Air pollution and individuals’ mental well-being in the adult population in United Kingdom: A spatial-temporal longitudinal study and the moderating effect of ethnicity

Background Recent studies suggest an association between ambient air pollution and mental well-being, though evidence is mostly fragmented and inconclusive. Research also suffers from methodological limitations related to study design and moderating effect of key demographics (e.g., ethnicity). This study examines the effect of air pollution on reported mental well-being in United Kingdom (UK) using spatial-temporal (between-within) longitudinal design and assesses the moderating effect of ethnicity. Methods Data for 60,146 adult individuals (age:16+) with 349,748 repeated responses across 10-data collection waves (2009–2019) from “Understanding-Society: The-UK-Household-Longitudinal-Study” were linked to annual concentrations of NO2, SO2, PM10, and PM2.5 pollutants using the individuals’ place of residence, given at the local-authority and at the finer Lower-Super-Output-Areas (LSOAs) levels; allowing for analysis at two geographical scales across time. The association between air pollution and mental well-being (assessed through general-health-questionnaire-GHQ12) and its modification by ethnicity and being non-UK born was assessed using multilevel mixed-effect logit models. Results Higher odds of poor mental well-being was observed with every 10μg/m3 increase in NO2, SO2, PM10 and PM2.5 pollutants at both LSOAs and local-authority levels. Decomposing air pollution into spatial-temporal (between-within) effects showed significant between, but not within effects; thus, residing in more polluted local-authorities/LSOAs have higher impact on poor mental well-being than the air pollution variation across time within each geographical area. Analysis by ethnicity revealed higher odds of poor mental well-being with increasing concentrations of SO2, PM10, and PM2.5 only for Pakistani/Bangladeshi, other-ethnicities and non-UK born individuals compared to British-white and natives, but not for other ethnic groups. Conclusion Using longitudinal individual-level and contextual-linked data, this study highlights the negative effect of air pollution on individuals’ mental well-being. Environmental policies to reduce air pollution emissions can eventually improve the mental well-being of people in UK. However, there is inconclusive evidence on the moderating effect of ethnicity.

and/or bad odours [26]. In a recent systematic literature review of 178 published articles, air pollution was shown to decrease happiness and life satisfaction substantially, and to increase anxiety, annoyance, mental problems, suicidal ideation, and coping approaches such as avoidance behaviour and migration [2].
Air pollution has been also linked to stress and experiential anxiety emerging from worrying feelings about one's physical health and future [27]. Due to the more conclusive research about the effect of air pollution on physical health including cardiopulmonary, immune system and cancer diseases [8,9,28], people living in highly polluted areas might experience stress and worrisome feelings of getting physical illness, which impairs their mental well-being.
Despite the establishment of linkages between air pollution and mental well-being in the literature, results are generally inconclusive and suffer from methodological drawbacks related to the chosen study design and methods of estimating air pollution [16]. Most of the studies are either cross-sectional studies or longitudinal studies that lack spatial-temporal specificity and a lengthy follow-up time. To date, no study has tried to address the association between long-term (11 years) air pollution exposure and mental well-being using a spatial-temporal (between-within) longitudinal design. A between-within analysis can determine the spatial time-constant cross-sectional (average 11 years air pollution) effect of air pollution on mental well-being between different geographical areas (e.g., local authorities or census output areas) as well as the temporal time-varying longitudinal (yearly air pollution deviation from the 11 years average) effect of air pollution on mental well-being within each geographical area. In other words, this analysis reveals whether living in more polluted local authorities or census output area is the driving cause for poor mental well-being (between) or whether it is the fluctuation in air pollution across time within each local authority or census output area (within) that is causing poor mental well-being. This type of analysis combines both cross-sectional and longitudinal designs; thus establishing a more robust measurement of the effect of air pollution [29].
Furthermore, published research has not yet covered population subgroups and the potential moderating effect that key demographic groups might have on the association between air pollution and mental well-being. To date, only age (young vs elderly population) and gender have been reported as effect modifiers for the association between air pollution and mental well-being. In Korea, a more pronounced risk of stress and depression from air pollution exposure was observed among men than among women and among people aged less than 65 [6]. In China, exposure to increased concentrations of ambient air pollution showed a greater risk for mental health problems and general well-being among female college students [30]. Therefore, examining how the effect of air pollution on mental well-being varies by other key demographic characteristics such as ethnicity can provide a more conclusive explanation for the association between air pollution and mental well-being. Earlier literature has shown that ethnic minorities suffer from relatively higher levels of stress, depression, and self-harm [31,32]. This could be attributed to their lower socio-economic status and to living in more deprived ethnic communities with poor housing and neighbourhood conditions [33][34][35]. Ethnicity was also examined in the literature from the lens of migration and being a non-native resident. A systematic literature review in Sweden showed increased risks of depression and psychotic problems among immigrants compared to the native population [36]. Given the higher observed risk of mental problems among ethnic minorities and immigrants, the effect of air pollution on mental well-being among ethnic groups should be investigated.
Accordingly, this study investigates longitudinally the overall and the spatial-temporal (between-within) effects of long-term (11 years) exposure to NO 2 , SO 2 , PM10, and PM2.5 air pollution in the UK on individuals' reported mental well-being measured using the 12 items General Health Questionnaire (GHQ12) scale. Unlike other studies that assess the effect of air pollution on well-being using one geographical scale, our study aims to assess the effect of air pollution exposure on mental well-being at two geographical scales, coarse local authorities (council areas in Scotland) and detailed Lower Super Output Areas (LSOAs; data zones in Scotland). This will allow us to compare the results between the two geographical scales and explore in more detail the local-contextual patterns of the effect of air pollution on mental well-being. Additionally, our study aims to consider both space and time by determining whether living in more polluted geographical areas (local authorities and LSOAs) is the driving cause for poor mental well-being (between) or whether it is the variation in air pollution across time within each geographical area (within) that is causing poor mental well-being; thus providing detailed spatial-temporal evidence for policymaking decisions. Finally, we aim to investigate whether ethnic minorities including Pakistani/Bangladeshi, Indians, Black/African/ Caribbean, mixed and other ethnicities as well as non-UK born individuals suffer from a more pronounced risk for mental well-being with increasing concentrations of the four air pollutants compared to British-White ethnicity and UK-born individuals, respectively.

