Assessment of environmental injustice in Korea using synthetic air quality index and multiple indicators of socioeconomic status: A cross-sectional study

ABSTRACT

Despite the existence of the universal right to a healthy environment, the right is being violated in some populations. The objective of the current study is to verify environmental discrimination associated with socioeconomic status in Korea, using synthetic air quality index and multiple indicators of socioeconomic status. The concentrations of NO₂ (nitrogen dioxide), CO (carbon monoxide), SO₂ (sulfur dioxide), PM₁₀ (particulate matter with an aerodynamic diameter <10 μm), and O₃ (ozone) in ambient air were integrated into a synthetic air quality index. Socioeconomic status was measured at individual level (income, education, number of household members, occupation, and National Basic Livelihood status) and area level (neighborhood index). The neighborhood index was calculated in the finest administrative unit (municipality) by performing standardization and integration of municipality-level data of the following: number of families receiving National Basic Livelihood, proportion of people engaged in an elementary occupation, population density, and number of service industries. Each study participant was assigned a neighborhood index value of the municipality in which they reside. Six regression models were generated to analyze the relationship between socioeconomic status and overall air pollution. All models were adjusted with sex, age, and smoking status. Stratification was conducted by residency (urban/rural). Moran’s I was calculated to identify spatial clusters, and adjusted regression analysis was conducted to account for spatial autocorrelation. Results showed that people with higher neighborhood index, people living with smaller number of family members, and people with no education lived in municipalities with better overall air quality. The association differed by residency in some cases, and consideration of spatial autocorrelation altered the association. This study gives strength to the idea that environmental discrimination exists in some socioeconomic groups in Korea, and that residency and spatial autocorrelation must be considered in order to fully understand environmental disparities.Implications: This is the first study that provides the possible evidence of the environmental injustice in Korea using air quality index. The findings suggested that air quality index was negatively correlated with several important socioeconomic status measured at either individual or area level. The main implication of this paper, therefore, is to provide another insight to environmental policy makers to consider environmental injustice problem into community intervention for resolving the public health problems by air pollution.

Introduction

It is stated in the Universal Declaration of Human Rights of the United Nations (UN) that everyone has the right to a healthy environment (UN, 2015). Many countries, including the Republic of Korea (Ministry of Government Legislation, 2013), state the right in their national constitutions or enforce legislations for the reduction of environmental hazards. However, the right is being violated due to undesirable land use and environmental pollution introduced by rapid industrialization and urbanization. Such violations led to some population, such as socioeconomic minorities, being exposed to unhealthy environments. The unequal distribution of environmental quality is conceptualized in synonymous terms such as environmental inequity, environmental injustice, environmental inequality, environmental discrimination, etc. (Briggs et al., 2008).

The issue of environmental inequity roots from environmental racism spotted by scholars and civil rights activists in the United States. The origin of environmental racism is seen as the civil movement to stop polychlorinated biphenyl-polluted soil from being dumped to a county (Warren County) with the highest proportion of African Americans in North Carolina in 1982 (Mohai et al., 2009). Afterwards, environmental injustice has been studied actively in America with regards to race being the most important factor of environmental discrimination (U.S. General Accounting Office, 1983), and studies were soon looking into other socioeconomic status variables (Brulle and Pellow, 2006). In other words, the disproportionate impact of environmental hazards and pollution was first observed in regards to race and then spread to minority communities regardless of racist extent.

Along with the environmental injustice issue, scientific evidences of environmental pollutions as risk factors for adverse health outcomes (Kampa and Castanas, 2008) pose further implications of health disparities. The socioeconomically deprived being prone to environmental pollution exposure and being more susceptible, end up with poor health, which is referred to as triple jeopardy (O’Neill et al., 2003). However, the complicated relationships between socioeconomic status (SES), environmental pollution, and health status make it difficult to disentangle the underlying true relationships. That is because SES of a person can influence various pathways, which may then lead to a specific health outcome. In other words, SES is necessarily associated with health through proximal exposure or risk factors (Blakely et al., 2004).

