As food and diet are implicated in the disproportionate burden that South Asian communities living in the UK bear with regard to chronic illness and type 2 diabetes in particular, we wanted to explore the intersections of faith, culture, health and food in these communities in order to generate evidence for addressing cultural difference in the design and provision of health services. Given that we were interested in the voices of South Asian women many of whom are largely invisible in public life on account of their ethnicity, gender and family structures and who are rarely represented in research, it was important to select an approach to data collection which was culturally in tune with women's lives and sensitive to their values and beliefs. The aim was to facilitate the kind of talk through which lay knowledge and beliefs are constructed and communicated, avoiding responses which participants believe are the ones researchers want to hear and which reflect public health messages rather than the 'lay theories' [58
] that are embedded in daily routines and cooking practices. The attention to cultural sensitivity, the notion that meaningful research must be founded on trust and commitment and on the ongoing relationship between researchers and communities are features which are shared with a number of approaches to research with seldom heard communities [18
]. However, the programme of research under the auspices of which this project was undertaken did not envision the type of community involvement in data collection, analysis and the generation of findings characteristic of Participatory Action Research [59
] or Community Based Participatory Research [60
]. Instead the focus of this study was the production of data rich in detail of social and cultural factors that may not have been generated through semi-structured interviews.
Reflections on participant recruitment and ethics
The Rangoli activity which provided the focus for the data collection was led by a visual artist who had previous experience of working within the health field. She has a South Asian background and uses her cultural and religious roots in her art. The lead researcher is a female health professional of white European descent and works in the discipline of medical sociology. Both were assisted by three female bicultural researchers of South Asian origin, one with a background in biomedical science and two in sociology. All were involved in the recruitment of participants which was carried out through a combination of personal contacts and word-of-mouth, desktop research to identify third sector organisations, and what we called 'shoe leather' research. This involved travelling in the local area and 'walking the patch', and consisted of meeting potential gatekeepers to appropriate women's groups at local events and open days, attending public events, speaking to people about our work informally, meeting community workers in the field, and generally raising our profile as researchers in the community. Such efforts were time consuming and outcomes were not easily defined in advance. They also did not always yield direct results in terms of recruitment. We found it particularly difficult to access Pakistani Muslim women. A cultural insider explained that this might be due to families' unwillingness to discuss what are deemed to be private family matters, namely the preparation and eating of food and the social and religious practices surrounding it, with strangers and for a public purpose, i.e. research. However, our outreach activities were very helpful in building relationships and in demonstrating our personal as well as professional commitment to ensuring that the research led to benefits for women in those communities. This phase of the project lasted around three months, but coincided with the first data collection sessions. We were able to invite a project lead, who was considering hosting a data collection session, to one that had already been organised to enable her to decide whether her group would be prepared to participate. In terms of success in recruitment, personal, family and community contacts by the members of the research team with a South Asian background were the most fruitful, followed by contacts with local faith organisations. We found that many community organisations that served the needs of women were organised around faith groups which in the Birmingham area are Sikh, Muslim and Hindu, and that group leads were receptive to activities promoting women's health and wellbeing. We were able to build trust with the group leads through our shared concerns about health and well-being in their communities.
Building relationships with gatekeepers and organisational or groups leads was essential in legitimising our work to women who were interested in participating in our project. When we had a positive response from a group lead, we arranged a meeting at the group's premises or community venue which was attended by the artist, the lead researcher and by one or two of the researchers. These meetings were crucial in terms of, first, building trust and confidence in the research team and the artist and, second, organising the data collection event and addressing the practicalities for potential participants. We stressed repeatedly that the event was to be cost-neutral to the group or organisation as there was anxiety over current and future funding. We also underlined that costs for room hire, materials, refreshments, travel and child care expenses would be met by the research team. The timing of the event needed to be carefully worked out with each group in order to reduce the burden of participation in the project and to ensure that as many women as possible were able to attend.
We avoided formal written consent procedures because of their potential to exclude people from participation in research. Our rationale for the decision not to use formal procedures was that we were planning to recruit a seldom heard group that is marginalised at many levels: they are part of a migrant community that tends towards collectivist rather than individualistic values [61
]; as women in their cultural group they are subject to patriarchal social control; the older generation of women tend to have fewer written language skills and also do not necessarily subscribe to the primacy of written consent [62
]. Furthermore, formal consent procedures requiring a signature would have obliged some women to ask their husband or male relative to sign on their behalf, a requirement which could have discouraged women from taking part. Participation may also have been refused by a husband or male relative. By keeping the project informal and part of the usual activities within their community group, we anticipated that women would be encouraged to take part. In running six data collection session over 12 months, we recruited over 120 women. Not everyone who participated in the Rangoli activity was able to contribute significantly to the data, but the majority added at least some individual or group data.
