Introduction

The last decades have been characterised by the rapid urbanisation of several parts of the world1. Consequently, recent research has explored the effect of growing up in urbanised environments on physical, social and psychological development2,3,4,5,6. Children and young people living in cities often have more access to health and social care services, better transportation, and better education opportunities compared to those living in rural areas1. However, they are also exposed to higher air and noise pollution and heat, fewer natural or green spaces, limits to personal space, and greater crime rates and socioeconomic disparities1,7. Adolescence is a sensitive period for the effect of the living environment on social and psychological wellbeing, and of vulnerability to the emergence of socioemotional disorders8,9. In this study, we sought to investigate the relationship between neighbourhood characteristics, perceived social support, and psychological wellbeing during adolescence.

Growing literature has identified compact urban forms as an important component of sustainable urbanisation10. Notably, however, living in densely populated environments (an element of compact urban forms) has also been related to chronic stress processing and greater symptoms of anxiety and psychotic disorders in adults11,12,13,14. While population density may contribute to wellbeing by offering greater opportunities for socialisation and establishing connections, this does not always translate to subjective experiences of social and psychological wellbeing15,16,17,18,19,20. Some studies have also reported weaker social relationships in high-density neighbourhoods20–22 and more feelings of insecurity and negative emotions towards neighbourhoods17,23. While most studies have included adult samples, living in environments that facilitate or hinder social cohesion might be particularly impactful for adolescents, as changes to their social environment influence their mood and anxiety to a greater extent than in other age groups24,25.

Urban neighbourhood disadvantage has also been associated with higher prevalence of emotional difficulties in young people12,26,27,28, with greater violence exposure and antisocial behaviour from peers contributing to higher depressive symptoms in adolescents living in more deprived urban neighbourhoods29. Other studies have found neighbourhood unpleasantness, low maintenance and perceptions of safety and crime to be associated with reduced independent mobility and exploration of neighbourhood spaces30–32. In turn, several indicators of neighbourhood deprivation (e.g. perceptions of safety, housing insecurity, and concentrated poverty) have been proposed to hinder social connection, such as increasing peer problems33,34 and feelings of loneliness32.

In contrast, natural features of neighbourhoods can positively contribute to young people’s wellbeing35,36. In fact, many guidelines now recommend having at least 3 well-established trees in view from homes, schools and places of work; no less than a 30% tree canopy in neighbourhoods; and no more than 300 m to the nearest public green space from the residential address38,39,40. Several mechanisms have been proposed to explain the beneficial health-related impact of urban green areas. For example, the availability of recreational green spaces is related to physical activity, stress reduction, and attention restoration in children4,36,41, and a higher percentage of tree canopy and tree diversity can have the potential to mitigate the impact of air pollution, heat and noise3,42. In addition, incorporating well maintained green spaces and green cover in neighbourhood design can also improve perceptions of safety and neighbourhood pleasantness44. Several studies have also shown that neighbourhoods with more green cover are associated with enhanced social cohesion and community building across the lifetime16,45. While evidence is mixed for child and adolescent samples46,47, one study showed that more frequent use of green spaces was associated with social connectedness with friends (quantity and time spent) in young people aged 10–18, and this relationship was particularly strong for adolescents aged 14 years and over48.

In line with this, neighbourhood characteristics may have different effects at different points in adolescence, although the evidence is not yet clear on which stage is most vulnerable. One reason to expect this is that exposure to urban stressors, such as noise, crowding, pollution, deprivation and safety concerns, can have cumulative effects on wellbeing. The longer young people are exposed to these stressors, the more likely they are to experience heightened psychological strain and poorer overall health5. Conversely, the prolonged availability of positive neighbourhood resources, such as access to green spaces, recreational facilities and supportive community environments, has been linked to improvements in health and wellbeing50. Because these effects accumulate over time, and adolescence is a sensitive period of development in which environmental input is particularly important, the impact of neighbourhood characteristics may become especially apparent in the psychological and social wellbeing of older adolescents34,48. In contrast, sociocognitive skills that underlie social stress processing and emotional regulation continue to develop throughout adolescence51. As a result, less developed sociocognitive skills earlier in adolescent development might heighten the susceptibility to the negative effects of urban stressors (e.g. experiences of exclusion) and the positive effects of natural areas (e.g. through physical activity and emotion regulation). Taken together, there might be age-related differences in the relationship between neighbourhood characteristics and social and psychological wellbeing, but the specific direction of these speculations requires further research.

