The Effect of Healthy Food Accessibility on Childhood Obesity Lauren Rosenshein
Nigel Waters
Dept. of Geography and Geoinformation Science George Mason University Fairfax, VA, 22030 United States
[email protected]
Dept. of Geography and Geoinformation Science George Mason University Fairfax, VA, 22030 United States
[email protected]
Abstract— The United States is facing an epidemic of childhood obesity, with obesity rates amongst children more than double what they were just 20 years ago. At the same time, research on the accessibility of healthy foods, especially in urban areas, has shown that certain populations are facing major barriers to a healthy diet. While there has been research on both the health consequences of childhood obesity and the deterioration of accessibility to healthy foods, there has been relatively little research on the relationship between the two problems. This paper will use a GIS based cluster analysis of Los Angeles County school level data on percentages of overweight 5th graders to expose clusters of overweight children in the area. Each school will be associated with a school polygon, which will represent the geographical area that the school serves. These school polygons will be used in the cluster analysis to locate areas with low accessibility to healthy foods. Several indicators will be used to establish an accessibility value, including distance to closest supermarket and the number of supermarkets within walking distance. The relationship between clusters of overweight 5th graders and low supermarket accessibility will then be examined. The results of the analysis indicate that a negative relationship exists between distance from the closest supermarket and childhood overweight. These results indicate the need for further research focusing on a smaller, more homogenous study area, a different measure of accessibility, or perhaps a regression technique that allows for spatial variation of the regression coefficients such as geographically weighted regression. Keywords- obesity; geographic clusters; accessibility; built environment
I.
INTRODUCTION
The United States is facing an epidemic of childhood obesity, with obesity rates amongst children more than double what they were 20 years ago [11]. At the same time, research on the accessibility of healthy foods, especially in urban areas, has shown that certain populations are facing major barriers to a healthy diet [1,3,6,7,8]. While much research has been done on both the health consequences of childhood obesity and the deterioration of accessibility to healthy foods [1,9,10] there has been relatively little research on the relationship between the two problems. This paper uses Los Angeles County school level data on percentages of overweight 5th graders to identify geographic clusters of overweight children, using spatial statistical
measures of clustering. Each school will be associated with a school polygon that will represent the geographical area that the school serves. A similar analysis will also be used to locate areas with low accessibility to healthy foods. The same school polygons will be used as the unit of analysis. Several measures will be used to establish an accessibility value, including distance to closest supermarket and number of supermarkets within walking distance. The relationship between clusters of overweight 5th graders and low supermarket accessibility will then be examined. It is expected that areas with high rates of overweight 5th graders also will be areas that have low access to the healthy foods found in supermarkets. The goal of the research is to reveal areas where the only accessible sources of food are convenience stores and fast food restaurants that have negative health impacts on children. This would have implications that could persuade policy-makers to provide incentives for the sale of healthy foods in such areas of low accessibility. This paper will briefly discuss the existing literature on the relationship between the built food environment and childhood overweight. The data used in the analysis will then be discussed, and the methods used will be explained, including the identification of geographic clusters of overweight 5th graders, a network analysis of the accessibility of healthy foods, and a regression analysis. The results will be reviewed, and the conclusions that can be drawn from those results will be outlined. Lastly, future research will be identified as a result of both the limitations of this study and the research concerns discovered due to the results of the analyses. II.
LITERATURE REVIEW
Research on food accessibility varies greatly in its methods, some studies relying more on the availability of affordable food, while others focus more on the distance and other physical accessibility attributes of supermarket chains [1,6,7]. The majority of research has concluded that accessibility varies greatly across space, especially in large urban areas with great income disparities [6,7]. Understanding that accessibility varies across space raises an important question, which is the effect of accessibility on the food intake of individuals.
There has been a large amount of research done on the relationship between accessibility of healthy foods and dietary intake [3,6,9,10]. Research has shown that individuals who live at a larger distance from large supermarkets (as opposed to convenience stores or other smaller food stores) consume significantly less fruits and vegetables than individuals who live closer distances [9]. Knowing that individuals’ eating habits are affected by healthy food accessibility, the research has now turned towards evaluating weight and Body Mass Index (BMI) variables as they relate to healthy food accessibility. Research on the impact of healthy food accessibility and weight has found that in many cases there is a positive relationship between accessibility and healthy weight of both adults and adolescents, meaning that the more accessible healthy foods are, the less likely a person is to be overweight or obese [4,5]. While there have been several studies on the issue, in order for major changes to be made to policies at the federal level there is still a need for an increased understanding of the relationship. Also, these differences in methods for evaluating accessibility lead to the need for studies that evaluate accessibility in different ways, specifically with more of an emphasis on network analysis and physical accessibility, where food price and retail store density have previously been the most popular measures [1,4]. The focus of several studies on adolescents in 7th through 9th grade [4,5] makes analysis of data on younger students a valuable addition to the body of research. III.
