DIFFERENTIAL EXPOSURE TO TRAFFIC IN THE
DENVER METROPOLITAN REGION by
EVAN GAYLORD ROSENLIEB B.A., Colorado State University, Fort Collins, 2011
A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Master of Urban and Regional Planning Urban and Regional Planning Program
EVAN GAYLORD ROSENLIEB
ALL RIGHTS RESERVED
This Thesis for the Master of Urban and Regional Planning degree by
Evan Gaylord Rosenlieb has been approved for the Urban and Regional Planning Program
Carolyn McAndrews, Chair Austin Troy Wes Marshall
Date: July 30, 2016
Rosenlieb, Evan Gaylord (M.U.R.R)
Differential Exposure to Traffic in the Denver Metropolitan Region Thesis Directed by Assistant Professor Carolyn McAndrews
Minority and low-income communities in the United States have been found to be significantly more likely to live close to major roads; this proximity to major roads is associated with quantifiable negative effects on health. While this marks traffic exposure as an important aspect of Environmental Justice, research to date has not been able to connect this differential exposure to policy. This thesis advances new methodology for examining these spatially occurring phenomena through the use of spatially explicit Spatial Durbin and Geographically Weighted Regression models. For the region as a whole disparities in respect to demographic and socioeconomic variables match those observed in in previous studies. Locally the associations are heterogeneous, with some locations showing the opposite relation observed in the region. The local model exhibits patterns at both the regional and neighborhood level and suggest that specific land use policy mechanisms may influence demographic disparities of traffic exposure at the local scale. This methodology represents an important first step in informing land use policies that could be used to combat differential exposure where it is detrimental to less privileged communities.
The form and content of this abstract are approved. I recommend its publication.
Approved: Carolyn McAndrews
First, Carey deserves acknowledgment for her role as adviser, research partner, and friend. Austin and Wes helped brainstorm approaches to the problem and provided feedback of results throughout the process. I thank all three for these roles as well as sitting on my committee. The funding provided by the Mountain Plains Consortium was instrumental to the process, and is greatly appreciated. Much credit is due to the open source community that provided me with the tools I used to conduct the research, specifically the maintainers of GNU R, QGIS, and LibreOffice. Finally, I would like to thank Reenie and my parents for their undying support.
TABLE OF CONTENTS
Context of the Environmental Justice Movement.................1
Traffic and Environmental Justice.............................3
II. RESEARCH QUESTION...............................................8
Goal of Research..............................................8
IV. MODEL RESULTS.................................................18
Distribution of Exposure to Traffic..........................27
Global and Local Models......................................27
Regional and Neighborhood Processes..........................28
Traffic Exposure and Smart Growth Policies...................29
Context of the Environmental Justice Movement
Activist roots. In October of 1991, the First National People of Color Environmental Leadership Summit convened in Washington D.C., signifying the official genesis of the environmental justice movement, a movement that had been operating in fragmented localities throughout the country. One such location was in South Central Los Angeles in the mid 1980s, where a community grassroots effort to prevent the location of a high capacity trash incinerator in a low income, predominately African American neighborhood succeeded. The local citizens had become concerned about how the environmental and social costs of public investments such as the trash incinerator were being systematically and disproportionately born by low-income neighborhoods and neighborhoods of color. Activists pointed out that this was true even while, in the case of the trash incinerator, those neighborhoods were generally less of contributors to solid waste and users of less of the power produced per capita compared to higher income neighborhoods (Di Chiro, 1996).
Given that trash incinerators were well known to give off potent environmental pollutants, including but by no means limited to dioxins and fluorocarbons, activists at the time were hopeful that they may be able to seek aegis from the powerful environmental organizations at the time such as the Sierra Club or the Environmental Defense Fund. The environmental movement lead by these organizations had racked up an impressive string of environmental policy victories in the previous fifteen years and
had fought to regulate the same toxins in more wild environments. To the contrary, activists were shocked and dismayed that such organizations, focused as they were on preserving endangered habitat and otherwise pristine wilderness, did not deem the plight of such neighborhoods as being related enough with the environment to take legal issue (Di Chiro, 1996).
The activists concerns were not purely anecdotal. The landmark study Toxic Waste and Race in the United States: A National Report on the Racial and Socioeconomic Characteristics of Communities with Hazardous Waste Sites in 1987 found that not only was race correlated with the location of toxic waste facilities, but that it was actually the leading identifiable factor (Toxic Wastes and Race in the United States, 1987). Armed with knowledge of the problem and without support from the leading environmental organizations, the urban activists were forced to gain their own traction in the public policy sphere. At the First National People of Color Environmental Leadership Summit, leaders of the movement gave it a name by releasing the Principles of Environmental Justice, which outlined the ways in which environmental hazards and man made environmental costs were being disproportionately born by people of color and outlined the way towards an environmentally just future(Di Chiro, 1996).
Academic interest in Environmental Justice. While the Environmental Justice movement did not find much solidarity with the broader Environmental movement, the desires of the movement aligned naturally with research within the fields of geography, sociology, urban planning, and public health. Particularly, the movement appealed to researchers within these fields who study how social, economic, and political systems structure inequalities opportunity and quality of life broadly in cities and in health
specifically in cities and metropolitan regions.
Some of the most established areas of research related to this stem from the study of the magnitude of racial segregation within cities and regions in America as well as the processes that caused them (Massey, 1990). Naturally, much of the research has centered around questions of the way that segregation affects economic opportunity and the availability of social capital, perhaps the most visible effects of segregation (Briggs, 2005). However, there has been increasing interest in how segregation and other local, spatially occurring phenomena influence racial discrepancies seen in health metrics e.g. preterm births (Osypuk & Acevedo-Garcia, 2008, 2010).