Environmental injustice has been studied with respect to race/ethnicity (Faber and Krieg, 2002), poverty (Faber and Krieg, 2002; Mitchell and Dorling, 2003), and other SES variables (Goodman et al., 2011). Some studies suggest that local exposure to undesirable land use, such as waste facilities and industrial pollution (Faber and Krieg, 2002), and air pollution (Mitchell and Dorling, 2003; Goodman et al., 2011), are unequally distributed in a way that socioeconomic minorities are more likely to be exposed to. However, most of the studies on environmental justice had been conducted in Western countries, and the results remain inconclusive (Ringquist, 2005; Deguen and Zmirou-Navier, 2010). The existence of environmental injustice varies depending on the type of environmental exposure. In the case of air pollution, most of the previous studies were conducted on single pollutants, whereas people are rarely exposed to single pollutants (Dominici et al., 2010). Also, the relationship between SES and air pollution differ by the type of indicators used to represent SES and the geographic scale in which the SES variables were measured in (Briggs et al., 2008). That is because SES can be represented with multiple indicators (e.g., race, income, education, etc.), measured in multiple scales (e.g., individual-level, area-level), and each indicator holds different implications. Furthermore, it is known that the characteristics of environmental discrimination may vary by the spatial scale and scope of analysis (Cutter et al., 1996; Baden et al., 2007) and the geographic patterns and characteristics of a community (Brulle and Pellow, 2006). Consideration of spatial autocorrelation is also known to alter the relationship between air pollution and SES (Jerrett et al., 2001), whereas only recently has such concept been statistically considered in the analyzing procedures. Such inconsistency in results implies the need for further studies to be conducted with multiple indicators of SES and air pollution, and in different regions regarding spatial autocorrelation and the local characteristics to strengthen the evidence of environmental injustice.

Limited number of studies has been conducted on environmental justice in Korea. Previous literature in Korea focused on finding environmental determinants of a disease, with SES being considered either as a confounder (Son et al., 2010) or an effect modifier (Kim et al., 2007). Only few studies (Chu, 2009; Byun et al., 2010) examined the relationship between unequal distribution of air pollution in relation to SES. The previous studies focused on the environmental injustice in respect to single air pollutants, which limits the inference of environmental inequalities in relation to immeasurable or unmeasured air pollution and overall air quality. Also, study areas were limited to single cities, and indicators or indices to represent SES were arbitrary. Therefore, further analysis need to be conducted to enable interpretation on environmental injustice associated with the overall air pollution.

In the current study, we examined the relation of SES to overall air quality in the Republic of Korea, using nationwide data. The goals were to examine the existence and scope of environmental injustice in Korea using multiple indicators of SES and synthetic air quality index, and to provide a basis for further studies on environmental justice.

Materials and methods

Study population was defined as 229,229 participants of the 2010 Community Health Survey of the Korea Centers for Disease Control and Prevention, covering 242 municipalities in Korea. Individual-level data on residential address (municipality of residence), smoking status, age, sex, whether the person was living in an urban or a rural area (residency), education, monthly income (1,000,000 ₩), occupation, number of household members, and National Basic Livelihood (NBL) status were derived from the database. NBL is a governmental aid designed to secure minimum standards of living for those who are incapable to living on their own earnings (Jung, 2005).

Exclusions were made for the individuals who had missing covariates or resided in municipalities with missing information on air pollution. The restrictions resulted in 117,588 persons residing in 123 municipalities (Figure 1).

Figure 1. Study area and data availability in the final target population. Study area is South Korea, and among a total of 242 municipalities in South Korea, air quality data were available for 133 municipalities.

SES variables

Individual level

Education, income, occupation, number of household members, and NBL status were selected as the individual-level indicators based on previous literature (Lanjouw and Ravallion, 1994). Race was not selected because although race is conventionally thought of as a representative SES indicator, such indicator is not as useful in homogeneous nations (Ma, 2010). Since the proportion of foreigners residing in Korea makes up only 1.99% of the total population of Korea (KOSIS, 2015), race is not likely to be very representative in Korea.

