Elsevier

Health & Place

Volume 16, Issue 2, March 2010, Pages 409-412
Health & Place

Short Report
Comparing the effects of neighbourhood characteristics on all-cause mortality using two hierarchical areal units in the capital region of Helsinki

https://doi.org/10.1016/j.healthplace.2009.10.008Get rights and content

Abstract

We examine how the choice of areal unit affects the estimation of neighbourhood effects on mortality using two different areal units. We used register data of 70,936 individuals aged 25–64 years residing in the capital region of Helsinki, Finland. Results from the multilevel Poisson regression show that the clustering of mortality was slightly stronger when using smaller area units. The differences disappear when account was taken of known individual-level characteristics of the residents. This was also the case for the effect of the proportion of manual workers in the area on mortality. Our results imply that the choice of area scale will not lead to serious underestimation of neighbourhood effects in mortality.

Introduction

Various studies have shown statistically significant variation between neighbourhoods in mortality and morbidity even after accounting for individual characteristics of the residents. (For reviews, see Diez-Roux, 2003, Macintyre and Ellaway, 2003 and Riva et al., 2007). However, in most studies the effects of neighbourhood characteristics have been modest. This has led researchers to ask whether we are underestimating neighbourhood differences by not paying enough attention to defining neighbourhood units used in the analyses. Defining boundaries such that areas are heterogeneous in terms of determinants of health may hide area differences. (Stafford et al., 2008). Furthermore, it is unlikely that the mechanisms producing area effects on health would operate at the same scale and follow the same neighbourhood boundaries that were used for data collection. (Manley et al., 2006, Haynes et al., 2007).

This problem is closely related to the modifiable areal unit problem (MAUP), which specifies that the characteristics of the areal units to which the data are grouped can affect the results of the analysis. (Openshaw and Taylor, 1981, Manley et al., 2006). The phenomenon is composed of two interrelated problems. The results can be affected by the scale problem depending on the number of areal units used in the analysis of a given area or by the aggregation problem depending on how the boundaries between areas are defined to form a certain number of areal units. (Openshaw, 1984.)

Few previous studies have explored what kind of impact the choice of areal unit has on the estimation of neighbourhood effects on health. In this paper we concentrate on the scale problem of MAUP. The mean population of the areal units compared varies a lot in previous studies but the pattern seems to be that the clustering of ill-health is stronger when using smaller areal units. (Reijneveld et al., 2000, Haynes et al., 2008, Ross et al., 2004, Boyle and Willms, 1999, Lovasi et al., 2008). Reijneveld et al. (2000) and Ross et al. (2004) are the only studies concerning adult health that have examined area effects using different areal units after controlling for various individual-level confounders. In these two studies the estimated neighbourhood effects were so weak or nonexistent at both areal unit levels that the differences between levels were also nonexistent. Furthermore, the scale problem has had only a very small effect on the magnitude of health differences by area deprivation (Oliver and Hayes, 2007; Lovasi et al., 2008; Reijneveld et al., 2000). All the aforementioned studies except Lovasi et al. (2008) have used multilevel modelling strategies and survey data on health. There are also some studies concentrating on these issues but not using multilevel modelling (e.g. Flowerdew et al., 2008; Cockings and Martin, 2005). Overall, it seems in the light of the results of the previous studies that the MAUP's scale problem has a rather limited impact on the estimation of neighbourhood effects on ill-health and mortality. However, to our knowledge, no studies have concentrated on how the scale problem impacts on the estimation of neighbourhood effects using register-based mortality data even though there is a wide range of studies concentrating on neighbourhood effects on mortality. The need to experiment with different definitions of areas has also been acknowledged by several authors (e.g. Tatalovich et al., 2006, Martikainen et al., 2003).

Our aim was to examine whether the choice of the areal unit affects neighbourhood differences and possible neighbourhood effects in all-cause mortality using data from the capital region of Helsinki. Most studies examining neighbourhood effects on mortality have defined the neighbourhoods using administrative or census boundaries and it is not always possible to obtain data that allow free definitions. Therefore it is important to study to what extent the different administrative areal units affect the results. Haynes et al. (2007) claim that modifiable areal units are problematic only when area definitions are arbitrary. In our study, areal units have boundaries based on zoning decisions and health-care centre districts. This is interesting because the local availability of health-care services is a possible source of contextual effects along with the psychosocial factors related to the area characteristics such as social networks and social cohesion. (Flowerdew et al., 2008; Macintyre and Ellaway, 2003). These psychosocial factors are likely to have stronger effects in smaller areas than districts and therefore might produce some differences between areal units.

