Abstract
Objectives. We assessed how traffic and mobile-source air pollution impacts are distributed across racial/ethnic and socioeconomically diverse groups in port-adjacent communities in southern Los Angeles County, which may experience divergent levels of exposure to port-related heavy-duty diesel truck traffic because of existing residential and land use patterns.
Methods. We used spatial regression techniques to assess the association of neighborhood racial/ethnic and socioeconomic composition with residential parcel-level traffic and vehicle-related fine particulate matter exposure after accounting for built environment and land use factors.
Results. After controlling for factors associated with traffic generation, we found that a higher percentage of nearby Black and Asian/Pacific Islander residents was associated with higher exposure, a higher percentage of Hispanic residents was associated with higher traffic exposure but lower vehicle particulate matter exposure, and areas with lower socioeconomic status experienced lower exposure.
Conclusions. Disparities in traffic and vehicle particulate matter exposure are nuanced depending on the exposure metric used, the distribution of the traffic and emissions, and pollutant dispersal patterns. Future comparative research is needed to assess potential disparities in other transportation and goods movement corridors.
Residential proximity to heavy traffic has been associated with adverse health effects, including asthma, reduced lung function, cardiac and pulmonary mortality, and adverse birth outcomes.1–3 Previous research suggests that non-White and lower income individuals may be exposed to higher levels of traffic-related air pollution4–8 and that disparities vary with social gradients associated with higher susceptibility to pollution.9,10 Environmental justice concerns are heightened in goods movement corridors in which substantial volumes of heavy-duty diesel trucks (HDDTs) transport shipping containers on arterials near residences and sensitive land uses through lower socioeconomic status communities.11,12
Significant questions remain, however, regarding the existence and magnitude of race- and income-based disparities in traffic and air pollution exposure.13–16 Some studies have found little association between air pollution exposure and socioeconomic status after controlling for confounding factors17; others found greater air pollution and traffic exposure for higher socioeconomic groups.18,19 Such discrepancies could arise because of methodological differences and challenges in assessing inequities at various scales.11,20–22 Scale could be an important consideration in assessing traffic impacts because vehicle-related pollutants are highly localized, with pollutant concentrations decaying to background levels within 200 to 300 meters during the day,23–26 and because ambient air quality monitoring data are likely insufficient to characterize near-roadway pollutant gradients.
Our study provides an important environmental justice case study by assessing how traffic and mobile-source air pollution impacts are distributed across groups in port-adjacent communities in southern Los Angeles County, which contain substantial racial/ethnic and socioeconomic diversity and may experience divergent levels of exposure to port-related HDDT traffic because of existing residential and land use patterns.12,27,28 We have contributed to the environmental justice literature by examining exposures at the parcel property assessment level to determine impacts at a finer spatial resolution,29 by using spatial regression techniques to account for spatial dependence of data when assessing disparities,30–33 and by using 3 parcel-level metrics of exposure that could have different spatial distributions and population impacts: total nearby vehicle miles traveled (VMT), nearby truck VMT, and the modeled concentrations of emissions from vehicles on neighborhood roadways. We hypothesized that the first 2 traffic exposure measures would provide a distance-based assessment of near-roadway exposure to traffic-related noise and air pollution and that the third would account for the air pollution “plume” after accounting for the geographic and temporal variation in traffic, wind, and other meteorological patterns.
The study area covers approximately 35 square miles immediately adjacent to the ports of Los Angeles and Long Beach in southern Los Angeles County, California, and is transected by a roadway network that carries substantial passenger and diesel truck traffic (Figure 1). The I-110 freeway on the western edge of the port complex carries substantial commuter traffic and about 12% HDDTs; the I-710 freeway on the eastern edge carries about 25% container truck traffic.28 Substantial port-related HDDT traffic travels through the study communities en route to and from these freeways, truck facilities, and transfer yards.12
FIGURE 1— Study area and distribution of vehicle particulate matter for residential parcels: Southern Los Angeles County, CA, 2005.
