Evaluating and evidencing asset-based approaches and co-production in health inequalities: measuring the unmeasurable?

ABSTRACT We provide new empirical and methodological findings problematising evidence, evaluation and measurement when using asset-based approaches and co-production to improve health and address inequalities. The Assets Model was applied to develop and implement a co-produced methodological evaluation framework – Asset-Based Indicator Framework – measuring impacts of creative community engagement on health and inequalities. Our three-phased approach synthesised data from multiple sources over five years. Phase One identified existing health assets in Black and Minority Ethnic (BME) groups using a six-month ethnography; interviews with community organisation representatives (n=35); three BME focus groups (n=27). Phase Two identified conditions/actions for assets to be activated to maximise health for ‘Roma’ using community-based participatory action research (n=50). Phase Three turned to evaluation and indicators. The Asset-Based Indicator Framework was co-produced with self-identified Gypsies following an extensive literature review; expert interview; three participatory action-research workshops with professionals (n=20); two participatory action-research workshops with community members and professionals (n=49). Asset-based evaluation and measurement is complex due to narrow conceptualisations of evidence; epistemological and methodological incompatibilities; and lack of capacity and resources for sustained engagement. Understanding the granularity and interrelation of assets at individual, community and structural levels; identifying mechanisms through which change happens; and applying participatory and empowering methods to capture actions on assets leading to community-defined outcomes may not lead to meaningful changes without systemic change. The paper fills a clearly-defined gap addressing evaluation of community engagement within complex systems rather than clinical-medicine interventions with wide-ranging implications for international research, practice and policy.


Introduction
This paper addresses calls to 'revitalise' the evidence-base and progress evaluation for asset-based approaches to tackle health inequities (Rippon & South, 2017;Morgan & Ziglio, 2007, p. 7). This is compelling given global health challenges including widening inequalities; poverty; non-existent health systems in some nations, culturally inappropriate ones in others; and aging populations (Marmot, 2005). Solutions have included setting high-level global targets such as United Nations Millennium Development Goals, and an array of interventionswith limited evidence of success (Clegg, 2015;Smith & Eltanani, 2015).
Dominant neoliberal discourses and emphasis on individual choice rather than structural drivers of inequalities are recurrently blamed (Smith, Bambra, & Hill, 2015) themes that have fractured public health over whether we should focus on individual or communities' needs or their strengths (assets) to improve health and reduce inequalities.
Our ongoing empirical and methodological research in disadvantaged communities since 2013 situates itself between the two paradigms. It fills a clearly defined gap in the health assets evidence-based research agenda (Rippon & Hopkins, 2015), and advances debates on evaluations of community engagement within complex systems rather clinical medicine interventions (South & Phillips, 2014). The paper problematises evidence, evaluation and measurement when using assetbased approaches and co-production to improve health and address inequalities. It does this by applying the Assets Model (also referred to as 'the model') (Morgan, 2012) to develop and implement a co-produced methodological evaluation framework -Asset-Based Indicator Framework (also referred to as 'the framework')to measure impacts of creative community engagement on health and inequalities. A critique of findings shows how evaluation and measurement is far from straightforward when using these approaches due to narrow conceptualisations of evidence; epistemological and methodological incompatibilities; and a lack of capacity and resources for sustained engagement.
The paper begins with explanations and critiques of asset-based approaches, followed by an explanation of the model's three-phased approach to evidence, action and evaluation (Morgan & Ziglio, 2007). We apply the three phases to empirical research in a disadvantaged Scottish neighbourhood then critique findings in relation to limitations of evidence, evaluation and measurement in asset-based approach interventions. The paper concludes with implications and international relevance of these approaches.

