Intended for healthcare professionals

Rapid response to:

Editorials

Income inequality and population health

BMJ 2002; 324 doi: https://doi.org/10.1136/bmj.324.7328.1 (Published 05 January 2002) Cite this as: BMJ 2002;324:1

Rapid Response:

Spatial Blind Spots Obscure Inequality-Health Relationship

Professor Mackenbach's assessment of dissipating evidence on the
correlation between income inequality and population health1 overlooks two
key questions. First, has the scale of analysis captured the underlying
social-spatial processes that may lead to the health effects of
inequality? Second, does current research adequately control for spatial
autocorrelation in the residuals of models used to test the inequality-
health association? The answer to the first question is “probably not,”
and the answer to the second is “definitely not.” We cannot conclude the
inequality-health debate with any certainty until these questions are
answered.

First, I must ask whether existing research has modelled the right
concept at the wrong spatial scale. Most of the analyses in this area have
been conducted at the country, state/province, or metropolitan level. In
the most recent issue of the BMJ, Osler et al.2 use a finer scale, and
they find no significant relationship between income inequality and
mortality after applying control for individual income (more on this study
below). Most studies, however, have focused on larger units of analysis.
At the metropolitan level of aggregation, we see considerable spatial
variation in the social determinants of health within the unit of
analysis. In Hamilton, Canada, for example, the determinants of health
have distinct and complex intra-urban spatial patterns.3 Research that
aggregates to coarser scales such as the metropolitan or state level may
miss these finer spatial patterns and, in the process of averaging over
large areas, may obscure, rather than reveal, the actual relationships.

When I use small-area census tract units (average population of about
4500) to compute directly standardized premature mortality rates (ages 0-
74) and Gini coefficients for income inequality in Hamilton, I find the
Gini coefficient exerts the largest effect on premature mortality compared
to 16 other indicators of social status, including high-school educational
attainment and median household income. I also tested smoking rates and
air pollution exposure and found these produced lower relative risks of
mortality than the inequality metric used in most studies. Men living in
areas with the highest income inequality have a relative risk (RR) of
premature mortality 2.92 times greater than men living in areas with the
lowest income inequality. For women, the corresponding figure is lower (RR
= 2.18). It must be remembered that this finding comes from Canada, the
same country where researchers found no association at the metropolitan
level.4

A related methodological problem arises because past research has
lacked control for "spatial autocorrelation." This commonly observed
phenomenon suggests that near observations (say metropolitan areas) tend
to have attribute values that are more similar than those that are farther
apart. If there is positive autocorrelation in the residuals of a model,
this violates the independent observation assumption, and the results of
significance tests become questionable.5 Autocorrelation in residuals may
also signal specification error in the form of missing confounders or
effect modifiers. Recent research using multilevel survival models to test
the association between air pollution and mortality in the United States
(US) demonstrated that spatial autocorrelation can have a large impact on
the size and statistical uncertainty of observed relationships at the
metropolitan level.6-7 This analysis, which relied on a large survey
administered by the American Cancer Society (N ~ 550,000), indicates a
significant positive association between income inequality and mortality —
even after controlling for 24 individual level risk factors. Although the
relationship withstood control for so many individual confounders,
accounting for spatial autocorrelation reduced the income inequality
effect among 151 metropolitan areas of the US. In some models, control for
autocorrelation rendered the inequality-mortality relationship
insignificant. The frequently cited study on metropolitan areas of the US
that showed a positive correlation between inequality and mortality,8
while deftly executed in many areas, neglected this issue. Thus, we cannot
be certain that the existing inequality findings would be robust to
control for spatial autocorrelation.

On the Osler et al. study,2 the authors do not report whether
collinearity between individual income and area income equality influenced
their results. My results from Hamilton suggest this possibility because
the zero-order correlation between the Gini coefficient and the ecologic
variable measuring median household income is quite high (r = 0.8). In
multilevel models, variables with lower measurement error (in this case,
individual income) may tend to dominate those that are, for reasons of
aggregation, measured less precisely (i.e., area inequality).9 I also note
that the authors did not control for possible autocorrelation in their
otherwise excellent analysis.

In summary, current research may underestimate the importance of
contextual income inequality because the scale of analysis fails to match
the spatial processes at work. Smaller units of analysis may reveal
patterns similar to what I have observed in Hamilton, Canada. As for the
relationships in metropolitan and higher-scale research, none of this work
— with the exception of our study that treated inequality as a potential
confounder of the air pollution-health association — has applied spatial
regression analysis to test the inequality-health association. It is
unknown whether reported relationships in the inequality-health literature
have been influenced by violation of the independent observations
assumption. Given these spatial blind spots, we cannot yet conclude that
the relationship between income inequality and health is either
dissipating or stronger than once believed. We need sound research that
addresses these spatial questions before delivering a requiem for the
inequality-health hypothesis.

References

1. Mackenbach, JP. Editorial: Income inequality and population
health: evidence favouring a negative correlation between income
inequality and life expectancy has disappeared. BMJ 2002; 324: 1-2.

2. Osler M, Prescott E, Grønbæk M, Christensen U, Due P, Engholm G.
Income inequality, individual income, and mortality in Danish adults:
analysis of pooled data from two cohort studies BMJ 2002; 324: 1-4.

3. Luginaah I, Jerrett M, Elliott S, Eyles J, Parizeau K, Birch S,
Hutchinson B, Veestra J, Giovis C. 2002. Health profiles of Hamilton:
characterizing neighbourhoods for health investigations. The GeoJournal.
In press.

4. Ross NA, Wolfson MC, Dunn JR, Berthelot J, Kaplan G, Lynch JW.
Relationship between income inequality and mortality in Canada and in the
United States: cross sectional assessment using census data and vital
statistics. BMJ 2000; 320:898-902.

5. Griffith D, and Layne L. 1998. A Casebook for Spatial Statistical
Data Analysis. New York: Oxford University Press.

6. Krewski D, Burnett R, Goldberg MS, Hoover K, Siemiatycki J,
Jerrett M, et al. 2000. Reanalysis of the Harvard Six-Cities Study and the
American Cancer Society Study of Air Pollution and Mortality, Phase II:
Sensitivity Analysis. Cambridge, MA: Health Effects Institute, 295 pages.

7. Jerrett, M, Burnett R, Willis A, Krewski D, Goldberg MS, De Luca
P, Finkelstein N. Spatial analysis of the air pollution-mortality
relationship in the context of ecologic confounders. J Toxicol Environ
Health. 2002; In press.

8. Lynch JW, Kaplan GA, Pamuk ER, Cohen RD, Heck K, Balfour JL, et
al. Income inequality and mortality in metropolitan areas of the United
States. Am J Public Health 1998; 320: 898-902.

9. Burnett R, Ma R, Jerrett M, Goldberg MS, Cakmak S, Pope III A,
Krewski D. The spatial association between community air pollution and
mortality: a new method of analyzing correlated geographical cohort data.
Environ Health Perspect 2001;109 (S3): 375-380.

Competing interests: No competing interests

16 January 2002
Michael Jerrett
Assistant Professor of Geography and Health Studies
School of Geography and Geology and Health Studies Program, 1280 Main Street West, Hamilton, Canada