Note. PM = particulate matter.
Our study is the first environmental justice study to our knowledge to assess disparities using parcel-level data. We obtained the geographic boundaries and characteristics (use type and the year the structure was built) from the Los Angeles County tax assessor.34 Previous studies have analyzed variation in exposure to urban air pollutants that disperse at the neighborhood level and regional level using zip code and census tracts, block groups (BGs), or blocks.32,33,35,36
Parcel-level data could more precisely assess the population impacts of near-roadway pollutant concentrations that decay to background levels within 200 to 300 meters during the day.23–26 When possible, the scale used for analysis should match the geographic patterns of the generation and diffusion of the hazard.17,21,22,31
We developed 3 measures of exposure to assess the robustness of findings across multiple metrics. We developed our first 2 exposure measures, total nearby VMT and total nearby truck VMT, on the basis of a consolidated traffic database previously described.37 In our analysis, we generally define “nearby” to be within 250 meters, a distance threshold that corresponds closely to the distance from roadways at which vehicle-related air pollutants drop to near-background concentration levels.26 This consolidated traffic database incorporates passenger vehicle and HDDT counts for freeways and major arterials and was derived from state and city departments of transportation, port authorities, transportation studies, and truck route designations. Accounting for HDDTs is important in the study area because the California Air Resources Board has declared diesel exhaust particulates emitted from HDDTs a toxic air contaminant,38 HDDTs have substantially higher particulate emission rates than do gasoline vehicles,39 about 70% of cancer risk from air toxins in Southern California is attributed to diesel particulate emissions,40 and about 84% of containers leaving the port complex are transported via HDDT.28
Our third exposure measure represents the previously described37 parcel-level–modeled concentration of emissions from vehicles on neighborhood roadways, which we derived from a modified CALINE4 line dispersion model of vehicle-related pollution including particulate matter (PM) less than 2.5 μm (PM2.5) on the basis of traffic volumes, vehicle class, and meteorological conditions. Line dispersion models use Gaussian plume equations to estimate pollution concentrations with increasing distance from an emission source, such as a roadway, by accounting for factors such as traffic volume, emission factors by vehicle type, meteorological conditions, atmospheric mixing heights, and topography.41
The California Department of Transportation and the US Federal Highway Administration developed the CALINE4 model.42 The model employs a mixing zone concept to characterize pollutant dispersion over the roadway. We ran the CALINE4 model simulations to estimate parcel-level PM2.5 concentrations from local traffic emission within 3 kilometers of a residence in a summer (August) and a winter (January) month in 2005 using vehicle emission factors from the California Air Resources Board’s EMFAC2007 and 2005 meteorological data from the National Weather Service at the Long Beach Airport, which is located at the eastern edge of the study region. Figure 1 shows the distribution of the modeled parcel-level concentrations of PM2.5 from vehicle traffic emissions in the study area. As previously reported, our model suggests that local traffic near the port complex contributes almost a fourth of total fine PM in the study area.37
We conceptualized variables for the nearby transportation infrastructure, land use, employment, and parcel-level characteristics to exert a direct influence on the level of nearby traffic- and vehicle-related pollution. That is, they are likely directly related to the presence of nearby traffic and the volume of pollution generation.41,43 We expected the total mileage of nearby truck routes and major nontruck roadways, which we derived from a previously described37 consolidated traffic database, to have a direct influence on the volume of nearby traffic and level of associated pollution.
We obtained 2005 land use data and 2008 InfoUSA business location and employment data from the Southern California Association of Governments to account for proximity to potential traffic-generating land uses such as commercial districts, job centers, and mixed-use areas (≥ 25% nearby residential and ≥ 25% commercial use).44 We used a previously described45 firm classification scheme to identify firms that were neighborhood-serving businesses on the basis of a firm’s standard industrial classification code to identify nearby land uses that could be associated with more localized, shorter vehicle or walking trips.
