Background
Recent years have seen a growing attention to the significance of the built environment for changing individuals’ energy balance and weight status. Considering low-income and minority individuals are routinely found to be less engaged in physical activity than whites and given the presence of persistent residential segregation by income and race/ethnicity in the United States, it follows to argue that the built environment may contribute to socioeconomic and racial/ethnic differences in physical activity. However, little research has been done to determine whether this is, in fact, true. To begin with, spatial inequality in the built environment has not been well understood. Evidence of a national pattern as to how neighborhood income and minority compositions are linked to built environmental features is particularly lacking.

Objectives
Focusing on two specific “environmental goods,” namely parks and green spaces, this study examined ecological correlations of accessibility of parks and green spaces with neighborhood SES and minority composition at both the census tract and the county levels in the United States. Based on previous work, we hypothesized that socioeconomically deprived and/or minority-concentrated neighborhoods were underexposed to parks and green spaces “environmental goods”.

Methods
Data and Measures
This study is an ecological study of cross-sectional associations of accessibility of parks and green spaces with SES and minority composition at both the census tract and the county levels. Socio-demographic variables were constructed from the 2000 census data. Three SES variables were obtained including percent of college graduates, percent of households in poverty, and median household income. Two demographic variables were constructed including population density, defined as number of residents per square miles, and percent of rural population, reflecting a place’s position on the rural to urban scale.

A measure of park accessibility was constructed from the park GIS layer in ESRI ArcGIS9.3 Data DVD. It was created in 2008 with 35,436 public park or forest units in the 50 states and DC. The park dataset includes national, state, and local parks and forests. Park size and within-park centroids were generated. We adopted an innovative method to calculate census tract or county’s park accessibility. Specifically, we identified seven closest parks to a centroid of the focal place (i.e., tract or county) and calculated population-weighted average distance from the centroid to these seven parks as a spatial measure of park (in) accessibility.

A green space or greenery measure was derived from the tree canopy data set in the National Land Cover Database 2001. This data set provides tree canopy density at a spatial resolution of 30 meters. The tree canopy density is represented as the percentage of area covered by tree canopy within each 30m pixel. Using this data set, aggregate greenery measures were generated at the county and census tract levels. The greenery measure represents the average of the percentages of tree canopy coverage associated with pixels that fall in each geographic unit (i.e., tract or county).

Analyses
Geographic information system techniques were employed to construct accessibility measures of parks and green spaces in ArcGIS 9.3. Pearson product-moment correlation analyses and Ordinary Least Square regression analyses were performed to test our hypotheses using Stata 11. All the tract-level and county-level variables were standardized in regression models.

Results
For park accessibility, models using different geographic units did not produce qualitatively different results. The tract-level regression analyses showed that lower SES, measured by poverty rate, was inversely associated with park accessibility. However, neighborhoods with higher proportion of minority residents had better park accessibility than whiter neighborhoods. The findings from the county-level analyses echoed those from the tract-level analyses.

For green space accessibility, multi-scale analyses generated different results. At the tract-level, the results were all expected and consistent with our hypotheses. Poverty and minority concentrated neighborhoods were underexposed to green spaces. However, county-level analyses found that counties of higher poverty rates actually were better covered by green spaces and this association was not attributable to the county’s population density or rurality.

Conclusions
In this nationwide multi-scale ecological study, we expanded the environmental justice research by addressing environmental benefits at two geographic scales, census tracts and counties. The patterns revealed from our study are mixed with respect to our a priori hypotheses. While race and class are indeed important factors of spatial distribution of parks and green spaces, they do not always work in expected ways. The take-home message of this study is that poorer and minority-concentrated neighborhoods do not always lack health promoting resources and are sometimes in favorable situations compared to more advantaged and whiter neighborhoods.

Support/Funding
This research was supported by an NIH grant (R01CA140319-01A1) to the first author.