Asset-based approaches
The recognition of assets aims at redressing 'the balance between meeting needs and nurturing the strengths and resources of people and communities' (Glasgow Centre For Population Health, 2012, p. 2). Assets can be 'any factor that enhances the ability to create or sustain health and wellbeing, such as the resources that promote self-esteem and the coping abilities of individuals and communities' (Brooks & Kendall, 2013, p. 128), from personal characteristics to interpersonal relationships (Kretzmann & McKnight, 1993). Co-production involves equal and active involvement of partners including communities, health practitioners and voluntary sector in the design and delivery of services, programmes, or initiatives (Batalden et al., 2016).
While Rippon and Hopkins (2015) note that asset-based approaches are sometimes not easily understood by practitioners whose work is often driven by their intuition of what is right, Roy (2017, p. 462) reports on: the agency of practitioners in resisting, (de-) constructing and utilising policy ideas and discourseseven those that could reasonably be argued as being ostensibly neoliberal in natureto suit their own agenda; to benefit the individuals and communities they exist to support.
He responds to critiques of the assets movement being a 'tool of neoliberalism' (Roy, 2017, p. 462) by masking structural issues and institutions that generate and propagate inequity (Friedli, 2013), and focusing on 'values of individualisation, marketisation and privatisation of public life' (MacLeod & Emejulu, 2014, p. 432). Others point to the lack of evidence on asset-based approaches to demonstrate successful prevention of ill-health (Glasgow Centre For Population Health, 2012).
In Scotland, however, these approaches have not only been embraced by its former Chief Medical Officer but are entrenched in the policy landscape (Glasgow Centre For Population Health, 2012;Scottish Government, 2010). Under the Community Empowerment Act 2015, there is legal obligation to mobilise assets, decentralise decisions and move towards effectively focused local policies so communities can have stronger voice and influence over issues that matter to them. National outcomes, determined with people representing communities, 'must have regard to the reduction of inequalities of outcome which result from socio-economic disadvantage' (Scottish Government, 2015).
Communities can be based on identity, common interest or geography. While the latter is straightforward, defining the common interest and identity of marginalised groups is complicated particularly with Black and Minority Ethnic (BME) communities as ethnicity is multifaceted and could be conceived as commonalities within a community or variances from 'other' communities. While ethnic self-identification categories for population surveys aim to acquire reliable and consistent measures of ethnic identity, they may not capture communities' own perceptions of the construct and lived experiences (Institute for Social & Economic Research, 2008). This research therefore sought to gather specific BME (Pakistani, Polish, Slovakian, 'Roma') groups' perceptions of health and explore how asset-based approaches and co-production could be used to engage so called 'hard-to reach' communities.