Study design and population
This study utilises individual-level data from the "Understanding Society: The UK Household Longitudinal Study (UKHLS)" [37]. The UKHLS is a rich longitudinal dataset that is composed of 10 data collection panels/waves over a period of 12 years from (2009-2020) with about 40,000 households enrolled at the first wave from the four UK nations: England, Wales, Scotland, and Northern Ireland.
The dataset contains yearly information on the self-reported general health and mental well-being of individuals and on their socio-demographic characteristics including gender, age, educational qualification, marital status, occupation, perceived financial situation, ethnicity, and country of birth. In addition, the dataset collects yearly information on individuals' lifestyle factors such as cigarette smoking and on contextual factors such as the local authority/ council area and the Lower Super Output Areas (LSOAs)/data zones where households are located [37].
The UKHLS main survey sample is composed of four data collection sub-samples as summarised in Fig 1 and described in detail in other publications [37][38][39].

Air pollution data.
Raster data of annual mean concentrations of NO 2 , SO 2 , PM10, and PM2.5 air pollutants available up to the year of 2019, measured in μg/m 3 , and estimated using air dispersion models at a 1x1 km spatial resolution were downloaded from the "Department for Environment Food and Rural Affairs (DEFRA)" online data repository [51]. These air pollution maps at 1x1 km resolution are modelled each year by DEFRA under the "Defra's Modelling of Ambient Air Quality (MAAQ) contract" and are used to provide policy support in the UK and to fulfil the UK's reporting obligations to Europe [51]. The 1x1 km air pollution raster data are the finest spatial resolution data that can be downloaded from DEFRA and are sufficient to obtain good modelling estimates [52,53].
These data were first projected using the UK National Grid projection system in ArcGIS Pro software. Then for each of the 391 local authorities/council areas in the UK, we computed the average concentrations of NO 2 , SO 2 , PM10, and PM2.5 pollutants from all the 1x1 km raster cells that fell within the boundaries of the respective local authorities/council areas for each year from 2009 up to 2019. Next, we linked the average concentrations of air pollution at the local authority level to the UKHLS dataset for each individual and each year of follow up (2009-2019).
To minimise exposure bias and establish more robust results from a spatial perspective, we also linked the 1x1 km raster air pollution data to the UKHLS data at the level of Lower Super Output Areas (LSOAs; data zones for Scotland and Super Output Areas for Northern Ireland), a finer geographical scale, for each individual and each year of follow up (2009-2019). LSOAs are used to decompose England and Wales based on the population size into areas with a minimum population size of 1000 people and are the lowest level of geography offered by the UKHLS dataset. The LSOAs in England and Wales are equivalent to data zones in Scotland and to Super Output Areas in Northern Ireland. For simplicity we refer to the joint LSOAs, data zones, and Super Output Areas as LSOAs. Using these smaller spatial units, we ran our analysis at a smaller geographic scale than local authorities, which allowed us to explore local patterns of the effect of air pollution.
A map showing the local authorities in the UK (council areas in Scotland) and an enlarged subset of 20 local authorities in the south-east of UK with an example of PM10 concentrations at 1x1 km resolution for the year of 2017 for Tower Hamlets local authority and its corresponding LSOAs is used to clarify the process of linking air pollution to the UKHLS dataset at the two geographical scales, the coarse local authorities and the detailed LSOAs (Fig 3). Table 1 was selected a priori for this study based on what is available in the UKHLS dataset and based on the potential confounders considered by relevant literature. Most of the researchers in the field of air pollution and mental well-being have considered age [4,6,16], gender [2,4,6,16], education [4,6], marital status [6], socio-economic deprivation and occupation [16,56], ethnicity [1,7,16], cigarette smoking [16,56], alcohol drinking [4,16], physical exercise [4,16], and body mass index [4] as potential confounders. Specifically, poor mental well-being, stress, and depression have been associated with younger or older ages, women, cigarette smoking, alcohol drinking, physical inactivity, lower education, divorced/widowed individuals, lower household income, and belonging to an ethnic minority group [6,[31][32][33], which in turn confounds the association between air pollution and mental well-being outcomes. In addition, environmental-contextual factors such as noise pollution [1,16], neighbourhood disorder [16], green spaces availability [25,57], air temperature [57], and seasonality [16,58] have been also considered as potential confounders in the association between air pollution and mental well-being.

