Published online December 1, 2005
PEDIATRICS
Vol. 116
No. 6
December 2005, pp.
1542-1545
(doi:10.1542/peds.2005-1568)
Is Region of Country a Useful Variable for Child Health Studies?
Mark F. Guagliardo, PhD and
Cynthia R. Ronzio, PhD
Department of Prevention and Community Health
Department of Epidemiology and Biostatistics
George Washington University School of Public Health and Health Services
Washington, DC 20052
Center for Health Services and Community Research
Children's National Medical Center
Washington, DC 20010
Abbreviations: CHI, child health index • MAUP, modifiable areal unit problem
In their article "The Health Status of Southern Children: A Neglected Regional Disparity" (in this month's Pediatrics electronic pages), Goldhagen et al1 have raised the intriguing question of whether US region is a significant factor for child health outcomes. Although much attention has been given to contextual influences on health,2–4 most studies have considered local contextual effects such as neighborhood characteristics.5 Few have considered region as a contextual level.2,6 Therefore, Goldhagen et al's conclusion that region is a stronger predictor of poor outcomes than other variables commonly used is quite remarkable. A careful examination of their methods is in order before their recommended research agenda is undertaken. We have 2 levels of concern: the general approach to the definition of health regions and a specific concern about the geographic unit of analysis.
 |
DEFINING HEALTH REGIONS
|
|---|
The definition of geographic regions for health studies can be achieved either by an impartial exploration of the geographic distribution of health statistics or through historical, sociocultural, or policy-relevant considerations. Goldhagen et al frame their article as an examination of the South, which suggests a preference for the latter approach. This focus might stem from an interest in the legacies of slavery or in current-day clustering of state policies and funding for public health and health care. Yet their methods are primarily an exploration of health statistics. They mapped the child health index (CHI) and identified the Deep South based on the revealed geographic distribution. Their methods for selecting member states are not precisely explained, but it seems that state selection was influenced somewhat by a desire to achieve a contiguous Deep South. Nevertheless, the CHI and its components were the variables used to define regions. To confirm the validity of the CHI-defined Deep South, they contrasted it with other regions to test for differences in CHI. This circularity could have been avoided if the Deep South were defined solely by historical or policy considerations. It also could have been avoided by a purely objective exploration of the spatial distribution of CHI.
To illustrate the latter, we analyzed the state CHI values from their first table. A histogram of the values suggested 5 natural categories with breakpoints that we defined by using the Jenks natural-breaks function7 in ArcGIS 9.0.8 The function seeks to minimize the squared intraclass deviations from each class mean. The resulting map of the 5 CHI levels is presented in Fig 1. It shows that "South" and "Deep South" can be elusive concepts in terms of child health outcomes. For example, to include Florida and Texas in the general South, one would also have to include 29 states stretching as far as Idaho and Michigan. On the other hand, if one sought to obtain the Deep South, it would consist of only 4 noncontiguous states (darkest shade) or 11 noncontiguous states (darkest 2 shades) that are not entirely southern.
If we ignored these objectively defined break points and used contiguity as a criterion for region membership, then we could reassign South Carolina from the worst-outcomes group to second worst group extending from North Carolina to New Mexico and Wyoming. This would still leave us with contiguity problems in the rest of the country, but it would yield a tidy Deep Central South region consisting of Louisiana, Mississippi, and Alabama. These 3 states lag significantly in child health outcomes, and clearly it would be fruitful to focus child health interventions on the more impoverished areas within these states. However, that has been clear for some time from basic state health statistics.9 That they are contiguous neither increases nor decreases the need within those states. Furthermore, none of the state combinations discussed above is equivalent to the Deep South defined by Goldhagen et al.
The point of this exercise has been to show that defining regions on which to build research programs and policies is challenging and subjective and may have less value than some realize. We also note (and Goldhagen et al would probably agree) that children in low-ranking nonsouthern states such as Wyoming are no less deserving of outcome improvements than the children in higher-ranking southern states such as Florida simply because Florida can be construed to belong to a southern region.
|
GEOGRAPHIC UNIT OF ANALYSIS
|
|---|
Notwithstanding the exercise above, we question the use of large bordered areas such as states to discover and understand causes of child health outcomes. We recognize that states independently set health policies and have varying per-capita resources to maintain public health and health care programs. However, that very independence might partially underlie the discontiguous interstate health patterns revealed in Fig 1. Furthermore, much more localized factors affect health, and most states have heterogeneous and unevenly distributed populations that make "state" too coarse of a geographic unit to provide nuanced insight into the contextual influences on health.
Spatial analysts refer to this as the modifiable areal unit problem (MAUP).10 MAUP arises from the imposition of artificial units of spatial reporting (eg, states) on continuous or highly localized geographical phenomena, resulting in the generation of artificial and potentially misleading spatial patterns. Although rarely considered in the health literature,11 MAUP has been investigated thoroughly in other fields of study. A recent example from political science, the 2004 presidential election, parallels the Goldhagen et al study in terms of geographic scope and unit of analysis and illustrates the MAUP well. The 2004 electoral map of the lower 48 states shows a vast, contiguous swath of 30 Republican (red) states reaching outward from the South into the Midwest and West. Democratic (blue) states are confined to 3 contiguous areas: the West Coast, Central North, and Northeast. The map shows that region is a strong predictor of state majority vote. However, state of residence is a very poor predictor of an individual's or community's vote. This was demonstrated by researchers at Princeton12 and the University of Michigan,13,14 who produced national maps of counties shaded by vote proportions. Changing the geographic unit dramatically changed the interpretation. Their maps proved that (1) nearly the entire nation was some shade of purple (as opposed to blue or red) in 2004, (2) there was a great deal of variation within nearly all states, and (3) continuous degrees of shading give a more accurate picture than 2 discrete, shaded groups.
