Spatial Analysis Of Inequity: Methods, Models, and Case Studies Jayajit Chakraborty Professor of Geography Department of Sociology and Anthropology University of Texas at El Paso El Paso, Texas, USA jchakraborty@utep.edu 1
BACKGROUND: WHAT IS ENVIRONMENTAL JUSTICE (EJ)? The EJ movement in the U.S. originated in the struggles of Black and Hispanic residents against toxic waste dumps and hazardous waste facility siting in their communities. Environmental (in)justice now defined in multiple ways: Unequal exposure of racial/ethnic minorities and low-income individuals to environmental pollution and adverse health risks. Unequal environmental protection and environmental quality through laws, regulations, governmental programs, enforcement, and policies. 2
QUANTITATIVE EJ RESEARCH Focus on testing the inequity hypothesis: whether environmental risk burdens are distributed evenly across society, or if individuals of racial/ethnic minority and lower socioeconomic status are disproportionately exposed. Based on the application of various statistical and spatial analytical techniques, including geographic information systems (GIS). Evidence from 25+ years of EJ research in the U.S.: Most studies have found unequal exposure to various types of environmental health hazards with respect to race, ethnicity, and poverty status, at the national scale and in specific urban areas. 3
PRESENTATION OUTLINE Spatial Inequity Analysis: Methods, Models, & Case Studies Proximity and Exposure to Environmental Risk Population Characteristics of Areas at Risk Emerging Geostatistical Techniques Limitations and Recommendations 4
PROXIMITY AND EXPOSURE TO ENVIRONMENTAL RISK Methods used in previous EJ studies to measure proximity or potential exposure can be classified into three categories: Spatial Coincidence Analysis Distance-Based Analysis Pollution Plume Modeling 5
SPATIAL COINCIDENCE ANALYSIS Potential exposure to environmental hazards defined spatially by boundaries of pre-defined administrative or census units (e.g., ZIP codes, census tracts, statistical local areas) containing a hazard. Most widely used method referred to as unit-hazard coincidence: Identify exact locations of environmental hazards on a map. Classify spatial units based on presence or absence of a hazard. Compare socio-demographic characteristics of units containing a hazard (host units) to those that do not (non-host units). Case studies from EJ research literature: United Church of Christ (UCC) 1987; Burke 1993; Hird 1993; Anderton et al. 1994; Goldman & Fitton 1994; Been 1995; Been & Gupta 1996; Cutter et al. 1996; Boer et al. 1997; Daniels & Friedman 1999; Fricker & Hengartner 2001; Boone 2002; Taquino et al. 2002; Walker et al. 2006; Baden et al. 2007; UCC 2007; Burwell-Naney et al. 2013. 6
UNIT-HAZARD COINCIDENCE : EXAMPLE USING CENSUS TRACTS 7
SPATIAL COINCIDENCE ANALYSIS: EXTENDING UNIT-HAZARD COINCIDENCE Instead of treating all host spatial units equally, several EJ studies have extended the basic approach by estimating: Total number or density of hazards: Burke 1993; Cutter & Solecki 1996; Ringquist 1997; Tiefenbacher & Hagelman 1999; Fricker & Hengartner 2001; Mennis & Jordan 2005. Total quantity of emitted pollutants: Bowen et al. 1995; Krisel et al. 1996; Boer et al. 1997; Tiefenbacher & Hagelman 1999; Daniels & Friedman 1999; Bolin et al. 2000. Toxicity-weighted quantity of pollutants: Bowen et al. 1995; Perlin et al. 1995; McMaster et al. 1997; Brooks & Sethi 1997; Bolin et al. 2000; Ash & Fetter 2004; Abel 2009; Sicotte & Swanson 2007. 8
SPATIAL COINCIDENCE ANALYSIS: EXTENDING UNIT-HAZARD COINCIDENCE Addressing the edge effect problem: Extending the boundary of the host spatial unit by a specific distance (e.g., 1 km) to include facilities located near the edge or border of the unit. Can be used to estimate number of sites, volume of emitted pollutants, or toxicity-weighted volume of pollutants. Case study example: Australia s first national level quantitative environmental justice assessment of industrial air pollution (Chakraborty and Green 2014) 9
SPATIAL COINCIDENCE ANALYSIS: EXTENDING UNIT-HAZARD COINCIDENCE 10
SOCIAL CHARACTERISTICS OF COMMUNITIES WITH AND WITHOUT NATIONAL POLLUTANT INVENTORY (NPI) SITES 11
CHARACTERISTICS OF COMMUNITIES WITH AND WITHOUT NATIONAL POLLUTION INVENTORY (NPI) SITES 12
SPATIAL COINCIDENCE ANALYSIS General assumptions and limitations: All individuals residing in a host spatial unit is equally exposed to pollution. Only those individuals residing in (or near) the host unit are exposed to pollution. Exact location of pollution source within the host unit not considered. Potential exposure to pollution is distributed uniformly within and confined to the boundary of the host spatial unit. 13
DISTANCE-BASED ANALYSIS: CIRCULAR BUFFER Address limitations of spatial coincidence approach by constructing buffers of uniform radius around pollution sources. 14