For the ease of interpretation, selected indicators were modified so that an increment implies higher SES. The reciprocal of the number of household members was used because increase in the number of household members is conventionally thought to represent low SES (Lanjouw and Ravallion, 1994). Occupational categories of 13 in the original data were recategorized into 8 by conducting analysis of variance (ANOVA) to group occupations with similar income (Table 1).

Table 1. Occupational categories in the original data and the modified data.

Occupational Categories in the Original Data	Recategorization of the Occupation
Administration	Administration
Specialized job	Specialized or office job
Office job
Student	Student, soldier, salesman
Soldier
Salesman
Craftsman	Craftsman, machine operator, service worker, housewife
Machine operator
Service worker
Housewife
Agriculture and fisheries	Agriculture and fisheries
Simple labor	Simple labor
None	None

Area level

Neighborhood index was developed to represent the area-level SES. Neighborhood index consisted of socioeconomic factors and accessibility to resources. For each municipality, five indicators (households receiving NBL, population density, employed persons 15 years of age and older, persons engaged in simple labor, number of service industries) were normalized and averaged to calculate neighborhood index (Table 2). Each person was assigned with a neighborhood index value of the residential municipality.

Table 2. Indicators used to calculate the neighborhood index.

Category	Indicator	Normalization Method	Reference
Wealth	Households receiving NBL (%)		2010 mean-tested benefit recipients
	Population density		Household registration statistics 2010 Korea statistical yearbook 2010
Occupation	Employed persons 15 years of age or older (%)		2010 population and housing census
	Persons engaged in simple labor (%)		2010 population and housing census
Accessibility to resources and services	Number of service industries		Total service industry survey 2005

Notes: NBL = National Basic Livelihood; X_i = percentage of households receiving NBL in the ith municipality; minX = minimum value among X_is; Y_i = population density of the ith municipality; minY = minimum value among Y_is; Z_i = percentage of employed persons 15 yr of age or older in the ith municipality; maxZ = maximum value among Z_is; V_i = percentage of persons engaged in simple labor in the ith municipality; minV = minimum value among V_is; W_i = number of service industries in the ith municipality; maxW = maximum value among W_is.

Synthetic air quality index

Synthetic index is useful to generalize the overall trend and to summarize a complex situation into a single figure (Bruno et al., 2007) and has been applied in previous studies (Wcislo et al., 2002). We analyzed the relationship between synthetic air quality index and the SES variables to better understand the relationship between SES variables and the overall air qualities. The synthetic air quality index (S_Air) were derived with the municipality-level concentrations of NO₂ (nitrogen dioxide), CO (carbon monoxide), SO₂ (sulfur dioxide), PM₁₀ (particulate matter with an aerodynamic diameter <10 μm), and O₃ (ozone) in ambient air obtained from the Annual Report of Ambient Air Quality in Korea, 2010.

For each municipality, the municipality-level concentrations of NO₂, CO, SO₂, PM₁₀, and O₃ in the air were normalized and integrated into S_Air. Normalization and integration (Zhou et al., 2006) were conducted according to eqs 1 and 2, and then multiplied by 100. As the increase in air pollution concentrations implies worse air quality, the second equation in eq 1 was used for the normalization.

(1)

(2)

where S_Air is the S_Air in the ith municipality; Y_ij is the value of the jth indicator in the ith municipality; Y_{ij_norm} is the normalized value of the jth indicator in the ith municipality; maxY_ij and minY_ij are the maximum and minimum values of the jth indicator in the ith municipality, respectively; and n is the number of indicators.

Statistical analysis

Regression analysis was conducted to examine the relationship between SES and air pollution. The effects of sex, age, and smoking status were adjusted in the regression model. Stratified analysis was conducted by residency, and a total of six models were generated (eq 3):

(3)

where y_i is S_Air; x_1i is the SES variable (neighborhood index, income, education, job, National Basic Livelihood status, reciprocal of number of household members); x₂ is sex; x₃ is age; x₄ is smoking status; and ε_i is the error term

Spatial autocorrelation was tested by calculating global Moran’s I and Getis-Ord general G. Both statistics are generally used to test spatial autocorrelation by either employing the sums of products (G statistics) or the covariances (Moran’s I) (Getis and Ord, 1992). Anselin Local Moran’s I and Getis-Ord G_i* were calculated to identify and map the regions of spatial clusters. The row-standardized spatial weights used in the analysis were generated with the principle that only neighboring polygon features that share a boundary will influence computations for that feature.

