Elsevier

Health & Place

Volume 5, Issue 2, June 1999, Pages 157-171
Health & Place

Assessing ecologic proxies for household income: a comparison of household and neighbourhood level income measures in the study of population health status

https://doi.org/10.1016/S1353-8292(99)00008-8Get rights and content

Abstract

This paper examines the validity of using ecologic measures of socioeconomic status as proxies for individual-level measures in the study of population health. Based on a representative 5% sample of households in a Canadian province, the study integrated three sources of information: administrative records of individual health care utilization, records of deaths and 1986 census records which contained information on household income and average neighbourhood income. Thirteen measures of health status were developed from these sources of information. The hypothesis that risk estimates derived from ecologic income measures will be attenuated relative to estimates obtained from household income was not supported. These results provide evidence for the use of ecologic-level measures of income in studies which do not have access to individual-level income measures.

Introduction

The socioeconomic characteristics of individuals, whether indicated by income, education or occupation, are known to influence both health status and the use of health services (Haan et al., 1987, Feldman et al., 1989, McLaughlin et al., 1989, Wilkins et al., 1991, Anderson et al., 1993, Pappas et al., 1993, Mustard et al., 1994Mustard et al., 1997, Davey Smith et al., 1996). Routinely collected administrative health information is frequently limited in its ability to describe these relationships because of the absence of measures of individual socioeconomic status. To overcome this obstacle, many researchers have attempted to supplement individual-level health information with measures of socioeconomic characteristics of the area of residence, which are typically derived from census or tax filing information (Satin and Monetti, 1985, Wissow et al., 1988, McLaughlin et al., 1989, Logue and Jarjoura, 1990, Wilkins et al., 1991, Krieger, 1992, Anderson et al., 1993Anderson et al., 1997, Mustard et al., 1994, Carstairs, 1995, Geronimus et al., 1996).

When individual-level measures of health status or health service utilization and area-based information on socioeconomic characteristics share complementary geographic identifiers, the merging of such data is relatively straightforward, providing an efficient method of substituting for the absence of individual-level socioeconomic measures (Wilkins, 1993). However, little research to date has examined the validity of this methodology in health research (Piantadosi et al., 1988, Krieger, 1991Krieger, 1992, Geronimus et al., 1996, Anderson et al., 1997). In the use of aggregate measures as proxies for individual level measures, at least two distinct issues of validity arise: the first concerns measurement validity and the second concerns construct validity, specifically, the theoretical specification of health effects influenced by characteristics at both the individual and the ecological level.

If socioeconomic status is considered to be primarily an indicator of individual attributes and resources, yet only group-level measures of socioeconomic status are available, the assignment of a group-level mean socioeconomic status or rank to a group of individuals will introduce error in the representation of individual scores relative to each individual's true socioeconomic status. As the level of aggregation increases at which mean socioeconomic attributes are measured, for example from a neighbourhood of 200 households to a community of 2000 households, the socioeconomic heterogeneity of the group will also tend to increase, raising the proportion of error in each individual's score. Random measurement error produced by this process is likely to attenuate the magnitude of relationships Morgenstern, 1982, Greenland and Morgenstern, 1989. Measurement theory predicts that a group level measure, when used as a proxy for a direct household measure, will contain more measurement error than the direct household measure. Accordingly, the relationship between the group level measure and a health outcome will be attenuated in comparison to the true (and unknown) relationship relative to the direct household measure of socioeconomic status (Nunnally, 1978).

Health is generally understood to be jointly determined by individual attributes and behavioral practices as well as by resources and exposures present within the larger social and physical environments. Because both individual and ecologic characteristics may effect health status, there is the potential for misspecification of important theoretical relationships when either individual-level or group-level socioeconomic measures are used exclusively. The independent effects of individual and ecologic characteristics are central to the theoretical frameworks of epidemiology and sociology Schwartz, 1994, Susser, 1994 and there is increasing interest in investigating the multi-level effects of socioeconomic status on health and the utilization of health care services Wong and Mason, 1985, Koepsell et al., 1991, Von Korff et al., 1992, Jones and Duncan, 1995, Ecob, 1996.