Relatedly, some young people may encounter greater barriers in accessing health-promoting resources, depending on their family affluence or sex. For example, young people from disadvantaged backgrounds often have reduced access to healthcare services (e.g., private healthcare, specialised mental health support) and fewer opportunities to participate in structured, enriching activities such as after-school clubs, sports teams or summer programmes52. These opportunities are more readily available to young people from more affluent families, who can afford fees, transportation and equipment. As a result, we could speculate that the detrimental effects on wellbeing of neighbourhood factors such as a lack of green space and high neighbourhood deprivation might disproportionately affect young people who have less access to—and less ability to compensate from—specialised mental health support and enriching activities53. In addition, sex-related differences also shape how young people engage with their neighbourhood environment. Girls often face greater barriers to using and exploring public spaces independently, partly due to heightened safety concerns (e.g., fear of harassment, parental restrictions)55. This can limit their participation in certain forms of social and physical activity, such as using public recreational grounds, cycling or spending time with peers in parks or community spaces, especially after daylight hours56. In contrast, boys may be less restricted in their mobility and social use of neighbourhood spaces, which could lead to differences in peer network formation, physical activity levels and overall wellbeing. Taken together, these patterns suggest that the same neighbourhood characteristics can have uneven effects across groups.

The current study

The current study applied linear mixed effects models to combined data from the Childhood and Environment cohort study57 (1492 total observations from 970 unique participants, comprising n = 743, ages 9–12 years; n = 528, ages 12–17 years; n = 221, 14–17 years), collected in three regions of Spain between 2016 and 2022. The first aim was to investigate whether several neighbourhood characteristics have an association with psychological and social wellbeing in adolescents. We hypothesised that higher residential greenness and green space availability would be related to higher psychological wellbeing and higher perceived social support. In addition, we hypothesised that neighbourhood deprivation would be related to lower psychological wellbeing and lower perceived social support. As the evidence relating population density to wellbeing is mixed7, we hypothesised that population density could be positively or negatively related to psychological wellbeing and social support.

As previous literature has consistently found social support to be related to psychological wellbeing in young people58–60, the second aim of this study was to investigate whether family and peer support play a role in the relationship between neighbourhood stressors and resources and psychological wellbeing. We hypothesised that higher perceived social support would mediate the relationship between higher residential greenness and green space availability and higher psychological wellbeing. We also hypothesised that lower perceived social support would mediate the relationship between neighbourhood deprivation and lower psychological wellbeing. In addition, we hypothesised that perceived social support would mediate the relationship between population density and psychological wellbeing. Note that mediation hypotheses do not imply longitudinal mediation, but instead compare variance in psychological wellbeing explained by differences in neighbourhood characteristics and perceived social support.

Finally, the impact of social and environmental settings on wellbeing might vary according to development and sex (i.e. age and sex differences)35,48, and access to health-promoting resources could mitigate the impact of living environments (i.e. family affluence differences)61,62. As such, our third aim was to explore, in a series of exploratory models, whether the relationship between neighbourhood characteristics, perceived social support and psychological wellbeing differ according to demographic characteristics, including age, sex and family affluence.

Methods

Sample

The sample included participants ages 9–17 years who are part of the INMA project (www.proyectoinma.org)57. The INMA project investigates the effects of environmental exposures on physical, social, and neuropsychological development from conception onwards. Between 2004 and 2008, 2644 women in the first trimester of pregnancy (over 16 years of age) were recruited from the public health service in several Spanish regions, including Gipuzkoa, Sabadell and Valencia (see Dadvand et al., 2012 for biogeographic details of regions)63. Follow-up data from families and children has been collected sequentially every two to three years, with slight age and time variations by region. The ethics committees of the hospitals involved in each region approved the project and supervised that the project was performed in accordance with their guidelines, and informed consent was obtained from all parents in each follow-up.

In the current study, we used combined data from the following regions: Gipuzkoa, 358 participants who took part in 2018–2019 (ages 10–12 years) and 261 participants who returned in 2021 (ages 12–14 years); Sabadell, 385 participants who took part in 2016–2018 (ages 9–12 years) and 267 participants who returned in 2020–2022 (ages 13–17 years); and Valencia, 221 participants who took part in 2019–2022 (ages 14–17 years; see Table S1 in supplemental material SM1 for sample differences between regions). Note that participants from Valencia did not have available social and psychological wellbeing questionnaire data at Time 1 and were therefore only included at Time 2. Participants who did not have complete data in all measures were excluded from the study (see missing data details in Table S2, Table S3 and Figure S1 in supplemental material SM1). There was a total of 1492 observations in the sample (from 970 unique participants).