Methods
We applied the model's three-phased approach to evidence, action, and evaluation (Morgan & Ziglio, 2007) to co-produce the framework (supplementary material Figure 1). An abductive (Peirce, 2014), systematic combining (Dubois & Gadde, 2002) approach to conceptualisation and theory-building combined inductive and deductive techniques and continuous backwards and forwards analysis to integrate the original model and evolving framework. Literature was reconsidered between phases to move 'between an empirical and a model world' (Dubois & Gadde, 2002, p. 554). The iterative process identified key themes by reading and coding data, literature and emergent theory. Discussions between researchers were ongoing to identify discrepancies, clarify meanings, and establish emergent coding themes (Dey & Teasdale, 2016;Saldana, 2015) to inform framework development (supplementary material Figure 2).
To identify assets and understand beliefs and behaviours from an insider view, data in Phase One were collected through a six-month ethnography, semi-structured interviews and focus groups with communities and representatives from community organisations (supplementary material Figure 1). Recruitment took place through local formal and informal networks; active participation in neighbourhood events; and visiting shops, community centres, libraries, and shisha cafes. Interviews and focus groups were recorded and professionally transcribed for analysis, and a reflexive journal captured ethnographic notes taken in situ or straight after exchanges (Foley, 2002).
All BME groups expressed interest in sustained creative community engagement with a view to coproducing, implementing, and evaluating services, though the 'Roma' emerged as a priority group as they are often socially isolated and excluded from participation. 'Roma' communities experience challenges in accessing healthcare including communication barriers, unfamiliarity with medical procedures and services, and infrequent contact with health providers (Lane, Spencer, & Jones, 2014;Poole & Adamson, 2008). Accommodation insecurity, poor living conditions and low community participation contribute to poor health outcomes, so joined-up working at national and international levels is needed to address wider social determinants of 'Roma' health (House of Commons, 2017).
In Phase Two, the focus was on what needed to be achieved to activate identified community assets and how organisations could work in partnership to mobilise solutions to address emergent issues. As communities identified creative events with food as preferred mediums for engagement, community-based participatory action research was conducted using Theatre of the Oppressed. Performers created scenes or images in which the protagonist was failing to achieve their own desires or goals and audience members stopped the dramatic action at any time to replace the protagonist and perform alternative actions (Boal, 1993). 'Roma' community members met for a 7-hour workshop to discuss and enact Phase One findings, then performed these to professionals during a second seven-hour workshop (supplementary material Figure 1).
A key Phase Two finding was to build on identified assets to develop and pilot a framework to monitor the effectiveness of asset-based approaches to engage communities and co-produce services leading to improved health outcomes and reduced inequalities. This aligns with the third phase of the Assets Model, which turns to evaluation and identification of asset-based indicators linked to improvement in communities' environments (Morgan & Ziglio, 2007).
Phase Three calls for a 'new set of asset indicators with multi-method evaluations to assess the effectiveness of community-based approaches to tackling health inequities' (Morgan & Ziglio, 2007, p. 18). It commenced with an extensive literature review to analyse existing asset-based evaluation methods, and identify overlaps between concepts related to asset-based approaches to establish framework perimeters. Three key interconnected concepts influenced by the same or similar individual, community, and structural assets were identified (supplementary material Table 1)well-being, social capital and resilience. Thirteen indicators were drawn from the literature. Put simply, if these measures shift then so too should health and well-being (de Andrade & Angelova, 2017). These served as a template to co-produce the framework with practitioners, professionals and Gypsies (community members wanted to be referred to as Gypsies rather than 'Roma' in Phase Three). Data were collected through a 90-minute face-to-face interview with a practitioner using the personaloutcomes approach, which focuses on what matters to service users (Miller & Barrie, 2016). Further fieldwork included three action-research workshops with professionals and two seven-hour actionresearch sessions with communities and professionals (supplementary material Figure 1). Data in Phases Two and Three were gathered through video, creative evaluations using drawings, flipcharts, feedback questionnaires from professionals and a reflexive journal.
Both researchers have experience working with vulnerable BME and 'Roma'/Gypsies. Initially, some participants did not feel comfortable signing consent forms due to mistrust but were comfortable with verbal consent. By spending time in the community and working closely with trusted partners, participants went on to sign consent forms. Once trust was established, there was an openness to share data in all forms. Community members were not only comfortable being filmed and photographed, but had a strong desire for their views to feed into policy and practice. This candidness was delicately managed to protect participants from exploitation and exposure to further stigma and marginalisation. To minimise potential risks, the research team worked with ethics and advisory groups to frame and communicate work appropriately. Ethical approval was granted by the School of Health in Social Science Research Ethics Committee at the University of Edinburgh.

Findings
Phase One: assets for knowing

Interpersonal relationships and trust
There is no such thing as a 'Roma' community. Far from homogenous, this group is comprised of several sub-cultures; each with identifying features. In the studied geographical community, two prominent 'Roma communities' surfaced: Romanian and Slovakian. Interpersonal relationships and trust emerged as essential assets for both. Community organisation representatives noted how 'self-appointed folk' or self-determined 'local champions come forward' as brokers of trust between organisations and communities (interview2). If organisational representatives are trusted by these autonomous individuals, then 'word spreads' without any formal communication: 'no fliers, no emails, just word of mouth' as 'there is a tradition of gathering on street corners for sharing information and just that's what you do'. However, 'if you don't work with [informal community leaders]. . . it'll be difficult to get any engagement' (interview2). Some Slovakians explained that faith in friends and family means they go to them for support and information but 'no one else', while others include doctors in this circle of trust. There are those who do not trust anyone, 'only God'a view echoed by several Romanians who 'believe and trust just God, not people' (ethnographic notes).

Spirituality and happiness
The asset of trust for health and well-being maximisation is inextricably linked to spirituality for these communities. As one Romanian put it: 'God knows what will happen in the future. . . you [health workers] cannot know if everything is decided by God'. Another added: 'only God can change our lives'. Alongside piety, Romanian community members stressed the importance of happiness: 'when you're laughing, life is passing faster. . . with music, dancing, laughter' (ethnographic notes).