A parcel’s residential use type may be related to nearby traffic levels because multifamily parcels may generate more vehicle trips. Also, older housing structures tend to have higher levels of nearby traffic, which raises concerns because these building types tend to have higher rates of indoor exposure to outdoor pollutants, including intrusion of motor vehicle exhaust.6,46
We used the city or municipality a parcel was located in as a control variable because services such as public amenities and schools could vary substantially across jurisdictions and could impact residential location choices (Figure 1).47 Although some previous studies have raised concerns regarding the inclusion of regional dummy variables in spatial regression models,32 our likelihood ratio tests showed that adding the city dummy variables could significantly improve performance of our models. We also estimated models with and without the city dummy variables and found that most of the dummy variables returned highly significant coefficients. The incidental problem,48 which can endanger the use of regional dummy variables, was not a concern for our study because we set infill asymptotics with a very large sample size and a small, fixed number of city dummy variables.49
We hypothesized that the demographic and socioeconomic variables would exert an indirect effect on exposure after controlling for transportation, land use, employment, and parcel-level characteristics. That is, although these factors are not as directly associated with traffic generation as the infrastructure and land use factors that we hypothesized to have direct effects, they could influence nearby housing affordability, community resources and cultural amenities, and residential location choices. We derived a parcel’s neighborhood characteristics using the most localized geographic scale available from the census. We obtained a parcel’s neighborhood racial/ethnic composition from the 2010 US Census BG data and obtained its socioeconomic indicators (including poverty, home ownership, and foreign-born status) from 2005–2009 US Census American Community Survey tract data.
We used spatial regression models to assess associations between exposure and socioeconomic variables after controlling for confounding variables, not to infer causality.36 Quantitative environmental justice studies have traditionally used ordinary least squares regression to evaluate community impacts of environmental hazards, but the use of spatial modeling techniques is becoming more common in environmental justice research because of their ability to address problems of spatial autocorrelation.30,31,50 Spatial autocorrelation occurs when the values of one area are influenced by the values of their neighbors, violating the assumptions of independence that ordinary least squares regression assumes. We tested the model residuals for spatial autocorrelation using the univariate Moran’s I and found spatial autocorrelation for all models. Next we ran the Lagrange multiplier diagnostic test to determine the spatial regression technique that was most appropriate for addressing spatial autocorrelation. Spatial lag models can be used to address spatial dependence in the dependent variable, and spatial error models can be used to address spatial dependence in the error terms.32,33
Both the Lagrange multiplier lag test statistic and Lagrange multiplier error test statistic suggested that spatial dependence effects may exist in both the dependent variable and the error terms for all 3 models (total VMT, truck VMT, and vehicle PM). Moreover, the robust versions of the Lagrange multiplier lag and error tests suggested that an appropriate approach for truck VMT and vehicle PM should account for potential threats of spatial autocorrelation in both the dependent variable and the error terms. This test did not indicate that spatial dependence in the error terms was a problem for the total VMT model. We did, however, estimate separate spatial lag and spatial error models for all dependent variables to understand the sensitivity of results over different modeling techniques. For consistency, the final spatial regression model reported in the Results section for all dependent variables used the Cliff-Ord approach, which adjusts for spatial dependence in both the dependent variables and the error terms.51,52 This modeling approach takes the following form: Y is an n × 1 vector of the dependent variable, and n is the total number of observations in the sample; W is an n × n spatial weight matrix, which describes the pattern of spatial dependent effects; X is an n × k matrix of independent variables, and k is the total number of independent variables; ε is an n × 1 vector of original error terms in which spatial dependence is not taken account of; u ∼ N (0, σ2In) is an n × 1 vector of error terms in which spatial dependence effects are taken account of; and λ and ρ are spatial coefficients.
