ABSTRACT
James Y. Nazroo and George Davey Smith Introduction Differences in health across ethnic groups, in term s of both morbidity and m ortality, have been repeated ly docum ented in both the U nited States (D epartm ent of H ealth and H um an Services 1990, Rogers 1992, Sorlie et al. 1992,Krieger etal. 1993, Rogot etal. 1993, Davey S m ith s al. 1998a, P am u k ^ al. 1998) and Britain (M armot et al. 1984, Rudat 1994, H arding and Maxwell 1997, Nazroo 1997a, b). However, the factors underlying such differences rem ain contested. In particular, the role th a t socio-economic position may play is the subject of considerable debate, with some claiming that it makes m inim al or no con tribu tion to ethnic inequalities in hea lth (Wild and McKeigue 1997); others suggesting that even if it does contribute, the cultural and genetic elem ents of ethnicity m ust also play a role (Smaje 1996); and o thers argu ing th a t e th n ic in eq u a litie s in h ea lth are p redo m inan tly determ ined by socio-economic inequalities (Navarro 1990, Sheldon and Parker 1992). In this chapter we will use evidence on m ortality and morbidity rates from both the U nited States [the M ultiple Risk Factor Intervention Trial (MRFIT) study - Davey Smith et al. 1996a, b, 1998a] and Britain (the Fourth N ational Survey of E thnic M inorities - Nazroo 1997a, b) to explore the contribution of socio-economic factors to ethnic inequalities in health, and to point to the methodological and conceptual pitfalls tha t beset such work.In Britain the possible contribution of socio-economic position to ethnic inequalities in health has, for a long time, been ignored by the majority of investigators. This may have its roots in the now classic study of im m igrant m ortality rates by M arm ot et al. (1984). Published shortly after the Black report (Townsend and Davidson 1982) firmly placed inequalities in health on the research agenda, and at the same time as the Policy Studies In stitu te ’s th ird survey of ethnic m inorities (Brown 1984) showed the poverty of ethnic minority groups living in Britain, M arm ot et al. (1984) used a combination of British census and death certificate data to explore the relationship between country of b irth and m orta lity ra tes. C en tra l to th e ir analysis was an
assessment of the contribution that occupational class made to the differences between country of birth groups. But, given the context just described, they surprisingly found no relationship between social class and m ortality for im m igrant groups. These findings led them to conclude ‘(a) that differences in social class distribution are not the explanation of the overall different m ortality of m igrants; and (b) the relation of social class (as usually defined) to m ortality is different among im m igrant groups from the England and Wales p a tte rn .’ (M armot et al. 1984: 21). It was not until 1997 that socio-economic position reappeared in published national data exploring the relationship between ethnicity and health in Britain (Harding and Maxwell 1997, Nazroo 1997a).One of the places where socio-economic position has reappeared in data on ethnicity and health in B ritain is in the most recent exam ination of im m igrant m ortality rates by H arding and Maxwell (1997). This analysis was conducted in the context of the British inequalities in health decennial supplem ent (Drever and W hitehead 1997) and, again used census and death certificate data to explore m ortality rates by country of birth. In contrast to M arm ot et a V s (1984) findings, H arding and Maxwell (1997) showed clear socio-economic gradients in m ortality rates for m igrant groups. Despite this, like M arm ot et al. (1984) they also found that controlling for occupational class made no contribution to the differences seen between the different country of birth groups. This led H arding and Maxwell (1997) to conclude that ‘Among the non-white ethnic groups, the relationship between social class and m ortality is becoming apparent in the 1990s for groups who have settled here for some time. O ur overall conclusion, however, supports the earlier ones that social class is not an adequate explanation for the patterns of excess m ortality observed’ (Harding and Maxwell 1997: 120). So, although the most recent evidence from im m igrant m ortality studies suggests that there are socio-economic gradients in health for all ethnic groups, they also continue to suggest th a t differences in socio-economic position do not contribute to ethnic inequalities in health in Britain.A lthough w ith in the U n ited S tates there is also a long trad itio n of investigating ethnic inequalities in health (see Trask 1916 for an early exam ple), the relationship between this work and the investigation of socioeconomic inequalities in health is quite different from that in Britain. In recent decades in the U nited States there has been g reater in terest in health differences according to ‘race’, than in socio-economic differentials in health perse. This is partly due to the limited data on socio-economic position available in routine Am erican health statistics. Thus the USDHHS report of 1990, ‘H ealth status of the disadvantaged’, largely presents health data tabulated by ‘race’, ra ther than by socio-economic indicators. Implicit in this approach is the assum ption tha t ethnic m inority groups in the U nited States are disadvantaged and that their poor health status is, to some extent, due to this disadvantage. However, it also suggests that ethnic m inority groups are somehow homogeneous, and therefore socio-economic differentials in health
