Article Text


Explanations for differences in health outcomes between
neighbourhoods of varying socioeconomic level
  1. TNO (Netherlands Organisation of Applied Scientific Research), Institute of Prevention and Health
  1. PO Box 2215, 2301 CE Leiden, the Netherlands (SA.Reijneveld{at}
  1. Department of Health Services, State of California
  2. Department of Health Studies, University of Chicago
  1. Kate Pickett (kpickett{at}
  1. Department of Health Services, State of California
  2. Department of Health Studies, University of Chicago
  1. Kate Pickett (kpickett{at}

Statistics from

Editor,—With much interest, I read the review of Pickett and Pearl regarding the effect of neighbourhood socioeconomic level on health outcomes.1 They conclude that there is fairly consistent evidence for modest neighbourhood effects on health, because 23 of the 25 reviewed studies report a statistically significant association between at least one measure of social environment and a health outcome, after adjusting for individual level socioeconomic status.

I agree with the conclusion of the authors that most studies show only modest differences in health outcomes between neighbourhoods of varying socioeconomic level. However, I am far less sure than they are that this is a real neighbourhood effect. Incomplete adjustment for individual socioeconomic status may be a much more likely explanation for the modest differences as found. For instance, regarding mental health, differences between areas of varying socioeconomic level in Amsterdam, the Netherlands, become small and without statistical significance if individual socioeconomic status if adjusted for by several measures jointly. In contrast, adjustment for separate measures of individual socioeconomic status leaves modest differences between neighbourhoods of varying socioeconomic level. Some of these previously published results are shown in table1.2 Similar effects have been found for other measures of health like self reported health, health complaints and obesity, and to a lesser degree regarding smoking and long term functional limitations.3 In the same way, incomplete adjustment for individual socioeconomic status may explain the modest differences between areas as Pickett and Pearl found in their review.

Table 1

Odds ratios (and 95% confidence intervals) comparing the prevalence of apoor mental health (that is, an increased score on the General Health Questionnaire) for tertiles of Amsterdam boroughs, grouped by deprivation; crude, adjusted for age and gender, and additionally adjusted for individual socioeconomic status

Pickett and Pearl1 propose another explanation for the findings as presented in the preceding section.2 3 They explain the mostly negative findings in these two publications by a lack of power, because of a limited number of areas (that is, 22). However, they seem to be unaware of a later reanalysis of the same data that contradicts their explanation.4 This reanalysis yields very similar (mostly small) differences in health outcomes after adjustment for individual socioeconomic status for boroughs (n=22), neighbourhoods (n=92) and postcode sectors (n=76). At least for these data, a lack of power thus seems to be no valid explanation for mostly lacking area differences in health outcomes after rigorous adjustment for individual. I invite Pickett and Pearl to examine the impact of an incomplete adjustment for individual socioeconomic status in all studies that they included in their review: does this explanation hold, or do the results of studies on area differences in health outcomes vary because of other reasons?


Authors' reply

Editor,—We appreciate Dr Reijneveld's interest in our review of neighbourhood socioeconomic level and health outcomes.1-1

Dr Reijneveld suggests that much of the “effect” attributed to neighbourhood level socioeconomic processes could be explained by a lack of control for individual level socioeconomic measures. We agree that neighbourhood studies are at risk for overestimating effects, and emphasised the importance of measuring individual level SES on page 116 of the article, and again in our discussion on pages 119 and 120. As we specifically mentioned, in general, adjusting for more measures of individual SES is associated with smaller effect sizes. However, as we also discussed, adjustment for individual level SES may in fact remove a true neighbourhood effect if individual level SES is affected by neighbourhood level socioeconomic circumstances. There is substantial evidence in the sociological literature showing that educational attainment is strongly influenced by neighbourhood level factors.1-2 Despite this, our table's summary of published studies shows persistent neighbourhood level associations after simultaneous statistical adjustment for multiple individual level indicators (for example, Shoulset al 1996, Jones and Duncan 1995, Robert 1998, Haan et al 1987). In addition, the evidence put forth in the above table regarding mental health does not make a strong case for multiple-indicator adjustment. The lack of statistical significance seems to be attributable to adjustment for income alone, rather than multiple adjustment (compare OR=1.21 (1.01, 1.46) adjusting only for income, to OR=1.18 (0.98, 1.42) adjusting for income, education and occupation). If area level socioeconomic factors act as proxies for individual level characteristics, it is probable that they are proxies for individual level income, which is lacking in many studies. Again, however, almost all of the reviewed studies that included individual level income information revealed some neighbourhood level associations (for example, Curryet al 1993, Robert 1998, Jones and Duncan 1995, Diez-Roux 1997, Waitzman and Smith 1998). We believe the statistical “effect” is real, although any interpretation is still unclear. While residual confounding is always a possible explanation, the body of evidence suggests that alternative explanations are also likely.

We are grateful to Dr Reijneveld for referring us to his reanalysis of neighbourhoods and health outcomes,1-3 which was published after the completion of our review. Indeed it seems that lack of statistical power in this case was not the explanation for the modest impact of area level deprivation on self reported health. An alternative explanation might be the relative homogeneity of socioeconomic status in the Netherlands. For example, among Dr Reijneveld's sample of 92 neighbourhoods, the least deprived neighbourhoods in terms of low income contained 35% low income residents, while the most deprived contained 52%. We suspect that the range between advantaged and deprived neighbourhoods would be far greater in many American cities and that neighbourhood “effects” might be much stronger at the extremes of the distribution.


  1. 1-1.
  2. 1-2.
  3. 1-3.
View Abstract

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.