where
We implemented the Cliff-Ord model using the R Project “sphet” package (Gianfranco Piras, Ithaca, NY),53 and we used modeling methodologies previously described.54–56 The model implementation relies on the instrumental variables and the generalized moments of methods estimators. We did not chose the maximum likelihood estimators because the Monte Carlo simulation of Arraiz et al.54 suggests that the maximum likelihood estimator can be substantially biased if the error terms are heteroskedastic. We relied on the commonly used Delaunay triangulation technique to create the spatial weight matrix that defines the spatial dependence effects among sample observations. Neighbors sharing a Delaunay triangle with one parcel have equal weights of impact on this parcel. We experimented with other forms of spatial weight matrices and obtained similar results.
The study area contains substantial racial/ethnic and socioeconomic diversity, and understanding population distributions in the study area in relation to major roadways and truck routes provides important context for assessing potential disparities in exposure. The 2010 US Census data indicated that the study area was home to more than 370 000 residents and had a population density greater than that of the county as a whole (about 10 300 vs 2405 persons/square mile). The densest study subareas were the northern and southern portions of Long Beach east of the I-710 freeway (Figure 1). The study area was composed of 56% Hispanic residents compared with 48% for Los Angeles County as a whole; less than one fifth (17%) of the residents were non-Hispanic White compared with almost one third (28%) for the county (Table 1). The Wilmington area of Los Angeles had the highest composition of Hispanic residents (75%), and the San Pedro area of Los Angles had the highest composition of non-Hispanic Whites (37%).
TABLE 1— Study Area and Subarea Demographic and Socioeconomic Characteristics, Census Block Groups: Southern Los Angeles County, CA, 2005–2010.
| Characteristic | Carson | Los Angeles, San Pedro | Los Angeles, Wilmington | Long Beach, North | Long Beach, South | Long Beach, West |
| Population density (persons/square mile)a | 4991 | 11 082 | 7746 | 14 884 | 25 377 | 8602 |
| Racial/ethnic composition, %a | ||||||
| Non-Hispanic White (single race) | 13 | 45 | 14 | 12 | 14 | 5 |
| Non-Hispanic Black (single race) | 8 | 5 | 4 | 17 | 14 | 12 |
| Non-Hispanic API (single race) | 34 | 5 | 9 | 24 | 13 | 32 |
| Hispanic | 39 | 36 | 68 | 39 | 54 | 41 |
| Socioeconomic and housing characteristics (tract), 2005–2009, % | ||||||
| Foreign-born persons | 39 | 20 | 35 | 36 | 37 | 49 |
| Persons in poverty | 6 | 11 | 18 | 23 | 31 | 14 |
| Owner-occupied housing units | 83 | 50 | 51 | 41 | 18 | 55 |
| Parcel characteristics | ||||||
| Multifamily residential parcel type (1/0) | 3 | 27 | 22 | 30 | 67 | 15 |
| Structure built before 1960 (1/0) | 59 | 77 | 68 | 87 | 86 | 87 |
| Land use type of parcels (within 250 m) | ||||||
| % area residential | 77 | 77 | 68 | 76 | 71 | 76 |
| % area commercial | 3 | 4 | 6 | 8 | 12 | 3 |
| Mixed-use area (1/0)b | 1 | 2 | 2 | 5 | 10 | 0 |
| Roadway type within 250 m | ||||||
| Truck route miles | 0.08 | 0.02 | 0.05 | 0.01 | 0.00 | 0.09 |
| Major (nontruck) route miles | 0.29 | 0.73 | 0.52 | 0.57 | 0.89 | 0.48 |
| Nearby employment, 2008 (BG) | ||||||
| Jobs per square mile/1000 | 2.1 | 1.8 | 2.2 | 3.2 | 3.6 | 1.2 |
| % jobs in neighborhood businesses | 48 | 50 | 49 | 46 | 51 | 51 |
| Exposures | ||||||
| Mean total VMT/100 (within 250 m) | 90.1 | 80.5 | 87.4 | 106.4 | 122.8 | 183.3 |
| Mean truck VMT/100 (within 250 m) | 4.1 | 2.4 | 2.9 | 2.9 | 2.2 | 25.7 |
| Mean vehicle PM (parcel level, μm/m3) | 3.9 | 1.9 | 2.8 | 5.8 | 3.6 | 7.0 |
Note. API = Asian/Pacific Islander; BG = block group; PM = particulate matter; VMT = vehicle miles traveled.