Section snippets

Data sources and methods

We used mortality data for 1991–2002 linked to various registers obtained from Statistics Finland. After removing people living in institutions our 15 percent sample consists of 70,936 individuals aged 25–64 years residing in the four cities (Helsinki, Vantaa, Espoo, Kauniainen) of the capital region of Helsinki, among whom there were 4338 deaths. The data contain two hierarchical areal units for each individual. There are 70 districts in the Helsinki region which are aggregated from 258

Results

The crude death rate calculated for the whole study population was 52.8 per 10,000 person years and for districts its 10th and 90th percentiles were 39.6–73.9 and for sub-districts 43.0–66.1. We constructed 2-level models where each areal unit level is used with individual-level variables (Table 2). There were statistically significant differences in mortality between the areas when age and sex were controlled for. The ARD at this point was estimated to be 16% (for the model containing

Discussion

In the age and sex adjusted multilevel models we observed strong mortality differences between areas. The differences in the clustering of mortality between the two areal unit levels were small but the variation was slightly larger among sub-districts. When all the individual level variables were included in the model there was no significant area-level variation and therefore no differences between the areal units used. The results are somewhat consistent with previous studies concerning the

Acknowledgements

We thank Statistics Finland for permission to use the data and the Academy of Finland for funding the study. The Social and Public Health Sciences Unit is jointly funded by the Medical Research Council (MRC) and the Chief Scientist Office (CSO) of the Scottish Government Health Directorates. AHL was funded by the Chief Scientist Office as part of the “Measuring health, variations in health and the determinants of health” programme, wbs U.1300.00.001.

References (25)

  • A.V. Diez-Roux

    The examination of neighborhood effects on health: conceptual and methodological issues related to the presence of multiple levels of organization

  • M. Huisman et al.

    Socioeconomic inequalities in mortality among elderly people in 11 European populations

    Journal of Epidemiology and Community Health

    (2004)
  • Cited by (22)

    • Long-term trends in urban-neighbourhood inequalities in cause-specific mortality and hospitalisation – multilevel analyses among individuals nested in Finnish post-code areas, 1991–2018

      2023, SSM - Population Health
      Citation Excerpt :

      This may dilute the true area-level effects and bias our estimates downwards. However, as reported in an earlier study, changing neighbourhood definitions in the Finnish context did not have a substantial impact on disparities in neighbourhood mortality (Tarkiainen et al., 2010). On the other hand, it is unlikely that we were able to account for all relevant individual-level characteristics in our analysis.

    • Neighborhood socioeconomic deprivation and psychiatric medication purchases. Different neighborhood delineations, different results? A nationwide register-based multilevel study

      2021, Health and Place
      Citation Excerpt :

      Although previous reviews have highlighted the importance of studying mental health conditions among alternative methods for designating neighborhoods, several reviews also emphasize that the majority of published studies use only a single set of areas, and most rely on administrative areas (Jivraj et al., 2019; Mair et al., 2008; March et al., 2008; Richardson et al., 2015). Some previous studies have systematically compared the use of different area delineations for various health outcomes, with studies finding that changing the areal units affected the results (e.g., Chaix et al., 2006; 2005; Cockings and Martin, 2005; Flowerdew et al., 2008; Franzini and Spears, 2003; Jablonska et al., 2020; Messer et al., 2012; Parenteau and Sawada, 2011) and other studies finding no major differences between areas (e.g., Cabrera-Barona et al., 2018; Lovasi et al., 2008; Root, 2012; Tarkiainen et al., 2010). This difference may be explained by the fact that the studies examined different health outcomes and contextual factors and that some studies only compared different administrative areas, while other studies looked at alternative area delineations.

    • Changes in regional variation in mortality over five decades – The contribution of age and socioeconomic population composition

      2021, SSM - Population Health
      Citation Excerpt :

      However, findings were similar, and we thus show results incorporating all municipalities. Furthermore, since previous results on mortality differentials in the Finnish capital region suggest that the choice of area unit does not significantly affect the estimation of area effects (Tarkiainen et al., 2010), we believe our main results are robust to the area level chosen. Nevertheless, further longitudinal studies using different area definitions are needed.

    • Neighborhood social capital and crime victimization: Comparison of spatial regression analysis and hierarchical regression analysis

      2012, Social Science and Medicine
      Citation Excerpt :

      From their results, it was suggested that area-level variables (such as racial segregation, proportion of female elderly persons living alone and proportion of elderly persons in poverty) differed according to the areal units of analysis. Tarkiainen, Martikainen, Laaksonen, and Leyland (2010) also examined the relationship between neighborhood characteristics such as the proportion of manual workers in the area and mortality in Helsinki using two alternative areal units. Although they found that the effects of neighborhood-level characteristics on mortality was slightly stronger when using smaller area units, it was suggested that the choice of area scale did not lead to serious underestimation of neighborhood effects on mortality.

    View all citing articles on Scopus
    View full text