, co-design processes yield long-term benefits like community empowerment, increased social capital, and context-sensitive interventions11,12,46. These outcomes underscore the importance of not only tangible outputs but also softer, process-based benefits. As demonstrated in the CLDs, shifting feedback dynamics require investment in resources, leadership, and creative, locally responsive delivery mechanisms. This is crucial as the extent and method of community participation are shaped by different local planning settings, legislative structures, and the diverse planning traditions across municipalities. For instance, strategies fostering social capital (R3) could be adapted to enhance green space co-design in rapidly urbanising regions like Southeast Asia, where community cohesion is vital for sustainable urban development. This should also allow for variations between large and small settlements and between different regions, each having distinctive perspectives on urban design and greening, as these lead to discrepancies in how the co-design model is applied47.

Participation patterns vary across demographic and socio-economic groups, shaped by factors such as gender, age, and cultural norms. Intergroup differences should be considered to navigate potential conflicts during deliberately diverse group contact48. Being mindful of the experiences of under-represented groups is equally critical to avoid feelings of exclusion or being ‘othered’ in the space38. For example, low-income women living in developing countries might face challenges in expressing their opinions and participating in co-design activities within mixed-gender groups due to reasons such as traditional roles dictated by their society and multiple household obligations. These challenges aggravate the power imbalance between men and women, thus reducing women’s contributions in co-design sessions49. Addressing these disparities aligns with global equity goals, such as UN Sustainable Development Goal 11 (Sustainable Cities and Communities), by ensuring inclusive urban planning processes that empower all community members.

This study further highlights the need to expand the foci of co-design practices from purely enhancing the capacity at the individual level to facilitating change at the organisational and systemic levels, since these three levels in combination form a ripple effect throughout the co-design process as demonstrated in the CLDs. Reinforcing feedback loops (e.g., R3) emerge when supportive organisational arrangements, informed participants, and systemic policies align, amplifying outcomes like empowerment. Conversely, obstacles at any level trigger balancing loops (e.g., B3, B4), diminishing impacts over time. Designing with this systems awareness can support more resilient, transformative change in green space planning and beyond.

Whist the CLDs indicate different outcomes and impacts generated through the interactions across variables and from effective co-design practices, we recognise that these might not materialise in the short term or at all levels, including individual, organisational, and systemic. Hence, there need to be well-timed, appropriate reflective and evaluation sessions as well as resources to ensure and sustain changes. Integrating stakeholder engagement across all co-design phases, from objective-setting to evaluation, ensures interventions align with intended outcomes.

Co-design practices must also remain flexible and responsive to systemic changes, treating public engagement as an ongoing process rather than a fixed endpoint. Developing interactive digital tools, such as CLD maps, can enhance stakeholder engagement by allowing dynamic exploration of causal links, though accessibility must be ensured through offline alternatives or training. This adaptability enables co-design to accommodate diverse planning contexts, facilitator capacities, and community motivations38. Actively involving stakeholders throughout the process, supported by CLDs, enhances intervention relevance and facilitates impact assessment, ensuring progress in co-design development. This approach can bridge variations between large and small settlements or regions with distinct urban design traditions, ensuring co-design remains inclusive and effective across diverse settings.

Despite its strengths, this study has limitations that inform future research directions. The sample size (n = 38, representing ~19.5 hours of audio data) is specific to LBHF, potentially limiting generalisability to other urban contexts, such as rural areas or non-UK settings. Snowball sampling introduced a selection bias, overrepresenting engaged residents (e.g., 15/23 White interviewees vs. 63% LBHF population), and underrepresenting less-engaged groups (e.g., elderly >65 at 18/38 vs. 18% UK-wide). The collage exercise, while enhancing engagement, may bias toward visual learners or those comfortable with creative tasks. The relatively small dataset (n = 38) and lack of longitudinal data constrained the ability to capture evolving co-design dynamics, such as long-term stakeholder relationships or sustained empowerment impacts.