Measures

Neighbourhood characteristics

The first neighbourhood characteristic was residential greenness, calculated using the Normalized Difference Vegetation Index (NDVI). The NDVI is a greenness measure derived using satellite imagery (see further details in supplemental material SM2) with values ranging from − 1 to 1: low values (0–0.1) indicate areas of barren rock, sand or snow; moderate values (0.2–0.5) indicate sparse vegetation (e.g. grasslands), and high values (0.6–1) indicate dense vegetation64. Negative values correspond to water and were removed before calculating residential greenness. Residential greenness was computed as the 5-year NDVI average in the greenest seasons only (May through August) in a 300 m buffer surrounding the home address. We use a 300 m buffer as this has been found to be an appropriate marker of residential surrounding greenness65 and is predictive of health outcomes in other studies using the INMA data66,67. Higher NDVI scores represent greater percentage of residential greenness.

The second neighbourhood characteristic was green space availability, computed as a binary measure measuring access to major green spaces (e.g. parks or countryside) surrounding the home address. This measure was obtained by calculating the straight-line distance from the home to the nearest green space with an area greater than 5000 m2 using the Europe-wide Urban Atlas68. Green space availability was determined by whether there was a major green space available within a 150 m buffer of the home address (1 = yes, 0 = no)69. We use a 150 m buffer in this study because 85.5% of the sample had a major green space available within a 300 m buffer, which is the distance recommended by the World Health Organisation39.

The third neighbourhood characteristic was neighbourhood deprivation, which is an index of deprivation derived from the 2011 Population and Housing Census of Spain70. This deprivation index is computed from six socioeconomic indicators: percentage of unemployed population, percentage of manual workers, percentage of casual workers, percentage of youth without education and number of homes without access to the internet. The index was classified into ordered tertiles (1 = low deprived, 2 = medium deprived, 3 = high deprived).

The last neighbourhood characteristic was population density, computed as the number of inhabitants per square kilometre surrounding the home address. This measure was derived as the point-to-raster assignment between geocodes and population density grid maps (250 m2 resolution) obtained from the 2015 Global Human Settlement Layer (GHS-POP R, 2015). Scores were transformed to 100 inhabitants per square kilometre for interpretability. Higher scores denote higher population density.

Household-level socioeconomic measures

This study also included two household-level socioeconomic measures: family affluence and maternal education. Family affluence was collected at birth using the EU Statistics on Income and Living Conditions (EUSLIC)72. The EUSLIC is a measure of net disposable household income, standardised for household composition and log-transformed to better approximate a normal distribution73. Higher values denote higher family affluence. In addition, this study used the International Standard Classification of Education 1997 to classify maternal education during pregnancy. This is a measure of the highest level of education completed by the mother and is ordered in three categories (1 = primary or no education, 2 = secondary education, 3 = university education).

Psychological wellbeing and perceived social support measures

The INMA study collected data on the self-reported 27-item version of the Kidscreen (Spanish version)74, which is a measure of health-related quality of life developed by a collaboration of 13 European countries76,77,78. The Kidscreen-27 includes five subscales measuring different dimensions of quality of life: physical wellbeing, psychological wellbeing, autonomy & parent relations, peers and social support, and school environment. All items are rated on a five-point response scale (0 = “not at all” to 4 = “extremely” or 0 = “never” to 4 = “always”). Items on each subscale are aggregated to create sum scores for each dimension which are later transformed into T-scores ranging from 0 to 100 using population norms74. In this study, we use the psychological wellbeing subscale, which measures positive emotions and life-satisfaction (7 items); the autonomy & parental relations subscale, which measures the quality of interactions with the family and feelings of autonomy (7 items); and the peers and social support subscale, which measures the quality of relations with peers (4 items). Higher T-scores on these subscales in general denote greater psychological wellbeing, family support & autonomy and peer support, respectively. This tool has been used and validated in several European samples in children aged 8–18 years old75.

Statistical analysis

Main analyses

First, to determine whether perceived social support and psychological wellbeing replicated previous literature60, psychological wellbeing, family support & autonomy and peer support were regressed separately on age and family affluence, both treated as continuous measures, and sex, treated as a dichotomous measure (female = 0, male = 1). In addition, psychological wellbeing was regressed separately on family support & autonomy and peer support. Note that these analyses do not test any hypotheses but rather check that the data replicate previous literature.