The Anselin Local Moran’s I provides five categories of clusters: HH (high regions surrounded by high regions), HL (high regions surrounded by low regions), LH (low regions surrounded by high regions), LL (low regions surrounded by low regions), and NS (spatial cluster not statistically significant). Multiple regression analysis was conducted with adjustment on spatial autocorrelation status by adding nonparametric terms representing the spatial clustering status to the six models described previously.

All regression analysis was conducted using the statistical program SAS 9.4 (SAS Institute Inc., Cary, NC, USA). Global Moran’s I, Getis-Ord General G, Anselin Local Moran’s I, Getis-Ord G_i*, and spatial weights were derived using ArcGIS (ArcGIS Desktop, Release 10, 2011; Environmental Systems Research Institute, Redlands, CA, USA).

Results

Characteristics of the study population

The study population consisted of 117,588 persons. The majority were aged 30–59 (59.72%), females (54.43%), and lived in urban areas (78.89%). The distribution of final education and job differed by residency (Table 3). The average overall air quality (urban = 52.22; rural = 58.59) was better in rural areas compared with urban areas, but variance was higher in rural areas. Stratification by residency was applied in further analysis.

Table 3. Descriptive analysis and characteristics of the study population, according to residence.

		Urban		Rural
Category	Subcategory	n	%	n	%
Sex	Female	50,545	54.48	13,457	54.22
	Male	42,225	45.52	11,361	45.78
Age	19–29	14,668	15.81	2,527	10.18
	30–39	19,590	21.12	4,029	16.23
	40–49	20,446	22.04	4,615	18.60
	50–59	16,961	18.28	4,580	18.45
	60–69	12,087	13.03	4,269	17.20
	70 and over	9,018	9.72	4,798	19.33
Smoking status	Nonsmoker	71,856	77.46	19,219	77.44
	Smoker	20,884	22.51	5,597	22.55
Education	None	5,484	5.91	3,906	15.74
	Elementary school	10,177	10.97	4,673	18.83
	Middle school	9,541	10.28	3,009	12.12
	High school	35,498	38.26	8,325	33.54
	College and higher	31,809	34.29	4,853	19.55
Job*	Administration	2,850	3.07	599	2.41
	Specialized or office jobs	18,808	20.27	2,702	10.89
	Students, soldiers, and salesmen	11,493	12.39	1,779	7.17
	Craftsmen, machine operators, and service workers	16,116	17.37	3,967	15.98
	Housewives	21,176	22.83	5,088	20.50
	Agriculture and fisheries	1,245	1.34	4,794	19.32
	Simple labor	6,247	6.73	1,549	6.24
	None	11,679	12.59	3,408	13.73
NBL status	Current recipient	2,991	3.22	955	3.85
	Former recipient	532	0.57	210	0.85
	Nonrecipient	89,204	96.16	23,632	95.22
Income (1,000,000 ₩)^‡		3.04 ± 2.21		2.50 ± 2.16
Neighborhood index^‡		0.34 ± 0.04		0.33 ± 0.03
Family size^‡		0.39 ± 0.22		0.43 ± 0.24

Notes: NBL = National Basic Livelihood. *Jobs listed in the order of high-income occupation to low-income occupation. ^‡Mean ± SD.

Relationship between SES and S_Air

A positive regression coefficient indicates a positive relationship between SES index and S_Air. In other words, positive regression coefficients imply that people with higher SES lived in municipalities with better air quality. Results showed that people with higher neighborhood status (urban: β = 16.87, P < 0.0001; rural: β = 81.16, P < 0.0001) and fewer household members (urban: β = 0.43, P < 0.0001; rural: β = 2.75, P < 0.0001) lived in municipalities with better air quality. Former recipients of National Basic Livelihood lived in municipalities with better air quality than current recipients (urban: β = 1.10, P < 0.0001; rural: β = 1.61, P = 0.02). People engaged in simple labor (urban: β = −0.20, P = 0.03; rural: β = −2.43, P < 0.0001) and housewives (urban: β = −0.17, P = 0.02; rural: β = −1.94, P < 0.0001) lived in municipalities with worse air quality compared with that of jobless people (Table 4).

