Papers

Postcodes as useful markers of social class: population based study in 26 000 British householdsCommentary: Socioeconomic position should be measured accurately

BMJ 1999; 318 doi: https://doi.org/10.1136/bmj.318.7187.843 (Published 27 March 1999) Cite this as: BMJ 1999;318:843

Postcodes as useful markers of social class: population based study in 26 000 British households

  1. John Danesh, Merton College junior research fellow (john.danesh{at}balliol.ox.ac.uk)a,
  2. Simon Gault, assistant statisticianb,
  3. Jo Semmence, statisticianb,
  4. Paul Appleby, research officerc,
  5. Richard Peto, professor of medical statistics and epidemiologya
  1. aClinical Trial Service Unit and Epidemiological Studies Unit, University of Oxford, Radcliffe Infirmary, Oxford OX2 6HE
  2. bFamily Resources Survey, Department of Social Security, London WC2N 6HT
  3. cImperial Cancer Research Fund Cancer Epidemiology Unit, Radcliffe Infirmary, Oxford OX2 6HE
  4. Department of Social Medicine, University of Bristol, Bristol BS8 2PR
  1. Correspondence: Dr Danesh
  • Accepted 4 December 1998

Markers of poverty or of low social class are associated with many diseases and potential causes of disease, but medical studies often fail to record sufficient information on socioeconomic status.1 Postcodes of individuals are, however, often available in Britain, and commercial software exists that estimates household income from the postcode alone. We assessed how informative postcode income estimates are, either about reported household income or about other characteristics related to social class in a large, population based survey of British residents.

Methods and results

The family resources survey involves personal interviews with members of private households in England, Scotland, and Wales selected by stratified clustered probability sample.2 During 1995-6, 26 445 (70%) of 37 712 eligible households gave answers to questions on socioeconomic characteristics. Reported weekly household income was taken as the sum of all sources of pretaxation income (excluding housing benefit) reported by household members. During 1985 to 1993 members of 11 million households, or about half of all households in Britain, provided information to a marketing company about annual income and gave a complete address that included a full postcode—that is, 6 or 7 characters.

This information was used to produce commercial software that estimates household incomes from postcodes. After adjustments for regional variation and for inflation in reported income levels, the pretaxation incomes of at least six households were used to calculate a weighted average income for that postcode. When there were fewer than six responses, the income information was combined with the data for respondents with neighbouring postcodes until a reliable estimate could be made. Parts of this database are updated annually. We compared household income estimates obtained by FIND (a commercially available software program) with information reported in the family resources survey. Matching of the data was carried out at the Office for National Statistics. The investigators in this study were provided with columns of numerical data without any personal identifiers.

The overall correlation coefficient between postcode estimates and reported values of weekly household income for 26 282 individuals was moderate (0.40, 99% confidence interval 0.39 to 0.42; 2P<0.0001). When households were ranked in three equal sized groups on the basis of postcode income estimates, there were substantial and highly significant differences in reported weekly income, duration of education, home ownership, membership of higher social classes, and access to various consumer goods (2P<0.0001 for each) (see table).

Markers of social class by thirds of income estimates based on postcode for 26 282 individuals. Values are numbers (percentages) of individuals, unless stated otherwise

View this table:

Comment

Postcode income estimates are easily available in Britain and can be useful markers of social class. As UK postcodes are usually shared by only 15 to 20 households,3 these estimates should more accurately predict the social class of individuals than can more aggregated data, such as information derived from enumeration districts or electoral wards,4 or, in the United States, from zip codes,1 which are usually based on several hundred households.

Consequently, postcode estimates may serve various epidemiological purposes, particularly when it is not feasible to collect detailed information on social class in large studies. Firstly, they can be used to help standardise for the effects of social class, which may be important when the occurrence of a disease and its possible risk factors are both related to poverty. Looking at the strength of an association before and after such statistical adjustment can suggest how much full adjustment for social class, if that were possible, would have modified the association. This comparison might then lead investigators to make more detailed measurements of social class in a subset of participants. Secondly, income estimates derived from postcodes might help to assess bias when a proportion of eligible people refuse to participate in a study. Most epidemiological reports show that people who fail to return questionnaires or to give blood samples are of lower social class than volunteers.5 The effects of such biases are difficult to establish, especially if surveying the non-respondents again is impracticable, but postcode income estimates can help with this and other aspects of medical research in Britain.