Exposure to environmental pollutants, much like segregation, is necessarily local and spatial in its nature. As the inequalities observed by the Environmental Justice movement are due to the unique nexus of segregation and other socioeconomic spatial sorting processes and the spatial siting and dispersion of environmental pollutants, local land use policies have a special importance to the questions raised by the movement. As such, there has been extensive academic interest in how phenomena such as residential segregation and environmental racism operate through specific mechanisms such as zoning codes, housing markets, and decisions about the siting of hazardous land uses such as toxic waste facilities and freight centers in the hopes that understanding these processes can allow for public policy intervention to ameliorate the discrepancies (Bullard, 1990; Graif & Sampson, 2009; Wilson, Hutson, & Mujahid, 2008).
Traffic and Environmental Justice
Community based responses to traffic. Most attention in the Environmental Justice field has been, in continuance of its roots, on large "point sources" of pollutants
such as trash incenerators. This makes sense, as these instances are not only the most visible instances of such but are also the easiest instances in which to shape change in existing policy regimes (stopping the placement of a trash incenerator can be done through existing development review processes). However there has also been growing use of the concept of environmental justice in regards to more diffuse and stochastic forms of environmental exposure, e.g. exposure to hazards (Morse, 2008). Along these lines there has recently been more interest within the environmental justice movement in inequities from traffic air pollution, although once again often tied to specific locations where the effect is particularly visible such as the fight to regulate idling diesels in the Port of Oakland to limit the impacts on the surrounding West Oakland neighborhood (Dallmann, Harley, & Kirchstetter, 2011; Golub, Marcantonio, & Sanchez, 2013), or the resistance to the expansion of 1-70 in North Denver (Zaffros, 2015).
Academic research on traffic exposure. While community efforts still have reason to be focused on spatially confined instances of environmental justice, there has been plenty of interest within the academic sphere on the systematic inequities in exposure to traffic air polution in regards to demographic and socioeconomic factors in a more diffuse and spatially abstracted sense. Despite the fact that air pollution from traffic does not produce extremely toxic substances in the same way that incenerators and other heavy industrial sites do, a large catalog of public health literature has found that particulate, NOx, CO, and VOC concentrations from regular traffic air pollution can lead to asthma and other chronic respiratory illnessess among those who are exposed and particularly children (Oosterlee, Drijver, Lebret, & Brunekreef, 1996; van Vliet et al., 1997). In addition such exposure has been linked to greater incidence of cardiovascular
disease (Anderson, Thundiyil, & Stolbach, 2012; Babisch, 2014; Hoek et al., 2013), cancer (Brunekreef & Holgate, 2002), general mortality (Roemer & van Wijnen, 2001), injury (Morency, Gauvin, Plante, Fournier, & Morency, 2012) and birth defects or other adverse birth outcomes (Sapkota, Chelikowsky, Nachman, Cohen, & Ritz, 2010; Wilhelm & Ritz, 2002).
Additionally, traffic is known to be a barrier to walking, bicycling, and other forms of active transportation. As active lifestyles are a very important predictor of general health, this shows that the presence of traffic can have affects on health that reach outside of the obvious air quality issues (Marshall, Piatkowski, & Garrick, 2014). Beyond just health, the barrier caused by high traffic has been linked to broaded social health ills, contributing to social severance, social disorder, and decreased social capital (Anciaes, Jones, & Mindell, 2016; Appleyard & Lintell, 1972). As such, there is a clearly defined nexus between traffic and Environmental Justice concerns.
In the last 15 years, the research has mounted that air pollution due to traffic in particular has measurable health outcomes and significantly systematically affects less priviledged neighborhoods to a greater degree. In 2001 Rachel Morello-Frosch et al found that racial and socioeconomic factors were correlated with incidence of the types of cancers thought to be caused by air pollution in Southern California. As this was for the entire region and the association held controlling for amount of industry nearby, this strongly implied that more diffuse types of air pollution such as traffic were having measurable impact on health, and disproportionatly for non white and socioeconomically less priviledged areas (Morello-Frosch, Pastor, & Sadd, 2001). In 2003 RB Gunier et al found that there were racial and socioeconomic disparities in children located closest to
the highest traffic roadways in the entire state of California, with minority and low income block groups generally displaying twice the traffic density as the regional average (Gunier, Hertz, von Behren, & Reynolds, 2003). Another paper found similar relationships using differening methods at the scale of the entire country (Tian, Xue, & Barzyk, 2013). Another paper of the same year also found similar disparities for the country as a whole, but additionally that there were large differences between regions, states, and counties ranging from correlations many times that of the national average to statistically insignificant associations (Rowangould, 2013).
Land use and traffic exposure. The fact that these environmental injustices are continually found in these studies speaks both to the effects that air pollution due to traffic can have and to the presence systematic proceses that tend to colocate people of lower privilege with areas of high traffic density. There are certainly some easily indentifiable hypotheses. Of course, the fact that people with lower incomes would be more likely to live next to high traffic roads should be no surprise given then economic system we live in: houses next to high traffic roads tend to have less value for a variety of reasons (including, perhaps, fear of air pollution itself but perhaps more mundanely the noise, light, and parking aspects), and therefore in any given area those with less money would be likely to filter to the cheaper houses next to major roadways.
On a more macro scale, the suburbs were designed in part specifically to be distant from traffic, and for most of the American post war history the distance at which you could live away from the city was a privilege to be bought with money. However these sorts of factors can only explain the economic causes, not that race and ethnicity consistently seem to be significantly related to exposure to high traffic roads even when
controlling for economic aspects. In addition, the story is muddled by the fact that while large lots of land in quiet neighborhoods in the suburbs is still a privilege bought and enjoyed by many, in the last few decades there has been a counter trend of wealthy communities moving back to the urban core of cities and areas of poverty being pushed to inner-ring suburbs (Howell & Timberlake, 2014).
There are reasons to believe that there is more of a diverse patchwork of land use policy decisions at stake, concrete decisions with specific, spatially located consequences. This would certainly be supported by the spatial heterogeneity of inequities observed by Rowangould. After all, planners have certainly been guilty of as much before: much research has been conducted on the ways in which minority neighborhoods were disproportionately affected by freeway constructions during the interstate era in America, and indeed sometimes the freeways were used by those in power specifically to reinforce existing segregation (Connerly, 2002).