status within ethnic m inority groups are not considered. Indeed, until recently (Sorlie et al. 1992, Davey Smith et al. 1996b, Pamuk et al. 1998) there have been few exam inations of the socio-economic stratification of health within ethnic m inority groups in the U nited States. The contribution of socioeconomic factors to, in particular, Black-W hite health differentials in the U nited States has, on the other hand, been investigated in many studies (for example Rogers 1992, Sorlie al. 1992, S te rlin g s al. 1993, Rogers al. 1996). These studies have, however, been lim ited by small sample sizes or a reliance on adm inistrative data in which inform ation on factors other than ethnicity and socio-economic position was not available. Methods D ata from two studies will be used here. The first is the Multiple Risk Factor Intervention Trial (MRFIT), which was conducted in the United States, where 361,662 men were screened into this study between 1973 and 1975 at twenty centres in eighteen cities, using a variety of settings, including their homes and workplaces. Full details of the MRFIT study are reported in Neaton et al. (1984, 1987). The m en were assessed on a num ber of medical risk factors, such as cigarette smoking, blood pressure, serum cholesterol and previous medical history. M ortality rates over a 16-year follow-up period will be used here to assess ethnic inequalities in health , with both all cause m ortality and some cause-specific rates being used. For the purposes of the analysis of ethnicity in this paper, a simple divide has been made between W hite and Black men. O ther ethnic groups (such as Hispanic) were not included in the analysis as in total they consisted of only about 13,000 individuals. Overall, data were analysed for 300,685 W hite and 20,224 Black men.No socio-economic data were collected at the time of inclusion into the MRFIT study. So income data from the 1980 U nited States Census have been used to determ ine m edian family income of W hite and Black households in the area of residence of the individual at the tim e of recruitm ent into the study (Davey Smith et al. 1996a, b, 1998a). Area was determ ined using zip code, with 4,644 zip code areas being used for the W hite m en and 1,376 for the Black men.The second study used here is the British Fourth National Survey of Ethnic M inorities (FNS). In 1993 and 1994 a nationally representative sample of 5,196 ethnic minority and 2,867 W hite people was interviewed. The sample was identified using focused enum eration (Brown and Ritchie 1981, Smith and P rio r 1997) and covered those of C arib b ean , In d ian , P ak istan i, Bangladeshi and Chinese origin. Because the Chinese sample was small (214 people), they are not included as a separate group in the data presented here.Three health outcomes from the FNS are considered in this paper. The first was a general health question asking respondents to compare their health with that of others of the same age on a five-point scale. The responses to
this question have been dichotomised into those who said their health was good or very good and those who said it was fair, poor or very poor. The second was an assessment of possible ischaemic heart disease (IHD). This included respondents aged 40 or older who said that they had had a diagnosis of heart disease or angina, or who said that they had experienced severe chest pain lasting for half an hour or more. The third was a report of a diagnosis of diabetes mellitus.One of the great advantages of the FNS was its coverage of a range of topics related to the experiences of ethnic minority people in Britain that may be relevant to their health. These included ethnic identity, education, employment, income, household and family structure, social support, area of residence, and experiences of discrim ination and harassm ent. These are reported in full elsewhere (M odood^a/. 1997), although some of the relevant findings will be reported here. Results
Socio-economic inequalities in the United States and Britain Using income data from the 1980 U nited States Census, the areas in which the respondents to the MRFIT study lived were divided into five groups, based on the m ean household income (in US$) in the area (less than $12,500, $12,500-117,499, $17,500-$22,499, $22,500-$27,499 and $27,500 or more). The findings were very sim ilar to those shown in the British data for the poorest ethnic minority groups compared with W hite people. More than a third of American Black men lived in an area with the lowest income category, compared with less than 1 per cent of W hite men in similar areas. In contrast, 30 per cent of W hite men were in the top income category compared with less than 5 per cent of Black men. Furtherm ore, 94 per cent of W hite men were in the top three of the five bands, while 70 per cent of Black men were in the bottom two of the five bands (see Davey Smith et al. 1998a for further details).D ata published from the British FNS contained a num ber of indicators of socio-economic position, including occupational class, unemployment rates, housing tenure and housing quality, and income (Modood et al. 1997). These indicators generally showed the same pattern across the ethnic groups that were included in the study, with the Indian and Caribbean groups somewhat worse off than the W hite group, and the Pakistani and Bangladeshi groups much worse off than the W hite group. These differences are, perhaps, best summed up by differences in total household income adjusted for household size. Just over a quarter of White people had less than half the average income - an indicator of poverty - compared with about two-fifths of Indian and Caribbean people and more than four-fifths of Pakistani and Bangladeshi people. In Britain the Pakistani and Bangladeshi groups are by far the poorest,