CLDs, while effective for visualising system dynamics, can be complex and less accessible for participants unfamiliar with systems thinking50. The static presentation of CLDs may further limit stakeholder engagement. An interactive map, allowing users to explore causal links dynamically, could enhance understanding and engagement, though it requires technical development and consideration of digital access barriers51. To improve usability, future studies could employ simplified visual aids, interactive workshops, or pre-study training sessions to build participant capacity in systems thinking. Validation of CLDs relied on real-time Menti surveys (85% agreement) and triangulation with the theoretical framework17 and literature52, rather than traditional group model building, due to resource constraints and participant inexperience with such methods. This approach provided immediacy and reduced recall bias but lacked the iterative depth of group model building, which could refine stakeholder consensus on CLD variables.

Future research should expand the sample to 50+ participants across diverse UK boroughs, employing stratified sampling to include underrepresented groups, and conduct intersectional analyses to deepen equity insights. Longitudinal studies over 12–24 months could assess co-design’s temporal impacts. Employing group model building with trained facilitators could enhance CLD validation, while testing this framework in varied contexts (e.g., rural UK, urban Global South) could improve transferability. Incorporating interactive digital tools, such as CLD maps, and ensuring accessibility through offline alternatives or training could further improve stakeholder engagement. These steps will build on the success factors, constraints, and system dynamics identified in this study to advance inclusive, sustainable co-design practices.

Methods

Study design and definitions

This study investigates perceptions of co-design, defined as a collaborative process where stakeholders, including residents, academia, and decision-makers, jointly shape green space outcomes beyond consultation. Co-design was selected over related concepts like co-production or co-creation due to its emphasis on equitable, reciprocal knowledge exchange and joint decision-making, aligning with the study’s focus on inclusive governance and community ownership in UGI planning17,53. Unlike co-production, which often focuses on service delivery, or co-creation, which may prioritise innovation, co-design ensures stakeholders collaboratively shape tangible outcomes, enhancing local relevance and empowerment17,54,55. This approach is critical in urban green space planning, where traditional top-down methods often fail to address public heterogeneity, leading to tokenism, distrust, and cultural disconnects56. Here, green spaces refer to parks and open areas that enhance ecological and social value, such as biodiversity and community well-being.

Case study context

The case study of this research is the London Borough of Hammersmith and Fulham (LBHF), United Kingdom, addressing the gaps in EU public participation studies19. LBHF’s varied urban landscape, featuring parks like Wormwood Scrubs and Bishop’s Park, plus biodiversity initiatives (e.g., Tiny Forest initiatives)57, and council commitment to co-design (e.g., Disabled People’s Commission in 2016)58 make it ideal. Its diverse demography (mean age 34 years, 63% White compared to 82% nationally, 27% non-UK identity compared to 9.7%) enables exploration of inclusive engagement across social groups, enhancing finding generalisability59. These demographics informed the analysis by ensuring themes reflected diverse perspectives, such as age- and ethnicity-specific needs, to promote equitable co-design outcomes. Demographic comparisons (e.g., age, gender, nationality) were explored to ensure representativeness, with plans for future research to conduct intersectional analyses for deeper equity insights.

Interviews (n = 23) and focus groups (n = 15) inform causal loop diagrams, grounded in a systematic review17, to enhance co-design practice. Ethics approval was secured from Imperial College’s Research Governance and Integrity Team (RGIT) on 8 February 2023 (Reference Number: 6481813), ensuring voluntary and informed participation with no identifiable data collected.

Methodological framework

The study follows a three-stage framework (Fig. 3), informed by established systems-based co-design methodologies, such as Participatory Systems Mapping, to ensure systematic stakeholder engagement52,60,61. Stage 1 synthesises theory from a systematic review (ref.17). Stage 2 collects qualitative data on co-design perceptions via interviews and focus groups. Stage 3 maps findings into CLDs to inform practice. This framework was iteratively refined by the research team to ensure alignment with the study’s objectives and to enhance methodological coherence.