If state CHI were reduced to a dichotomous high-low variable such as electoral majority, then region would probably be as good a predictor of a state's CHI group as it is of electoral majority. Yet region would be just as poor a predictor of individual or community health as it is of individual or community vote. Although Goldhagen et al do not claim to have analyzed individual or community health, they repeatedly state that "living in the southern region" is a powerful predictor of children's health. It takes a careful reading to understand that in this context, "children's health" refers only to state ranking. The utility of region for understanding individual or community health is not demonstrated. If researchers are interested in the historical or sociopolitical sequelae of slavery, then a regional study is appropriate, and relevant theories about the Deep South should be presented and tested explicitly. On the other hand, if researchers are interested in improving child health, then analyses should focus on the proven determinants of health that act at the individual, local, and state levels until such time as independent region-level effects are demonstrated. Briefly stated, regional differences in health do not prove present-day regional effects, nor do they demonstrate the utility of region as a consideration for the development of child health interventions.
|
SETTLING THE QUESTION
|
|---|
To determine if region is an important factor for individual or community child health, a multilevel analysis such as hierarchical regression15 is an important next step. Such a model would use "individual child" as the unit of study and include factors that influence child health on a number of levels. Examples of individual- and family-level factors are income, insurance, education, and cultural/behavioral choices; community-level factors might include access to and availability of clean air and water, healthy affordable foods, health care facilities, and safe neighborhoods for outdoor activities; and state-level factors could be types of public health policies and programs, per-capita dollars allocated to those programs, Medicaid coverage rates, and dominant employment sectors. To test the validity of region as a determinant of health, the analysis then would include region as a clustering variable in a multilevel analysis, and it would have to show significant regional health effects that are independent of the other factors in the analysis.
In no way is our commentary meant to distract researchers and policy makers from the fact that child health statistics for southern states tend to be worse than for nonsouthern states, nor do we mean to suggest that state policies play no role in the production and maintenance of child health. We also recognize that state policies tend to cluster, probably because of shared cultural and historical backgrounds. However, until compelling data are presented, we do not believe it is helpful to use region of country as either a focus of blame for child health outcomes or an analysis or adjustment variable in child health studies. Developments in public health and health care research, policy, and practice should continue to be focused on individual-, family-, community-, and state-level factors.
|
FOOTNOTES
|
|---|
Accepted Jun 28, 2005.
Address correspondence to Mark F. Guagliardo, PhD, Center for Health Services and Community Research, Children's National Medical Center, 111 Michigan Ave NW, Washington, DC 20010. E-mail: mguaglia{at}cnmc.org
No conflict of interest declared.
|
REFERENCES
|
|---|
- Goldhagen J, Remo R, Bryant T III, et al. The health status of southern children: a neglected regional disparity. Pediatrics. 2005;116
(6). Available at: www.pediatrics.org/cgi/content/full/116/6/e746
- Hillemeier MM, Lynch J, Harper S, Casper M. Measuring contextual characteristics for community health. Health Serv Res. 2003;38
:1645
–1717[CrossRef][ISI][Medline]
- Diez Roux AV. Investigating neighborhood and area effects on health. Am J Public Health. 2001;91
:1783
–1789[Abstract/Free Full Text]
- Pickett KE, Pearl M. Multilevel analyses of neighbourhood socioeconomic context and health outcomes: a critical review. J Epidemiol Community Health. 2001;55
:111
–122[Abstract/Free Full Text]
- Brooks-Gunn J, Duncan GJ, Aber JL. Neighborhood Poverty. New York, NY: Russell Sage Foundation; 1997
- Andersen RM, Yu H, Wyn R, et al. Access to medical care for low-income persons: how do communities make a difference? Med Care Res Rev. 2002;59
:384
–411[Abstract/Free Full Text]
- Jenks GF. The data model concept in statistical mapping. In: International Cartographic Association, ed. International Yearbook of Cartography 7. Ulm, Germany: University of Ulm; 1967:186–190
- ArcGIS [computer program]. Version 9. Redlands, CA: ESRI, Inc; 2004
- Kids Count Project, Population Reference Bureau. Children at Risk: State Trends 1990–2000. Baltimore, MD: Annie E. Casey Foundation; 2002
- Openshaw S. The Modifiable Areal Unit Problem. Norwich, United Kingdom: Geo Books; 1984
- Guagliardo MF. Spatial accessibility of primary care: concepts, methods and challenges. Int J Health Geogr. 2004;3
:3[CrossRef][Medline]
- Vanderbei R. Election 2004 results. Available at:
www.princeton.edu/
rvdb/JAVA/election2004. Accessed June 13, 2005 - Gastner MT, Newman MEJ. Diffusion-based method for producing density-equalizing maps. Proc Natl Acad Sci USA. 2004;101
:7499
–7504[Abstract/Free Full Text]
- Newman M. Maps and cartograms of the 2004 US presidential election results. Available at: www-personal.umich.edu/mejn/election. Accessed June 13, 2005
- Diez-Roux AV. Multilevel analysis in public health research. Annu Rev Public Health. 2000;21
:171
–192[CrossRef][ISI][Medline]
PEDIATRICS (ISSN 1098-4275). ©2005 by the American Academy of Pediatrics
TOP
DEFINING HEALTH REGIONS
GEOGRAPHIC UNIT OF ANALYSIS