Second, to investigate the effect of neighbourhood characteristics on psychological wellbeing, a set of models regressed psychological wellbeing on the following neighbourhood characteristics (separately): residential greenness and population density, both treated as continuous measures; green space availability, treated as a binary measure (1 = yes, 0 = no); and neighbourhood deprivation, ordered in tertiles (1 = low deprived, 2 = medium deprived, 3 = high deprived). In addition, to investigate the effect of neighbourhood characteristics on perceived social support, another set of models regressed family support & autonomy and peer support separately on each predictor.

Third, to determine whether the relationship between the neighbourhood characteristics and psychological wellbeing was explained by perceived social support in our data, we re-ran those models that showed a relationship between the neighbourhood characteristics and psychological wellbeing, adjusting for family support & autonomy and peer support (separately). Adjusting for a mediator (i.e. controlling for an indirect effect) allows us to estimate the direct effect of a predictor on an outcome measure (direct effect = total effect–indirect effect)78. According to this mediation literature, a mediator is likely to have a role on the relationship between the predictor and an outcome if the direct effect is weakened compared to the total effect. Given the limitations of applying mediation analyses on hierarchical mixed models with complex clusters (e.g. 3-level structures), we did not calculate inferential statistics for indirect effects.

Using the R statistical software (R version 4.3.2)79, the combined data (1492 observations) were modelled in a set of linear mixed effects models80 using the lme4 (version 1.1–35.1)81 and lmerTest packages (version 3.1-3.1)82 with participant and region as random effects. Robust causal inference methodology was applied to determine the minimal covariate adjustment sets for each model by using a Directed Acyclic Graph (DAG; see Table S4, Table S5, Figure S2 and Figure S3 in supplemental material SM3 for details of the DAG and model-specific adjustments). Main effects and interactions (see below) were inspected using omnibus F-tests with Kenward-Roger approximations for degrees of freedom. All continuous measures were standardised (centred and scaled) prior to analysis.

Effect modification

We additionally explored whether demographic characteristics modify the relationship between neighbourhood characteristics, perceived social support and psychological wellbeing. To do so, we re-ran the models described in the main analyses, additionally interacting the neighbourhood characteristics with one of the following predictors: age, sex and family affluence. Any significant interactions were further inspected using post-hoc marginal mean-trend estimations with Kenward-Roger approximations for degrees of freedom83. Specifically, we explored whether pathways described in the main analyses varied with sex, and at the 25th, 50th and 75th percentiles of age and family affluence, using the emmeans and the emtrends function of the emmeans package (version 1.10.0)84. Scripts for the statistical analysis can be found in the Open Science Framework (https://osf.io/d24kt/).

Sensitivity analyses

Model diagnostics of all linear mixed effects model residuals approximated a normal distribution, but homogeneity of variance and linearity assumptions could be violated towards the extremes of the scales, potentially due to the presence of influential observations. Therefore, in a first set of sensitivity analyses, we re-ran all models excluding potential influential observations. In addition, to adjust for period-specific influences in the relationships, we re-ran a second set of sensitivity analyses additionally including timepoint (0,1) as a fixed effect to all models. Finally, to adjust for selective attrition of the missing data, we re-ran a third set of sensitivity analyses adjusting for participant-level weights derived from the inverse probability of inclusion predicted by age, family affluence, maternal education, timepoint and region. Omnibus tests reported in the results section are robust to sensitivity analyses adjusting for the presence of influential observations, period-specific effects and selective attrition, unless otherwise stated.

Results

Descriptive statistics are reported in Table 1. The results of the linear mixed effects models investigating the relationships between demographic characteristics, psychological wellbeing and perceived social support are presented in supplemental material SM5 (see Table S6 and Figure S4). Summary statistics and post-hoc comparisons of significant main effects showed that psychological wellbeing was lower in older participants (βstandardised = − 0.49, 95%CI [− 0.54 − 0.44], p <.001) and in females compared to males (contrastFemale−Male = − 0.23, SE = 0.05, p <.001). In addition, psychological wellbeing was associated with higher family support & autonomy (βstandardised = 0.48, 95%CI [0.44 0.52], p <.001; Fig. 1, left panel) and higher peer support (βstandardised = 0.28, 95%CI [0.24 0.32], p <.001; Fig. 1, right panel). The relationship between peer support and psychological wellbeing was stronger in females than males (contrastFemale−Male=0.10, 95%CI [0.02 0.17], p =.009) and increased with family affluence (βstandardised = 0.06, 95%CI [0.03 0.10], p <.001). The models also showed significant associations between age and family support & autonomy, age and peer support, and family support & autonomy and sex. However, these were not robust to adjusting for period-specific effects (see Table S7 in supplemental material SM6 for full details and results of sensitivity analyses).