In light of this line of thinking, there is no end to the possible land use decisions that could be driving both the observed inequities and the spatial heterogeneity of the observed inequities. Are cities still systematically locating their major roads in less priviledged areas? Are less priviledged people drawn closer to major roads because of greater transit service? Are metro regions exacerbating the issue by colocating large transit investments close to existing roadways because of the ease of right of way acquisition? Are cities inadvertently locating their affordable housing in areas of high traffic density through Transit Oriented Development? With so much evidence weighed in on the "what" of the problem, it would seem time to examine the "where" and "why" questions in more detail.
Goal of Research
For the reasons given above, it is clear that the general question of what processes drive differential exposure to traffic is of clear import both to the affected communities and to multiple strings of academic research that aim to study spatial inequities generally and health implications of traffic exposure specifically. However being able to prove the causality of differential traffic exposure, in a specific location or generally, is much more advanced than the state of current literature. Not all questions can be addressed at once, so it is necessary to decide and justify what the most important next step is given the current state of the literature.
Though as noted above there much research on differential health impact of traffic exposure generally, this research is built primarily upon four papers: Gunier, Hertz, von Behren, & Reynolds, 2003; Houston, Wu, Ong, & Winer, 2004; Tian, Xue, & Barzyk, 2013; and Rowangould, 2013. These papers all use some sort of estimation of traffic exposure based on density calculated from a road network and associated AADT counts. All find significantly higher exposure with respect to race and most find higher exposure with respect to income or other socioeconomic variables. The first two papers were focused on Southern California, while the later two examine the entire nation. In addition, the later two papers begin to examine spatial heterogeneity by examining how differential exposure is different in different regions, states, and counties across the country.
In particular, the most advanced paper, methodologically speaking, found that
while on a national and regional scale differential exposure along racial and socioeconomic lines was very consistent, at smaller scales there were plenty of counties with no apparent relationship, some with an inverse relationship, and a wide degree of variation in the degree to which differential exposure happens in different counties (Rowangould, 2013). This suggests that some of the processes that drive the observed disparities in exposure happen at a very local paper. However, nothing since that time has appeared in the literature that attempts to further model and examine the spatial pattern that these disparities display at a finer spatial scale.
Since many public health and most land use policies take place at the sub county scale, in order to start examining the relationship between policies and traffic disparities, the pattern of differential exposure at smaller spatial scales must happen. Therefore, the goal of this paper is to develop and test methods that can model exposure at these scales. This would represent a significant extension of the literature thus far and would help move the academic discussion from observed trends to possible policy recommendations. Study Area
The study area chosen to develop this methodology is the Denver Metro Region. The location of the Denver Metro Region is shown in Figure 1 below for context. The Denver Metro Region was chosen partially due to availability of data and familiarity with the region. However, the Denver Metro Region has many qualities that make it a good case study to examine these dynamics. First, it will only be the second region of the country that will have such an in depth analysis done after Southern California. Like Southern California, the region is a fast-growth sun belt city, and as such there are both development and traffic pressures. Unlike other cities in this category that had strong
growth in the 1990s but slowed down in the 2000s, particularly after the 2008 recession, Denver's growth has remained strong (Frey, 2012). In addition to general development pressures due to growth, the region has also invested heavily in transit projects. Voters approved tax increases to fund 40 miles of light rail along 1-25, the regions busiest corridor, in addition to an expansion to the interstate itself in 1999. Voters again supported a tax increase to expand the light rail system regionally in 2004. The transit investment has been paired with land use policy including zones for transit-oriented development (TOD). The TOD zones not only change the zoning and other land use regulations around transit stops but also provide access to funds for affordable housing credits not available outside of the zones.
I I Colorado o Other Major Cities
Interstate Highways Denver Regional Council of Goverments Boundary
250 500 km
Figure 1: Context of Denver Metro Region
These TOD zones are one of many land use instruments outlined in the Denver Regional Council of Government's (DRCOG) Metro Vision plans, including (nominally)
an urban growth boundary for the region and the designation of Regional and Neighborhood Centers where zoning allows for greater density than would otherwise be available. Together, these instruments have the goal of managing growth in a way inspired by the Smart Growth movement by simultaneously decreasing demand for automobile infrastructure and land for greenfield development. These goals can be seen as being rooted in historic context of anti-growth sentiment in the region as seen in the decision to turn down the Olympics in 1976 and the fight to defeat the 1-470 beltway proposed in the 1970's (Oglesby & Billingsley, 2015). In addition to the growth management policies pursued at the regional level, the region also contains some of the most austere growth management regulations in the nation in Boulder County. Boulder County residents approved the first green space preservation tax in the nation in 1967, and the County has been continually acquiring open space to prevent urban development since. This combination of development and land use policy in the region gives it the potential to give insight into the effects of such policies on traffic exposure.
To represent traffic, we used 2010 average annual daily traffic (AADT) estimates for each road segment in the ten-county Denver metropolitan region. DRCOG generates these estimates with an advanced activity-based regional travel model. The activity based model uses over 10,000 in depth travel journals in addition to land use, demographic and socioeconomic, and traffic count data to better model travel of households to job and activity centers according to their household characteristics. DRCOG serves as the Denver metropolitan region's Metropolitan Planning Organization, as mandated by federal transportation policy, and forecasting travel patterns for the region is one of its core functions.
Socioeconomic and demographic data for the region are from the American Community Survey (ACS) five-year block group estimates for 2006-2010 (Minnesota Population Center, 2011). Block groups are the smallest spatial unit available that include the socioeconomic information necessary for the study only demographic variables are available at the block level for the five-year estimates.