Table 4. Multiple regression analysis of SES and S_Air, stratified by residence.

		Urban				Rural
Variable		β^‡	P^‡	β^§	P^§	β^‡	P^‡	β^§	P^§
Neighborhood index		16.87	<0.0001	5.44	<0.0001	81.16	<0.0001	8.47	<0.0001
Family size		0.43	<0.0001	0.29	<0.0001	2.75	<0.0001	0.82	<0.0001
Income		–0.02	0.014	–0.02	<0.001	–0.13	<0.0001	0.05	0.01
Education	None	0.00	–	0.00	–	0.00	–	0.00	–
	Elementary school	–0.35	<0.001	–0.16	0.03	–1.60	<0.0001	–0.39	<0.01
	Middle school	–0.26	0.011	–0.003	0.97	–1.86	<0.0001	–0.56	<0.001
	High school	–0.32	0.001	–0.05	0.48	–2.56	<0.0001	–0.75	<.0001
	College and higher	0.08	0.433	–0.01	0.9	–2.41	<0.0001	–0.55	<0.001
Job*	None	0.00	–	0.00	–	0.00	–	0.00	–
	Simple labor	–0.20	0.027	1.95	0.05	–2.43	<0.0001	0.04	0.82
	Agriculture and fisheries	4.86	<0.0001	14.59	<.0001	2.94	<0.0001	0.74	<.0001
	Housewives	–0.17	0.021	0.48	0.63	–1.94	<0.0001	0.36	0.01
	Craftsmen, machine operators, and service workers	0.22	0.003	2.87	<0.01	0.18	0.466	0.45	0
	Students, soldiers, and salesmen	0.18	0.021	1.78	0.07	–1.07	<0.001	0.22	0.2
	Specialized or office jobs	0.11	0.149	0.88	0.38	–1.25	<0.0001	0.3	0.05
	Administration	0.18	0.121	2.88	<0.01	–2.51	<0.0001	0.49	0.04
NBL status	Current recipient	0.00	–	0.00	–	0.00	–	0.00	–
	Former recipient	1.10	<0.0001	–0.19	0.01	1.61	0.022	–0.01	0.95
	Nonrecipient	–0.27	0.009	1.15	<0.0001	–1.00	0.001	0.41	0.31

Notes: NBL = National Basic Livelihood. *Jobs listed in the order of low-income occupation to high-income occupation. ^‡Adjusted for sex, age, and smoking status. ^§Adjusted for sex, age, smoking status, and spatial autocorrelation.

Negative relationships between SES and S_Air were observed as follows: a unit increase in income indicated a significant decrease in S_Air of urban and rural municipalities by 0.02 and 0.13, respectively. People with no education lived in municipalities with less air pollution than did those with higher levels of education. Nonrecipients of National Basic Livelihood lived in municipalities with more air pollution than current recipients (urban: β =−0.27,P = 0.01; rural: β = −1.00, P = 0.001) (Table 4).

Spatial autocorrelation

The two methods in global statistics and local statistics showed similar results. Therefore, we present the results from global Moran’s I and Anselin Local Moran’s I in the paper, and the results by Getis-Ord G in Supplemental Material. Spatial autocorrelation existed in S_Air (Moran’s index = 0.56, Z-score = 7.86, P < 0.0001). There were three types of spatial clusters in S_Air: 7 HH, 1 HL, and 12 LL (Figure 2).

Figure 2. Spatial clusters of S_Air identified in the study area. The spatial clusters were identified with the calculation of Local Moran’s I, and classified as high-high, high-low, low-low, and not significant clusters.

Adjustment for spatial autocorrelation significantly altered the association between SES indicators and overall air pollution. The relationships between several SES indicators (neighborhood index, education, and family size) and S_Air were attenuated. However, the relationship between NBL status and S_Air was reversed. Regardless of residency, people with higher neighborhood index, people living with smaller number of family members, people with occupation, and nonrecipients of the NBL lived in municipalities with better overall air quality. However, the relationship was reversed for people with education (Table 4).

