Assessing the socio-demographic dynamics of urban agriculture development in New York City, Reynolds notes that while low-income communities, immigrants and people of color often bring significant knowledge, energy and enthusiasm to the development and maintenance of gardens, these groups tend to have less access to the resources, networks and cultural capital required to build and defend community gardens in the urban landscape. As earlier chapters have demonstrated, in order to attract resources and legal status for their gardens, community garden programs must legitimize themselves according to some of urban agriculture’s potential benefits at the expense of others. In all three cities, urban agriculture advocates have made claims about the role of gardens in helping people in need, but they have also emphasized arguments about the economic benefits of community gardens. Economic benefits like neighborhood development and elevated property values can be in tension with serving the needs of the marginalized, whose interests are often left behind in the flow of capital through cities . On the one hand, community gardens may be easier to establish where vacant land is more abundant, that is, in neighborhoods with depressed property values—often those with higher proportions of people in poverty, immigrants, and/or residents of color. On the other hand, marginalized communities may have a harder time marshalling the resources needed to defend community gardens from rising property values and increased neighborhood development, if and when these potential economic benefits of urban agriculture materialize. This tension is ubiquitous in urban agriculture . However, ebb and flow flood table researchers have to date paid little attention to the role that citywide community gardening organizations can play in mitigating neighborhood inequalities by amassing and equitably distributing the resources needed to build, maintain, and defend urban agricultural spaces.
In this chapter, I draw on historical datasets developed from review of organizational documents for each city to conduct a longitudinal spatial analysis of garden accessibility. To fully understand who is benefitting most from community gardens, multiple types of data ought to be considered. Ethnographic research or extensive surveys would be needed to determine who is actually using the gardens within a program, how much nutritional, recreational, social and cultural benefit participants are receiving, and what collective benefits the local community is realizing from the presence of a garden. With the exception of a few documents summarizing survey results from Seattle’s P-Patches in the 1990s, my data do not provide this type of detail about usage or measured outcomes. However, extensive review of the historical documents from each program does enable another important approach to understanding equity in garden access: the proximity of gardens to different neighborhoods. Mapping the gardens that each program invested in over time and using spatial analysis to assess the gardens’ accessibility to marginalized communities, we can understand the historical trajectory of each program’s impact on the urban environment, a perspective that would not be possible with ethnographic or cross-sectional survey methods. A spatial approach is especially relevant for understanding urban agriculture as a land use as well as a social practice. Urban researchers have used spatial analysis to assess whether community gardens are alleviating food deserts and to identify the neighborhoods in a city which would benefit most from urban agriculture , but have yet to analyze the extent to which existing community gardens in a city actually serve the neighborhoods with the highest need.
In this chapter, I first summarize the methodology used to map the gardens in each city over time. Then, I describe the results of my spatial analysis in detail, connecting them to key points from the qualitative historical analysis laid out in preceding chapters. I conclude by highlighting the ways that organizational decisions over time are evident in how gardens are and have been distributed across each city.In order to show how the citywide gardening programs in Milwaukee, Philadelphia and Seattle have expanded over time and how accessible their gardens have been to marginalized communities, I built an original historical dataset2 , mapped the gardens that were associated with each program in 1980, 1990, 2000, 2010, and 2019, and conducted a series of spatial analyses on the relationships between garden locations and neighborhood demographic characteristics. During my review of documents from the main garden programs in each city, I compiled a database with the name, location, and years active for each garden mentioned over the programs’ histories. Records such as annual reports and garden maps tended to provide complete snapshots of the gardens included in a program at a particular time, while newsletters and newspaper articles offered supplementary information to date the creation or closure of some gardens. Together, the documents available for each city furnished enough information for a detailed, if not perfectly complete, picture of how the programs expanded in urban space as their budgets grew and they were able to develop new gardens—and how and where the programs contracted under the pressure of changing budgetary or real estate market conditions.In order to understand the relative accessibility of each programs’ gardens for marginalized groups, I acquired neighborhood demographic information for each city at the Census tract level. Using Geolytics, I downloaded a dataset with relevant variables for 1980, 1990, 2000, and 2010 fit to the 2010 Census boundaries. Using the software program R, I then downloaded equivalent values from the 2015-2019 American Community Survey . Given the salience of urban agriculture’s potential benefits for immigrants, low-income and people of color, I obtained counts and calculated percentages for each tract’s poverty rate, percent foreign born, percent nonHispanic white, percent non-Hispanic Black, percent Hispanic, and percent non-Hispanic Asian or Pacific Islander.