a Relates to 2010 BG.
b Mixed use was defined as > 25% residential and > 25% commercial.
According to 2005–2009 US Census data, about 16% of residents in the study area had an income below the federal poverty level (vs 15% for the county) and about 33% of residents were foreign-born (vs 35% for the county). The southern area of Long Beach had the highest poverty level, the lowest homeownership rate, the highest percentage of multifamily parcels, and parcels with the highest percentage of nearby commercial uses (Table 1). By contrast, the Carson area had the lowest poverty level, the highest homeownership rate, and the lowest percentage of structures built before 1960. These factors could be related to potential exposures because multifamily parcels and nearby commercial uses and employment centers could be associated with greater nearby traffic generation; also, multifamily and older housing structures may have higher rates of indoor exposure to outdoor pollutants.46 Parcels in Carson and Long Beach had the highest levels of nearby truck routes, whereas the San Pedro area of Los Angeles and the southern portion of Long Beach had the highest level of nontruck roadways.
Although the parcel exposure measures were significantly correlated (0.59–0.67), their spatial distribution varied across the study area in ways that could differentially affect nearby populations. The highest total VMT exposures occurred in the Long Beach study areas, perhaps because of proximity to the I-710 freeway (Table 1). The highest truck VMT exposures occurred in the western Long Beach study area, which has major truck routes on its western and eastern boundaries. Western and northern Long Beach had the highest levels of parcel-level vehicle PM exposures.
As expected, the means of nearby truck route miles increased sizably from parcels in the lowest exposure quartile to those in the highest quartile for all exposure metrics, but this did not hold for nontruck roadway miles (Table 2). Parcels in the highest quartile for both traffic exposure measures had lower residential use and higher commercial use and job density, but these patterns did not hold for vehicle PM exposure.
TABLE 2— Characteristics of Parcels by Magnitude of Exposure: Southern Los Angeles County, CA, 2005–2010
| Total VMT/100 (Within 250 m), Quartile Means | Truck VMT/100 (Within 250 m), Quartile Means | Vehicle PM (Parcel Level), Quartile Means | |||||||||||
| Characteristics | All | 1st | 2nd | 3rd | 4th | 1st | 2nd | 3rd | 4th | 1st | 2nd | 3rd | 4th |
| Total parcels | 46 242 | 11 560 | 11 561 | 11 561 | 11 560 | 11 560 | 11 561 | 11 561 | 11 560 | 11 560 | 11 561 | 11 560 | 11 561 |
| Direct factors: built environment characteristics | |||||||||||||
| Roadway type within 250 m | |||||||||||||
| Truck route miles | 0.038 | 0.005 | 0.025 | 0.040 | 0.084 | 0.000 | 0.004 | 0.009 | 0.141 | 0.014 | 0.038 | 0.036 | 0.065 |
| Major (nontruck) route miles | 0.597 | 0.397 | 0.574 | 0.663 | 0.755 | 0.424 | 0.661 | 0.705 | 0.598 | 0.635 | 0.577 | 0.619 | 0.558 |
| Parcel use type | |||||||||||||