Data analysis
Individuals' socio-demographic and lifestyle factors were described using percentages for each wave (waves 1 to 10) of the UKHLS sample.
The mean of NO 2 , SO 2 , PM10, and PM2.5 concentrations and the Pearson's correlation between the four pollutants was computed at both geographical scales, the coarse local authorities and the detailed LSOAs. Given the high observed correlations between NO 2 , PM10, and PM2.5 pollutants (Pearson's coefficient � 0.7 [59]; Tables 3 and 4), the association of NO 2 , SO 2 , PM10, and PM2.5 with individuals' reported mental well-being was analysed in separate regression models. Nevertheless, low to moderate correlation was observed between SO 2 and each of the other three pollutants which enabled the construction of bi-pollutant models adjusting the NO 2 , PM10, and PM2.5 models for the SO 2 pollutant.
Intraclass correlation coefficients (ICCs) were also calculated to assess the homogeneity in the mental well-being scores within individuals and household clusters; with an ICC of greater than 0.3 indicating moderate to fair homogeneity [60]. Given the presence of 42% and 49% homogeneity (ICC = 0.42 and ICC = 0.49; Table 5) for the mental well-being scores of GHQ0-12 and GHQ0-36 within individuals' clusters across time, respectively, the mean score of mental well-being was computed from predictions of mixed effects linear models, adjusting for age in fixed effects and for the individual ID in random intercept.
The association between the three binary measures (GHQ0-36 � 12; GHQ0-12 � 2; GHQ0-12 � 4) of individuals' reported mental well-being and each of NO 2 , SO 2 , PM10, and PM2.5 pollutants (linked at two geographical scales: once at the LSOAs and once at the local authorities level) was examined separately and in bi-pollutant models adjusted for SO 2 using three-levels (repeated individual observations across time nested within LSOAs or local authorities) mixed-effect logit models, adjusting for the socio-demographic and lifestyle covariates and for the year (2009-2019) dummies. This type of analysis was chosen as it fits the longitudinal panel design of the study which involves repeated individual responses across time linked to air pollution data at the LSOA or local authority level whereby repeated individual responses are nested within LSOAs or local authorities. The individual-level random intercept is necessary in the multilevel models given the high homogeneity in the individual's responses across time (ICC = 0.42 and ICC = 0.49; Table 5), while the local authority or LSOAs random intercept is needed to allow for less biased assessments of the contextual-linked air pollution effect on mental well-being [61]; resulting in three-levels mixed-effect logit models. In a supplementary analysis, we also show the association of individuals' reported mental well-being with each of the socio-demographic and lifestyle covariates ( Table 1 in S1 File). It should be noted, however, that our models did not account for the household clustering in random intercepts due to the low observed homogeneity in the well-being responses within each household cluster (ICC = 0.16 and ICC = 0.18; Table 5).
In additional analysis, we decomposed the overall effect of air pollution (linked at two geographical scales: LSOAs and local authorities) on mental well-being into between (spatial) and within (temporal) effects. Between effects were used to examine the spatial time-constant effect  of air pollution between different geographical areas (local authorities and LSOAs) by calculating the average concentrations of NO 2, SO 2, PM10 and PM2.5 pollutants across the 11 years of follow up (2009-2019) for each local authority and each LSOA. On the other hand, within effects were used to assess