An estimated 970 million people – or 1 in every 8 – around the world live with a mental health disorder (World Health Organization, 2022). Poor mental health can have negative and long-lasting consequences for the individual as well as significant costs for society in terms of health services and economic losses. The need to reduce the prevalence of mental health problems is a public health priority and research to better understand risk and protective factors plays a vital role in this by informing prevention strategies. In addition to well-established risk factors such as genetics (Burmeister et al., Reference Burmeister, McInnis and Zöllner2008), trauma (Danese & Baldwin, Reference Danese and Baldwin2017), and poor physical health (Daré et al., Reference Daré, Bruand, Gérard, Marin, Lameyre, Boumédiène and Preux2019; Doherty & Gaughran, Reference Doherty and Gaughran2014), there has been growing interest in how the neighborhood environment matters for people’s mental health.
Neighborhood deprivation refers to a lack of access to opportunities and resources among residents including material/physical (e.g., food, accommodation, clothing) and social (e.g., crime, education, community services) aspects (Office for National Statistics, 2013), though studies vary in how they operationalize this. Limited education and employment opportunities and limited access to sufficient food, medical services and other collective resources are theorized to leave individuals prone to poor health while poor living conditions can also contribute to stress that undermines mental health (Kröger et al., Reference Kröger, Pakpahan and Hoffmann2015; Stafford & Marmot, Reference Stafford and Marmot2003).
Indeed, meta-analyses of longitudinal studies of adults have shown that living in more deprived neighborhoods (defined using composite measures) is associated with higher levels of mental disorders including depression (Sui et al., Reference Sui, Ettema and Helbich2022). Moreover, there is evidence of long-term effects of childhood exposure to neighborhood deprivation on mental health problems such as depression and anxiety in adulthood (Elovainio et al., Reference Elovainio, Vahtera, Pentti, Hakulinen, Pulkki-Raback, Lipsanen, Virtanen, Keltikangas-Jarvinen, Kivimaki, Kahonen, Viikari, Lehtimaki and Raitakari2020; Pearce et al., Reference Pearce, Cherrie, Shortt, Deary and Ward Thompson2018) suggesting that childhood and adolescence may be an important period for exposure. Indeed, growing up in a deprived neighborhood has also been linked with a greater risk of internalizing and externalizing problems in childhood and adolescence (Flouri et al., Reference Flouri, Midouhas and Francesconi2020; Leventhal & Brooks-Gunn, Reference Leventhal and Brooks-Gunn2000; Visser et al., Reference Visser, Bolt, Finkenauer, Jonker, Weinberg and Stevens2021) with socio-economic aspects of neighborhood deprivation potentially driving this association (Flouri et al., Reference Flouri, Midouhas and Francesconi2020). Findings suggest that the effect of neighborhood deprivation is typically modest in size, explaining 5%–10% of the variation in child and adolescent outcomes and is robust to adjustment for family-level disadvantage (Leventhal & Brooks-Gunn, Reference Leventhal and Brooks-Gunn2000). The latter is important as families are not randomly selected into neighborhoods, rather individuals with similar characteristics often come to live in similar areas. Therefore, ensuring that associations between neighborhood deprivation and poor mental health are robust to factors that influence both is critical to limit reverse causation in observational studies.
There remain important gaps in our knowledge. First, most studies of neighborhood deprivation measure exposure only once – rather than repeated measures over time – which may over- or under-estimate the effect of neighborhood deprivation on mental health (Helbich, Reference Helbich2018; Sui et al., Reference Sui, Ettema and Helbich2022). Second, a review of studies published since 2009 that examined the effects of neighborhood deprivation on young people’s mental health showed that the majority had been conducted in the USA, with few examining the UK context (Visser et al., Reference Visser, Bolt, Finkenauer, Jonker, Weinberg and Stevens2021). Although both are western, educated, industrialized, rich and democratic countries, they have differing social, demographic, and political contexts that may impact neighborhood deprivation and its effects on mental health. Third, UK studies using large, representative samples have typically focused on children (Flouri et al., Reference Flouri, Tzavidis and Kallis2010, Reference Flouri, Mavroveli and Midouhas2013) with less attention to adolescent mental health. Adolescence and the transition to adulthood is commonly the period in which a variety of mental health problems have their onset (Solmi et al., Reference Solmi, Radua, Olivola, Croce, Soardo, de Pablo, Shin, Kirkbride, Jones, Kim, Kim, Carvalho, Seeman, Correll and Fusar-Poli2021) and for many this will signal the start of recurrent mental illness (Caspi et al., Reference Caspi, Houts, Ambler, Danese, Elliott, Hariri, Harrington, Hogan, Poulton, Ramrakha, Rasmussen, Reuben, Richmond-Rakerd, Sugden, Wertz, Williams and Moffitt2020). Furthermore, neighborhood deprivation may be particularly salient for adolescent mental health as the strength of the effects may accumulate with sustained exposure since childhood. A better understanding of the relationship between neighborhood deprivation during childhood and mental health problems in adolescence in a contemporary UK population sample is therefore needed. Finally, studies of neighborhood deprivation have typically assessed child and adolescent mental health outcomes using parent-reported problem behaviors (e.g. Flouri et al., Reference Flouri, Midouhas and Francesconi2020) such as the Strengths and Difficulties Questionnaire (Goodman, Reference Goodman1997) and the Child Behaviour Checklist (Achenbach, Reference Achenbach1991). It remains unclear whether similar associations are found for mental health problems based on adolescents’ own reports.