Acknowledgments

Emily Banks, Robert Clarke, Rory Collins, Jeyanthi John, and Martin Vessey commented helpfully. Charles Lound of the Office for National Statistics helped with data management, Rom Rahman of QAS Systems provided postcode estimates, and Ian Liddicoat of Market Information Consultancy provided details about “FIND” methodology.

Contributors: JD initiated the study, performed the statistical analyses, interpreted the data, and drafted the paper. All authors discussed the main issues, interpreted the data, and edited the draft. JD is guarantor for the study.

Footnotes

  • Funding JD was supported by Merton College and a Frohlich award.

  • Competing interests None declared.

References

Commentary: Socioeconomic position should be measured accurately

  1. Yoav Ben-Shlomo, senior lecturer in clinical epidemiology,
  2. George Davey Smith, professor of clinical epidemiology
  1. aClinical Trial Service Unit and Epidemiological Studies Unit, University of Oxford, Radcliffe Infirmary, Oxford OX2 6HE
  2. bFamily Resources Survey, Department of Social Security, London WC2N 6HT
  3. cImperial Cancer Research Fund Cancer Epidemiology Unit, Radcliffe Infirmary, Oxford OX2 6HE
  4. Department of Social Medicine, University of Bristol, Bristol BS8 2PR

    Measures of deprivation based on area of residence have been used for several decades as ecological markers of a person's socioeconomic position. In the United States, median income of residents of areas such as census tracts and zip codes have been used; in the United Kingdom researchers have mainly relied on aggregate deprivation scores based on census measures, such as housing tenure, car ownership, and social class. The study by Danesh et al represents one of the first attempts in Britain to use income data rather than a composite, census based index. A single, direct measure is theoretically attractive if one argues that income underlies social inequalities in health through increased access to better living conditions and a healthier lifestyle.1Furthermore, the authors have classified areas to postcode level rather than to larger areas such as enumeration district or ward; this reduces the likelihood that any observed association is influenced by the “ecological fallacy.” 2

    Their work fails to address two fundamental questions. Firstly, is this new measure any better than existing methods? The authors' results show only a moderate correlation between their postcode aggregate income measure and individual data. It would have been useful if they had compared this with census scores based on enumeration district and ward. Smaller is not always better with respect to area based measures. Geronimus and Bound found little improvement in predicting self reported health when comparing data at census tract level (5000 individuals) with zip code data (25 000 individuals)3 The optimal population size for categorising the contextual nature of areas will depend on the nature of this contextual effect, and this cannot be assumed to be better indexed by aggregate measures for areas with smaller populations. At a postcode level one must also be concerned with possible sampling error and more importantly the systematic bias introduced by respondents who reply to a commercial survey.

    Secondly, can researchers manage without individual measures of socioeconomic position? Sometimes individual data are simply not collected or the quality of such data is extremely poor—for example, routinely collected health services data. Here, the use of a postcode based measure is invaluable for testing whether area based deprivation may be related to access to health care.4 However, the use of area based measures is less justifiable when researchers are prospectively collecting data as part of a large trial or observational study. Many studies show that both individual and area based measures seem to have independent effects on health outcomes, possibly as a result of the contextual effects of residing in poor neighbourhoods. To measure one and not the other will result in an underestimation of potential effects associated with socioeconomic position. Analyses based solely on an area measure of socioeconomic position can be highly misleading, especially if other risk factors are measured at an individual level. For example, it has been argued that variations in mortality by area based deprivation can be almost fully accounted for by smoking.5 However, applying the relative risk associated with individual smoking behaviour to mortality differences by area based deprivation underestimates the importance of individual socioeconomic position. Individual measures produce much steeper gradients of mortality risk than area based deprivation measures. We hope that the area based income measure introduced by Danesh et al will not be used in this way.

    Researchers should, when possible, continue to measure both individual and area based measures of socioeconomic position. Relying on ecological measures alone rather than using both would be analogous to asking people whether they smoked but not measuring how many years or the number of cigarettes they smoked.

    References

    View Abstract