DATA
This study draws on school level data for 1,149 schools in Los Angeles County that participated in the Physical Fitness Test. For each school the percentage of 5th graders that are outside of the Healthy Fitness Zone (HFZ) for body composition is recorded. The HFZ for body composition is based on both the BMI of students, as well as a body fat percentage. The data was acquired from the California Department of Education [14]. Fig. 1 shows the locations of all of the schools that are included in the study. The percentages are shown on the map in Fig. 2. The polygons that are used are school zones related to each participating elementary school. Each block group was assigned to the school it was closest to, and each group of block groups was then aggregated into a “school zone”. The school zones will be used for the analysis of accessibility of healthy foods as well. The data on supermarkets is from Dun and Bradstreet, and was acquired from the California Department of Public Health’s Network for a Healthy California GIS Map Viewer [13]. Only those supermarkets that are classified as Large-Chain supermarkets were included in the analysis.
Figure 1. All schools participating in the Physical Fitness Test in Los Angeles County, and associated school zones
IV.
METHODS
A) Measuring Clustering of Overweight 5th Graders Fig. 2 is a map of the percentages of overweight 5th graders by school zone in LA County. It appears from a visual analysis of the map that there are clusters of high percentages in certain areas. South-central, north-central, and mid-western LA County all appear to be areas of clustering of high percentages of overweight 5th graders. Potentially, this clustering would indicate that there are environmental or community factors that play a role in explaining the prevalence of overweight adolescents.
Figure 2. Percentages of overweight 5th graders by school zone
Of course, a visual analysis of clustering is not enough to make any strong recommendations for research, or to state that there are, in fact, clusters of high percentages. To determine if the clustering that is visually apparent on the map is in fact statistically significant, there are several statistical methods that need to be implemented, including the Getis Ord General G Statistic as well as the Anselin Local Moran’s I test for clustering [2]. The purpose of the Getis Ord General G Statistic is to determine if there is spatial clustering of values in the study area, in this case across the entire county of LA. The results of running the Getis Ord General G Statistics on the school level (point feature) data are shown in Table I. The positive Z score of 2.545593 shows that there are clusters of high values in the dataset, and the low p-value means that there is less than a 0.05 chance that the clustering is random. The results of running the Getis Ord General G Statistic on the school zone features are shown in Table II. The high positive Z score of 10.139442 means that there is clustering of high values in the study area, and the extremely low p-value means that there is less than a 0.01 chance that the clustering of high values in our dataset has occurred randomly. The p-value is even lower for the school zones, meaning that the results are more reliable, although both are considered statistically significant. This means that the clustering that was observed through a visual analysis is in fact statistically significant. TABLE I. Observed General G Expected General G Variance Z Score p-value TABLE II. Observed General G Expected General G Variance Z Score p-value
GETIS ORD GENERAL G, SCHOOL LEVEL DATA 0.000034 0.000031 0.000000 2.545593 0.010909 GETIS ORD GENERAL G, SCHOOL ZONE LEVEL DATA
0.000031 0.000029 0.000000 10.139442 0.000000
Figure 3. Anselin Local Moran’s I results with Z-score rendering
It is not enough to look at the Z score though, as without knowing the p-value (or probability that the cluster has occurred randomly), Z scores are not meaningful. Fig. 4 shows only the clusters that have a Z score less than 0.05, meaning that there is less than a 5% chance that the clusters occurred randomly. Fig. 4 and Fig. 5 reveal that there are statistically significant clusters of both high values and low values in several areas of the study area. These clusters are present both when analyzing the individual schools for clusters, as well as the aggregated school zones. South-central LA definitely stands out as the largest cluster of high values, although there are several others outside of the area.
Now that it is clear that there is a statistically significant clustering of high percentages of overweight 5th graders in the study area, the analysis will benefit from understanding the local nature of those clusters. The visual analysis suggests that the clustering of high values does not have a global character (for instance, the clustering is not north to south, or east to west), but rather the clusters appear to be local in nature (for example, there are pockets of clusters throughout the study area, seemingly without a global trend). In order to test for local clusters in the study area, the Anselin Local Moran’s I was used. Fig. 3 shows the results of the test using Z scores. High positive Z scores represent local clusters of high values, and high negative Z scores represent local clusters of low values. It is clear from Fig. 3 that there are several clusters of high values throughout the study area.