Despite evidence that neighborhood deprivation increases the risk of mental health difficulties, outcomes do vary among those exposed to socio-economic disadvantage (Bakolis et al., Reference Bakolis, Murray, Hardy, Hatch and Richards2023; Kim-Cohen et al., Reference Kim-Cohen, Moffitt, Caspi and Taylor2004) with some individuals having better than expected outcomes. These people are sometimes referred to as showing “resilience” (Luthar et al., Reference Luthar, Cicchetti and Becker2000; Masten et al., Reference Masten, Best and Garmezy1990). Resilience is not a personal trait (though it may be facilitated by personal qualities such as self-esteem) but a dynamic process encompassing positive adaptation in the presence of significant adversity (Luthar et al., Reference Luthar, Cicchetti and Becker2000). Thus, by definition, resilience requires there to have been exposure to some form of adversity and factors that mitigate this risk resulting in adaptive outcomes – so called “protective factors” – can come from personal qualities or may be external to the individual such as from family, friends, and their environment/community (Rutter, Reference Rutter1985). Analyses of children in the UK Millennium Cohort Study (MCS) have highlighted factors such as self-regulation (Flouri et al., Reference Flouri, Midouhas and Joshi2014a), cognitive ability (Flouri et al., Reference Flouri, Midouhas and Joshi2014a), green space (Flouri et al., Reference Flouri, Midouhas and Joshi2014b), and early aspirations (Flouri et al., Reference Flouri, Midouhas, Joshi and Sullivan2016) as buffering some negative effects of family-level poverty. However, relatively few studies of neighborhood-level deprivation have investigated factors that are protective for mental health. A study of Welsh adults found high levels of social cohesion attenuated the association between neighborhood deprivation and poorer mental health (Fone et al., Reference Fone, White, Farewell, Kelly, John, Lloyd, Williams and Dunstan2014). Additionally, studies of UK children have shown verbal and non-verbal cognitive ability to moderate the association between neighborhood deprivation and pre-schoolers’ peer problems (Flouri et al., Reference Flouri, Mavroveli and Tzavidis2012); and general intelligence to moderate the effect of neighborhood poverty on behavioral and emotional problems at some, but not all, ages examined during childhood (Flouri et al., Reference Flouri, Midouhas and Joshi2015). No studies, to our knowledge, have focused on adolescents. However, an understanding of potential protective factors is central to improving knowledge of resilience during adolescence and may inform targets for interventions to better support young people who grow up in deprived neighborhoods and prevent them from developing mental health difficulties as they transition to adulthood.