Table 1 Descriptive statistics of the sample.
Full size table
Fig. 1
figure 1

Relationship between perceived social support and psychological wellbeing. Dots show participant psychological wellbeing as a function of perceived social support, including family support & autonomy (left panel) and peer support (right panel). Lines of best fit show the main effect of family support & autonomy (in purple; left panel) and peer support (in pink; right panel) on psychological wellbeing, as estimated by the linear mixed effects models. The shaded areas are the 95% confidence intervals. The asterisks indicate significant main effects of the perceived social support measures: ***p <.001.

Full size image

Relationship between neighbourhood characteristics and psychological wellbeing

The results of the linear mixed effects models investigating the relationship between the neighbourhood characteristics and psychological wellbeing, as well as exploratory interactions with demographic characteristics, are presented in Table 2. Significant interactions are presented in bold.

Table 2 Results of the linear mixed effects models for psychological wellbeing.
Full size table

Residential greenness

As reported in Table 2, the analyses of the linear mixed effects models showed no main effect of residential greenness on psychological wellbeing. However, a separate exploratory analysis showed a positive interaction between residential greenness and age (βstadardised=0.08, 95%CI [0.03 0.13], p =.003). In support of our hypotheses, exploratory post-hoc marginal mean-trend estimations at the 75th percentile of age (15.13 years) showed that the relationship between higher residential greenness and higher wellbeing in older individuals was small and significant (Q75: βstandardised=0.15, 95%CI [0.01 0.30], p =.038; Fig. 2, Panel A). In contrast, exploratory post-hoc marginal mean-trend estimations at the 25th (10.84 years) and 50th percentile of age (12.79 years) did not show a relationship between residential greenness and psychological wellbeing in younger individuals (Q25: βstandardised= − 0.01, 95%CI [− 0.14 0.12], p =.896; Q50: βstandardised=0.06, 95%CI [− 0.06 0.19], p =.312). Finally, separate exploratory analyses did not show an interaction between residential greenness and sex, or residential greenness and family affluence, on psychological wellbeing. These models included population density, neighbourhood deprivation, green space availability, maternal education and family affluence as covariates, as determined by the DAG minimum covariate adjustment set.

Population density

As reported on Table 2, the analyses of the linear mixed models showed no main effect of population density on psychological wellbeing. However, a separate exploratory analysis showed a small negative interaction between population density and age (βstandardised = − 0.07, 95%CI [− 0.12 − 0.03], p =.002). Exploratory post-hoc marginal mean-trend estimations at the 25th (10.84 years), 50th (12.79 years) and 75th percentiles of age (15.13 years) showed that, in line with our hypothesis, there was a weak and significant relationship between higher population density and lower wellbeing in older individuals (βstandardised = − 0.07, 95%CI [− 0.14 0], p =.044; Fig. 2, Panel B). Population density was not related to psychological wellbeing in younger individuals (Q25: βstandardised = 0.07, 95%CI [− 0.01 0.15], p =.090; Q50: βstandardised = 0.01, 95%CI [− 0.06 0.07], p =.845). Note however that the association between higher population density and psychological wellbeing in older adolescents was weak and no longer significant in sensitivity analyses adjusting for period-specific effects (βstandardised = − 0.07, 95%CI [− 0.14 0], p =.053; see SM6 and Table S7 for full details and results of sensitivity analyses). Further, separate exploratory analyses did not show an interaction between population density and sex, or population density and family affluence, on psychological wellbeing. These models included maternal education and family affluence as covariates, as determined by the DAG minimum covariate adjustment set.

Neighbourhood deprivation and green space availability

As reported in Table 2, the analyses of the linear mixed effects models showed that neighbourhood deprivation and green space availability were not related to wellbeing. In addition, separate exploratory models showed no differences in these relationships according to age, sex or family affluence. Therefore, our hypotheses were not supported for neighbourhood deprivation and green space availability.

Fig. 2
figure 2

Age-related differences in the relationship between neighbourhood characteristics, psychological wellbeing and perceived social support Top panels. Circles show grand means of psychological wellbeing as a function of residential greenness (percent cover; Panel A) and population density (inhabitants per square km; Panel B), colour-coded by age (in years; 9–17 years; light to dark tones). Bottom panels. Circles show grand means of family support & autonomy (Panel C) and peer support (Panel D) as a function of residential greenness (percent cover), colour-coded by age (in years; 9–17 years; light to dark tones). All panels. Circle area is proportional to the number of participants; the keys show examples for reference. Lines of best fit show the marginal mean-trends at the 25th (10.84 years; light tone), 50th (12.79 years; medium tone) and 75th (15.13 years; dark tone) percentiles of age, as estimated by the linear mixed effects models. The shaded areas are the 95% confidence interval. The asterisk indicates a significant marginal mean-trends of age: ***p <.001; *p <.05.

