IJE Advance Access originally published online on September 28, 2006
International Journal of Epidemiology 2006 35(5):1361-1363; doi:10.1093/ije/dyl211
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Letter to the Editor |
Neighbourhood effects and the real world beyond randomized community trials: a reply to Michael J Oakes
1 Community Medicine and Public Health, Department of Clinical Sciences in Malmö, Faculty of Medicine, Lund University, Malmö, Sweden
2 Unit of Health and Health Care Epidemiology, Department of Health and Health Care Management, Region Skåne, Sweden
3 UMR-S 707 (INSERM—Université Pierre et Marie Curie-Paris6), Paris, France
* Corresponding author. Community Medicine and Public Health, Department Clinical Sciences, Malmö University Hospital, 205 02 Malmö, Sweden. E-mail: juan.merlo{at}med.lu.se
We appreciate the positive opinion of Michael J Oakes1 on our article2 and his recognition of the contribution of our research group to the field of contextual epidemiology. We also understand that Oakes' critics are not specifically directed at our publication but express a general concern on the use of observational approaches in research on neighbourhood and health.
In the absence of randomization the possibilities for confounding are theoretically enormous.3 Recognizing this may lead to a paralysing nihilism or to a radical ‘trialism’ (overemphasis of the advantages of randomized trials).4,5 Conversely, rather than rejecting observational studies, we believe it is our effort to reduce bias and confounding that constitutes a condition for the progress of epidemiology. Randomized trials, in spite of their theoretical suitability for causal inference, do not offer a solution to many epidemiological questions.3
Severe socioeconomic stratification, inferential support, and structural bias in observational multilevel analysis
Oakes argues that the individual characteristics leading to the ‘selection’ of individuals into a specific neighbourhood are entirely different from those leading to the selection of individuals into other contrasted neighbourhoods. This complete separation would produce a situation of structural confounding impossible to eradicate: specifying our individual-level regression model to get unconfounded neighbourhood variance, we would simultaneously eliminate all neighbourhood variance, leaving no variability to explain with neighbourhood variables.6
Oakes' scenario of absolute separation, as a good example of ‘paralysing nihilism’, is theoretically possible but unrealistic. Whether socioeconomic stratification produces such an absolute confounding is an empirical question to address in each study.7,8 In our database, 34% of residents of deprived areas had a low income, as compared with 19% of residents of affluent areas. Therefore, our data do not support Oakes' picture of absolute separation of individual attributes in contrasted neighbourhoods. Also, we were able to separate the population density and area socioeconomic effects, since the distribution of individuals across cells of combined area indicators (Table 1) was much less unbalanced than that approximated by Oakes.1 In any case, such estimation problems can be mitigated by the use of large databases as the one we employed and Markov chain Monte Carlo estimation approaches.9
|
Oakes argues6 that multilevel analysis is suitable for investigating school effects on pupil performance but not for neighbourhood effects. The reason would be that teachers constitute exogenous forces influencing the pupils, in opposition to most neighbourhood effects that would be endogenous (i.e. determined by the neighbourhood composition in terms of various individual characteristics). We do not agree with Oakes' argument. Indeed, many neighbourhood factors are exogenous forces influencing the residents (e.g. healthcare and sport facilities, urban design, etc.). Moreover, even endogenous neighbourhood effects can be properly investigated as long as social stratification is not as strong as in the extreme case considered by Oakes.
The reason for analysing within-neighbourhood clustering
Measures of neighbourhood variance complement information obtained from classical measures of association.10 Multilevel and spatial measures of variance are useful to assess whether health phenomena have a contextual dimension and the geographical scale of variations.10–16 We consider these indicators as descriptive but interesting measures of geographic heterogeneity but have never proposed such tools as a way to perform causal inference on ‘pure’ neighbourhood effects as Oakes seems to imply.
We see within-neighbourhood clustering of health outcomes as the result of a mixture of processes including selective migration driven by external forces of segregation and residential choices, integral external forces (e.g. urban design, healthcare facilities, air pollution) and contextual endogenous effects emerging from social interactions (e.g. sense of community, collective efficacy). Advances in the identification of these processes can only be expected by combining quantitative epidemiological approaches10–19 with hermeneutic qualitative methods and social observation.20
Are randomized community trials the alternative to observational studies?
