Choice of indicator is important

EDITOR—The paper by Sacker et al,1 and the associated editorial by Vågerö,2 discuss differences in health inequality between men and women measured with two alternative schemes. We have been investigating a similar problem—the indicators that are most sensitive for measuring health inequalities in an older population.

Sacker et al included only those people in paid work in 1981. The 1981 census shows that 20% of men and 49% of women in the age groups they consider were not working, so this restriction will have resulted in many people in the sample being excluded. Such exclusions are known to result in bias.3 In the older population the use of classifications based on current occupation is problematic. Other common indicators of socioeconomic status, such as income and education, also present difficulties. This is because of the strong association between income and employment status and because most of today's older population (particularly women) left school at the minimum age with no academic qualifications.

The data we use come from a nationally representative sample of 55-69 year olds first surveyed in 1988.4 The dataset includes lifetime occupational histories, which have allowed us to compute a social class measure based on usual occupation and exclude only 4% of men and 9% of women. Information on qualifications (including those obtained through on the job training), income, and two indices of relative deprivation is also available.

The table shows the proportions of men and women reporting “not good” or “fair” (rather than good) health by these indicators. Differentials among men and women using the measure of social class were broadly similar, as were differentials by quarter of income and education. In both sexes the Townsend score produced the greatest differentiation; over three fifths of the sample, however, fell into the most advantaged category. It might be valuable if the score was used in conjunction with an indicator such as education, which, in older groups, has the opposite problem-that is, lack of differentiation at the bottom of the distribution.

Choice of an indicator will depend on theoretical models of relations between socioeconomic factors and particular health outcomes as well as on apparent sensitivity. The principle of inclusiveness is also extremely important and can be met in some circumstances by the collection of retrospective data.

Authors' reply

EDITOR—We are pleased to see that other researchers are also measuring different dimensions of inequality. It is important to know about their varying effects on health and whether effects are different at different stages of life.

In our paper we asked why it might be that health inequality seems greater in men than women in some studies, though not in others. One reason that has been put forward is that studies use different measures of inequality. This was particularly interesting to us, as the wider programme of work from which our paper emerged investigated the possibility that different forms of inequality could influence health through different pathways. These suspicions were borne out, and the results are published elsewhere. 1 2

Our aim was limited to a single hypothesis. We did not aim to locate the most inclusive measure of social inequality, although we agree that such an aim is indeed an important one in other studies and for other research questions. We suspected that social class based on occupational relations and conditions would not be as powerfully related to health in women as in men of working age. The strongest test of this could be carried out with women who had an occupation at the time of the census. It would have been superfluous to look at past occupation because this would have been most unlikely to change the answer to this particular research question.

The Office for National Statistics' longitudinal study is based on the census and does not include measures of past occupation. A more serious problem for some purposes is that the longitudinal study does not contain data on income, education, access to consumer durables, or health related behaviours. The dataset used by Holt and Grundy is far richer. It enables them to ask such questions as, Is income related to health independent of education (which other studies show both increases earning potential and improves health behaviours)?3 Unfortunately, at present there are no datasets that include all these items of information and are linked to mortality.