Full size image

Relationship between neighbourhood characteristics and perceived social support

The results of the linear mixed effects models investigating the relationship between the neighbourhood characteristics and perceived social support, as well as exploratory interactions with sociodemographic characteristics, are presented in Table 3. Significant main effects and interactions are presented in bold.

Table 3 Results of the linear mixed effects models for perceived social support.
Full size table

Residential greenness

Family support and autonomy

As reported in Table 3, the analyses of the linear mixed effects model showed a significant main effect of residential greenness on family support & autonomy. In line with our hypotheses, higher residential greenness was in general related to higher family support & autonomy (βstandardised = 0.16, 95%CI [0.06 0.26], p =.044). In addition, a separate exploratory analysis showed a positive interaction between residential greenness and age on family support & autonomy (βstandardised = 0.08, 95%CI [0.03 0.14], p =.006). In support of our hypotheses, exploratory post-hoc marginal mean-trend estimations at the 25th (10.84 years), 50th (12.79 years) and 75th percentile of age (15.13 years) showed that the overall relationship between higher residential greenness and higher family support & autonomy was small and significant for middle and older individuals (Q50: βstandardised = 0.16, 95%CI [0.01 0.30], p =.035; Q75: βstandardised = 0.25, 95%CI [0.08 0.41], p =.004), but was not significant for the youngest individuals (Q25: βstandardised = 0.08, 95%CI [− 0.07 0.23], p =.271; Fig. 2, Panel C). These models included population density, neighbourhood deprivation, green space availability, maternal education and family affluence as covariates, as determined by the DAG minimum covariate adjustment set.

Peer support

The analyses of the linear mixed effects model showed no main effect of residential greenness on peer support. A separate exploratory analysis showed a weak and nonsignificant positive interaction between residential greenness and age on peer support (βstandardised = 0.06, 95%CI [0 0.12], p =.052). Exploratory post-hoc marginal mean-trend estimations at the 25th (10.84 years), 50th (12.79 years) and 75th percentile of age (15.13 years) showed that, while the relationship between residential greenness and peer support was not significant in younger individuals (Q25: βstandardised = 0.06, 95%CI [− 0.08 0.21], p =.392; Q50: βstandardised = 0.12, 95%CI [− 0.03 0.26], p =.103), higher residential greenness was related to higher peer support in older individuals (Q75: βstandardised = 0.18, 95%CI [0.02 0.34], p =.028; Fig. 2, Panel D). The interaction between residential greenness and age was significant after adjusting for period-specific effects (βstandardised = 0.08, 95%CI [0.01 0.14], p =.034) and for selective attrition (βstandardised = 0.06, 95%CI [0 0.12], p =.047). These models included population density, neighbourhood deprivation, green space availability, maternal education and family affluence as covariates, as determined by the DAG minimum covariate adjustment set.

Residential greenness and psychological wellbeing, adjusting for perceived social support

Follow-up analyses explored the remaining direct interaction of residential greenness and age on psychological wellbeing, after adjusting for family support & autonomy. In line with our hypotheses, the interaction between residential greenness and age on psychological wellbeing was no longer significant when the model was adjusted for family support & autonomy (F(1,872.84) = 2.60, p =.107). Therefore, higher family support & autonomy could play a role in the relationship between higher residential greenness and higher psychological wellbeing in older individuals. In addition, follow-up analyses explored the remaining interaction of residential greenness and age on psychological wellbeing, after adjusting for peer support. The interaction between residential greenness and age weakened but remained significant when the model was adjusted for peer support (F(1,828.62) = 4.89, p =.027). Therefore, higher peer support might only play a weak role in age-related differences in the relationship between residential greenness and higher psychological wellbeing in our data.

Population density, neighbourhood deprivation, and green space availability

Finally, contrary to our hypotheses, the linear mixed effects models investigating the relationship between population density, neighbourhood deprivation and green space availability and perceived social support did not show a relationship with family support & autonomy or with peer support. In addition, while the relationship between population density and psychological wellbeing varied according to age, the relationship between population density and perceived social support did not vary according to age.