and they are poorer than both W hite pensioners and W hite lone parents (see Tables 5.6, 5.7 and 5.8 in Berthoud 1997). Inequalities in health Figure 4.1 shows all cause m ortality rates, stratified by m ean family income in the area of residence, over the 16-year follow-up period in the United States MRFIT study. There is a very clear gradient for both W hite and Black men that is sim ilar for both groups. The extent of the socio-economic inequality in health is shown by the twofold difference in m ortality rates between those in the top and bottom income groups for both Black and W hite men.Figure 4.2 contains data from the British FNS. It shows rates of reporting ‘fair or poor’ health for three ethnic minority groups (in order to achieve large enough sam ple sizes the Pakistan i and B angladeshi groups are combined), all of the ethnic minority groups combined and the comparative W hite sample. Each of the groups is stratified by occupational class of the head of the household, with a simple distinction drawn between m anual and non-manual households and a third group, those where there was no fulltime worker in the household, also included. As in the American data on all cause mortality, Figure 4.2 shows a clear relationship between reported general health and socio-economic position for each ethnic group in Britain.Figure 4.3 also shows British FNS data, but this time each ethnic group is stratified by tenure, with owner-occupiers and renters separated. Again, for each group a very clear socio-economic gradient is evident.The overall impression given by the one figure on the American data and the two on the British data is of clear socio-economic gradients for each ethnic
group, with poorer people having poorer health and higher m ortality rates. The pattern for both ethnic m inority and W hite groups is very similar. Standardising for socio-economic position As in the analysis of im m igrant m ortality in Britain by H arding and Maxwell (1997), controlling for socio-economic position does not elim inate ethnic differences in hea lth in the B ritish FNS. People in the Pakistan i and
Bangladeshi groups reported worse health than any other group (Nazroo 1997a) and, while Figures 4.2 and 4.3 show a clear socio-economic gradient in health for them , in each socio-economic category they continue to have much worse health than W hite people. For exam ple, Figure 4.2 shows that about a third of non-m anual Pakistani and Bangladeshi people reported their health as ‘fair or poor’, compared with less than a quarter of non-manual W hite people.Although this might suggest that socio-economic factors do not contribute to ethnic inequalities in health, it is im portant to recognise that the process of standardising for socio-economic position when making comparisons across groups, particularly ethnic groups, is not as straightforw ard as it m ight at first sight seem. As Kaufm an et al. (1997, 1998) point out, the process of standardisation is effectively an attem pt to deal with the non-random nature of sam ples used in cross-sectional studies - controlling for all relevant ‘extraneous’ explanatory factors introduces the appearance of random isation. However, attem pting to introduce random isation into cross-sectional studies by adding ‘controls’ has a num ber of problems. These have been sum m arised by K aufm an et al. (1998) as follows, ‘W hen considering socio-economic exposures and m aking comparisons between racial/ethnic groups ... the m ateria l, behavioral, and psychological circum stances of diverse socioeconomic and racial/ethnic groups are distinct on so many dimensions that no realistic adjustm ent can plausibly sim ulate random ization.’ (K aufm an^ al. 1998: 147).Indeed, an analysis of ethnic differences in income within class groups in the FNS emphasises this point. Table 5.2 in Nazroo (1997a) showed that while total household income adjusted for household size followed the class gradient for each ethnic group, w ithin each class group ethnic minority people had a sm aller income than W hite people. The differences were particularly large for the Pakistani and Bangladeshi group, who, as was shown earlier, are the poorest group and also the group with the poorest health. W ithin each class group they had, at most, half the income of the W hite group and those in social classes I and II had a lower average income than W hite people in social classes IV and V Nazroo (1997a) also showed that a sim ilar pa tte rn exists for other indicators of socio-economic position. For exam ple, unemployed W hite people had been unemployed for a shorter period than their ethnic m inority equivalents, and within tenure groups W hite people had be tte r housing than some ethnic minority groups.Similarly, census data in the U nited States show that while m edian family income increases with increasing num ber of years of education for both W hite and Black people, w ithin each education group Black men and women have a substantially lower income than their W hite equivalents. Indeed, Black men and women had a household income equivalent to that of W hite men and women in the education group below them (Pamuk et al. 1998).One way of addressing this problem is to use alternative indicators of socioeconomic position that more accurately reflect ethnic differences to adjust
for socio-economic differences when m aking com parisons across ethnic groups. However, most studies do not contain sufficient m aterial on socioeconomic circumstances to do this. Fortunately, as described earlier (p. 43), we were able to add income data to the MRFIT study and the FNS contained considerable detail on socio-economic position. The rich m aterial in the FNS was used to construct an index of ‘standard of living’. Full details of this index can be found in Nazroo (1997a), but it was based on the following items: • num ber of people per room;• absence of, or shared, basic am enities in household, including bath or shower; bathroom; kitchen; inside toilet; hot w ater from a tap; central heating;• possession of consum er goods, including TV; video; fridge; freezer; washing m achine; tum ble drier; dishwasher; microwave; CD player; PC; phone;• num ber of cars.
