Social inequalities in neighborhood conditions: spatial relationships between sociodemographic and food environments in Alameda County, California

Abstract

Previous research suggests that neighborhoods in the United States with high concentrations of poverty or of racial/ethnic minorities have lower access to healthy foods and greater access to unhealthy foods, compared with higher income or predominantly White, non-Hispanic neighborhoods. Lower access is thought to influence dietary habits and resulting health consequences, such as obesity. While most studies have focused on either neighborhood SES or features of the built environment, few have explicitly examined both. Using data from the Geographic Research on Wellbeing study, we map the spatial relationships between sociodemographic characteristics (poverty trajectories, racial/ethnic/nativity composition) and food environments in Alameda County, California. Our map presents poverty trajectories and racial/ethnic/nativity composition at the tract level, as well as maps depicting accessibility to healthy, unhealthy, and a composite of both, based on rasterized maps and a network analysis of food types within a quarter-mile walking distance. We found that neighborhoods that have experienced long-term poverty have the greatest access to both healthy and unhealthy food outlets compared with more economically advantaged neighborhoods. We also found that predominantly Black/Latino neighborhoods had the greatest access to healthy foods compared with other neighborhoods with a different race/ethnicity/nativity composition. Neighborhoods experiencing long-term affluence, as well as predominantly White neighborhoods, had the lowest access to any of the food types, which likely reflects their surburban locations. Results suggest that spatial relationships between sociodemographic characteristics and food access at the neighborhood level depend upon place and urbanization.

Keywords:

• social inequality
• food environment
• spatial relationship

1. Introduction

Obesity in the United States is epidemic and getting worse, contributing to many chronic conditions. Each year, 400,000 deaths and $117 billion in costs are attributable to obesity (Bassett & Perl, 2004). The problem is particularly evident in Black and Latino populations, as well as populations of low socioeconomic status. Although many factors contribute to obesity, recent epidemic increases suggest major roles for environmental factors, broadly defined to include the socioeconomic and built environment, resulting in an obesity-promoting society (Bassett & Perl, 2004; Hill & Trowbridge, 1998).

The United States is a highly segregated society, whereby racial and ethnic groups, and the rich and poor, tend to live in separate neighborhoods (Booza, Cutsinger, & G, 2006; Iceland, Weinberg, & Steinmetz, 2002). Living in socioeconomically disadvantaged neighborhoods has been hypothesized to increase the risk of unhealthy behaviors in part because of lower exposure to health-promoting residential environments (e.g., healthy affordable food and availability of recreational activities), and/or higher exposure to health-compromising residential environments (e.g., crime and density of fast food restaurants), compared with more advantaged neighborhoods (Block, Scribner, & DeSalvo, 2004; Booth et al., 2001; Cubbin, Hadden, & Winkleby, 2001; R. Lee, Booth, Reese-Smith, Regan, & Howard, 2005; R. E. Lee, Cubbin, & Winkleby, 2007; Morland, Wing, Diez Roux, & Poole, 2002; Sallis & Glanz, 2006). In the absence of direct measures of specific neighborhood characteristics, some studies in the United States have primarily relied upon census-based measures to characterize neighborhoods (e.g., poverty rates), providing evidence that neighborhood characteristics are independently associated with dietary habits, physical activity, and obesity (Cubbin et al., 2001; Cubbin et al., 2006; R. E. Lee et al., 2007). Other research has examined how the physical characteristics of neighborhoods – also referred to as the ‘built environment’ – affect dietary habits, physical activity, and obesity (Booth et al., 2001; Sallis & Glanz, 2006). While most studies have focused on either neighborhood SES or features of the built environment, few have explicitly examined how both jointly influence behaviors.

Using data from the recently funded Geographic Research on Wellbeing Study, we mapped the spatial relationships between sociodemographic characteristics (poverty trajectories, racial/ethnic/nativity composition) and food environments in Alameda County, California, which is located in the San Francisco bay area.

