The survey of white (n = 227), Indian and Pakistani (n = 233) and African Caribbean (n = 213) adults aged 18–59 years living in Leeds, UK was part of a larger study which aimed to examine the relationship between ethnicity, health and SEP carried out as part of the Economic and Social Research Council Health Variations Program. Details of the study sample have been reported previously. Electoral wards were divided into three groups by Townsend score, which combines measures of unemployment, car ownership, home ownership and household crowding. Seven high, medium and low deprivation wards were selected. General practice lists were seen to provide the most appropriate, reliable and up to date sampling frame. In addition, such lists contained the information needed to stratify the sample by age-band and gender. A total of 30 practices were selected; of these 20 agreed to give us access to their practice lists for sampling purposes. The 10 practices that declined to participate were not markedly different than those that did. The practice response rate did not vary between wards.
There is no statutory requirement for primary care to collect data on ethnicity. In the absence of ethnic monitoring, different approaches needed to be taken to identify people to be included in the white sample and the Indian and Pakistani sample, compared to the African-Caribbean sample. In the current study, names were used to allocate individuals to the White and the Indian and Pakistani groups from which random samples were drawn. Patients with South Asian names were identified with the aid of a software package (Nam Pehchan sensitivity 88–96% and Positive predictive validity of 59–67% against names from Yorkshire),  with additional manual confirmation by team members familiar with South Asian names. In the current study, patients were allocated to the African-Caribbean sample based on recall by practice staff.
Following difficulties in reaching the required numbers, additional interviews were undertaken using quotas to ensure that the sample had an appropriate level of representation of each group. This involved targeting potential interviewees living around a number of identified sampling points through random walks as well as snowball sampling in the electoral wards. Response rates for the pre-selected sample were 27.1% and 41.3% for the quota sample. The overall response rate was 33%. The proportion of people selected by each recruitment method varied between ethnic group (Primary sample: white = 42.5%, Indian Pakistani = 57.1%, African Caribbean = 50.5%). There was no interaction between ethnicity and sample strategy on any socioeconomic variable except home ownership. Among white people the primary sample were slightly more likely than people in the quota sample to own their own homes but the opposite was true for minority ethnic groups.
The ethnicity designation of participants was confirmed using self-identified categorisations that were then mapped into the 2001 census categories. The results are presented here for White, African-Caribbean groups, and, due to small numbers, for Indians and Pakistanis as one group. People of mixed origin were generally assigned to minority rather than white groups. Of the 247 White participants, 241 identified as British, 2 as Irish, 1 as French and 3 as other White. Of the 232 Indian and Pakistani participants, 92 identified themselves as Indian, 129 as Pakistani, 3 as African, 2 as any other Asian background, 1 as Any other mixed background; 1 as British, 1 as British Sikh 1 as East African Asian and 2 as White and Asian. Of the 212 African-Caribbeans, 167 identified themselves as Caribbean, 21 identified themselves as African, 3 as Black British, 7 as British Caribbean and 6 as of other black backgrounds. The remainder identified themselves as being of mixed origin, including 8 as White and Black Caribbean. Each person was only allowed to be in one category. People identifying in two categories where one was a minority were allocated to the minority group. Small sample size meant that there were only sufficient numbers for three groups; White, African Carribbean and Indian and Pakistani. This may limit the generalisability of the results and potentially subgroup differences within ethnic groups.
Instruments and Procedure
The instrument used in the community survey was developed following preliminary analysis of the qualitative data that had been collected in the previous phase of the research and a detailed review of existing published and unpublished survey instruments used to assess socioeconomic position. The questionnaire included sections on ethnicity, socioeconomic position, social resources, discrimination and health. Demographic and health questions were informed by the Census and Health Survey for England. Household and individual income questions were taken from Breadline Britain. An Urdu version of the questionnaire was available where required.
The survey was conducted by a commercial market research company using face to face interviews to complete each questionnaire. Interviewees and interviewers were matched on the basis of language and gender. Most interviews took place in people's homes.
All analyses were conducted in Intercooled Stata version 10.0. Binomial logistic and linear regressions were used to examine ethnic differences in the response rate to income and education (secondary or below, post-Secondary (non-university) and university) questions by home ownership, car ownership, savings of more than £1000 and can not afford household goods. All SEP measures were coded dichotomously.
Binomial logistic and linear regressions were used to examine ethnic differences in car ownership, home ownership, poor quality housing, worry about losing home, can not afford household goods, investments, no debts, able to get £10 000, owe money to family/friends, lent money to family/friends, welfare and employment/self-employment. All variables were coded dichotomously. People who reported that their house was too damp, uncomfortably cold in winter or too small were classified as having poor housing quality. The measure for can not afford household goods was based on whether the respondent said their household lacked the following goods because they could not afford it; telephone, washing machine, freezer/fridge, dishwasher, mobile phone, cable/satellite television, video recorder, central heating, tumble drier/washer, burglar alarm, compact disc player or home computer. Investments included personal pension, PEP/TESSA/stocks/shares, savings, jewellery, property other than home, community savings scheme and other. Debts included credit card, catalogue, hire purchase, bank loan, family/friend loan, other loan, overdraft and mortgage/rent arrears. The welfare measure included family benefit, income support, job seekers allowance, housing benefit.
Further analysis was conducted to assess the effect of ethnic difference on these SEP measures taking education into account. Rates and means adjusted for age and sex were calculated using adjprop and adjmean procedures. Age was coded into three categories (18–29, 30–44, 45–59 years). Age was further reduced to two categories for some variables, because there were no cases among younger people. Age (and, where appropriate gender) adjusted rates as well as odd ratios are reported, to clarify differences between groups. The white group was used as the reference group for the analysis of ethnic group differences to enable comparison with other studies. Thus p values represent differences between the minority ethnic groups and the white group. Reported differences were significant at the 0.05 level.