Census questions about racial and ethnic categories have changed slightly over the last 50 years, and the groups above were chosen for this study because they can be calculated consistently across the 5 decades of interest while speaking to the patterns of racial inequality and marginalization most commonly observed in US cities. Because the ability to create, maintain and preserve community gardens is influenced by socioeconomic characteristics such as real estate values and supporters’ cultural capital, I also obtained tractlevel data on education levels , median household income, and median monthly housing costs. After compiling a dataset with the independent variables of interest for the 2010 Census tracts across all 5 decades, I calculated measures of garden accessibility for each Census tract in each decade. I used the Google API to geocode the garden addresses into latitude and longitude, and then georeferenced the coordinates to align with the Census tract coordinate reference system. Overlaying the gardens’ geographic information onto the 2010 Census tracts, I obtained counts for the number of gardens in each tract in 1980, 1990, 2000, 2010 and 2019. Since most tracts had zero gardens and very few had more than one, this measure had significant skew; I then created a binary variable indicating whether a tract contained at least one garden in a given year. There is a great deal of variation in the size of Census tracts, hydroponic drain table and the boundaries between tracts do not represent firm restrictions on residents’ activities. To address these concerns, I created additional dependent variables based on distance rather than tract boundaries. For each tract and year, I calculated the distance from the tract centroid to the nearest garden, and I created another binary variable indicating whether at least one garden was within a one-mile radius of the tract centroid. Before mapping and modeling garden accessibility, I conducted exploratory data analysis to refine my variable specification. I ran correlations of all variables and found several strong correlations that risked weakening the models through multicollinearity. First, the two variables for education were strongly negatively correlated. I chose to model percent with a college degree and leave out percent with less than a high school education, given the role of cultural capital in successful creation and preservation of community gardens that earlier studies have identified . Next, median household income was strongly positively correlated with housing costs and strongly negatively correlated with poverty rates, but the correlation between housing costs and poverty rates tended to be much weaker. I chose to include housing costs and poverty rates in the models while removing household income to reduce multicollinearity.
Retaining the poverty and housing variables, both the accessibility of gardens for low-income communities and the threat to gardens from high land values can be represented in the model. Due to a consistently strong negative correlation between percent white and percent Black, I chose to remove percent white from the models and retain focus on gardens’ proximity to people of color. I also found strong positive correlations between percent foreign born and percent Hispanic in Milwaukee and Philadelphia, and between percent foreign born and percent Asian or Pacific Islander in Seattle. Due to the theoretical importance of understanding garden accessibility both for racial minorities and for immigrants, I chose to retain all three variables in my models and test the outcomes when each of them was removed to see if multicollinearity was impacting the results.After testing for multicollinearity, I tested for spatial autocorrelation—that is, whether high or low values for any of the variables were clustered in adjacent Census tracts. For each city, I made three matrices defining neighboring tracts: queen contiguity, 2-nearest, and 3- nearest neighbor weights matrices. Then I calculated Moran’s I for all variables in each city and year, using each of the three neighbor weights matrices. Regardless of the matrix used, Moran’s I values were greater than 0.3 for almost all of the independent variables, indicating substantial spatial autocorrelation. In other words, neighborhoods show clustering in characteristics such as poverty rates, racial and ethnic composition, and education levels. This finding is unsurprising, given what we know of neighborhood effects and the legacies of residential racial segregation, yet it is important to note due to its potential impact on any spatial models. For all cities, years, and neighbor weights matrices, Moran’s I showed significant spatial autocorrelation in the distance based measures of garden accessibility . However, the tract-boundary measures of garden access had Moran’s I values close to 0 in Milwaukee and Seattle for all years, indicating that the gardens themselves are not generally clustered in these cities. Only Philadelphia appeared to have statistically significant clustering in the locations of gardens. Mapping the dependent variables for each city and year showed that much of the clustering in the distance-based measures of garden accessibility was due to a complete lack of gardens in certain areas of the city, where adjacent tracts logged progressively larger distances to the nearest garden. Figure 1 illustrates the typical appearance of this pattern, with the areas of northwest and south Milwaukee and northeast and south Philadelphia hosting zero gardens from their cities’ respective garden programs. The clustering of distance-based garden accessibility variables in Seattle is not as visible when mapped, but it nonetheless registered as significant in the Moran’s I tests for all neighbor weights matrices and years.Given that garden programs are working to administer multiple sites across a city with limited resources, the lack of gardens in far-flung regions may be understandable. Still, when large areas of a city remain unserved by a citywide program, the lack of service to these areas is notable. For this reason, I chose not to treat the far-flung tracts as “outliers” and remove them from the models altogether. However, in practical terms, the progressively larger distances to the nearest garden that result from this pattern can skew the dependent variable in a way that interferes with the overall model fit and accuracy. Therefore, I developed a “corrective” variable giving the distance from each tract centroid to City Hall, a measure approximating the resources required to travel to the tract from garden program offices5 . I chose to run models with and without distance to downtown in order to assess how well it corrected for skew from the far-flung tract values and whether it impacted results in any other way.Given the potential impact of multicollinearity and spatial autocorrelation on regression modeling, I ran a series of Ordinary Least Squares regressions and diagnostics to test the impact of controlling for distance to downtown, to assess the influence of correlations between race and immigration variables, and to determine whether OLS or spatial models would be more accurate.