Creative community engagement
Creativity came up frequently including singing, music, food, parties, arts (ethnographic notes), 'media training', and 'football' as a 'key driver' for giving 'BME groups a platform to become healthier' (interview9). One community organisation representative added: we've talked about things around food and music which bring people together extensively to talk about. . . something that is going on in the community or perhaps just to find out what people think are the gaps in the community themselves and how they think we should design things for the community. (interview2) Phase Two: assets for doing Attention turned to how organisations could work with communities to mobilise this knowledge and develop asset-based programmes to address specific issues. As Romanian and Slovakians asked for creative engagement, a theatre director facilitated Phase Two where community members enacted Phase One findings to explore what could 'activate' identified assets.

Rebuilding trust
The need for practitioners to rebuild trust with 'Roma' communities who 'may not want to engage due to past [negative] experiences' with the health service (questionnaire1), emerged as a strong asset of doing. Community members acted out past undesirable involvements with the health servicenot knowing how to book doctor appointments then being turned away; not having or knowing how to get national insurance numbers; translators sent to appointments using the wrong language (for example, Romanian rather than Slovakian); lack of awareness of available services (such as stop-smoking services and screenings); and the belief that migrants are ignored, mistrusted or stigmatised.
Community members created images in groups portraying mistrust between them and health practitioners. To explore how solutions could be activated, they were granted three wishes to improve their images and show how situations could be 'better'. Changes included practitioners 'appearing interested' in community members' lives; open body language; offering cups of tea to make them feel welcome; water in the waiting room; smiling; and providing information leaflets in relevant languages (video, journal).

Understanding cultural differences and empathising
Practitioners reflected on how they could improve services 'by having greater understanding of cultural differences' (questionnaire2). 'Family friendly services' (questionnaire3); 'working together' and 'listening' (questionnaire4) to communities' experiences were identified as ways to support community members by 'being on the same level' and simply introducing themselves and the service (question-naire5). This was linked to empathising with communities. One practitioner said there was a clear need to 'understand better how [Roma] families are feeling when they don't understand the language as this makes them more vulnerable' (questionnaire5). Another added there is a need for practitioners 'to continue to be seen in these areas and build up trust' (questionnaire6).

Partnership working through creative community engagement
Partnership working across services was emphasised by public sector workers: 'it would be fantastic to have representatives from primary care and benefit agencies in a similar event, as it seems a lot of issues are linked to difficulties families face on a daily basis' (questionnaire7). Community members and professionals reflected on how creative engagement using theatre could be used as a methodological tool for involving communities to explore health inequalities. This creates opportunities to use asset-based approaches to 'look at the needs of diverse communities but also consider and respect their strengths as well', and track approaches for evaluation (questionnaire7).

Phase Three: evaluation and indicators
Attention then turned to how Phases One and Two could inform development of a co-produced methodological evaluation framework to measure impacts of asset-based approaches on health and inequalities applied for the first time with the 'Roma' community. The Asset Model cites several evaluation approaches including Pawson and Tilley's (1997) realistic evaluation to capture: linkages between the context (the necessary conditions for an intervention to trigger mechanisms), mechanisms (what is it about a particular intervention that leads to a particular outcome in a given context) and outcomes (the practical effects produced by causal mechanisms being triggered in a given context). (Morgan & Ziglio, 2007, p. 21) Drawing from evaluation approaches using methods determined by the purpose, context, setting, suitability of measurement tools, and community participant inputs (Koelen, Vaadrager, & Colomer, 2001), the Asset-Based Indicator Framework consists of five stages, namely context, process, indicators, evaluation, and policy (supplementary material Figure 3).