Discussion

The current study analyzed the relationship between SES and overall air quality in Korea, using multiple indicators. In particular, we explored the effect modification of residency and considered spatial autocorrelation. The results showed that the relationship between SES and air exposure was influenced by residency. Also, spatial autocorrelation existed and modified the relationship between SES and air pollution. People with higher neighborhood index, people living with smaller number of family members, and people with occupation lived in municipalities with better air quality. However, inverse relation was found between education and air quality.

In the current study, residency influenced the strength of association between SES variables and S_Air. The overall air quality was better in rural areas compared with urban areas. Possible explanations could be that rural areas in Korea consist mostly of green land and mountains, whereas urban areas are highly populated and filled with roads. The relationship was stronger for neighborhood index in rural areas and job and NBL in urban areas. Stronger relationship observed for neighborhood index in rural areas can be explained by the fact that the overall air quality is better and its variance is larger in rural areas. Also, inference can be made that the municipality of residence and the SES at municipality level is more important in explaining the environmental injustice. However, in terms of job and NBL where the most socioeconomically deprived are compared with the least socioeconomically deprived, the associations were stronger in urban areas. Persons engaged in agriculture and fisheries lived in municipalities with better overall air quality, and the nonrecipients lived in municipalities with better overall air quality than the recipients of NBL. One explanation could be that even though the NBL recipients and the jobless in rural areas are exposed to relatively bad air quality when compared with others in rural areas, they live in municipalities with relatively good air quality when compared with the same SES groups residing in urban areas because rural areas have better overall air quality compared with urban areas in general. Heterogeneity in the strength of associations by residency observed in the current study is similar to the results of a previous study in Hong Kong (Fan et al., 2012). In the previous study, SES was measured with an area-level index defined as Social Deprivation Index (SDI). An association between SDI and vehicular air pollution was only found in stratified analysis by residency. Current study results add to the previous study that effect modification of residency is present in wider settings (e.g., multiple SES variables, overall air quality, different country) and imply the importance of considering residency in exploring environmental injustice.

Spatial autocorrelation was present in the current study. A lack of consideration for spatial autocorrelation could lead to a biased results, increasing the probability of false-positive findings (Jerrett et al., 2001). Adjustment of spatial autocorrelation resulted in attenuation of the strength of association in several SES variables, which is in concordance with previous literature (Jerrett et al., 2001; Sheppard et al., 1999). The influence of spatial autocorrelation did not only attenuate, but also amplified the associations in the current study. Therefore, spatial autocorrelation must be taken into account in studies regarding environmental disparities.

People living in municipalities with high neighborhood index and living with less number of family members lived in municipalities with better overall air quality. This is in concordance with previous studies where people with low SES are more likely to be exposed to environmental pollution, including air pollution (Pearce and Kingham, 2008; Briggs et al., 2008). However, education was negatively associated with the overall air quality. The results suggest that those with low SES are not always subject to worst overall air quality. This could be explained by several factors. For example, final education is a more proximal factor to exposure than living in a municipality with high neighborhood index or living with less number of family members. Also, exposure misclassification and residual confounding may be present because individual measurement on air quality was not conducted and factors such as mobility was not considered in the current study.

Environmental injustice existed in Korea when SES was represented by neighborhood index, number of family members, and occupation, which is in concordance with previous studies conducted in Korea (Chu, 2009; Byun et al., 2010). In the previous studies, increase rate of NO₂ was positively related to composite deprivation index, and low SES characterized by paternal education, region, type of housing, floor of residence, year of home construction, average monthly household expenses, and the number of children contributed to higher indoor PM₁₀ concentrations. The current study not only reconfirmed the existence of environmental injustice in Korea, but also provides further implications unavailable in the previous studies. It can be inferred from the current study that environmental injustice is present in Korea, nationally, in regards to overall air quality and multiple SES variables.