We had been concerned that some women might be uncomfortable about audio-recordings and either decline to participate or refuse to be audio-recorded which would have required the removal of the recorder from their table. We were prepared for one member of the research team to take notes instead. However, none of the women refused audio-recording either in the whole group, or when we asked them individually. 'You give us Rangoli, we give you talk' was how one of the participants expressed her satisfaction with the reciprocal arrangement we had made. We also wanted to take some photographs of the session as it progressed, focusing on the women's hands as they handled the materials and built up the Rangoli patters. Some of the women requested not to have their faces photographed, but were happy for their hands to be included. Consent has been given for the publication of the photographs accompanying this article. One of the researchers spoke to each of the women to collect anonymous demographic data including age range; country of origin or parents' country of origin; whether they had been given a diagnosis of type 2 diabetes or pre-diabetes; and whether they had family members diagnosed with the condition, and if so, how they were related and if they lived in the same household. We used several digital recorders: one was placed on each of the tables around which the Rangoli activity took place, and each researcher carried one.
Contrary to our previous experience of running focus groups, we found that many more women turned up to the data collection sessions than had originally signed up. This was the result of last minute invitations to female family members or neighbours who had heard about the session and wanted to come along. Thus the data collection sessions were social occasions for the women, reinforced by the hospitality arranged by the research team so that data collection sessions taking place in the evening were preceded by a light supper whereas those taking place in the morning finished with lunch. Water, juice and hot drinks were available throughout the sessions. The hospitality contributed to a convivial, relaxed atmosphere.
Reflections on data and language
As a result of the interactive nature of the data collection technique, the data we collected were heterogeneous and very different to those gathered through semi-structured interviews. The artist created a relaxed, sometimes noisy and exuberant atmosphere which encouraged women to talk to each other and to the researchers. This presented a challenge to audio-recording as the background noise led to problems with the sound quality which made transcribing the data difficult. Different languages and dialects being spoken also added another layer of complexity to the data. Most of the women spoke to the researchers in English although they would speak to each other in their mother tongue. The first generation women struggled sometimes with their English language skills, but because they found themselves in an all-female, informal environment, they wanted to speak to the white researchers in English. The younger women spoke English to the researchers and well as to each other.
Some direct translation was possible after the sessions on listening to recordings and transcribing the data, when researchers with the necessary language skills were able to translate verbal interactions between the women, or between a participant and researcher. Sometimes daughters interpreted for their mothers during the sessions, and in one group the centre lead and second generation women working for the organisation interpreted researcher questions into the participant's native language and the participant responses back into English. However, we did not check for the accuracy or the conceptual equivalence [63
] of the translations provided by these lay translators. We also did not account for the effects of translators on the data. This is a methodological limitation that needs to be addressed in this kind of study in order to increase the what has been termed '"cross-language trustworthiness" which shows that the researcher systematically accounted for factors that would compromise the credibility [and] transferability ... of translated data [[64
]; p. 285].
Some data were unusable due to the poor sound quality as a result of background noise whereas other data were clearly not related to the research question. The latter included conversations about the technical aspects of the Rangoli activity, such as how much glue to use or what patterns to fill in next, and included private talk about social events or family business which was captured by the microphone, but was clearly not intended for research use.
Apart from the observational data which we drew on to shed light on the social, cultural and community context, all other were digitally recorded. The diversity of data with regard to speakers and interactions taking place, made data transcription, management and analysis challenging and required some decisions about the data in order to organise and reduce them and also make them amenable to systematic analysis.