Estimation of the traffic exposure surface. Previous research on differential exposure to traffic has computed traffic density for census blocks or block groups by drawing a buffer of 200 or 250 meters around the census unit before summing vehicle-miles of travel (VMT) within it (Gunier, Hertz, von Behren, & Reynolds, 2003;
Rowangould, 2013). This buffer method accounts for traffic on roads that are within dispersion distance. However, this method does not account for how exposure to the negative effects of traffic (e.g., noise, pollutants, nuisance) decreases with distance from the roadway. With the buffer method, a road that bisects a census unit, forms its border, or runs parallel to its border outside of the census unit would all contribute equally to the calculated traffic density of the unit. To represent the decreasing intensity of exposure to traffic with distance, we created a traffic density surface using a bisquare kernel density function with a 300m bandwidth. We used a 6m by 6m output cell size, a resolution that is computationally tractable while being much smaller than the census block group unit of analysis. The unit of the output was standardized to units of VMT per square mile of the block group. We do not investigate environmental factors, such as wind direction, that influence the dispersion of specific pollutants because we are primarily interested in exposure to traffic as its own multidimensional environmental hazard.
Extraction of exposure to block groups. Census blocks groups are standardized to have similar populations over time, and as such have highly variable areas depending on the population density of the area. Large non-urban block groups have dimensions much larger than the scale at which many types of land use patterns that affect exposure, such as the presence of green buffers around major roads, take place. Therefore the estimate of traffic exposure at the unit of the block group could be susceptible to certain biases based on the size of the block group; estimates of exposure in suburban block groups could be underestimated relative to urban ones due to the larger size of the block groups not being able to model land use patterns in suburbs that tend to locate residences away from major arterials. These issues that occur from extracting values from the traffic
density surface to the block group are aspects of the modifiable areal unit problem. The modifiable areal unit problem is a problem arising from the imposition of artificial units of spatial reporting on continuous geographical phenomenon resulting in the generation of artificial spatial patterns (Heywood, Cornelius, & Carver, 1998). In order to minimize these sorts of errors, we used population counts from census blocks to weight exposure at the block group level. Census blocks are much smaller than block groups, thus being more congruent with the scale at which patterns of exposure occur. In addition, as they are defined by road networks as opposed to population, they have a more consistent spatial size. We define traffic exposure at the block group level as
In order to minimize these sorts of errors we used population counts from census blocks in order to weight the estimation of exposure at the the block group level. Because the boundaries of census blocks are determined by the physical street network the population data is of a much more congruent scale to that of both traffic exposure and the small scale land use policies that affect it. The calculation of traffic exposure at the block group level was done as such:
1 r bg
n = Number of Blocks within Block Group
EBg = Weighted Exposure of Block Group
Eb = Exposure of Block
PB = Population of Block
Peg = Population of Block Group
The exposure are the block level was calculated by taking the average value of the traffic exposure surface over the block using a zonal statistics algorithm.
Statistical modeling. We used multiple regression with traffic density as the dependent variable and three socioeconomic and demographic variables as independent variables with the census block group as the unit of analysis. The three independent variables were: (1) percent of persons in the block group that are not non-hispanic white; (2) percent of block group households living at or below the local poverty level; and (3) percent of persons in the block group with a college degree. We selected these independent variables based on their theorized association with exposure to traffic, previous literature, and because they were significant predictors of traffic density.
We examined the correlations and collinearity among the independent variables and found high correlations among them (Pearsons correlation coefficients between .42 -.66), but not enough collinearity to violate the assumptions of the ordinary least squares model with the highest condition index of 9.7 being well bellow the suggested threshold of 30 (Belsley et al., 1980). All analyses were weighted by block group population. We assessed the normality of traffic density, poverty, education, and race variables and found a high amount of skew that was not normalized with a natural logarithm or other common transformation. Because there was no a priori distribution that any of the variables should follow, we normalized them using Box-Cox family methods to estimate the proper transformation. Though at a sacrifice to interpretability, the exact lambda coefficients of the normality diagnostics were used due to the fact that the smaller sample sizes of the localized regression used in the Geographically Weighted Regression are sensitive to skew.
Because of the high spatial clustering present in the independent variables and the error term in the model (global Moran's I statistics were significant for all at a=.01), we tested for spatial error dependence and a missing spatially lagged dependent variable using Langrange multiplier diagnostics. These Lagrange multiplier statistics test whether the presence of spatial autocorrelation among variables violates the assumptions of the OLS model (Anselin & Rey, 1991). Both statistics were significant at a = 0.01, indicating the need to correct for both types of spatial autocorrelation. Therefore, we used a mixed spatial error-lag model (spatial Durbin model).
We used a geographically weighted regression (GWR) to examine the local spatial patterns in the relationship between traffic density and poverty, college, and race. We used an adaptive kernel bandwidth due to the irregular size of the block groups, and the bandwidth selection algorithm found the optimal bandwidth to be 80 neighbors using a bisquare kernel. The use of an adaptive kernel means that all local regressions have the same sample size, whereas a non-adaptive kernel would have larger sample sizes in dense, urban places and too small of sample sizes in low density area.
We ran a Monte Carlo simulation to test for spatial nonstationarity of the local coefficients of the model. All three indepedent variables were found to be statistically significantly nonstationarity, with percent college and percent minority being significant at the a = .01 level, and with percent poverty being significant at the a = .10 level. This means that the relationship between the dependent and independent variables varies by space for all variables. This very strongly suggests that localized, neighborhood level processes are very important in the relationship of socioeconomic and demographic variables and traffic exposure. The fact that the poverty variable has a lower p-value may
indicate, although not statistically rigorously, that the processes that drive increased exposure due to economic status are somewhat less driven by local processes and somewhat more driven by global processes. As a result of these tests, all three independent variables were included in the GWR.
Table 1 presents a descriptive summary of the untransformed variables in the regression model, which have asymmetric distributions around the mean. The unit for traffic density is 1,000 VMT per square mile, and the unit for population is persons.