Strengths of the current study are as follows. First, S_Air were used to represent the overall air quality, where previous studies conducted analysis on single pollutants. However, people are rarely exposed to single pollutants, rather exposed to mixture of multiple pollutants (Dominici et al., 2010). Also, it is difficult to make further inference on unmeasured air pollution with the results of single pollutants (Wcislo et al., 2002). Therefore, utilization of S_Air enables further inference on environmental inequalities associated with the overall and unmeasured air pollution. Second, multiple indicators were used to represent SES. It is difficult to capture SES with a single indicator (Pearce et al., 2010), and SES provides different inferences depending on the spatial scale that is measured in. Individual-level SES provides inference on the individual, and area-level SES provides inference at the aggregate level (Faber and Krieg, 2002) or as a surrogate measure for personal SES (Crowder and Downey, 2010). Area-level SES may be useful to characterize the availability of resources, which may then serve as proxy of the SES of a community in a manner that personal SES cannot (Diez Roux, 2001). Living in a community of high SES may be beneficial to an individual regardless of the individual’s personal SES (Poortinga, 2006). Therefore, presenting area-level SES along with individual-level SES could be useful. Third, nationwide data were used to analyze environmental injustice in Korea where previous studies were conducted in single cities. Lastly, current study can be further applied to understand the relationship between SES and overall environmental pollution by considering multiple dimensions of environmental pollution (e.g., noise pollution, light pollution, water quality, waste generation, industrial plants).

The current study holds several limitations. First, there is a possibility of measurement errors because we were unable to conduct individual monitoring to estimate air pollution due to limited access to data of residential address. Although it would be best to conduct personal monitoring of air pollution exposures, many epidemiological studies perform exposure assessment at aggregate level (Samet et al., 2000) or apply spatial interpolation (inverse distance weighting, kriging, land-use regression) (Nuckols et al., 2004). Therefore, further studies incorporating methodologies to improve the validity of individuals’ exposure status is needed. Second, municipalities with missing air pollution data were excluded. The exclusion resulted in large reduction of the study population and may have caused selection bias. Third, the concentrations of PM_2.5 were not considered due to data availability issues, although it is a known risk factor for many adverse health outcomes. Lastly, cross-sectional analysis was conducted in the current study. Cross-sectional analysis is useful in understanding a correlation, but temporal and causal inference cannot be confirmed (Gordis, 2009). Therefore, further studies incorporating complicated study design and analysis methods (O’Neill et al., 2003) are needed to better understand the true relationship between SES and air pollution.

Studies on environmental justice could help integrate environmental injustice problem into community interventions (Baron et al., 2009) and create frameworks for policies and decision-making (Pearsall and Pierce, 2010). Previous studies have shown how study results on environmental justice could help policy makers set priorities and modify childhood asthma policies, which could resolve asthma discrimination (Kreger et al., 2011). As the environmental injustice has been confirmed in the current study, future studies examining more specific population with enhancement in methodological issues are needed, to be used as a basis for decision-making and developing detailed policy plans.

Conclusions

Overall, we confirmed that the distribution of overall air pollution was associated with SES and identified the existence of environmental injustice in the Republic of Korea. The extent of environmental injustice differed by residency (urban/rural). Spatial autocorrelation also altered the relationship between SES and air pollution. People with higher neighborhood index, people living with smaller number of family members, and people with occupation lived in municipalities with better air quality. However, the association between SES and air quality differed depending on the type of SES variables used. Therefore, careful interpretation is needed in understanding environmental injustice. Further studies should be conducted in different regions with multiple indicators of SES and overall air quality to strengthen the evidence of environmental injustice.

Funding

This study was supported by an Institute of Health Science Grant, Korea University.

Supplemental Material

Supplemental data for this article can be accessed on the publisher’s Web site.

Additional information

Funding

This study was supported by an Institute of Health Science Grant, Korea University.

Notes on contributors

Giehae Choi

Giehae Choi and Seulkee Heo were MPH students in Korea University. Giehae Choi is interested in environmental epidemiology and its policy implications. Seulkee Heo is interested in pollution and climate change.

Jong-Tae Lee

Jong-Tae Lee is a professor at the Division of Health Policy and Management, Korea University, and is interested in air pollution, climate change, health impact assessment, and environmental epidemiology.