The current study addresses the identified knowledge gaps by extending existing analyses of the MCS to answer the following research questions: (i) is high neighborhood deprivation in childhood associated with age-17 mental health problems and (ii) which factors protect against age-17 mental health problems for children exposed to high neighborhood deprivation? Drawing on the aforementioned factors that have been found to be protective of mental health in young children from deprived neighborhoods (Flouri et al., Reference Flouri, Mavroveli and Tzavidis2012, Reference Flouri, Midouhas and Joshi2015) and among children or adolescents exposed to other forms of adversity (Crush et al., Reference Crush, Arseneault, Jaffee, Danese and Fisher2018a; Crush et al., Reference Crush, Arseneault, Moffitt, Danese, Caspi, Jaffee, Matthews and Fisher2018b; Flouri et al., Reference Flouri, Midouhas and Joshi2014b; Masten et al., Reference Masten, Best and Garmezy1990; Meng et al., Reference Meng, Fleury, Xiang, Li and D’arcy2018; Sharma et al., Reference Sharma, Mustanski, Dick, Bolland and Kertes2019), we examine the potential protective role of biological sex, infant temperament, cognitive skills, self-esteem, family income, parent-child relationship, parenting behavior, access to green space, and social support. We test the robustness and generalisability of our findings using the Environmental Risk (E-Risk) Longitudinal Twin Study, an independent UK sample born in a similar time period, by investigating: (iii) are the associations between high neighborhood deprivation and mental health problems found in MCS comparable in E-Risk, and do they extend to clinically relevant psychiatric diagnoses at age 18? And (iv) are the putative protective factors found in MCS comparable in E-Risk?
Methods
Samples
MCS is a longitudinal survey of UK children that draws its sample from all live births between 1 September 2000 and 31 August 2001 (in England and Wales) and between 24 November 2000 and 11 January 2002 (in Scotland & Northern Ireland). Detailed information about the cohort is available at https://cls.ucl.ac.uk/cls-studies/millennium-cohort-study/ and in Supplementary Material. Briefly, eligible children were identified using government child benefit records. The sample was constructed to be representative of the total UK population and was disproportionately stratified to oversample certain subgroups, namely families living in areas of high child poverty, children of ethnic minority backgrounds, and living in the smaller nations of the UK (Plewis, Reference Plewis2007). At baseline, 18,552 families (18,827 children; 51% male) were included and an additional 692 eligible families were included at sweep 2 (total cohort = 19,244 families). Currently, there are seven sweeps of data, the first took place when children were around 9 months of age, and then they were followed up at ages 3, 5, 7, 11, 14 and 17 years. For comparability with the E-Risk cohort, we selected only MCS families who lived in England or Wales at baseline (N = 15,170) and for families with two children participating (N = 194; this includes twin and non-twin siblingsFootnote 1), one child was randomly selected to prevent family-level clustering. Furthermore, we analyzed cases with complete data for neighborhood deprivation, self-reported adolescent mental health, ethnicity, and family income (N = 5,422).
Ethical approval for MCS was obtained from the National Health Service Multi-Centre Ethics Committee up to age 14 years and the National Research Ethics Service at age 17 years. Study parents (and children after age 11 years) gave informed consent before data collection took place.
E-Risk tracks the development of a nationally representative birth cohort of 2,232 British children. The sample was drawn from a larger birth register of twins born in England and Wales in 1994–1995 (Trouton et al., Reference Trouton, Spinath and Plomin2002). Full details about the sample are reported elsewhere (Moffitt & E-Risk Study Team, Reference Moffitt2002) and in Supplementary Material. Briefly, the E-Risk sample was constructed in 1999-2000 when 1,116 families (93% of those eligible) with same-sex 5-year-old twins participated in home-visit assessments. Sex was evenly distributed within zygosity (49% male). Follow-up home-visits were conducted when the children were aged 7, 10, 12 and 18 years (participation rates were 98%, 96%, 96%, and 93%, respectively). There were 2,066 E-Risk participants (47% male) assessed at age 18, who represent the full socioeconomic distribution of the UK population (Supplementary Figure S1). We analyzed participants with complete data for neighborhood deprivation, self-reported adolescent mental health, ethnicity, and family income (N = 1,920).
Each phase of the E-Risk Study was approved by the Joint South London and Maudsley and the Institute of Psychiatry Research Ethics Committee. Study parents gave informed consent and children gave assent between 5–12 years and then informed consent at age 18.
Measures
Neighborhood deprivation was assessed in MCS at sweeps 1–6 (ages 9 months – 14 years) with the UK Government’s 2004 Index of Multiple Deprivation (IMD). IMD measures the relative levels of deprivation in small geographical areas using a weighted aggregation of several specific dimensions of deprivation. In England, the dimensions are income deprivation, employment deprivation, health deprivation and disability, education, skills and training deprivation, barriers to housing and services, living environment deprivation, and crime (Noble et al., Reference Noble, Wright, Smith and Dibben2006). The IMD ranks every small area in England (so-called Lower-Layer Super Output Areas, LSOAs, containing approximately 650 households or 1,500 individuals) from 1 (most deprived area) to 32,844 (least deprived area). Participants’ IMD rank could change between study sweeps only if they moved to a different LSOA. There is no unified measure of IMD across the different UK countries, therefore, for our analyses using MCS we standardized IMD rank within country at each study sweep and then averaged these to create an overall childhood IMD rank for each participant (Supplementary Table S1 shows how many participants’ IMD changed quartile during childhood). Because we investigated potential protective factors following exposure to high levels of neighborhood deprivation we defined “high” neighborhood deprivation as an average childhood IMD rank below the 25th centileFootnote 2.