Discussion

The current study investigated the relationship between neighbourhood stressors and resources, perceived social support and psychological wellbeing in adolescents aged 9–17 years from the Spanish INMA cohort. In support of our hypotheses regarding residential greenness, the age-related results suggest that higher residential greenness was related to higher psychological wellbeing in older adolescents (aged 15–17 years), but not in younger children and adolescents (aged 9–14 years). In addition, higher family support and feelings of autonomy in older adolescents living in neighbourhoods with higher residential greenness played a role in this relationship. There was weak evidence that the relationship between higher residential greenness and higher peer support played a role in psychological wellbeing. In addition, the results showed age-related differences in the relationship between population density and psychological wellbeing, with older adolescents living in more densely populated neighbourhoods reporting lower wellbeing. Contrary to our hypotheses regarding neighbourhood deprivation and green space availability, there was no relationship between these neighbourhood characteristics, perceived social support, and psychological wellbeing. Finally, and in contrast to our exploratory hypotheses, none of the significant relationships varied according to sex and family affluence.

Residential greenness, perceived social support and psychological wellbeing

Based on the growing literature indicating an association between urban living and worse physical and mental wellbeing across the lifespan5,7,12,41, this study investigated whether living in areas with more greenness was related to improved psychological and social wellbeing in young people. In line with our hypotheses, higher residential greenness was related to higher psychological wellbeing in older adolescents (aged 15–17 years). Given that socioemotional difficulties typically worsen throughout adolescence8, our findings could contribute to an understanding of how greener living environments can reduce age-related vulnerabilities in wellbeing. Conversely, and in line with literature showing mixed or no associations between green environments and cognitive and emotional development in children85, the relationship between residential greenness and psychological wellbeing was not significant in children and young adolescents. Previous literature has also reported age-related effects in the relationship between residential greenness and psychological wellbeing35,48. While the effects reported here are small, small effects are still meaningful in understanding health inequalities, as they accumulate over populations and over time86. It is possible that the age-related differences discussed below might become increasing relevant with rapidly changing urban environments.

There are at least a couple of developmental explanations for the results reported in this study. First, feelings of autonomy and learning how to navigate complex social environments outside the home environment are key developmental tasks for adolescents in industrialised societies87,88, and these are related to positive affect and social connectivity89. In line with our hypotheses, our findings showed that higher residential greenness was in general related to higher family support & autonomy and this relationship played a role in higher psychological wellbeing in older adolescents living in environments with higher residential greenness. Future research could extend the current findings to understand how neighbourhood design and the spatial distribution of residential greenness can contribute to feelings of autonomy in young people44, perhaps by enabling independent mobility and exploration, and therefore contributing to overall psychosocial adaptation of young people2.

Second, belonging to peer groups and social evaluative concerns are particularly salient during adolescence25, with quality of friendships being an important determinant of positive self-perceptions and psychosocial adaptation during this period90,91. In line with research suggesting that residential greenness offers more spaces to build social ties16,46,48, our results show a weak but nonsignificant relationship between higher residential greenness and higher peer support in older adolescents. While further evidence is needed to support a positive role of residential greenness on peer support, it is possible that socialising and connectedness benefits of neighbourhoods characterised by residential greenness could be particularly impactful later in adolescence when peer networks become increasingly important92.

The health-promoting association of residential greenness, psychological wellbeing and perceived social support did not extend to green space availability, measured as whether or not there was a green area larger than 5000 m2 within a 150 m distance from the home. While large green spaces contribute to recreational uses of urban green, the benefits to subjective social and psychological wellbeing might depend rather on the daily use of these areas, which was not captured in this measure.

Population density and psychological wellbeing

Another aim of this study was to investigate whether population density was related to psychological wellbeing in children and young people. The results showed some support for our hypotheses, such that older adolescents (15–17 years) living in densely populated neighbourhoods showed lower psychological wellbeing than those in less densely populated neighbourhoods. This effect was weakened when adjusting for period-specific effects (p =.053). It is possible that the measure of population density used in this study captures a wider array of factors related to urban living that could independently contribute positively and negatively to wellbeing, weakening the overall effect. For example, densely populated urban environments are typically characterised by more traffic, noise and air pollution1, as well as by some positive factors, including more opportunities for participating in leisure activities41. Future research could investigate how specific factors in densely populated urban neighbourhoods might positively or negatively contribute to wellbeing and how sensitivities might differ throughout childhood and adolescence.

Neighbourhood deprivation and social inequalities in perceived social support and psychological wellbeing

Contrary to our hypotheses, there was no relationship between neighbourhood deprivation, perceived social support and psychological wellbeing. In addition, family affluence was not related to differences in psychological wellbeing and young people from disadvantaged backgrounds did not benefit to a greater extent from social and health-related benefits of residential greenness and green space. These results are inconsistent with the literature on social inequalities94,95 and the notion that young people from lower socioeconomic backgrounds are more likely to live in neighbourhoods that accumulate risky urban exposures (e.g. less natural space97).