DISTANCE-BASED ANALYSIS: CIRCULAR BUFFER Buffer radii in EJ studies have ranged from 100 yards (91.4 meters) to 3 miles (4.8 km), but distances of 0.5 (0.8 km) and 1.0 mile (1.6 km) used most frequently: Glickman 1994; Zimmerman 1994; U.S. GAO 1995; Chakraborty and Armstrong 1997; Neumann et al. 1998; Bolin et al. 2000; Baden and Coursey 2002; Boone 2002; Harner et al. 2002; Mohai and Saha 2006; Maantay 2007; Kearney and Kiros 2009; Mohai et al. 2009. Several studies have used multiple circular rings at increasing distances from hazard sources: Neumann et al. 1998; Perlin et al. 1999; Sheppard et al. 1999; Atlas et al. 2002; Perlin et al. 2002; Fitos 2004; Pastor et al. 2004; Walker et al. 2006 ; Chakraborty & Zandbergen 2007; Wang 2010; Burwell-Naney et al. 2013. 15
DISTANCE-BASED ANALYSIS: CIRCULAR BUFFER Advantages: Does not assume that the exposure effects are restricted solely to the boundaries of spatial units hosting the hazard. Easily implemented and visualized using GIS software. Makes statistical comparisons between potentially exposed (inside circle) and non-exposed (outside circle) areas convenient. Limitations: Buffer radius is selected arbitrarily and identical for all hazards. Assumes adverse effects of a hazard are limited only to the specified circular area or distance. Properties, quantities, and operational parameters of toxic emissions rarely considered. Adverse effects are equal and uniform in all directions (isotropic) from the hazard. 16
POLLUTION PLUME MODELING Geographic plume analysis: Integrate air dispersion modeling with GIS to accurately estimate areas and populations exposed to air releases of toxic substances (Chakraborty and Armstrong 1996; 1997; 2004). Dispersion models combine data on the quantity and physical properties of a chemical with information on circumstances of release and local meteorological conditions to estimate pollutant concentrations downwind from an emission source and delineate a plume footprint. 17
GEOGRAPHIC PLUME ANALYSIS: CASE STUDIES Sources: Chakraborty & Armstrong 1996, 1997, 2004. 18
POLLUTION PLUME MODELING Various pollutant fate-and-transport models used in EJ studies: Areal Locations of Hazardous Atmospheres (ALOHA): Chakraborty & Armstrong 1996, 1997, 2001, 2004; Chakraborty 2001; Margai 2001. Industrial Source Complex Short Term (ISC-ST) model: Dolinoy & Miranda 2004; Fisher et al. 2006; Maantay 2007; Maantay et al. 2009. Noise pollution model: Chakraborty et al. 1999; Most & Sengupta 2004. 19
POLLUTION PLUME MODELING Advantages: Allows concentration of toxic pollutants released from a hazard source and their health risks to: (a) vary according to wind direction; and (b) decline continuously with increasing distance from the emitting source. Addresses problems of assuming that residing either within a spatial unit containing a hazard (spatial coincidence) or a specific distance from a hazard (distance-based) results in potential exposure and health risks. Limitations: Dispersion models typically required large volumes of data on emission parameters, as well as site-specific and facility-specific information. Some models assume topography is flat and do not provide accurate concentration estimates when atmosphere is stable or wind speeds are low. Creation of plume modeling data to include all toxic facilities and chemical emissions in a large area is a time-consuming and expensive process. 20
PRESENTATION OUTLINE Spatial Inequity Analysis: Methods, Models, & Case Studies Proximity and Exposure to Environmental Risk Population Characteristics of Areas at Risk Emerging Geostatistical Techniques Limitations and Recommendations 21
POPULATION CHARACTERISTICS OF AREAS POTENTIALLY EXPOSED TO RISK Methods used to estimate the number and characteristics of people exposed to risk can be classified into two broad categories: Point Interpolation Areal Interpolation 22
POINT INTERPOLATION Can be used only when the addresses of all individuals or households relevant to the study are available and can be located as points on a map. Case studies from EJ research: Mohai & Bryant 1992; Chakraborty & Armstrong 2001; Bevc et al. 2007; Chakaborty & Zandbergen 2007; Mohai et al. 2009. 23
AREAL INTERPOLATION Becomes necessary when population data are available at the level of pre-defined administrative units or census areas. GIS-based techniques used in previous EJ studies: polygon containment, centroid containment, and buffer containment. 24
GIS-BASED AREAL INTERPOLATION TECHNIQUES Polygon Containment Centroid Containment Buffer Containment Aggregation of census units either within or in contact with the buffer. Also known as boundary intersection method. Variation: cut-off criteria to limit census units partially enclosed (e.g., 50% area containment). Aggregation of census units whose geometric centers (centroids) fall within the buffer. Assumes a point (centroid) represents entire spatial unit in terms of population characteristics. Population of each census unit weighted by the % of its area inside the buffer. Assumes uniform distribution of population characteristics within unit. Also known as areal apportionment. 25