2. Methods

2.1 Data sources

Data on currently existing food outlets in Alameda County (accessed in October, 2011 and last updated in September, 2011), came from Reference USA (http://www.referenceusa.com). We categorized food outlets as either ‘healthy’ (based on their high probability of stocking nutritious food options) or ‘unhealthy’ (based on their high probability of stocking energy-dense, nutrient poor foods). ‘Healthy’ food outlets included: Fruit and vegetables markets (Standard Industrial Classification, or SIC, code 543101), grocery stores (SIC code 541105), and food markets (SIC code 541101). ‘Unhealthy’ food outlets included: Fast food (a subset of SIC codes 581206 and 581208), pizza places (a subset of SIC codes 581206, 581208, and 581222), and convenience stores (SIC code 541103). For examples of the types of outlets included in each category, see Table 1.

Table 1. Examples of types of healthy and unhealthy food outlets.

Healthy food outlets	Unhealthy food outlets
• Fruit and vegetable markets • Grocery stores (e.g., Safeway, Lucky, Grocery Outlet) • Food markets (miscellaneous small markets)	• Fast food (e.g., Taco Bell, Wendy's, McDonald's) • Pizza places (e.g., Domino's, Pizza Hut) • Convenience Stores (e.g., 7-Eleven, Circle K)

Data on sociodemographic characteristics came from the Neighborhood Change Database (NCDB) (http://www.geolytics.com/USCensus,Neighborhood-Change-Database-1970-2000,Products.asp) and the 2005–2009 American Community Survey (ACS) (http://www.census.gov/acs/www/). The NCDB was used to calculate poverty rates (percentage of persons with incomes below the federal poverty level) at the census tract level from 1970 to 2000, which had been normalized to census 2000 boundaries so that comparisons can be made on the same geographic boundaries with the ACS data. ACS data were used to calculate poverty rates and racial/ethnic/nativity composition (percent Asian, not Hispanic/Latino; Black, not Hispanic/Latino; Hispanic/Latino; White, not Hispanic/Latino, and percent foreign-born) from 2005–2009. The Reference USA food outlets were geocoded using the latitude and longitude coordinates from Reference USA. Street centerlines were obtained from the 2010 TIGER\Line shapefiles and tract boundaries were obtained from the 2011 TIGER\Line shapefiles.

2.1.1 Analyses

Latent class growth modeling (LCGM) was used to define poverty trajectories at the tract level. Using fixed slopes and intercepts, LCGM identifies distinct subgroups of the sample that follow a similar pattern of change over time on any given variable, in this case, poverty rates (Andruff, Carraro, Thompson, Gaudreau, & Louvet, 2009). In addition, latent class analysis (LCA) was used to identify subgroups based on racial/ethnic/nativity compositions. LCA is a statistical method for identifying unobserved subgroups of the sample using observed variables. We found three distinct poverty trajectories (stable, affluent; stable, moderate poverty; stable, concentrated poverty) and four distinct classes of racial/ethnic/nativity composition (non-Hispanic Black and Hispanic or Latino [‘Black/Latino’]; predominantly Hispanic or Latino and high proportion of foreign-born [‘Latino/immigrant’]; mixed race/ethnicity and high proportion of foreign-born [‘Mixed/immigrant’]; and predominantly non-Hispanic White [‘White’]).

To calculate food accessibility zones, we selected ‘walkable’ streets according to their Master Tiger Feature Class Code (S1400, S1500, S1640, S1710, S1730, S1740, S1750, S1780), basically excluding primary and secondary roads (http://www.census.gov/geo/www/tiger/tgrshp2010/TGRSHP10SF1AF.pdf). We then created an accessible quarter-mile network area for each of the six categories of food outlets (three healthy, three unhealthy): The network area was rasterized and re-classified as a ‘1’ if it was within a quarter-mile network walk of the food outlet category and ‘0’ if it was not. Quarter-mile areas were then added together for the three healthy categories and three unhealthy categories (range 0–3 for each, depending on how many areas overlaid each other). The resulting healthy food outlet raster is thus a visualization of the density of types of healthy food outlets that are accessible within a quarter-mile walk. It does not distinguish between the number of outlets within the same category (e.g., if the area was within two grocery stores, that area would still be assigned a ‘1’ for that category). The ‘composite’ map was created by subtracting the healthy raster from the unhealthy raster, so that a value of −3 would indicate that the area had walkable access to all three categories of healthy food outlets (and no walkable access to the unhealthy food outlet categories). Finally, we calculated the mean composite accessibility score (−3 to 3) for each tract from the underlying raster to come up with a tract-based accessibility score.