The lack of a relationship between family affluence and wellbeing might result from the fact that the main socioeconomic measures available in the INMA dataset were collected during pregnancy or in early childhood, including the measures included in this study (family affluence, maternal education and neighbourhood deprivation). Past studies using the INMA dataset have shown that parental social class, parental education, and employment-related measures collected at birth were all predictive of cognitive outcomes in children aged 5–6 years98 and 7–8 years99. In addition, maternal education at birth, but not neighbourhood deprivation, was related to differences in parent-reported internalising and externalising difficulties in children aged 6–12 years67. One explanation for the null results in this study is that objective measures of family affluence might not completely represent subjective experiences of socioeconomic contexts, which might be particularly important for subjective wellbeing during adolescence100.

Limitations

The findings in this study should be interpreted considering certain limitations. First, some measures included in this study might have been affected by participant biases or measurement errors. For example, we used self-report measures of perceived social support and psychological wellbeing. These could have been affected by participant-specific affective states such as mood, which might have led to biased responses. Nonetheless, while the correlations between subscales were moderate to high (r =.37 to r =.57), this range is not high enough to be problematic101. In addition, the measures of perceived social support and psychological wellbeing were collected concurrently. Therefore, while this study investigated the effect of perceived social support on psychological wellbeing, higher psychological wellbeing could in turn influence perceived peer support, family support and feelings of autonomy, which is in line with research reversely showing better psychosocial adaptation (e.g. sociocognitive abilities) in young people with better mental wellbeing59. Also, our socioeconomic variables were collected at pregnancy, birth or estimated in early childhood and might only be representative of a short timeframe.

Second, while we used a directed acyclical graphic to map out the complex empirical relationships between the relevant covariates and confounder variables, it is possible that the overall small relationships found in our study reflect a combination of complex relationships from unmeasured variables and confounders not included in the causal model. For example, ethnicity and migrant status have been related to a heightened risk of emotional disorders and psychosis around Europe102, but there was insufficient variance to include these factors in our model. Parental mental health and genetic-related factors are also important predictors of child mental health8,103, and these have been shown to influence the likelihood of families migrating to urban centres104. In addition, while we cluster participants by region of study, a better representation of the residential neighbourhood environment would be postcodes or censual districts. Notably, however, the participants in this study might not only be exposed to residential environments, but might also be exposed to other environments (e.g. school). This is known in epidemiological research as the uncertain geographic context problem105,106. Future research could account for smaller and diverse geographic contexts.

Finally, most of the Time 2 follow-up data was collected during or after the coronavirus pandemic, and therefore the age-related differences in the results presented in this study could have been amplified by lockdown restrictions, which were particularly strict in urban settings. To adjust for period-specific influences in our results, we ran a set of sensitivity analyses including timepoint as a fixed effect. These sensitivity analyses suggested that: (i) the age-related effect of higher residential greenness on higher psychological wellbeing and perceived social support was robust to adjustment for period-specific effects; (ii) the age-related effects were mostly in older adolescents (aged 15 years and older) and not in middle adolescents (under 15 years), who were also tested during the pandemic; and (iii) these age-related differences were not driven by greater variance in psychological wellbeing responses in older adolescents (i.e. homogeneity of variance assumptions were met). In addition, the unexpected main effect of age on lower family support & autonomy and lower peer support might have been driven by period-specific effects. Importantly, however, timepoint and age are highly correlated in our data, resulting in a high variance inflation factor. Therefore, while it remains important to understand how age-related differences can emerge in the relationship in neighbourhood characteristics and psychological and social wellbeing, our results should be interpreted in the context of the pandemic.

Conclusion

The current study suggested that certain neighbourhood characteristics, including residential greenness and population density, may be related to adolescent wellbeing. Particularly, the findings suggested that older adolescents living in neighbourhoods with higher residential greenness experienced higher feelings of autonomy and perceived social support, and these were related to higher psychological wellbeing. In addition, we found that older adolescents living in densely populated urban environments reported lower psychological wellbeing. Future research could build on the current findings to investigate whether certain neighbourhood environments (e.g., greener neighbourhoods) might enable exploration and independence, contributing to healthy integration into friendship and larger peer groups. Finally, adopting a developmental framework, and taking into account biological, cognitive and social differences across childhood and adolescence, could be beneficial for studying population health effects in epidemiological studies, as these could disentangle how different urban exposures are related to wellbeing at different points in development (e.g. sensitive periods for specific urban stressors).