| % area residential | 0.740 | 0.775 | 0.782 | 0.724 | 0.680 | 0.784 | 0.789 | 0.737 | 0.651 | 0.767 | 0.746 | 0.699 | 0.748 |
| % area commercial | 0.056 | 0.012 | 0.032 | 0.069 | 0.110 | 0.012 | 0.043 | 0.071 | 0.097 | 0.021 | 0.054 | 0.084 | 0.063 |
| Mixed use (1/0) (> 25% residential and > 25% commercial) | 0.032 | 0.003 | 0.004 | 0.024 | 0.095 | 0.003 | 0.007 | 0.034 | 0.083 | 0.006 | 0.016 | 0.068 | 0.037 |
| Nearby employment (BG) | |||||||||||||
| Jobs per square mile/1000 | 2.276 | 1.642 | 1.995 | 2.560 | 2.909 | 1.678 | 2.209 | 2.589 | 2.631 | 1.359 | 2.532 | 2.730 | 2.485 |
| % neighborhood jobs | 0.491 | 0.513 | 0.472 | 0.487 | 0.494 | 0.502 | 0.486 | 0.483 | 0.494 | 0.489 | 0.482 | 0.496 | 0.499 |
| Parcel characteristics | |||||||||||||
| Multifamily residential parcel type (1/0) | 0.269 | 0.115 | 0.230 | 0.336 | 0.393 | 0.130 | 0.300 | 0.342 | 0.303 | 0.185 | 0.346 | 0.320 | 0.223 |
| Structure built before 1960 (1/0) | 0.761 | 0.708 | 0.787 | 0.787 | 0.763 | 0.736 | 0.804 | 0.749 | 0.756 | 0.724 | 0.744 | 0.737 | 0.841 |
| Indirect factors: demographic and socioeconomic characteristics | |||||||||||||
| Nearby racial/ethnic composition (BG), % | |||||||||||||
| Non-Hispanic Black, 2010 (single race) | 0.085 | 0.066 | 0.074 | 0.092 | 0.108 | 0.073 | 0.081 | 0.096 | 0.090 | 0.034 | 0.061 | 0.107 | 0.137 |
| Non-Hispanic API, 2010 (single race) | 0.160 | 0.174 | 0.164 | 0.140 | 0.162 | 0.178 | 0.138 | 0.160 | 0.165 | 0.045 | 0.120 | 0.218 | 0.257 |
| Hispanic, 2010 | 0.469 | 0.446 | 0.434 | 0.479 | 0.518 | 0.431 | 0.454 | 0.426 | 0.566 | 0.363 | 0.583 | 0.502 | 0.429 |
| Nearby socioeconomic status, 2005–2009 (tract), % | |||||||||||||
| Poverty | 0.162 | 0.117 | 0.151 | 0.185 | 0.196 | 0.123 | 0.170 | 0.169 | 0.187 | 0.090 | 0.179 | 0.204 | 0.175 |
| Home ownership | 0.501 | 0.625 | 0.524 | 0.437 | 0.417 | 0.602 | 0.469 | 0.451 | 0.482 | 0.571 | 0.457 | 0.479 | 0.496 |
| Foreign-born | 0.334 | 0.328 | 0.320 | 0.326 | 0.360 | 0.326 | 0.312 | 0.321 | 0.376 | 0.205 | 0.349 | 0.379 | 0.402 |
Note. API = Asian-Pacific Islander; BG = block group; PM = particulate matter; VMT = vehicle miles traveled.
Parcels in the highest quartile for both traffic exposure measures consistently had a higher percentage of nearby Black, Hispanic, and poor residents. Parcels in the highest quartile for vehicle PM exposure had a higher percentage of nearby Black and Asian/Pacific Islander (API) residents, but this pattern did not hold for the percentage of Hispanic and poor residents. Parcels in the lowest quartile for all exposure metrics had higher home ownership, and those in the highest quartile had a higher percentage of foreign-born residents.
We specified 6 regression models to assess potential racial/ethnic and socioeconomic disparities in exposure after accounting for land use, built environment, and infrastructure factors that could be associated with traffic generation (Table 3). For each dependent variable (total VMT, truck VMT, and vehicle PM), we have reported a Cliff-Ord model that accounts for spatial autocorrelation in both the dependent variables and the error terms. The λ and ρ variables in the Cliff-Ord models were statistically significant.