Table 2 presents the means of the healthy, unhealthy and composite accessibility scores across census tracts within each poverty trajectory and race/ethnicity/nativity class. The mean healthy score was highest in the stable, concentrated poverty class, indicating that census tracts in this class had, on average, highest accessibility (via walking) to the three categories of healthy food outlets. The mean healthy score was second highest in the stable, moderate poverty class and lowest in the stable, affluent class. ANOVA comparisons indicated that the mean healthy score differed significantly across poverty trajectory classes (F = 83.635, p < 0.001), and post-hoc comparisons showed significant differences between each class. The mean unhealthy score was also highest in the stable, concentrated poverty class; second highest in the stable, moderate poverty class; and lowest in the stable, affluent class. Unhealthy scores also differed significantly across poverty trajectory classes (F = 17.501, p < 0.001); however, post-hoc analyses demonstrated that only the stable, affluent class differed significantly from the other two classes. The mean composite score was negative in the stable, concentrated poverty and stable, moderate poverty classes, indicating somewhat greater accessibility to healthy compared with unhealthy outlets. In contrast, the mean composite score was positive in the stable, affluent class, indicating greater accessibility to unhealthy outlets. Results of ANOVA comparisons across classes were significant (F = 11.062, p < 0.001), but post-hoc analyses demonstrated that only the stable, affluent class differed significantly from the other two classes.

Table 2. Number of census tracts (N) within poverty trajectory and race/ethnicity classes and mean (SD) Healthy Accessibility Score, Unhealthy Accessibility Score, and Composite Accessibility Score across census tracts within each class.

	N	Healthy Mean (SD)	Unhealthy Mean (SD)	Composite Mean (SD)
Poverty trajectory class
Stable, concentrated poverty	31	0.84 (0.46)	0.72 (0.66)	−0.11 (0.64)
Stable, moderate poverty	77	0.54 (0.44)	0.52 (0.49)	−0.02 (0.44)
Stable, affluent	213	0.17 (0.23)	0.31 (0.33)	0.14 (0.24)
Race/ethnicity class
Black/Latino	52	0.58 (0.41)	0.39 (0.38)	−0.18 (0.45)
Latino/immigrant	35	0.45 (0.48)	0.44 (0.33)	−0.01 (0.34)
Mixed/immigrant	97	0.24 (0.38)	0.40 (0.51)	0.16 (0.29)
White	137	0.25 (0.31)	0.40 (0.42)	0.15 (0.33)

Healthy score means were highest in the Black/Latino tracts and lowest in the White and Mixed/immigrant tracts (Table 2). ANOVA analyses showed that the mean Healthy score differed significantly between racial/ethnic/nativity classes (F = 12.640, p < 0.001). Post-hoc analyses demonstrated that the White class differed significantly from the Black/Latino and Latino/immigrant classes, and that the Black/Latino class differed significantly from the Mixed/immigrant class. Unhealthy score means were similar and did not differ significantly across classes (F = 0.096, p = 0.962). Composite score means were negative in the Black/Latino class (indicating greater accessibility to healthy outlets), close to zero in the Latino/immigrant class (indicating equal access or no access to both), and positive in the White and Mixed/immigrant classes (indicating greater accessibility to unhealthy outlets). These scores differed significantly across classes (F = 14.446, p < 0.001). Post-hoc analyses showed that the Black/Latino class differed significantly from the White and Mixed/immigrant classes.