Figure 4. Anselin Local Moran’s I for schools with only statistically significant hot spots of high values and low values
supermarkets has much less of an impact on childhood overweight than other variables such as the percentage of the population with a bachelors degree. TABLE III. Accessibility Income Diversity Index Education
REGRESSION RESULTS
Std. Error 0.000 0.000 0.030 7.157
Beta -0.068 -0.026 0.085 -0.310
V.
Figure 5. Anselin Local Moran’s I for school zones with only statistically significant hot spots of high values and low values
B) Measuring Accessibility of Healthy Foods In order to determine the accessibility of healthy foods in Los Angeles County, ESRI’s Network Analyst extension for ArcGIS 9.3 was used. The Closest Facility functionality was used in order to determine the distance between the centroid of each census block group and the closest large chain supermarket. The resulting distances ranged from 0.71 km to 22.5 km, with an average distance for the entire study area of 1.42 km. These census block group distances were used to calculate the average distance from the centroid of each census block group to the nearest supermarket for each school zone. These distances were then used as an explanatory variable in the regression analysis described below. C) Explaining Percentages of Overweight 5th Graders using an Accessibility Index Using the percentages of overweight 5th graders in each school zone as the dependent variable, a regression analysis will determine if the accessibility measures described in the previous section play a role in explaining overweight 5th grader prevalence. An Ordinary Least Squares regression analysis was used to determine what, if any, relationship exists between accessibility to healthy foods and overweight 5th graders while controlling for several socioeconomic and demographic variables including income, race/ethnicity, and education levels, as is discussed in the literature [4,5,7,12]. Table III shows that while the accessibility variable is statistically significant, its coefficient of -0.068 is the opposite of what would be expected based on the literature. The observed negative relationship means that as the distance from large chain supermarkets decreases, the percentage of overweight 5th graders in each school zone actually increases. This goes against what much of the literature has found, and definitely leads to the need for more analysis. The coefficient is also a very low number, meaning that the distance to
t -2.432 -0.561 1.754 -6.076
Sig. 0.015 0.575 0.080 0.000
RESULTS
The analysis of the clustering of high percentages of overweight 5th graders in LA County has revealed that there are statistically significant clusters of overweight 5th graders in various areas of the county. Understanding where there is significant clustering is a first step in the research, but it does not answer the important questions that can help lead to policy changes. Now that it is known where clusters of overweight 5th graders exist, the next issue is why clustering is happening in those areas. One hypothesis is that the accessibility of healthy food in certain areas of the county is leading to unhealthy diets in certain communities, and thus negative health outcomes for adolescents. The closest facility analysis demonstrated that accessibility defined by distance to the closest supermarket varies greatly depending on location. The range of distances of 0.71km to 22.5km is a large range the represents the potential for a major impact on the choices that people make. That being said, some of the areas with the largest distance are actually some of the highest income areas, which may be playing a role in the unexpected coefficient for accessibility. Excluding those high-income areas (which are also in much less denselypopulated areas) may result in outcomes that are closer to expectations. An attractive alternative would involve the construction of geographically weighted regression models that considered these spatial variations in income and accessibility [15]. The result of the regression analysis has shown that accessibility of healthy foods actually has a negative relationship with rates of overweight 5th graders, with a coefficient for the accessibility variable of -0.068. The negative relationship is found when accessibility is defined as the average distance to the nearest supermarket, which is only one way of defining an accessibility variable. It may become clear that rather than distance, a density variable may show the positive relationship that was initially expected. VI.