the temporal time-varying effect of air pollution within each geographical area by calculating the annual air pollution deviation from the 11 years average air pollution for each local authority and LSOA. Therefore, two sets of four multilevel mixed effect logit models (one for each pollutant) were used to examine the overall (Eq 1) and the betweenwithin (Eq 2) effects of air pollution on individuals' mental well-being, respectively, at the two geographical scales (coarse local authorities and detailed LSOAs). ln ln Where Y tij is the mental well-being outcome for individual i, in local authority or LSOA j at year t; β 1 , β 2 . . .. β 12 are the slopes of fixed effects; β 0 is the fixed intercept; U 0ij is level 2 random intercept of individuals nested in local authorities or LSOAs; U 0j is level 3 random intercept of local authorities or LSOAs.
To assess the effect modification of ethnicity and country of birth on the association between air pollution and mental well-being, we added interaction terms between ethnicity and each of NO 2 , SO 2 , PM10, and PM2.5 pollutants and between country of birth and each of the four pollutants. These interaction terms were added into the models investigating the overall effect of air pollution and into the models examining the between and within effects of air pollution at the two geographical scales of local authorities and LSOAs, separately (i.e., one interaction term at a time, once with the between effect and once with the within effect). The interaction results were visualised with coefficient plots.
To estimate the cohort effect and to balance the follow up time, we repeated the same multilevel mixed effect logit modelling in a sensitivity analysis only for individuals recruited during the first wave of the UKHLS.
STATA software (StataCorp. 2015. Stata Statistical Software: Release 14. College Station, TX: StataCorp LP) was used for statistical analysis and ArcGIS Pro software was used for spatial pre-processing of air pollution data. Odd ratios (ORs) and 95% confidence (CIs) for every 10 μg/m 3 increase in air pollution were used to report the study findings. Statistical significance was considered at a P-value < 0.05.

Ethical considerations
This paper was granted ethical approval on the 14 th of May 2020 by the authors' affiliated institution (School of Geography and Sustainable Development Ethics Committee, acting on behalf of the University Teaching and Research Ethics Committee (UTREC) at the University of St Andrews). The paper uses secondary adult (age 16+) fully anonymised data from the "Understanding Society: The UK Household Longitudinal Study (UKHLS)" and authors did not have access to potentially identifying information; thus, obtaining participants' informed consent is not applicable and was waved by the authors' institution ethics committee. The University of Essex responsible for the UKHLS data collection and management has already obtained written informed consent from all the study participants [37]. Requesting consent for health record linkage was approved at Wave 1 by the National Research Ethics Service (NRES) Oxfordshire REC A (08/H0604/124), and at Wave 4 by NRES Southampton REC A (11/SC/0274). Approval for the collection of biosocial data by trained nurses in Waves 2 and 3 of the main survey was obtained from the National Research Ethics Service (Understanding Society-UK Household Longitudinal Study: A Biosocial Component, Oxfordshire A REC, Reference: 10/H0604/2). Description of the individuals' socio-demographic and lifestyle characteristics for the 10 data collection waves of the UKHLS are summarised in Table 2. For all waves, the majority of individuals were females, belonged to the middle-aged group (34-58 years), were married, had