Table 1: Descriptive Summary of Untransformed Model Variables
N = 2,029 Census Block Groups
Traffic density (1,000 VMT / mi2) Non-white (%) Poverty No College (%) (%) Total population
Minimum 0.1 0.0 % 0.0 % 0.0 % 92
First quartile 24.8 11.8% 2.3% 44.3 % 900
Median 53.4 23.7 % 7.0 % 63.4 % 1,266
Mean 74.1 31.3% 12.1 % 61.5% 1,387
Third quartile 94.1 45.9 % 17.9 % 79.3 % 1,708
Maximum 667.3 100.0 % 100.0 % 100.0 % 8,582
Moran's I 0.68 0.65 0.44 0.7
The spatial distribution of the four untransformed variables is shown in Figure 2. For interpretability, the figure does not include the full extent of the region that was included in the regression model.
A) 1,000 VMT/SqMi B) % Persons
C) % Households D) % Persons _ _
0 5 10 15 20 km
0-40 I | 0-15 0-6 0-34
40-89 i I 15-31 6-14 I I 34-52 A
89-168 31-50 14-27 52-68 r--1 Denver City Limits f
168-293 50-72 27-41 68-83 (excluding airport) N
293-566 72-100 41-100 83-100 Major Highways
Figure 2: Spatial Distribution of Untransformed Variables
Table 2 presents the results of the Durbin regression. The model is significant at a = .01. These global model results are consistent with those of previous studies: variables that indicate lower socioeconomic status and racial/ethnic identities of less privileged populations are associated with higher traffic exposure. Specifically, block groups with higher percentages of minority and poor populations are associated with higher traffic exposure, while block groups with higher percentages of college educated populations are associated with less exposure. For contrast, we also estimated a model that did not correct for the spatial autocorrelation of the variables, and the percent college coefficient was significantly positive rather than significantly negative, which underscores the importance of using spatially-explicit models to model spatially occurring phenomena.
Table 2: Output of Durbin Model
Estimate Std. Error z-value p-value
Intersect 50.0 2.61 19.2 < 2.2e-16
PctMinority 0.372 0.151 2.47 0.014
PctNoCollege 0.034 0.014 2.41 0.016
PctPoverty 0.774 0.142 5.44 5.2e-08
Lag.PctMinority 0.466 0.392 1.19 .234
Lag.PctCollege -0.129 0.033 -3.84 1.2e-04
Lag.PctPoverty 2.86 0.390 7.34 2.2e-13
College variable. The adjusted coefficient for the college variable is negative, indicating that block groups with high concentrations of college educated persons have a tendency to be protected from traffic exposure. The results of the GWR, however, show that this protective effect of a college education is not observed evenly throughout the
region. The negative associationthe protective effectbetween college and traffic exposure tends to be locally statistically significant in outer suburbs, where residents of block groups with the highest traffic density also have a significant tendency of having less than a college degree. The majority of the urban core of the metropolitan region, roughly coincident with the City and County of Denver, does not show the same protective effect. Indeed, certain areas in the core show a positive association between
exposure to traffic and college education.
Localized Significance of College Education in Predicting Traffic Exposure
Geographically Weighted Regression Coefficient t-value
t <-1.96 I I -1.96 1.96
Positive values indicate association between lack of college education prevelance and traffic exposure
A: Locality including New Urbanist Stapleton Redevelopment as well as older, working class suburbs
B: Locality including the University of Denver as well as many relatively well-to-do and white neighborhoods that border I-25.
Denver City Limits (excluding airport)
Figure 3: College Coefficient of GWR model
One location, marked as A in Figure 3, includes the new urbanist Stapleton
Airport redevelopment, which is a neighborhood of both high college education and high
traffic density. The location marked as B in Figure 3 included the areas around the University of Denver, which included many block groups of both high college education and high traffic density. The ring pattern around the urban core in which the GWR college education coefficient is negative shows that the spatial heterogeneity of the relationship between college education and traffic density has a regional pattern that suggests regional processes. The areas of positive association highlighted in Figure 3 suggest that more localized effects may also play a part.
Minority variable. The sign of the racial and ethnic minority coefficient in the global spatially-adjusted regression is positive, indicating that minority racial and ethnic groups have systematically higher exposure to traffic. In contrast to the relationship between college and traffic density, which had a strong regional pattern, the spatial pattern of exposure by race and ethnicity does not appear to be regional, and instead shows stronger neighborhood-level effects.
For instance, the GWR finds that in certain neighborhoods whiteness is associated with increased exposure, shown in area A in Figure 4. This area includes both Denver's central business district and historically high minority neighborhoods. Although all of area A has high traffic density relative to the region, the very highest traffic densities are co-located with new luxury apartment developments. City policy has targeted such infill development in areas that are transitioning from industrial land uses.
Localized Significance of Minority Populations in Predicting Traffic Exposure
Geographically Weighted Regression Coefficient t-value
t < -1.96
-1.96 < t < -1.645 -1.645 < t < 1.645
1.645 < t < 1.96
t > 1.96
Positive values indicate association between percentage minority population and traffic exposure
A: Area around Downtown Denver, where some of the most traffic dense block groups in the entire region are also relatively white.
B: Tech Center area with almost all multifamily housing located close to freeways in TOD zones.
r_J Denver City Limits (excluding airport)
0 5 10 15km
Figure 4: Minority Coefficient of GWR Model
Area B in Figure 4 is an area around the Denver Tech Center, the second largest concentration of jobs in the region after the Central Business District. The vast majority of housing is in the form of low-density single family housing. In addition, there are newer transit-oriented developments along the light rail corridor next to Interstate 25, the region's busiest roadway. An example of this is shown in Figure 5 below. The zones of transit-oriented developments contain a large portion of the area's multifamily housing. As such, while the area is very white relative to the region, the few block groups with larger minority populations also have a very strong tendency to have multifamily housing directly adjacent to the busiest roads. In this area, whiteness is protective against traffic
Figure 5: Example of Transit-Oriented Development Adjacent to Freeways Poverty variable. The global coefficient of the poverty variable is positive, meaning higher poverty in a block group is associated with increased traffic exposure. The coefficient is the largest of the three dependent variables. Even while accounting for the variable transformations, poverty has the greatest effect on a block group's expected traffic exposure. This is not surprising given the easily identifiable economic processes of land value that reduces the price of residential land close to major roadways. The strength of poverty's association with traffic exposure appears in the GWR analysis as well. The localized coefficients, shown in Figure 6, are significant for vast swaths of the region as opposed to the smaller patches seen with the other variables. In addition, there are very few areas that show a significant negative association.