CONCLUSION
The fact that percentages of overweight 5th graders are not randomly distributed throughout the county indicates that there are local, community-level attributes that may be playing an important role in the health outcomes of adolescents. In order to understand better what those attributes are, a regression analysis was used to evaluate the relationship between accessibility of healthy foods and overweight 5th graders. A positive relationship between accessibility and healthy weight
would indicate that by increasing the accessibility of healthy foods in areas that are currently dealing with accessibility issues could have a positive impact on adolescent weight issues and all of the associated negative health outcomes. The results of the analysis showed that using distance as the measure of accessibility yields a negative coefficient. This indicates that there are potentially aspects of accessibility that need to be considered other than distance. One possible reason for this is the fact that some regions of Los Angeles county are both high income and low population density areas. These areas may be leading to the unexpected results that show that increased distance from supermarkets leads to lower percentages of overweight 5th graders. The fact that the initial results of the analysis indicate that there is a negative relationship between accessibility and overweight 5th graders leads to several other questions. Some of the issues that should be considered are changing the study area to include mainly the clusters of overweight 5th graders, for instance in South LA, instead of including the entire county. The major differences between different areas of the county may be affecting the analysis, so that should be the next step for this research. Also, it is possible that by using only largechain supermarkets some other major sources of healthy foods are missing that would change the results to be more in line with existing literature. VII. FUTURE RESEARCH One issue that should be considered is the scale of the data analyzed in this analysis. Considering school level data may miss out on some of the larger-scale community characteristics that are blurred together by looking at school zones. Individual scale analyses may find an even stronger relationship between community characteristics and health outcomes. The use of school zones based on proximity to schools, as opposed to the actual zones that are used by schools to determine enrollment may also produce different results. For the purpose of this study, with the data available, the school zones were considered the closest possible proxy. In future research, the zones used by the Department of Education would be preferred. Rather than using supermarkets, future research might look at establishments such as convenience stores and fast food restaurants as a way of determining the accessibility of unhealthy foods, rather than assuming healthy foods are bought and sold at supermarkets. By focusing on the distribution of unhealthy food the results may be very different, and certainly would provide some additional insight into the issue.
As mentioned earlier, it may also be useful to consider other regression techniques that allow for spatial variation in the relationships between variables, such as geographically weighted regression [15]. An analysis that allows the relationship between the distance to the closest supermarket and childhood overweight may reveal different results, due to the very diverse neighborhoods being considered in this study. REFERENCES [1]
[2]
[3] [4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
Apparicio P, Cloutier MS, Shearmur R: The case of Montreal’s missing food deserts: Evaluation of accessibility to food supermarkets. Int J Health Geogr 2007, 6(4). Chen Y, Yi Q, Mao Y: Cluster of liver cancer and immigration: A geographic analysis of incidence data for Ontario 1998-2002. Int J Health Geogr 2008, 7:28. Black JL, Macinko J: Neighborhoods and obesity. Nutrition Reviews 2007, 66(1):2−20. Powell LM, Auld C, Chaloupka FJ, O’Malley PM, Johnston LD: Associations between access to food stores and adolescent body mass index. Am J Prev Med 2007, 33(4S). Nelson MC, Gordon-Larsen P, Song Y, Popkin BM: Built and social environments: Associations with adolescent overweight and activity. Am J Prev Med 2007, 31(2). Larson NI, Story MT, Nelson MC: Neighborhood Environments: Disparities in access to healthy foods in the U.S. Am J Prev Med 2009, 36(1). Algert SJ, Agrawal A, Lewis, DS: Disparities in access to fresh produce in low-income neighborhoods in Los Angeles. Am J Prev Med 2006, 30(5). Liu G, Wilson J, Qi R, Ying J: Green neighborhoods, food retail and childhood overweight: Differences by population density. Am J Health Promot 2007, 21:317−325. Rose D, Richards R: Food store access and household fruit and vegetable use among participants in the US Food Stamp program. Public Health Nutr 2004, 7:1081−1088. Cheadle A, Psaty BM, Curry S, et al: Community-level comparisons between the grocery store environment and individual dietary practices. Prev Med 1991, 20:250−261. Institute of Medicine. Childhood obesity in the Unites States: facts and figures. 2004. Available at htto://www.iom.edy/Object.File/Master/22/606/FINALfactsandfigures2. pdf. Accessed 10 March 2009. Lutfiyya MN, Garcia R, Dankwa CM, Young T, Lipsky MS. Overweight and obese prevalence rates in African American and Hispanic children: An analysis of data from the 2003-2004 National Survey of Children’s Health. J Amer Board Fam Med 2008;21:191−99. Dun and Bradstreet. California Food Retailers. Network for a Healthy California. Metadata described at http://www.cnngis.org/metadata/foodretailers.htm. 2008. Educational Data Systems for the California Department of Education. California Physical Fitness Test Summary Data Files. Available at http://www.eddataonline.com/fitness/data/main.aspx?year=2008. Fotheringham, Stewart A., Brunsdon, Chris, and Charlton, Martin.. Geographically Weighted Regression: the analysis of spatially varying relationships. John Wiley & Sons, 2002.