This paper, however, has a more limited objective: to examine evidence for the validity of using ecologic measures of socioeconomic status in the study of population health in the absence of individual-level measures of socioeconomic status. In the work reported here we have focused exclusively on assessing measurement validity, adopting the heuristic assumption that household and ecologic measures of income are operationalizing the same underlying socioeconomic construct.

The study was designed to test two related hypotheses. The first hypothesis proposes that the magnitude of associations between discrete measures of health status and ecologic measures of socioeconomic status will be less than observed for associations with socioeconomic status measured at the household level. The second hypothesis predicts that the magnitude of the association of ecologic measures of socioeconomic status and health status will be diminished for rural populations relative to estimates obtained for urban populations.

The second hypothesis is based on the following propositions. Urban populations are strongly stratified on the basis of the cost of housing with the ability to afford housing in turn determined by income. Most households buy or rent as much housing as they can afford. The consequence of this sorting by housing value in urban populations is that the use of a group level measure of neighbourhood income may be a valid proxy for household income in urban environments. While the mean neighbourhood income may be only a poor approximation of an individual household in that neighbourhood, within-neighbourhood incomes are more similar than across-neighbourhood incomes.

However, there is good reason to be suspicious of this methodology in geographic settings with substantial rural populations. There may be less sorting of households by housing cost in rural areas, suggesting that there is greater within-neighbourhood income variance in rural communities than in urban communities.

Section snippets

Study sample

This study is based on a representative sample of approximately 5% of the Manitoba population in 1986, consisting of 47,935 individuals in 16,627 private or collective dwellings. The 1,050,000 residents of the province from which this study sample was drawn were universally insured for a comprehensive range of medically necessary health care services. Comprehensive, computerized records of health care utilization for each person in the sample are maintained by the Manitoba Health Services

Results

The incidence of mortality, births and nursing home admissions during the 36 month observation period are reported in Table 1. Table 1 also reports the prevalence of self-reported disability in June 1986 and the 36 month treatment prevalence for arthritis, major mental health disorders, minor mental health disorders and diabetes. There are a number of differences between urban and rural populations in these crude rates reported in Table 1. A higher proportion of residents of urban areas were in

Discussion

The objective of this study was to examine the validity of ecologic measures of income in predicting differences in population health as a proxy for income measured at the household level. The study has also contrasted the performance of group-level and household level income measures in urban and rural populations. As such, this study has focused exclusively on the issue of measurement validity and has not tested hypotheses which derive from theoretical propositions concerning independent and

Acknowledgements

Portions of this work were supported by grant 6607-1697-302 from the National Health Research and Development Program, Health Canada.

References (41)

  • J.J Feldman et al.

    National trends in educational differentials in mortality

    Am. J. Epidemiol.

    (1989)
  • A.T Geronimus et al.

    On the validity of using census geocode characteristics to proxy individual socioeconomic characteristics

    J. Am. Stat. Assoc.

    (1996)
  • S Greenland et al.

    Ecological bias, confounding, and effect modification

    Int. J. Epidemiol.

    (1989)
  • M Haan et al.

    Poverty and health: prospective evidence from the Alameda county study

    Am. J. Epidemiol.

    (1987)
  • Houle, C., Berthelot, J.M., David, P., Mustard, C.A., Roos, L.L., Wolfson, M.C., 1996. Project on Matching Census 1986...
  • N Krieger

    Women and social class: a methodological study comparing individual, household, and census measures as predictors of black/white differences in reproductive history

    J. Epidemiol. Commun. Health

    (1991)
  • N Krieger

    Overcoming the absence of socioeconomic data in medical records: validation and application of a census-based methodology

    Am. J. Publ. Health

    (1992)
  • J.P Leigh et al.

    Education level and rheumatoid arthritis: evidence from five data centers

    J. Rheumatol.

    (1991)
  • R.B Lipton et al.

    Determinants of incident non-insulin-dependent diabetes mellitus among blacks and whites in a national sample: the NHANES I epidemiologic follow-up study

    Am. J. Epidemiol.

    (1993)
  • C.G McLaughlin et al.

    Small-area variation in hospital discharge rates: do socioeconomic variables matter?

    Med. Care.

    (1989)
  • Cited by (263)

    View all citing articles on Scopus
    View full text