Localized Significance of Poverty in Predicting Traffic Exposure
CWR Coefficient t-value
t < -1.96
Positive values indicate association between percentage poverty population and traffic exposure
A: Area bordering the high traffic Tech Center as well as poorer inner ring suburbs.
B: Area around city of Boulder in Boulder County, both of which have some of the most restrictive smart growth policies in the nation.
B: Area in North Denver that contains most of the regions traditional heavy Industry.
GWR Beta Coefficient Value
P < 2.0 -2.0 < P < 0.0 0.0 < p< 2.0
2.0 < p< 4.0
4.0 < P < 6.0 P > 6.0
Positive values indicate association between percentage poverty population and traffic exposure
r_J Denver City Limits (excluding airport)
I I Boulder City Limits
--- Major Highways
0 5 10 15 km
The area exposure, labeled
Figure 6: GWR Percent Poverty Results
showing a negative significant association between poverty and A in Figure 6, borders more affluent areas along 1-25 to the east and
inner-ring suburbs to the west in the city of Aurora that have experienced an increase in poverty in recent decades that similar inner ring suburbs have experienced across the nation [cite]. This patch shows that these suburbs, while no longer primarily occupied by the upper-middle class that they were built for decades ago, still fulfill a function of protecting the residents from traffic. Though increasing poverty suburbs has had many deleterious effects, for example lower accessibility to services and community assets than more central places, the analysis shows that they may still gain benefits from the suburban form that also benefit affluent residents.
Figure 5 also shows the distribution of the beta coefficients of the GWR in addition to the t-values. Not only does the solid majority of the region show a significant positive correlation between poverty and traffic exposure, there is also a wide variation in how positive the coefficient is with some block groups having a coefficient approaching ten times the global value of .79. One such area, labeled B on Figure 6, is in the vicinity of the City of Boulder in Boulder County. The strict growth management policies in Boulder County that make residential land scarce may have the effect of amplifying the need for those who are poor to seek cheaper land close to high-traffic roadways.
The area labeled C in Figure 6 is Denver's heavy industry corridor. The area had historically weak zoning and other land use regulations, and residences are intermixed with traffic, rail, and other noxious uses. There are few residential neighborhoods in Denver that are less buffered from traffic than these, where, in some cases, houses and stores are literally in the shadow of the interstate.
Distribution of Exposure to Traffic
The GWR coefficients are an estimate of the spatial distribution of relationships between socioeconomic variables and exposure, but the exact boundaries of block groups that are seen to be significant are not as significant as the general patterns seen. A blockgroup that is assigned a significant positive relationship between percent college and exposure by the model does not necessarily have both high college and exposure itself. Instead, it is the center of an area on the neighborhood scale that displays the relationship. Because of this the model is sensitive to spatial contrast; areas that are significant are generally found along borders between different types of neighborhoods and thus sometimes seem offset from the apparent cause of the relationship. Also, it should be noted that areas where there are significant coefficients in the local model will tend to be collocated with the highest traffic roadways as areas around high traffic roads have the gradients necessary in traffic exposure (the dependent variable) in order to find significance. There are very likely similar relationships present around arterials in addition to the freeways, however they do not tend to provide the traffic density gradients necessary for the model to find significance with the sample size used for the local regressions.
Global and Local Models
The fact that all global model variables were significant and consistent with the results of previous studies reinforces the finding that less privilege in this U.S. is
systematically linked to higher exposure to traffic. Methods used in previous studies could identify differential exposure, but could not attribute this observed inequality to potential mechanisms, whether such mechanisms might be aspatial aspects of economic phenomena or inherently spatial aspects of local land use and planning. The spatial models used in this paper allow us to begin addressing this question.
Regional and Neighborhood Processes
The spatial pattern of both the college and minority coefficient strongly suggest that there are both regional processes and localized processes that drive inequalities in the exposure to traffic. The regional process as seen in the college variable, in which those of more education seem able to protect themselves from traffic in suburban areas but not in more urban areas, seems related to aspects of the general form of American cities and the so-called back to the city cultural trend. It would make sense that swaths of very low density single-family housing, typical of the American suburb, provides more opportunities for those to with the means to separate themselves from traffic. An additional reason that the same trend is not seen in the urban core of the region could be that a younger generation of college educated people that want to live in the city do not care as much about traffic exposure and are accordingly less willing to pay to distance themselves from it.
Traffic Exposure and Smart Growth Policies
Without additional study, particularly a longitudinal analysis that could examine effects of specific planning interventions, it cannot be proven that specific planning processes caused any of the areas of inequality identified from the results of the GWR analysis. However, certain areas identified in our study are characterized by planning
interventions. For instance, A in Figure 4 is linked to infill development strategies, and B in Figure 4 is linked to dense multifamily housing located near a light rail station. Based on these patterns, more attention should be given to the possible effects of these and other smart growth policies on traffic exposure. This is particularly important because contemporary transit infrastructure investments, which are usually central to smart growth ideas, are often co-located freeway infrastructure due to the availability of right of way in these locations (Loukaitou-Sideris, Higgins, Cuff, & Oprea, 2013). Limitations
GWR models are sensitive to spatial contrast. In this analysis, the GWR coefficients are estimates of the local relationships between socioeconomic variables and exposure, but the exact visual representation of areas with statistical significance need to be interpreted with care. Areas that have significant local coefficients are generally found along borders between different types of neighborhoods. For example, the area represented by A in Figure 3 is not centered on the Stapleton airport redevelopment, but rather is generated by the contrast between Stapleton and its surrounding areas.
Areas where there are significant coefficients in the local model will tend to be co-located with the highest traffic roadways because areas around high traffic roads have the gradients necessary in traffic exposure (the dependent variable) to find significance. There are very likely similar relationships present around major arterial roads in addition to the freeways, however they do not tend to provide the traffic density gradients necessary for the model to find significance with the sample size used for the local regressions.