TABLE 3— Multivariate Analysis of Exposure, Residential Parcels: Southern Los Angeles County, CA, 2005–2010
| Independent Variables | Total VMT/100 (Within 250 m), Model 1, Coefficient | Truck VMT/100 (Within 250 m), Model 2, Coefficient | Vehicle PM (Parcel Level), Model 3, Coefficient |
| Intercept | −33.42** | −12.15*** | 1.05*** |
| Direct factors: built environment characteristics | |||
| Roadway type (within 250 m) | |||
| Truck route miles/100 | 711.95*** | 86.11*** | 7.10*** |
| Major (nontruck) route miles/100 | 163.00*** | 7.47*** | 0.85*** |
| Land use type of parcels (within 250 m) | |||
| % area residential | 12.38 | 6.32*** | −0.21 |
| % area commercial | 182.37*** | −0.90 | −1.02* |
| Mixed-use area (1/0)a | −2.59 | 0.80** | 0.13** |
| Nearby employment, 2008 (BG) | |||
| Jobs per square mile/1000 | −0.60*** | −0.07*** | −0.01** |
| % jobs in neighborhood businesses | −11.60*** | −1.12* | −0.08 |
| Parcel characteristics | |||
| Multifamily residential parcel type (1/0) | −0.81 | −0.08 | −0.02* |
| Structure built before 1960 (1/0) | −0.88 | 0.08 | 0.03** |
| Indirect factors: demographic and socioeconomic characteristics | |||
| Nearby racial/ethnic composition, 2010 (BG), % | |||
| Non-Hispanic Black (single race) | 193.04*** | 19.83*** | 3.18*** |
| Non-Hispanic API (single race) | 58.17*** | −1.77 | 1.42*** |
| Hispanic | −37.63*** | −2.22 | 0.76*** |
| Nearby socioeconomic status, 2005–2009 (tract), % | |||
| Poverty | −101.59*** | 7.46 | −1.54*** |
| Home ownership | −2.19 | 5.24*** | −0.23 |
| Foreign-born | −40.68* | −0.12 | 0.65 |
| Municipal subareasb | |||
| Los Angeles, Wilmington area (1/0) | 32.43*** | 0.31 | 0.51*** |
| Long Beach, western area (1/0) | 96.85*** | 22.11*** | 5.24*** |
| Long Beach, northern area (1/0) | 42.12*** | −1.74 | 3.02*** |
| Long Beach, southern area (1/0) | 24.50*** | −0.18 | 1.47*** |
| Carson (1/0) | 4.84 | −4.28** | 0.97*** |
| Spatial lag coefficient on dependent variable (λ) | 0.01 | 0.00 | −0.09*** |
| Spatial lag coefficient for errors (ρ) | 0.90*** | 0.90*** | 0.90*** |
Note. API = Asian-Pacific Islander; BG = block group; PM = particulate matter; VMT = vehicle miles traveled. We have reported the direct effect coefficients (β in equation 1) for these Cliff-Ord models. According to Li and Saphores55 and Saphores and Li,56 the direct effect coefficients are very similar to the total effect coefficients in spatial regression models. The sample size was n = 46 242 parcels.
a Mixed use was defined as > 25% residential and > 25% commercial.
b The San Pedro area is the excluded reference category.
*P < .05; **P < .01; ***P < .001.
As expected, more nearby roadway and truck route mileage was associated with higher exposure for all measures. More nearby commercial land use was associated with higher total VMT exposure but unexpectedly lower vehicle PM exposure. More residential land use was associated with higher truck VMT exposure. Nearby mixed land use was associated with higher truck VMT and vehicle PM exposure, but more nearby neighborhood-serving businesses were associated with lower traffic exposure. Unlike descriptive results, parcels with more nearby job density were associated with lower exposure after controlling for other factors.