DASYMETRIC MAPPING Use ancillary data (e.g., land use/land cover) to redistribute population in a more accurate and logical manner. Cadastral dasymetric mapping found in recent studies to represent a substantial improvement on the use of aggregated census data: Maantay et al. 2008; Maantay & Maroko 2009; Montgomery & Chakraborty 2013. 26
PRESENTATION OUTLINE Spatial Inequity Analysis: Methods, Models, & Case Studies Proximity and Exposure to Environmental Risk Population Characteristics of Areas at Risk Emerging Geostatistical Techniques Limitations and Recommendations 27
WHY USE GEOSTATISTICAL TECHNIQUES? Most EJ studies use traditional statistical methods such as correlation or regression to examine the effect of relevant independent variables (e.g., race, ethnicity, and measures of socioeconomic status) on a dependent variable representing environmental risk (e.g., presence/absence of hazard sources, distance to hazard source, quantity of emitted pollutants, or estimates of health risk). Recent studies have questioned whether these traditional or non-spatial statistical methods are appropriate for analyzing spatial data and relationships relevant to EJ. 28
PROBLEMS WITH CONVENTIONAL STATISTICAL ANALYSIS OF SPATIAL DATA FOR EJ RESEARCH Spatial Dependence Classical statistical tests such as correlation or regression assume independently distributed observations and errors, but observations from nearby locations are often more similar than what can be expected on a random basis (positive spatial autocorrelation). Spatial Heterogeneity The use of a single or global regression model for the entire study area assumes model parameters or statistical relationships between the dependent and explanatory variables do not vary spatially within a study area (spatial homogeneity). 29
ADDRESSING SPATIAL DEPENDENCE AND NONSTATIONARITY Spatial autoregressive (SAR) models: consider spatial autocorrelation as an additional variable in the regression equation and estimate its effect simultaneously with the effects of other explanatory variables (Anselin 2005). A spatial weights matrix used to specify, for each location, which other locations are neighbors and potentially influence values at that particular location. Geographically weighted regression (GWR): a local spatial statistical technique for exploring how relationships differ from place to place within a study area (Fotheringham et al. 2002). Instead of generating a single global regression equation for an entire study area, a separate regression equation or a unique set of parameters is produced for each observation or spatial unit. 30
GWR EXAMPLE: EJ ANALYSIS OF EXPOSURE TO HAZARDOUS AIR POLLUTANTS IN FLORIDA, USA Dependent Variable: estimated cancer risk from exposure to hazardous air pollutants (census tract data ) Percent Hispanic Percent Below Poverty Persons per sq. mile Source: Gilbert & Chakraborty 2011. 31
PRESENTATION OUTLINE Spatial Inequity Analysis: Methods, Models, & Case Studies Proximity and Exposure to Environmental Risk Population Characteristics of Areas at Risk Emerging Geostatistical Techniques Limitations and Recommendations 32
SPATIAL INEQUITY ANALYSIS: CURRENT LIMITATIONS AND FUTURE NEEDS 1. Most studies rely on the latest environmental and census population data (outcome inequity): Need to examine historical data on environmental risks and population characteristics to evaluate the factors and processes that led to the current outcomes (process inequity). 2. Most studies focus only on nighttime exposure by using residential population data from the U.S. Census: Need to explore additional data sources and develop data sets that can be used to examine daytime distribution of various population groups. 33
SPATIAL INEQUITY ANALYSIS: CURRENT LIMITATIONS AND FUTURE NEEDS 3. Use of aggregated population data (lack of addressspecific and individual level data) leads to unreliable estimates of the population at risk: Need to use local household surveys and dasymetric mapping to enhance accuracy of estimates. 4. Most studies use traditional statistical methods (e.g., correlation and regression) to analyze spatial data: Need to incorporate emerging geostatistical techniques that are more appropriate for analyzing spatial data and relationships. 34
CONCLUDING POINTS Spatial inequity analysis has evolved from comparing the prevalence of minority or low-income residents in census units hosting hazardous sites to more rigorous methods that utilize GIS, distance-based buffer analysis, facilityspecific data, plume modeling, dasymetric mapping, and emerging geostatistical techniques. Current methodological and data deficiencies need to be addressed to ensure that EJ research and findings are not constrained by such limitations. This would lead to more reliable results, stronger evidence, and potentially more equitable policy solutions. 35
SPATIAL INEQUITY ANALYSIS: METHODS, MODELS, AND CASE STUDIES Thank you Jayajit Chakraborty Professor of Geography Department of Sociology and Anthropology University of Texas at El Paso El Paso, Texas, USA jchakraborty@utep.edu 36