More fundamentally, the cross sectional design of this study makes it impossible to identify the causal effects of policy and planning interventions, such as TOD, on traffic exposure. Future work needs to use research designs that allow for identification of these causal mechanisms. For example, a time-series analysis examining the effect of light rail investments and associated TOD on traffic exposure could identify potential causal relationships.
In addition, and despite being weighted by population, census block groups may still be too coarse a spatial resolution to examine built environment and design characteristics that influence exposure to traffic, such as the distance between buildings, and roadway design. A finer spatial unit of analysis, such as parcel-level data, would help examine the relationship between design features of the built environment and traffic exposure.
Our use of spatially-explicit models shows that it is possible to examine the spatial patterns of traffic exposure with respect to demographic and socioeconomic patterns in a way not previously explored. For the case of Denver, these associations between race and socioeconomic status and traffic exposure are extremely spatially dependent, and their analysis requires the use of spatial models to account for spatial auto correlation. The importance of spatially explicit models was shown in our model by the percent no college independent variable which showed a negative relationship with traffic exposure in the ordinary least squares model but showed a positive association with traffic exposure in the spatial Durbin model. This positive relationship is more consistent with previous research on income, another marker of socioeconomic status
The global model shows that the lives of those who are not white, poor, or without a college education in Denver Metro are significantly more afflicted by the nuisance of traffic than their white, affluent, and college educated neighbors. Considered with previous studies that have found similar relationships on a national scale in respect to race and income, this study suggests that the way in which metro regions are designed in America systematically places those of less privilege in proximity to traffic. This study finds that poverty is the most consistent predictor of traffic density in the region. This contrasts with Rowangould, who found that race was a more consistent predictor in the nation as a whole than income, as well as Tian et al., who found the effect of poverty to be very marginal. Whether this is unique to a the Denver region or a result of differences in variables and methods is unclear. In addition and in contrast to other studies, we find that at least in the Denver metro region lack of college education also helps predict exposure to traffic independently of race and poverty. The significance and consistency of this study with others of its type indicate that, regardless of cause, the time has come for policymakers to consider ways to ameliorate differential exposures to traffic.
The localized model reinforces the findings of the global model, but also adds complexity to the implications to policy. The results suggest that the relationship between social factors and traffic exposure may be affected by multiple larger economic and social phenomena, and potentially by specific local land use policies and practices. Effects are generally patchy and often appear at the neighborhood scale. However, there are also effects that display regional patterns. Exurbanization, redevelopment and revitalization projects, transit-oriented development, the increasing poverty in suburbs, and the siting of highways appear to be some of the mechanisms that affect disparities in traffic exposure.
Some of these mechanisms are at least partially in the purview of planning and policy at the local, regional, or even national scale, whereas others may be outcomes of larger economic and demographic trends that are not easily controlled by policy. While more work remains to be done with methods that can establish causes, effects, and magnitudes of different processes, we believe this represents and important step to better understand the spatial aspects of differential traffic exposure.
Anciaes, P. R., Jones, P., & Mindell, J. S. (2016). Community Severance: Where Is It Found and at What Cost? Transport Reviews, 36(3), 293-317. http://doi.org/10.1080/01441647.2015.1077286 Anderson, J. O., Thundiyil, J. G., & Stolbach, A. (2012). Clearing the Air: A Review of the Effects of Particulate Matter Air Pollution on Human Health. Journal of Medical Toxicology, 8(2), 166-175. http://doi.org/10.1007/sl3181-011-0203-l Anselin, L., & Rey, S. (1991). Properties of Tests for Spatial Dependence in Linear Regression Models. Geographical Analysis, 23(2), 112-131. http://doi.Org/10.llll/j.1538-4632.1991.tb00228.x Appleyard, D., & Lintell, M. (1972). The Environmental Quality of City Streets: The Residents Viewpoint. Journal of the American Institute of Planners, 38(2), 84-101. http://doi.org/10.1080/01944367208977410 Babisch, W. (2014). Updated exposure-response relationship between road traffic noise and coronary heart diseases: A meta-analysis. Noise and Health, 16(68), 1.
Briggs, X. de S. (Ed.). (2005). The geography of opportunity: race and housing choice inmetropolitan America. Washington D.C: Brookings Institution Press. Brunekreef, B., & Holgate, S. (2002). Air Pollution and Health. The Lancet, 360, 1233-1242.
Bullard, R. D. (1990). Dumping in Dixie: race, class, and environmental quality. Boulder: Westview Press.
Connerly, C. E. (2002). From Racial Zoning to Community Empowerment The Interstate Highway System and the African American Community in Birmingham, Alabama. Journal of Planning Education and Research, 22(2), 99-114. http://doi.org/10.1177/0739456X02238441 Dallmann, T. R., Harley, R. A., & Kirchstetter, T. W. (2011). Effects of Diesel Particle
Filter Retrofits and Accelerated Fleet Turnover on Dray age Truck Emissions at the Port of Oakland. Environmental Science & Technology, 45(24), 10773-10779. http://doi.org/10.1021/es202609q
Di Chiro, G. (1996). Nature as Community: The Convergence of Environment and Social Justice. In W. Cronon, Uncommon Ground: Rethinking the Human Place in Nature. W. W. Norton & Company.
Frey, W. H. (2012). Population growth in metro America since 1980. The Brookings Institution, Mars. Retrieved from
Golub, A., Marcantonio, R. A., & Sanchez, T. W. (2013). Race, Space, and Struggles for Mobility: Transportation Impacts on African Americans in Oakland and the East Bay. Urban Geography, 34(5), 699-728. http://doi.org/10.1080/02723638.2013.778598
Graif, C., & Sampson, R. J. (2009). Spatial Heterogeneity in the Effects of Immigration and Diversity on Neighborhood Homicide Rates. Homicide Studies, 13(3), 242-260. http://doi.org/10.1177/1088767909336728
Gunier, R. B., Hertz, A., von Behren, J., & Reynolds, P. (2003). Traffic density in
California: Socioeconomic and ethnic differences among potentially exposed children. Journal of Exposure Analysis and Environmental Epidemiology, 13(3), 240-246. http://doi.org/10.1038/sj.jea.7500276
Heywood, D. I., Cornelius, S., & Carver, S. (1998). An Introduction to Geographical Information Systems. Addison Wesley Longman.