Description of air pollution
We also observed high correlations (Pearson's coefficient � 0.7) between NO 2 , PM10, and PM2.5 pollutants (Table 3), which could be attributed to the source of emission and the atmospheric chemical reactions between these pollutants. For example, the major source of NO 2 and particulate matter (PM10 and PM2.5) emissions is traffic exhaust [62,63], while industrial processes and power plants are the major sources of SO 2 pollution [64].

Description of air pollution at the local authority level.
Similar to the air pollution at the LSOAs level, air pollution at the local authority level also showed fluctuations across time with lower concentrations seen in the last 5 years (2015-2019) of observation compared to previous years (2009-2014) for all four pollutants (Fig 5). Likewise, high correlations (Pearson's coefficient � 0.7) between NO 2 , PM10, and PM2.5 pollutants were noticed (Table 4).

PLOS ONE
Air pollution and individuals' mental well-being

Description of individuals' reported mental well-being
The mean score for mental well-being GHQ12 (0-12) scale was 1.8 (SD = 0.2) with 30% of responses having a GHQ12 score of 2 or more and 19% having a GHQ12 score of 4 or more. The mean score for GHQ12 (0-36) Likert scale was 11.14 (SD = 0.05) with 36% of responses having a score of 12 or more. The ICC was 0.42 and 0.49 for GHQ12 (0-12) and GHQ12 (0-36), respectively, indicating a moderate homogeneity in the well-being responses within the individual clusters over time, whilst low homogeneity (ICC = 0.16 and ICC = 0.18) was detected within the household clusters (Table 5).
We noticed similar results of higher odds of poor mental well-being with every 10 μg/m 3 increase in NO 2 , PM10, and PM2.5 pollutants in bi-pollutant models adjusted for SO 2 ( Table 7). ORs and 95%CIs are expressed in terms of 10 μg/m 3 increase in the air pollutants.

Air pollution and individuals' mental well-being
Decomposing the overall effect of air pollution on mental well-being into between (spatial: across LSOAs) and within (temporal: across years within each LSOA) effects, revealed significant associations with poor mental well-being for the between effects for all the four pollutants; while no significant associations were noted for the within effects despite the sign of the odds ratios being largely as expected. An exception was PM10 and PM2.5 pollutants which showed significant within effects on poor mental well-being (GHQ0-12) only in model 1; yet these significant effects disappeared after controlling for the sociodemographic and lifestyle covariates (Table 6). Therefore, living in more polluted LSOAs was the driving cause for poor mental well-being (between) rather than the variation in air pollution across time within each LSOA.
In a sensitivity analysis, results remained the same for the overall and for the between-within effects of the four pollutants on individuals' mental well-being for individuals starting at wave 1 of the UKHLS survey ( Table 2 in S1 File).

The effect of air pollution on individuals' mental well-being at the local authority level.
Similar to the air pollution results at the LSOAs geographical scale, higher odds of poor mental well-being were also observed with every 10 μg/m 3 increase in NO 2 (ORs ranging between 1.10-1.16), SO 2 (ORs ranging between 1.27-1.51), PM10 (ORs ranging between 1.13-1.22), and PM2.5 (ORs ranging between 1.21-1.36) pollutants at the local authority level (Table 8).
Higher odds of poor mental well-being were also observed at the local authority level with every 10 μg/m 3 increase in NO 2 , PM10, and PM2.5 pollutants in bi-pollutant models adjusted for SO 2 ( Table 9).
The between-within (spatial-temporal) analysis of the effect of air pollution on mental wellbeing at the local authority level revealed similar results to the between-within analysis at the LSOAs level (Table 8). Thus, living in more polluted local authorities was also the driving ORs and 95%CIs are expressed in terms of 10 μg/m 3 increase in the air pollutants. cause for poor mental well-being (between) rather than the variation in air pollution across time within each local authority. In a sensitivity analysis for individuals starting at wave 1 of the UKHLS survey, results remained unchanged for the overall and for the between-within effects of the four pollutants on individuals' mental well-being at the local authority level ( Table 3 in S1 File).