Context
Engagements must be context specific and may be expressed differently depending on circumstances (action-research1). Those applying the co-produced framework should allow participants to elaborate on their own understanding of indicators in specific contexts: how are they different in personal, community or professional settings? (action-research2).
Evaluation is linked to explanations of what community members view as important to their own health. The extent to which practitioners can understand community members' realities and capture conceptual and actual changes in their lived experiences is connected to the ability to evidence meaningful change for specific communities (action-research1, 2, 3). This process starts by identifying who the community is. The diversity of communities needs to be acknowledged, and communities should self-identify (action-research2). In this application, community members were identified by practitioners as a homogenous group ('Roma') but identified as Romanian or Slovakian in Phase One. As trust developed, they identified as Czech Roma, Romania Roma, Slovakian Roma, Polish Roma, and/or Gypsies. When asked if they felt comfortable defining themselves as a 'Roma community' for evaluations, none agreed, suggesting this label was associated with racism. One community member explained how he hid his identity from colleagues, who openly insult the 'Roma'. Another said calling the group 'Roma' was disrespectful as it illustrated little understanding of the prejudice of having to live with that label. Community members said they would not tick the 'Gypsy' box in the national census due to stigma associated with the categorisation, and being called 'Roma' is misleading as Roma is not a country (Polish Roma would say they are Polish when asked) (journal, video).
After discussions, musical sessions and lunch, the group finally agreed to be defined as 'Gypsy' as they are proud of their heritage and want to come out of the shadows of their hidden identity. It was agreed that racism has a negative impact on their individual and community well-being, so measures of success or improved health outcomes need to be linked to structural changes in equality. This highlights the importance of the first step of the process for framework application (answering the question 'who?')to acknowledge and reflect on how each participant selfidentifies and how this relates to the group; and to capture how this identification may influence evaluation and link to policies and outcomes determined and defined by the community.
The second step is to identify 'why?' community engagement is needed, and 'why now?' Services should be measuring impact in terms of changes in communities' experiences of wellbeing rather than driven by organisational targets and pre-identified outcomes (action-research2). Through a process of 'genuine' co-productionopen conversations without a predetermined agendacommunity assets can be identified along with ways to capture how changes unfold (or do not unfold) while asset-based approaches are implemented (action-research1).
In Phase One of this application, practitioners perceived the 'Roma' as a 'hard-to-reach' community with poor health outcomes. A key reason for engagement was understanding why community members were not accessing health improvement services and to get them to use these services. Reasons for engagement shifted in Phase Two when practitioners became aware of underlying causes for not accessing services including lack of awareness, mistrust, stigma and culturally inappropriate interventions. Practitioners recognised the need to engage simply to connect and learn more about the community's circumstances, needs, and wants. This openness extended to Phase Three where community members expressed what they hoped to gain from engagement with practitioners: a co-produced service ('The Void') where they could be supported as they tackled addictive behaviours, stress and transference of maladaptive coping mechanisms. They spoke about how they turn to unhealthy foods when trying to quit smoking or drink less alcohol, and suggested that 'The Void'a social space where community members can meet with interested and trustworthy practitioners, friends and familycould be helpful in improving health outcomes. Gypsies said food and music would be integral to 'The Void' (journal, video).
Professionals also highlighted the need to establish 'what?' the most appropriate method of engagement for this community is and 'how?' best to capture evidence from engagements. Creative methods were highlighted as preferable, particularly when working with marginalised groups. For example, using entertaining and informative videos to capture and share communities' stories and experiences of housing, education, and health (action-research2). They reflected on how mediums of engagement might be different depending on a community's preferencesgardening instead of singing; writing rather than cooking. Creative approaches for engagement and evaluation were described as 'practical and fun way[s] of linking and staying connected' (questionnaire7) by professionals and community members, who saw this as a way to overcome language barriers and build trust in safe environments that encouraged sharing. Views were not deemed to be simple insights, but crucial forms of evidence that need to be systematically recorded in appropriate ways as engagements progress (action-research1, 2, 3). Conversations with community members could be captured in reflective diaries (written, film or audio) for deeper reflections about the impact of engagements (action-research3).
The importance of asking communities to reflect on the change process at the beginning, during and end of engagement (if there is one) was emphasised irrespective of whether organisations use scale measures, 'soft scales' or other mechanisms for capturing data: 'Where do you think you are now? How did the change happen?' (interview). The limiting and misleading potential of numerical scales, however, was noted in all action-research sessions. Participants reflected on the subjective nature of well-being scales. Happiness means different things to different people at different times. Similarly, a 'one' rating on a scale of one to ten is meaningless without insight into a community member's lived experience (action-research1, 2, 3). The personal-outcomes practitioner gave an example of survey aimed at evidencing the efficacy of a care service. It asked the participant to respond on a scale: . . . since you've had this service, would you say 'I feel safer'? 'I have more social contact'?'. The response -'oh yes, well obviously a lot more'could suggest the person was happier and no longer socially isolated. However, she has carers coming in twice a day and that's not the kind of social contact she wants. (Interview) It was further proposed that the framework should be applied at baseline and systematically applied during and after project implementation to evidence potential change and impact (interview). By reflecting and deciding with communities how best to capture data at the start of engagements, practitioners are better positioned to record data continuously and rigorously throughout the process. In Phase Three of this application, Gypsies laughed when asked if they would complete questionnaires and surveys to evidence changes in health. However, they were willing for their views to be filmed, photographed and captured on flipcharts during 'storytelling' sessions where they gathered to share food and make music (journal, video).