Hoek, G., Krishnan, R. M., Beelen, R., Peters, A., Ostro, B., Brunekreef, B., & Kaufman, J. D. (2013). Long-term air pollution exposure and cardio- respiratory mortality: a review. Environmental Health, 12, 43. http://doi.org/10.1186/1476-069X-12-43
Houston, D., Wu, J., Ong, P., & Winer, A. (2004). Structural Disparities of Urban Traffic in Southern California: Implications for Vehicle-Related Air Pollution Exposure in Minority and High-Poverty Neighborhoods. Journal of Urban Affairs, 26(5), 565-592. http://doi.Org/10.llll/j.0735-2166.2004.00215.X Howell, A. J., & Timberlake, J. M. (2014). Racial and Ethnic Trends in the
Suburbanization of Poverty in U.s. Metropolitan Areas, 1980-2010. Journal of Urban Affairs, 36(1), 79-98. http://doi.org/10.llll/juaf.12030 Loukaitou-Sideris, A., Higgins, H., Cuff, D., & Oprea, D. (2013). Up in the Air: Urban Design for Light Rail Transit Stations in Highway Medians. Journal of Urban Design, 18(3), 313-339. http://doi.org/10.1080/13574809.2013.800448 Marshall, W. E., Piatkowski, D. P., & Garrick, N. W. (2014). Community design, street networks, and public health. Journal of Transport & Health, 1(4), 326-340. http://doi.Org/10.1016/j.jth.2014.06.002 Massey, D. S. (1990). American Apartheid: Segregation and the Making of the Underclass. American Journal of Sociology, 96(2), 329-357.
Morello-Frosch, R., Pastor, M., & Sadd, J. (2001). Environmental Justice and Southern Californias Riskscape: The Distribution of Air Toxics Exposures and Health Risks among Diverse Communities. Urban Affairs Review, 36(4), 551-578. http://doi.org/10.1177/10780870122184993
Morency, P., Gauvin, L., Plante, C., Fournier, M., & Morency, C. (2012). Neighborhood Social Inequalities in Road Traffic Injuries: The Influence of Traffic Volume and Road Design. American Journal of Public Health, 102(6), 1112-1119. http://doi.org/10.2105/AJPH.2011.300528
Morse, R. (2008). Environmental justice through the eye of Hurricane Katrina. Joint
Center for Political and Economic Studies, Health Policy Institute. Retrieved from http://www.stanford.edu/group/scspi/_media/pdf/key_issues/Environment_policy.
Oglesby, K., & Billingsley, G. (2015, October). Politics and Land Use: 30 Years on the Northwest Parkway. Presented at the 2015 APA Colorado Conference.
Oosterlee, A., Drijver, M., Lebret, E., & Brunekreef, B. (1996). Chronic respiratory symptoms in children and adults living along streets with high traffic density. Occupational and Environmental Medicine, 53(4), 241-247.
Osypuk, T. L., & Acevedo-Garcia, D. (2008). Are Racial Disparities in Preterm Birth Larger in Hypersegregated Areas? American Journal of Epidemiology, 167(11), 1295-1304. http://doi.org/10.1093/aje/kwn043 Osypuk, T. L., & Acevedo-Garcia, D. (2010). Beyond individual neighborhoods: A geography of opportunity perspective for understanding racial/ethnic health disparities. Health & Place, 16(6), 1113-1123. http://doi.Org/10.1016/j.healthplace.2010.07.002 Roemer, W. H., & van Wijnen, J. H. (2001). Daily mortality and air pollution along busy streets in Amsterdam, 1987-1998. Epidemiology, 12(6), 649-653.
Rowangould, G. M. (2013). A census of the US near-roadway population: Public health and environmental justice considerations. Transportation Research PartD: Transport and Environment, 25, 59-67. http://doi.Org/10.1016/j.trd.2013.08.003 Sapkota, A., Chelikowsky, A. P., Nachman, K. E., Cohen, A. J., & Ritz, B. (2010).
Exposure to particulate matter and adverse birth outcomes: a comprehensive review and meta-analysis. Air Quality, Atmosphere & Health, 5(4), 369-381. http://doi.org/10.1007/sll869-010-0106-3 Tian, N., Xue, J., & Barzyk, T. M. (2013). Evaluating socioeconomic and racial
differences in traffic-related metrics in the United States using a GIS approach. Journal of Exposure Science and Environmental Epidemiology, 23(2), 215-222. Toxic Wastes and Race in the United States: A National Report on the Racial and Socio-
economic Characteristics of Communities with Hazardous Waste Sites. (1987). Public Data Access.
van Vliet, P., Knape, M., de Hartog, J., Janssen, N., Harssema, H., & Brunekreef, B.
(1997). Motor Vehicle Exhaust and Chronic Respiratory Symptoms in Children Living near Freeways. Environmental Research, 74(2), 122-132. http://doi.org/10.1006/enrs.1997.3757
Wilhelm, M., & Ritz, B. (2002). Residential Proximity to Traffic and Adverse Birth
Outcomes in Los Angeles County, California, 1994-1996. Environmental Health Perspectives, 111(2), 207-216. http://doi.org/10.1289/ehp.5688 Wilson, S., Hutson, M., & Mujahid, M. (2008). How Planning and Zoning Contribute to Inequitable Development, Neighborhood Health, and Environmental Injustice. Environmental Justice, 1(4), 211-216. http://doi.org/10.1089/env.2008.0506 Zaffros, J. (2015, December 21). Highway injustice in Denvers Latino neighborhoods. High Country News. Retrieved from