After controlling for nearby built environment factors that could be associated with traffic generation, parcels in BGs with a higher percentage of Black residents were associated with higher exposures for all exposure measures, a higher percentage of nearby API residents was associated with higher VMT and vehicle PM exposure, and a higher percentage of nearby Hispanics was associated with higher vehicle PM exposure. Contrary to descriptive results, however, a higher percentage of Hispanic, poor, and foreign-born residents was associated with lower total VMT exposure after controlling for other factors. Higher home ownership was associated with higher truck VMT exposure, and more nearby poverty was associated with lower vehicle PM exposure.
We found racial/ethnic disparities in traffic and vehicle PM exposure in a major goods movement corridor after controlling for built environment and land use factors associated with traffic generation, particularly for parcels with a higher percentage of nearby non-Hispanic Black and API residents. Interestingly, a higher percentage of nearby Hispanic residents was associated with higher total VMT exposure but lower vehicle PM exposure, suggesting racial/ethnic disparities are nuanced depending on the exposure metric used, the distribution of the emission source, and pollutant dispersal patterns.
In contrast to most available distributional studies of traffic and mobile-source air pollution exposure,5,6,8,10 we found that lower socioeconomic status (more foreign-born and poor residents) tended to be associated with lower exposure and that higher socioeconomic status (more home ownership) tended to be associated with higher exposure. This finding, however, is in line with a handful of stationary- and mobile-source air pollution and traffic exposure studies that found little or no income-based disparity17,57 or higher exposures for higher income or nonpoor areas.18,19,58
Further research is needed to investigate factors underlying disparate exposures in goods movement corridors. Transportation infrastructure, land use, and residential patterns emerged historically in the context of structural inequalities, uneven development patterns, and residential and economic segregation.6,32 Future research should seek to understand local awareness of traffic and air pollution hazards, whether housing market constraints restrict the residential location choices of subpopulations, and whether some residents are more likely to accept higher levels of exposure to live in more affordable, accessible, or culturally diverse areas.
Nearby roadway designation may play an important role, considering that nearby truck route miles were a much stronger predictor of exposure than were nontruck roadway miles. We found some evidence that commercial strips and mixed land use areas may be associated with higher traffic and vehicle PM exposure, which raises concerns about whether policies promoting mixed-use land development could result in higher exposures.
Our study has several limitations. Our results cannot be readily generalized, and future research is needed to assess whether similar disparities exist in other corridors with divergent population, built environment, and land use geographic patterns. We assessed population impacts on the basis of a parcel’s BG and tract composition, but exposures may also vary systematically by the characteristics of parcel residents. Lastly, our vehicle PM exposure measure may underestimate cumulative parcel air pollution exposures because it does not account for nonvehicle sources, such as petroleum refineries and idling cargo ships and pollution transported from other regions.
Despite these limitations, this study makes several contributions to the environmental justice literature. First, it stresses the importance of understanding environmental disparities in transportation and goods movement corridors in ways that inform infrastructure and land use planning. Second, we have provided the first assessment to our knowledge of exposure disparities at the parcel level, a more geographically refined spatial resolution more appropriate for examining near-roadway impacts. Third, we used 3 metrics of exposure that reflect geographic differences in emission sources and pollutant dispersion patterns. Fourth, we used spatial regression techniques to account for spatial autocorrelation in our assessment of environmental inequities. Our results raise concerns that planning and policy mechanisms to lower ambient air pollution levels will likely not be sufficient to protect those most exposed to mobile-source pollution.
Acknowledgments
We are grateful to Paul Ong, who provided helpful suggestions on the study conceptualization, and Leah Brooks, who provided helpful assistance obtaining parcel data.
Human Participant Protection
No protocol approval was needed for this study because it relied on secondary data containing no personal identifying information.