Indicators
Next comes the identification of asset-based indicators linked to improvement in communities' environments. Indicators were seen as a starting point for framework application by practitionersthey should be adapted at baseline depending on what assets are important to a particular community (interview). Indicators were also understood and defined differently by different stakeholdersthere were even nuances in indicator definitions among practitioners from the same organisation (action-research1, 2, 3). Negotiation of the importance of indicators was also different across various community engagements. This highlights the need for those applying the framework to allow communities to elaborate on their own understanding of indicators in specific contexts (personal, community, professional or otherwise).
Furthermore, participants said 'community members might not naturally consider all indicators' and highlighted flexibility in the use of the framework (for example, excluding existing indicators or including new ones) (action-research1). Negotiations between professionals and communities on what assets or indicators are important to them is crucial for evaluation (action-research3), as is recording changes in indicators communities identify as important to them but services struggle to provide (interview).
In the framework application with Gypsies, community members selected and defined the following indicators as important: culture as 'religion, respect, God's will, rules (background), traditions, learning from each other, food, customs, music, singing, dancing'; happiness as 'sports, music, dance, parties, support, friends, helping people, holidays achieving your goals, feeling good about your children'; spirituality as 'inner peace, compassion, community spirit, belonging/not belonging, free feeling, connection, to be accepted'; health as 'healthy food, exercise'; and access to resources as 'finances sufficiency; helping; respect; education and money'. They included an additional indicator: satisfaction defined as 'all achievements' (journal, video).

Outcomes
Once communities have defined what assets and attributes are important to them, attention turns to how these would inform meaningful outcomes and provide measures of 'successful' health interventions for a particular community. Three types of outcomes where identified: quality of life outcomes (features of a community member's whole life that they are working to achieve or sustain), process outcomes (the way in which service and change happens) and change outcomes (traditional outcomes such as reduction of symptoms) (interview). Practitioners should look at 'patterns in changes of behaviour', and have conversations about 'what is going on [for communities]' consistently throughout the engagement (interview).
Services should be measuring impact in terms of change and well-being rather than key targets and pre-determined outcomes (action-research2): 'organisations should not prioritise organisational outcomes over outcomes that individuals identify as important' (interview). Further considerations in the evaluation of outcomes include building relationships with communities; looking at ways of developing trust; and ensuring conversations with communities are considered as data (action-research2).

Policy
In the framework application with Gypsies, community members linked their health outcomes to policy through the Community Empowerment Act. This Act explicitly references inequalities, and creates opportunities for discussions about justice for communities. Gypsies said exclusion and racism led to poor health (stress and depression), and suggested that key outcomes to feel more equitable and healthy included being heard in public forums and having more political representation. Having just one Gypsy representative in local government would be a marker of success. An increased sense of belonging, feeling free, connected to and accepted by other communities and majority ethnic groupswhich Gypsies linked to spiritualitywere also identified as health outcomes linked to equality (journal, video).
Professionals emphasised joint working as crucial to helping practitioners and policy-makers understand what issues are important to communities (action-research1). Although single community engagements might have effect on a local level, they will have stronger impact on reducing inequalities if outcomes feed into policy. Thus systematic engagement captured through the framework allows for outcomes to emerge and be determined by communities and inform local, national and international policies (action-research1, 2, 3).