The association of air pollution at the LSOAs level with individuals' mental wellbeing by ethnicity and country of birth.
At the LSOAs level, Pakistani/Bangladeshi, otherwhite, and other ethnicities group as well as non-UK born individuals showed higher odds of poor mental well-being compared to British-white and UK-born individuals, respectively, with every 10 μg/m 3 increase in SO 2, PM10, and PM2.5 pollutants only with the GHQ0-12 well-being measure (Fig 6). Nevertheless, most of these significant differences disappeared in a cohort sub-analysis for only individuals recruited at wave 1 of the UKHLS survey (Fig 1 in S1  File). In addition, we observed no significant ethnic differences for the between or within Table 8. The association of individuals' mental well-being with each of NO 2 , SO 2 , PM10, and PM2.5 air pollutants linked at the local authority level in separate models (N = 349,748 surveys from 60,146 individuals).

The association of air pollution at the local authority level with individuals' mental well-being by ethnicity and country of birth.
Similar to the LSOAs level, analysis at the local authority level showed higher odds of poor mental well-being with increasing concentrations of SO 2, PM10, and PM2.5 pollutants among people from Pakistani/Bangladeshi and other ethnicities origin compared to British-white and among non-UK born people compared to natives. People from an Indian origin also showed higher odds of poor mental well-being than the British-white with every 10 μg/m 3 increase in SO 2 pollution at the local authority level (Fig 7). Yet, these significant differences disappeared in a cohort sub-analysis for only individuals recruited at wave 1 of the UKHLS survey (Fig 2 in S1 File). Similarly, no significant ethnic differences for the between or within effects of air pollution at the local authority level on mental well-being were observed with exception of between effects for PM10 and PM2.5 acting as protective factors (ORs<1) against poor mental well-being for people from Indian, Black/African/Caribbean and mixed ethnicities origin (Figs 5 and 6 in S1 File). Table 9. The association of individuals' mental well-being with each of NO 2 , PM10, and PM2.5 air pollutants linked at the local authority level in bi-pollutant models adjusted for SO 2 pollutant (N = 349,748 surveys from 60,146 individuals).

Discussion
This study showed the negative impact of NO 2 , SO 2 , PM10 and PM2.5 air pollution (linked at two geographical scales: coarse local authorities and detailed LSOAs) on mental well-being in the UK for individuals followed from the year 2009 up to 2019. These results are supported by

PLOS ONE
Air pollution and individuals' mental well-being relevant literature whereby exposure to ambient air pollution has been shown to affect negatively individuals' mental well-being and contribute to increased rates of mental health problems such as autism spectrum disorders [13], schizophrenia [14], dementia [15], psychotic