Fig. 3: Three-step methodological approach of this research.
figure 3

This iterative process integrates theoretical and empirical insights to develop and refine causal loop diagrams (CLDs) for assessing co-design effectiveness. Step 1 synthesises theory from a systematic review to identify variables influencing co-design and inform a conceptual framework. Step 2 gathers qualitative data through in-depth interviews (n = 23) and focus groups (n = 15) to capture diverse stakeholder perceptions. Step 3 maps findings into CLDs, revealing causal relationships, reinforcing and balancing feedback loops, and leverage points. The process supports practical application and continuous refinement of the co-design framework.

Full size image

First, a PRISMA-guided systematic review17 identified six co-design themes, including organisational arrangements, structured systems, social capital, interrelationships, perceptions, and context (Table 1), shaping stakeholder dynamics and distinguishing co-design from consultation. This framework (Supplementary Fig. 1) guided the design of interview and focus group protocols and data analysis, grounding the study in collaboration and co-design principles.

Subsequently, we conducted 23 semi-structured, face-to-face interviews with LBHF residents, all conducted face-to-face to bolster open dialogue, recruited via snowball sampling and a prior UK-wide survey56. Snowball sampling was strategically employed to access hard-to-reach community members, with initial contacts diversified through community organisations and local councils to mitigate bias toward engaged residents, though its potential bias toward engaged residents is acknowledged. Participants were purposively selected to reflect LBHF’s demographic diversity, including age, ethnicity, and socio-economic status. The interview protocol included broad questions (e.g., “What green space features are most important to you?”) to establish context, and specific prompts (e.g., “How could collaboration enhance green space outcomes?”) to explore co-design perceptions. The protocol was piloted with five participants, selected purposively from diverse demographic groups (e.g., varying age, ethnicity, and co-design experience) within LBHF, to refine question clarity, reduce interviewer bias, and ensure alignment with the study’s focus on perceptions rather than active processes. Interviews were conducted by trained researchers between March and June 2023, lasting 30-40 minutes each, recorded using Microsoft Teams, and transcribed verbatim.

Three focus groups (n = 15, 4–6 per session) co-generated ideas on inclusive green spaces and co-design enhancements. Each session (90 min) was recorded and transcribed with consent. Participants addressed: (1) inclusive design for diverse/ non-users, and (2) collaboration improvements, hypothetically. Engagement was enhanced through real-time Menti surveys to gauge consensus and a 40 min group collage exercise (Fig. 4), designed to stimulate creative input, though potential visual bias is noted. Neutral prompts (e.g., “What would an inclusive green space look like?”) ensured participant-driven outcomes. Participants were recruited from the prior UK-wide survey, supplemented by interview participants and snowball sampling, aiming for diversity in co-design experience. This mixes targeted broader insights, though it may overrepresent engaged individuals. See Fig. 4 and Table 3.

Fig. 4: Structure and flow of focus group sessions with co-design elements highlighting group collage to co-generate inclusive green space solutions.
figure 4

The process began with preparatory steps, including the recruitment of participants with diverse co-design experience and a photo task to prompt engagement. Each session combined structured surveys (via Menti) with facilitated discussions to explore perceptions of green space, co-design approaches, and collaboration improvements. The central group collage exercise served as a participatory tool to co-generate ideas for inclusive green space design. Sessions concluded with impact assessment and reflective discussion, supporting stakeholder-led evaluation of co-design processes.

Full size image
Table 3 Demographic characteristics of the interviewees and focus group participants
Full size table

Data processing and analysis techniques

Data from 23 interviews ( ~ 15 hours of audio) and focus groups (n = 15, ~4.5 h excluding silences) were processed and analysed to map co-design perceptions in CLDs, adapting systematic qualitative techniques from Saldana, Hsieh and Shannon, and Braun and Clarke50,51,62. The total dataset comprised ~19.5 h of audio, transcribed with an accuracy rate of 98% using Microsoft Teams’ automated transcription, followed by manual correction by a trained researcher. A four-step process (Fig. 5) guided the analysis: (1) thematic coding, (2) theme synthesis, (3) CLD development, and (4) validation, iterating until saturation. No new themes emerged after 20 interviews and 2 focus groups, confirmed by redundancy across transcripts.