Discussion
This research has shown that through co-production and asset-based approaches, it is possible to work with marginalised groups to identify attributes, outcomes and actions that maximise health and well-being. It is also feasible for communities and practitioners to co-produce participant-led evaluation frameworks to measure impacts of asset-based approaches on health and inequalities, but this comes with several caveats.
The framework calls for a radical re-conceptualisation of what constitutes evidence in the design, delivery and evaluation of public health interventions. It builds on the Asset Model, which attests that a lack of evidence may not be the problem 'but the ways in which we conceptualise issues and where we look to find the evidence may be limiting effective action on health inequities' (Morgan & Ziglio, 2007, p. 12). We need to generate new 'kinds of evidence' that 'move beyond traditional approaches' to address health inequalities  through creative engagementinnovative approaches determined by communities which may include the arts, media, sport, food or other mediums (de Andrade, 2016).
As communities decide on means of engagement and data collection and what constitutes evidence for them, we stumble into our next dilemma. Preferred methodologies may differ throughout an evaluation and have conflicting epistemological underpinnings that need to be reconciled, as apparent in our research, where abductive analysis generated 'creative and novel theoretical insights through a dialectic of cultivated theoretical sensitivity and methodological heuristics' (Timmermans & Tavory, 2012, p. 180). Ethnography documents cultures and practices so we can view the world from communities' perspectives. Theatre of the Oppressed, however, goes beyond observation. All parties represented should apply their perspectives to debating ways forward to challenge injustice and promote social change (Schutzman & Cohen-Cruz, 1994). It provides a mechanism to facilitate the transfer of experience and knowledge between stakeholders (performers and audience) and creates a dialogue of knowledges (Freire, 1970). This exchange, in itself, is a radical re-conceptualisation of evidence that cannot be excluded from the evidence-base if applying asset-based and co-production principles. Community-based participatory action-research methods aim to give 'voice' to the marginalised; facilitate change in participants' situations; work with people to find tangible solutions to difficulties recognised by them; and raise critical awareness and analysis of participants' place in society (Dover & Lawrence, 2010;McIntyre, 2007). As such, they are compatible with the theoretical foundations of asset-based approaches.
Processes and their outcomes, however, cannot easily be measured in traditional quantitative ways (Barrie & Miller, 2015;de Andrade, 2018). Our research shows even validated psycho-social measures recommended as appropriate measurement tools for asset-based evaluation by national health bodies (Sigerson & Gruer, 2011), may misrepresent participants' views and provide inaccurate insights when contextualised.
Realistic evaluation is cited as a 'helpful' way 'to overcome the evaluation gap' (Morgan, 2012, p. 21) as it is method neutral, uses accessible data and values multiple methods aligned with the realist hypothesis being tested (Marchal, Belle, Olmen, Hoerée, & Kegels, 2012). When realistic evaluation has been used in health systems research, however, there has been considerable diversity in researchers' use of terminology and application of philosophical constructs. There are also intrinsic methodological challenges in ascribing changes in outcomes to complex interventions (Marchal et al., 2012), not least time and resource constraints (Redfern, Christian, & Norman, 2003;Rycroft-Malone, Fontenla, Bick, & Seers, 2010).
This leads to a further critique of asset-based evaluations, namely a lack of capacity and resources for sustained engagement. This research has been ongoing for five years pieced together by small grants and the willingness of communities and organisations to engage on topics they feel passionate about. With few exceptions, it has been possible to keep the same participants engaged throughout the three phases. However, this involved contact and presence in communities even when funding ended, and allocation of small financial incentives to disadvantaged community members when grants were renewed.
The latter raises ethical considerations including engagement with marginalised groups to coproduce services that may not be delivered due to a lack of resources and capacity in public service organisations. In our research, the Gypsy community could put in a participation request for their service (The Void) to be supported through the Community Empowerment Act. However, this assumes participants would have time, capacity and meet criteria for a community participation body. Language and literacy are additional barriers.