PLOS ONE
Air pollution and individuals' mental well-being experiences [16,17], cognitive disabilities [18], anxiety and major depressive disorders [19]. Evidence from China revealed an elevation in the rate of depressive symptoms and poorer selfreported mental well-being with long-term exposure to air pollution [65][66][67]. Similarly, exposure to NO 2 and particulate matter (PM10 and PM2.5) air pollution in the Netherlands was positively associated with poor mental health and prescription of anti-anxiety drugs [68]. In South Korea, increased exposure to NO 2, SO 2, and PM10 pollutants resulted in higher hazards for suicide death [69]. In addition, a recent systematic literature review and meta-analysis of 22 articles revealed that exposure to PM2.5 pollution increases the risk for depression and anxiety with a pooled odd ratio estimate of 1.10 for every 10 μg/m 3 increase in PM2.5 concentration [70]. The observed positive association between air pollution and poor mental well-being in this study can be explained by four factors. The first explanation is through the biological mechanisms of air pollutants on the human central nervous system and neuro-behavioural processes [1,5,9,10]. Air pollution particles of small diameters such as PM2.5 are capable of initiating oxidative stress and forming inflammatory cytokines that infiltrate the blood-brain barrier causing neurodegeneration and neuroinflammation [11]. The second explanation is through the aesthetic and odorous nuisance caused by air pollution, which results in avoidance behaviour and inhibition of psychological-supporting outdoor activities and sports. This in turn leads to reduced happiness and life satisfaction and to elevated levels of stress, anxiety, loneliness, and poor mental well-being [2,[23][24][25]. The third explanation is related to experiential anxiety and worrying feelings about one's physical health and future [27]. Through substantive research, people are made aware of the negative impact of air pollution on human's physical health and the higher risk for acute and chronic diseases such as cardiovascular, respiratory, cancer and immune system diseases [8,9,28]. People living in highly polluted areas might experience stress and anxiety about physical illness, which is reflected in poorer mental wellbeing. The last explanation is manifested in the indirect effect of air pollution on mental wellbeing through the physical health of individuals. People who already suffer from physical illness often also suffer from poor mental well-being. Based on an international study that utilises data from the World Health Organisation, 9% to 23% of patients with one or more chronic physical health conditions displayed symptoms of depression [71]. Thus, air pollution might be associated with poor mental well-being because individuals are also suffering from a physical health condition. Conducting a simple t-test on the UKHLS data showed a higher GHQ12 (0-12) mean score (mean = 2.44, SD = 3.46) among individuals with at least one physical health condition (e.g., asthma, arthritis, coronary heart disease, cancer, liver illness, chronic bronchitis, diabetes, blood pressure) in comparison to a GHQ12(0-12) mean score of 1.68 (SD = 2.92) for those with non-reported physical health condition (t-test P-value = 0.000).
Despite the existence of literature on the topic of air pollution and poor mental well-being, this study went a further step in analysing the spatial-temporal effects of air pollution on mental well-being using a between-within longitudinal design. Additionally, we carried out the analysis at two geographical scales, coarse local authorities and detailed LSOAs, which forms another novelty of the present study. The between-within analysis is extensively used in the fields of economics, behavioural finance, and strategic management [29], yet little used in health research [72]; and no previous study has examined the between-within effects of air pollution on mental well-being. Our study revealed significant between effects for NO 2, SO 2 , PM10 and PM2.5 pollutants on poor mental well-being at both the LSOAs and the local authority geographical scales, while no significant within effects were noted. Thus, individuals residing in LSOAs or local authorities with higher average concentrations of the four pollutants across the 11 years of follow up exhibited poorer mental well-being than individuals residing in LSOAs or local authorities with lower pollution concentrations. This shows the importance of the spatial dimension in the association between air pollution and mental wellbeing whether at the coarse local authorities or at the detailed LSOAs geographical level given that both geographical scales resulted in similar findings. Nevertheless, the non-significant within (temporal) effects could be related to the low variation of yearly air pollution concentrations across the 11 years of follow up, particularly for SO 2 pollutant as shown in Fig 4 for LSOAs and in Fig 5 for local authorities. Therefore, further research with longer follow up time is needed to allow for more temporal variation in air pollution which might result in significant within effects.
In the second part of this study, we attempted to examine the moderating effect of ethnicity and country of birth on the association between air pollution and individuals' mental wellbeing. We hypothesised that the lower socio-economic status of ethnic minorities and living in more disadvantaged neighbourhoods, near major roads and transportation networks can result in higher levels of stress, anxiety, and mental health problems among ethnic minority groups; thus, moderating the association between air pollution and mental well-being. However, our findings did not reveal much difference in the overall and in the between and within effects of air pollution on mental well-being across the ethnic groups; with exception for Pakistani/Bangladeshi, other-white (only at the LSOAs level but not at the local authority level), Indians (only for SO 2 pollution at the local authority level), other ethnicities, and non-UK born individuals who showed higher odds