Fig. 5: The four principal steps used to process and analyse data drawn from the systematic review, interviews, and focus groups.
figure 5

The analysis began with thematic coding of transcripts using a hybrid deductive–inductive approach (Step 1), followed by synthesis of themes based on prominence and depth (Step 2). In Step 3, themes were mapped into causal loop diagrams by identifying variables, feedback loops, and directional relationships. Step 4 involved validation through cross-checking against raw data, literature comparison, and confirmation of thematic saturation, ensuring robustness and reliability of the causal loop diagram models.

Full size image

In the first step, we coded ~19.5 hours of audio (transcribed via Teams) in NVivo 11 using thematic analysis50,51,62 to identify meanings and relationships. A hybrid deductive-inductive approach was employed to balance theoretical grounding with data-driven insights:

  • Deductive coding: We tested the theoretical framework’s six themes (e.g., ‘organisational arrangements’, ‘trust’) (Table 1) via a pre-defined codebook, developed from the PRISMA-guided systematic review17 and refined through pilot interviews. Codes were cross-checked against to validate or refine constructs (e.g., ‘social capital’ expanded to ‘cultural sensitivity’).

  • Inductive Coding: We identified novel themes (e.g., ‘sense of ownership’) from uncoded segments, ensuring findings emerged from data, not just the framework. Coding ceased at saturation, confirmed by redundancy across transcripts (e.g., no new codes after 90% of data).

    In the second step, themes were synthesised by prominence and depth.

  • Prominence was calculated as mention time divided by total audio time where higher percentages indicate greater centrality and lower percentages suggest lesser dominance. Frequencies were tallied across all transcripts.

  • Depth was assessed via discussion duration (e.g., >15 min), ensuring that rich but low-frequency themes were retained.

Evaluating both depth and prominence allows the recognition of themes that, while infrequent, sparked in-depth discussions, suggesting their complexity or importance. Conversely, frequently mentioned themes with limited discussion may indicate broad awareness but a lack of detailed exploration37.

Subsequently, three CLDs, including interview-specific, focus group-specific, and unified, were built in four sub-steps, adapting22,36,63 and using EDrawMax software for visualisation:

  1. 1.

    Variable Aggregation: Codes were reduced to core variables by thematic overlap (e.g., ‘clarity’ from related sub-codes), each supported by more than two quotes, with relationships centred on ‘effective engagement and participation.’

  2. 2.

    Causal Link Identification: In a CLD, interconnected variables are linked by arrows to represent their causal relationships. The dependence of variables is indicated by either (+) to denote a positive change when the independent variable changes, or a (-) in the opposite directions28,36. One thing to note that these positive and negative signs only infer the direction of change, irrespective of whether the change might be considered beneficial or detrimental. The interplay of positive and negative links can create reinforcing (‘R’) or balancing (‘B’) feedback loops. Reinforcing loops represent that the variables amplify each other, contributing to either unbounded growth or decline. Balancing feedback loops serve as stabilising forces that work to maintain equilibrium and dampen system deviations22. In this study, relationships were coded from participant statements (e.g., “This would encourage…”; “X is important for Y”; “this is important for”) using keywords (e.g., ‘boosts’, ‘limits’) and context. Polarity was assigned ( + for same-direction and - for opposite), logged with supporting quotes.

  3. 3.

    Separate CLD development: Interview data and focus group data were analysed independently, producing two initial CLDs. Shared themes were identified via overlap (e.g., 80% thematic consistency).

  4. 4.

    Unified CLD integration: Initial CLDs were merged via team consensus, retaining key variables and closed loops (R for reinforcing, B for balancing), excluding open chains from R/B labels20. Loops were traced across non-consecutive transcript segments.

Finally, the unified CLD was validated by: (1) cross-checking all variables and causal links against raw transcripts and audio recordings to ensure fidelity, (2) comparing the model against the systematic review and additional literature (e.g., see refs. 17,64) to confirm theoretical alignment, and (3) conducting a real-time Menti survey post-collage exercise with 85% participant agreement on key variables, minimising retrospective bias. Leverage points were identified as nodes with high connectivity or long-term impact, refined through iterative team discussions. This multi-layered validation process enhances the CLD’s scientific credibility, though limitations in participatory validation are noted.