Conclusion
This paper problematised evidence, evaluation and measurement when using asset-based approaches to improve health and address inequalities. Working with BME communities (especially Gypsies) and practitioners, it developed and critiqued a co-produced framework to define, implement and evaluate asset-based approaches for communities' own benefits, exploring why researchers have been 'fretting about how difficult it is' to assess these approaches (Hills, Carroll, & Desjardins, 2010, p. 97). We contribute to a paucity of literature about policies to reduce health inequalities experienced, in particular, by Roma/Gypsy communities. There are several implications for international practice, research and policy.
Policymaking for public health interventions is still largely informed by biomedical studies at the individual level that neglect to address health inequities embedded in social environments (The Health Foundation, 2018). Upstream interventions like those applying asset-based approaches, however, raise significant challenges. Evidence of communities' lived experiences may not come from traditional sources. In some instances, creative engagement may be the only way to gather data from marginalised groups so theatrical images or co-produced songs become valid forms of evidence. Much like evidence, validity needs to be radically re-conceptualised if the aim is 'to emancipate those involved in the research process by empowering them to take control of their own lives and challenge the status quo' (Sparkes, 2001, p. 543). Validity is 'more personal and interpersonal rather than methodological' (Reason, 1981, p. 244). Readers or the audience provide validation. We compared participants' lives to ours; looked for similarities and differences.
A further critique of asset-based approaches is that local changes cannot be scaled up to evidence improvements in population health and reductions in inequalities over time (Smith, 2008). This reflects a 'narrow conceptualisation' of community engagement as 'bounded' and 'standardised' with success measured by short-term outcomes and specific individual behaviour changes (South & Phillips, 2014, p. 693). Epistemological underpinnings of asset-based approaches necessitate a dynamic long-term inquiry and re-conceptualisation of evidence in whatever creative or relational form is deemed appropriate by community members, who assume equal and active roles in the design and delivery of services. A focus on complex systems and the 'non-linear, reciprocal relationship between community engagement processes and the determinants of health' (South & Phillips, 2014, p. 694) creates opportunity for communities to identify appropriate processes and outcomes thereby 'reshaping the system in favourable ways' (Rutter et al., 2017). These actions may not result in instant change so monitoring shifts in quality of life, process and change outcomes in the short, medium and long-term is important, otherwise engagements may be inaccurately deemed ineffective.
At a time of scarce public resources, a key question is whether asset-based approaches truly give communities voice or turn their gaze away from structural causes of inequality making communities responsible for issues that should be dealt with by policy-makers. Certain groups in society are privileged over others and have more power, resources and opportunities (Kincheloe & McLaren, 1994). By disclosing interests of marginalised groups, as asset-based approaches seek to do, there is the potential to disrupt the silence; bring frustrations and powerlessness out of the shadows so non-privileged groups may realise their potential, challenge the status quo, become more autonomous and ultimately liberated (McLaren, 2003).
While the Asset-Based Indicator Framework has been used by disadvantaged communities as a subversive tool to challenge structural causes of inequalities through creative media (de Andrade, 2018), much more action and political will is needed at the structural level to make a difference to the lives of participants in this study. Even when you know 'what is to be measured, by what means and by whom' (South & Phillips, 2014, p. 22)when communities themselves decide on appropriate measures and methodologies as illustrated in the frameworkprogress is limited without sufficient capacity and resource. As this research has shown, it is possible to understand the granularity of the interrelation of assets at individual, community and structural levels (Morgan & Ziglio, 2007); identify mechanisms through which change happens; and apply participatory and empowering methods to capture actions on assets leading to community defined outcomes (Foot, 2012). However, this may not lead to meaningful changes for communities without systemic change.

Disclosure statement
Funded by an ESRC Impact Accelerator Grant with NHS Greater Glasgow & Clyde (NHSGGC) & Romanet Multi-Agency Working Group (Glasgow). NHSGCC funded the original research. Funders had no role in the collection, analysis and interpretation of data; writing or decision to publish.