of poor mental well-being than the British-White and UK-born individuals with increasing concentrations of SO 2 , PM10, and PM2.5 pollutants. The poorer mental well-being observed among the Pakistani/Bangladeshi, other ethnicities group, and non-UK born individuals could be related to the socio-economic and lifestyle differences or to place-related contextual differences. Literature on ethnic inequalities in health has shown that ethnic minorities often live in more disadvantaged communities, and have lower socio-economic status, lower healthcare coverage and higher job/income insecurity [34,73,74], which increases their risk of physical and mental illness. Nevertheless, our analysis adjusted for the main socio-economic and lifestyle factors of individuals including age, gender, marital status, education, occupation, financial situation, and cigarette smoking. Therefore, the most likely explanation for the more pronounced effect of SO 2 , PM10, and PM2.5 pollution on poor mental well-being among Pakistani/Bangladeshi, other ethnicities, and non-UK born individuals could be related to the contextual factors and place of residence. Ethnic minorities and immigrants often choose to reside in large cities and highly urbanised regions, near major roads and key transportation networks to simplify their commuting and working conditions [75]. In addition, ethnic minorities often live in low-priced social housing offered by local authorities, which is often situated in more deprived ethnic concentration neighbourhoods [35]. These place-related factors can result in more pronounced effect of air pollution on mental well-being among ethnic minorities and immigrants due to greater exposure to air pollution resulting from vehicles, factories, and burning of fossil fuels. In a supplementary analysis, we show through Chi2 square tabulation that a very high percentage of non-UK born individuals (93%) and of ethnic minorities including Pakistani/Bangladeshi (99%) reside in urban areas; whereas this percentage is much lower for British-white (71%) and UK-born (75%) individuals ( Table 4 in S1 File). It should be noted, however, that decomposing the overall effect of air pollution into between (spatial) and within (temporal) effects at both the LSOAs and local authority levels did not show significant differences among the ethnic groups, not even for the Pakistani/Bangladeshi and non-UK born individuals. Thus, we cannot be conclusive that residing in more polluted areas is the key explanation for the more pronounced effect of SO 2 , PM10, and PM2.5 pollution on mental well-being among Pakistani/Bangladeshi and non-UK born individuals. Furthermore, in a cohort sub-analysis for individuals recruited at wave 1 of the UKHLS survey (Figs 1 and 2 in S1 File), the differences noted with respect to the Pakistani/Bangladeshi, other ethnicities group, and non-UK born individuals disappeared. Therefore, our study shows no conclusive evidence of ethnicity or nativity differences in the association between air pollution and poor mental well-being.
Despite the new insights provided by this study, it is important to discuss its limitations. The first limitation is related to the design of the study in which individual-level data from the UKHLS survey was linked to yearly air pollution contextual data at the local authority level. Therefore, individuals residing within a respective local authority and a respective year were assigned to the same value of air pollution exposure. However, we also linked the air pollution data to the UKHLS at the LSOAs level (the lowest available geography level at the UKHLS due to ethical considerations) which minimised the exposure bias. Air pollution linked at both, the local authority and the LSOAs level, revealed similar results. For future research, we recommended the usage of data sources that allow linkages of air pollution at the postcode level, the lowest available geography in the UK. Second, our analysis included all individuals recruited at different waves of the UKHLS survey, that had at least two observations through the follow up time (2009-2019). Hence, the time window for follow up differed between the study participants, some were followed for 11 years across all the 10 waves while others entered the study at later waves and were followed for a shorter period of time. It should be noted, however, that similar results were observed in a sensitivity analysis on wave 1 cohort, except for the analysis by ethnic groups whereby Pakistani/Bangladeshi, other ethnicities, and non-UK born individuals did not show significant differences in the association of air pollution with mental wellbeing at both the LSOAs and local authority geographical scales (Tables 2, 3 and Figs 1 and 2 in S1 File). This can be explained by the UKHLS sample design which involved ethnic minority boost samples at waves 1 and 6 of data collection to enable ethnicity-focused research. Therefore, by conducting analysis on only wave 1 cohort, we are missing the second ethnic minority boost sample at wave 6 which resulted in the observed differences. Finally, the UKHLS survey included longitudinal weights that adjust for the overrepresentation of some groups, such as the ethnic minority groups; thus, allowing for greater generalisation of the estimates. However, we could not add the longitudinal weights into our analysis as this requires that all individuals be followed until the last wave (wave 10) of the survey, which was not the case.

Conclusion
Using a longitudinal panel design that involves linking individual to context-level data at two geographical scales (coarse local authorities and detailed LSOAs) and a between-within analysis, this study highlights the negative effect of air pollution on individuals' mental well-being over space and time and emphasises the importance of the spatial dimension in the shaping of this association. Thus, environmental policies to reduce air pollution emissions with a core of spatial planning can eventually improve the mental well-being of people residing in the UK. There is, however, less conclusive evidence on the moderating effect of ethnicity.