We are less confident than Oakes that randomized interventional studies constitute the canon for understanding neighbourhood effects. Sorensen have noted that ‘the randomized controlled design is the widely accepted paradigm for assessing the effects of community interventions’.21 However, this does not imply as Oakes infers that ‘the randomized community trial is canonical design for neighbourhood effect studies’.6 A community intervention is an exogenous force suitable for randomization and trial evaluation. Conversely, many neighbourhood effects are generated by the internal dynamics of the neighbourhoods and must be investigated as such since they cannot be experimentally recreated.
First, for reasons of practicability and costs, randomized community trials often include few communities/neighbourhoods,21 precluding generalizability and limiting the benefits of randomization. Second, owing to feasibility and ethical considerations, many hypotheses on neighbourhood effects cannot be tested through randomized trials.3,21
Third, randomized community interventions may lead to inferential problems when multiple causal pathways are involved, which is rather common because of the multifaceted nature of interventions.22,23 In this case, the only inference to perform is on the combined effects of the various action strategies implemented, which does not bring specific knowledge on a given neighbourhood effect outside an intervention context. In fact, causal inference on everyday-world neighbourhood effects seems in contradiction with the objective of community trials: whereas causal inference on out-of-intervention-context neighbourhood effects would need one to neutralize social placebo effects (changes induced in individuals by simply being the focus of attention), intervention planners understandably may wish to maximize it by involving as much as possible the whole community.23
Finally, Oakes' point of view that the non-exchangeability of individuals between neighbourhoods must be solved through randomization leads to a dramatic simplification of our tasks. In our view, the non-exchangeability of individuals between neighbourhoods is due to differences in individual resources and vulnerabilities that may confound, but also mediate or interact with neighbourhood effects, something of direct relevance to neighbourhood research rather than a nuisance to dissolve by randomization.
As a final point, we share the general concerns of Oakes regarding good observational epidemiological practice, including having strong a priori theories and hypotheses, and evaluating the consistency of the findings through different modelling strategies.24 Oakes' insight is much appreciated and we, and surely our colleagues around the world, will consider his comments in future observational multilevel analyses.
References
1 Oakes JM. Commentary: Advancing neighbourhood-effects research—selection, inferential support, and structural confounding. Int J Epidemiol 2006;35:643–47.
2 Chaix B, Rosvall M, Lynch J, Merlo J. Disentangling contextual effects on cause-specific mortality in a longitudinal 23-year follow-up study: impact of population density or socioeconomic environment? Int J Epidemiol 2006;35:633–43.
3 Black N. Why we need observational studies to evaluate the effectiveness of health care. BMJ 1996;312:1215–18.
4 Rimm AA, Bortin M. TRIALISM: the belief in the Holy Trinity clinician—patient—biostatistician. Biomed Special Issue 1978;28:60–63.
5 Abel U KA. The mythology of randomization. Available at: http://www.symposion.com/nrccs/abel.htm (Accessed August 20, 2006).
6 Oakes JM. The (mis)estimation of neighborhood effects: causal inference for a practicable social epidemiology. Soc Sci Med 2004;58:1929–52.[CrossRef][Web of Science][Medline]
7 Diez Roux AV. Estimating neighborhood health effects: the challenges of causal inference in a complex world. Soc Sci Med 2004;58:1953–60.[CrossRef][Web of Science][Medline]
8 Subramanian SV. The relevance of multilevel statistical methods for identifying causal neighborhood effects. Soc Sci Med 2004;58:1961–67.[CrossRef][Web of Science][Medline]
9 Browne WJ. MCMC Estimation in MLwiN. 2nd edn. London: Centre for Multilevel Modelling, Institute of Education, University of London, 2004.
10 Merlo J. Multilevel analytical approaches in social epidemiology: measures of health variation compared with traditional measures of association. J Epidemiol Commun Health 2003;57:550–52.