Reflections on data quality and analysis
We began our thematic analysis as we transcribed and organised the transcripts from the data collection sessions, also including notes and transcripts from the researcher debriefing sessions. This resulted in a complex and heterogonous data set which presented us with analytic challenges and surprises. We had divided the data into two broad types - group/individual interactions and one-to-one interviews. Interactive data were obtained through (1) direct responses to direct questions and prompts from researchers to the group involved in Rangoli activity; (2) interactive talk between women; (3) interactive talk between women and researchers. These data were similar to that of focus group data, but different in two fundamental ways: in the Rangoli setting, women rarely spoke more than a sentence or two in one turn, and interruptions occurred frequently because the activity demanded participants' attention and collaboration. The result of these short interactive data sequences was what has been termed 'small stories' [65
]. 'Small stories' focus on the stories we tell each other in passing, in our everyday encounters with each other; they are 'narratives-in-interaction' [[68
]; p. 235]. They refer to material that is not neatly storied, but fragmented and sometimes even incoherent. They arise in everyday social interactions and reflect particular local practices [65
]. The excerpt below gives an illustration of two interwoven and familiar 'small stories' about type 2 diabetes: one story concerns a destiny out of the individual's control; the other is one of prevention, diet and exercise.
Woman 1: You can't ever cure it, can you? It's true though, isn't it?
Woman 2: What's not curable?
Woman 1: Sugar. Diabetes.
Woman 2: You can prevent it.
Woman 1: You can't prevent it if it runs through your family.
Woman 3: No, you can prevent it.
Woman 1: Not if your parents and your uncles got it and if your aunts have got it, everyone in your family has got it, then how can you prevent it?
Woman 3: You can prevent it. You can diet and exercise.
Woman 1: I know the healthiest people and they've got diabetes. It's their fate, isn't it?
'Big stories', on the other hand, emerged from the more narrative data derived from the 'off-line' conversations between individual participants and researchers which took place away from the main group. The data they yielded resembled conventional individual interview data, but as they were collected within the social context of the Rangoli session, frequent reference was made to what had been said previously in the group situation, or expanded upon what had already been mentioned. This data set included stories about the experiences of parents' migration, the effects of those experiences on their community and on second generation South Asians in the UK. The stories were often constructed with what Morse [[69
]; p. 3] has called 'shadowed data'. These are participants' discussions of not only their own experiences, but also of those of others. In this instance, participants relayed their interpretation of the experiences of their parents and those of their parents' generation, how their own experiences differ from theirs, and for what reasons. The young woman speaking here gives an account of her parents' migration and identifies their changing attitude to food and traditional eating practices:
At the beginning they worked really hard and they just settled down to fit in. So they worked their socks off to do that an, you know, when they started to make it then they realised that actually, we have a house, we have a job, and then they just chilled out. Then it was just like the family would come round and just eat and drink because they realised that they made it, that they were here ... they didn't have to worry about anything now, but they didn't realise the food they were eating then like ten, twenty years down the line was affecting their health. (...) It was a big thing for them to come here and it took 100 per cent out of them to come here and to take like all of the racism and to take all the headaches, you know, and the rubbish jobs they had to do when they came here. But then they made it and I think it was just a matter of finding "oh we can have this food and there are these restaurants and fast food now so you don't have to cook at home; why would you cook at home, why would you want to eat this basic Daal when you can have things like pizza and take-away?"
These stories, both 'big' and 'small', have been helpful in elucidating the themes that have been generated through the constant comparative analysis of data collected, both within and across data collection sites and groups. One of these themes which is most relevant to producing evidence for the design of future health services is that of the relationship between food, social practices and religion, and women's pivotal position in relation to all three. What the 'small stories' in particular were able to contribute to the development of this theme were the connections between food, and social and moral responsibilities: much of women's roles in the preparation of food was not just about technical skill and social duty, but also about their moral worth as women, wives and mothers in nourishing their family with health giving food. Honouring traditions handed down through generations of women was seen as a social as well as a moral obligation. In some instances, these obligations were also linked to religious practice. The women's accounts suggested that they are enmeshed in family and social structures which define who they are and how they prepare food. This indicates that what, why and how food is prepared is a deeply complex social and cultural phenomenon that is unlikely to be amenable to educational interventions which seek to address a biomedical knowledge deficit and aim to promote self-efficacy.
We propose that the use of techniques that are specifically tailored to attend not only to the research questions or objectives, but also to the values and cultural needs of the participants, has the potential to generate finely grained and nuanced data which is not just theoretically interesting or helpful in building descriptive categories. Rather, such culturally responsive methods also have the potential to increase the explanatory power of study results. Indeed, the findings from this project so far suggest reasons why educational interventions in isolation may be ineffective, and why strategies to engage South Asian women in educational sessions have met with limited success [35
]. These reasons are likely to be related to a mismatch between the conceptual underpinnings of educational intervention and the way women construct their identities as social and moral members of a family and community. Such findings have the potential to generate highly practical and applicable evidence for informing the design of culturally sensitive health care services.