2.1.2 Map production

Our map presents poverty trajectory classes, racial/ethnic/nativity classes, and healthy, unhealthy, and composite accessibility scores, all at the census tract level, for Alameda County, California. An additional map depicts the gridded composite accessibility surface. We also include a table with examples of the healthy and unhealthy food outlet categories, and mean values of healthy, unhealthy, and composite accessibility scores, according to poverty trajectory and racial/ethnic/nativity class.

3. Conclusions

Results from previous research in the United States suggest that neighborhoods with high concentrations of poverty or high concentrations of racial/ethnic minorities have lower access to healthy foods and greater access to unhealthy foods, compared with higher income or predominantly White, non-Hispanic neighborhoods (Block et al., 2004; Booth et al., 2001; Cubbin et al., 2001; R. Lee et al., 2005; Morland et al., 2002; Sallis & Glanz, 2006), which may influence dietary habits and resulting health consequences, such as obesity (Cubbin et al., 2001; Cubbin et al., 2006; Diez-Roux et al., 1999; R. E. Lee & Cubbin, 2002). Our results from Alameda County, California, suggest however, that those spatial relationships may depend on place and urbanization. We found that neighborhoods that have experienced long-term poverty have the greatest access to both healthy and unhealthy food outlets compared with more economically advantaged neighborhoods, suggesting perhaps that access is only one determinant of dietary habits, and that other factors may play important roles as well, such as time, convenience, and cost. As well, we also found that predominantly Black/Latino neighborhoods had the greatest access to healthy foods compared with other neighborhoods with a different composition based on race/ethnicity/nativity, but that each type of neighborhood had about equal access to unhealthy foods. Stable, affluent neighborhoods, predominantly White neighborhoods, and Mixed/immigrant neighborhoods had the ‘worst’ accessibility scores (on average, higher access to unhealthy compared with healthy foods). For the first two groups, this is probably reflected by their location in car-dependent, suburban locations, where fast food restaurants and convenience stores may be quite prevalent, with full-service grocery stores located further away from residential areas. The Mixed/immigrant neighborhoods in highly urban areas then may be most in need of targeted efforts to increase access to foods that promote healthy dietary habits. One limitation to our analysis is that while we had information on food outlets that we categorized as likely to offer relatively healthy or unhealthy foods, we were not able to measure what foods were actually offered. It may be that in the most economically disadvantaged or predominantly Black/Latino neighborhoods, ‘healthy’ food outlets did not stock substantial amounts of high quality, affordable healthy foods relative to the other offerings at their stores. If this is the case, the physical access to healthy foods for residents of those areas projected on the maps is an overestimate of actual access. Previous research has found mixed results on the cost of food according to neighborhood economic disadvantage or racial/ethnic composition (Andreyeva, Blumenthal, Schwartz, Long, & Brownell, 2008; Ball, Timperio, & Crawford, 2009; Dunn, Sharkey, Lotade-Manje, Bouhlal, & Nayga, 2011; Krukowski, West, Harvey-Berino, & Elaine Prewitt, 2010; Leone et al., 2011; Sooman, Macintyre, & Anderson, 1993; Troutt, 1993; Warren, Cubbin, & Winkleby, 2008; Zenk et al., 2006). Alameda is the seventh most populous county in California, with 14 incorporated cities, including Oakland, its largest city, and several unincorporated communities. The total population is estimated to be 1,510,271 as of April 2010. The county has varied geography and is characterized by a rich diversity with no majority racial/ethnic group and 53 languages spoken in the public school system in 2008–2009 [http://www.acgov.org/about/, accessed 6/21/2012]. Thus, our results may reflect relations between poverty, race/ethnicity/nativity and food access that are unique to Alameda County, but that may not prevail in less dense or less diverse cities, rural areas, or other regions of the United States.

Software

Statistical analyses were performed using SAS version 9.2 (SAS Institute Inc.) and Mplus version 6.12 (Muthén and Muthén). GIS network analysis, raster overlays, and maps were produced using ArcGIS 10 software from ESRI.

Acknowledgements

This work was supported by a grant from the American Cancer Society (RSGT-11-010-01-CPPB) to C. Cubbin. We thank Nicolas Welch for technical assistance on the map production.