11 Merlo J, Chaix B, Yang M, Lynch J, Rastam L. A brief conceptual tutorial of multilevel analysis in social epidemiology: linking the statistical concept of clustering to the idea of contextual phenomenon. J Epidemiol Commun Health 2005;59:443–49.
12 Larsen K, Merlo J. Appropriate assessment of neighborhood effects on individual health: integrating random and fixed effects in multilevel logistic regression. Am J Epidemiol 2005;161:81–88.
13 Merlo J, Chaix B, Ohlsson H et al. A brief conceptual tutorial of multilevel analysis in social epidemiology: using measures of clustering in multilevel logistic regression to investigate contextual phenomena. J Epidemiol Commun Health 2006;60:290–97.
14 Chaix B, Merlo J, Subramanian SV, Lynch J, Chauvin P. Comparison of a spatial perspective with the multilevel analytical approach in neighborhood studies: the case of mental and behavioral disorders due to psychoactive substance use in Malmo, Sweden, 2001. Am J Epidemiol 2005;162:171–82.
15 Chaix B, Merlo J, Chauvin P. Comparison of a spatial approach with the multilevel approach for investigating place effects on health: the example of healthcare utilisation in France. J Epidemiol Commun Health 2005;59:517–26.
16 Chaix B, Leyland AH, Sabel CE et al. Spatial clustering of mental disorders and associated characteristics of the neighbourhood context in Malmo, Sweden, in 2001. J Epidemiol Commun Health 2006; 60:427–35.
17 Merlo J, Yang M, Chaix B, Lynch J, Rastam L. A brief conceptual tutorial on multilevel analysis in social epidemiology: investigating contextual phenomena in different groups of people. J Epidemiol Commun Health 2005;59:729–36.
18 Merlo J, Chaix B, Yang M, Lynch JW, Rastam L. A brief conceptual tutorial on multilevel analysis in social epidemiology: interpreting neighbourhood differences and the effects of neighbourhood characteristics on individual health. J Epidemiol Commun Health 2005;599:1022–102.
19 Merlo J, Asplund K, Lynch J, Rastam L, Dobson A. Population effects on individual systolic blood pressure: a multilevel analysis of the World Health Organization MONICA Project. Am J Epidemiol 2004;159:1168–79.
20 Macintyre S, Ellaway A. Neighborhoods and Health: An Overview. In: Kawachi I, Berkman LF (eds). Neighborhoods and Health. New York: Oxford University Press, 2003.
21 Sorensen G, Emmons K, Hunt MK, Johnston D. Implications of the results of community intervention trials. Annu Rev Public Health 1998;19:379–416.[CrossRef][Web of Science][Medline]
22 Baranowski T, Lin LS, Wetter DW, Resnicow K, Hearn MD. Theory as mediating variables: Why aren't community interventions working as desired? Ann Epidemiol 1997;S7:S89–S95.[CrossRef][Web of Science]
23 Krieger J, Allen C, Cheadle A et al. Using community-based participatory research to address social determinants of health: lessons learned from Seattle Partners for Healthy Communities. Health Educ Behav 2002;29:361–82.
24 King G, Zeng L. When can history be our guide? The pitfalls of counterfactual inference. Available at: http://gking.harvard.edu/preprints.shtml (Accessed August 20, 2006).
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  J Merlo, H Ohlsson, K F Lynch, B Chaix, and S V Subramanian Individual and collective bodies: using measures of variance and association in contextual epidemiology J Epidemiol Community Health, December 1, 2009; 63(12): 1043 - 1048. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B Chaix, M Lindstrom, M Rosvall, and J Merlo Neighbourhood social interactions and risk of acute myocardial infarction J Epidemiol Community Health, January 1, 2008; 62(1): 62 - 68. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. Agyemang, C. v. Hooijdonk, W. Wendel-Vos, E. Lindeman, K. Stronks, and M. Droomers The association of neighbourhood psychosocial stressors and self-rated health in Amsterdam, The Netherlands J Epidemiol Community Health, December 1, 2007; 61(12): 1042 - 1049. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||