In E-Risk, neighborhood deprivation was assessed using the UK Government’s 2015 IMD based on participants’ address at age 12. Scottish (N = 6) and Welsh (N = 39) addresses were converted to English IMD equivalents using a published conversion (Abel et al., Reference Abel, Payne and Barclay2016; Payne & Abel, Reference Payne and Abel2012). For our analyses using E-Risk data we utilized IMD decile (rather than rank) due to IMD rank being uniquely identifying given the twin status of participants. We defined “high neighborhood deprivation” as the two most deprived (i.e. lowest) IMD deciles. This cut-point identified 24% of the E-Risk analysis sample as being exposed to high neighborhood deprivation and is thus comparable to the 25th centile cut used in MCS.
Adolescent mental healthwas reported separately by participants and parents in MCS at sweep 7 (age 17) using the Strengths and Difficulties Questionnaire (SDQ; Goodman, Reference Goodman1997), a 25-item questionnaire with item responses indicated on a three-point scale (ranging from 0 = not true to 2 = certainly true). For self- and parent-reports respectively, a total difficulties score was calculated by summing all items (self-report α = 0.75; parent-report α = 0.63); items measuring emotional symptoms and peer problems were summed to create a subscale of internalizing problems (self-report α = 0.88; parent-report α = 0.96); and items measuring conduct problems and hyperactivity/inattention were summed to capture externalizing problems (self-report α = 0.86; parent-report α = 0.95). Self- and parent-reported adolescent mental health problems were moderately correlated (total problems: r = 0.46; internalizing problems: r = 0.47; externalizing problems: r = 0.44. All p’s < .001).
In E-Risk, participants were interviewed at age 18 about past-year symptoms of mental disorders. These methods have been previously described (Schaefer et al., Reference Schaefer, Moffitt, Arseneault, Danese, Fisher, Houts, Sheridan, Wertz and Caspi2018) and are detailed in Supplementary Material. Briefly, symptoms of five externalizing-spectrum disorders (alcohol, cannabis and tobacco dependence, conduct disorder, attention-deficit/hyperactivity disorder (ADHD)), four internalizing-spectrum disorders (depression, generalized anxiety disorder, posttraumatic stress disorder, eating disorder), and thought disorder (delusions, hallucinations, unusual thoughts and feelings) were assessed. Using confirmatory factor analysis, we estimated a bifactor model specifying a general psychopathology factor and three factors (representing internalizing, externalizing, and thought disorder symptoms; Supplementary Figure S2). Factor loadings and model fit are provided in Schaefer et al. (Reference Schaefer, Moffitt, Arseneault, Danese, Fisher, Houts, Sheridan, Wertz and Caspi2018) and summarized in Supplementary Material. Here we utilize the general psychopathology, internalizing and externalizing factors as outcomes to align with the total problems, internalizing and externalizing problems measured in MCS.
To extend our analyses to clinically relevant psychiatric diagnoses, E-Risk participants were additionally classified using Diagnostic and Statistical Manual of Mental Disorders (American Psychiatric Association, 1994, 2013) criteria as having any internalizing, externalizing or thought disorder based on their symptom reports and an overall outcome of “any psychiatric disorder” based on these three domain specific classifications (see Supplementary Material). For consistency with our main analyses we utilize any psychiatric disorder, any internalizing disorder, and any externalizing disorder as outcomes.
Putative protective factorsassessed in MCS are detailed in Table 1. Comparable constructs identified in E-Risk are detailed in Supplementary Material.
Table 1. Description of putative protective factors in the Millennium Cohort Study
Confounders Informed by literature and using a directed acyclic graph (see Supplementary Figure S3) we accounted for the potential confounding effect of family income (Peverill et al., Reference Peverill, Dirks, Narvaja, Herts, Comer and McLaughlin2021; Townsend, Reference Townsend1987) and ethnicity (Zuccotti, Reference Zuccotti2019).
Family income was assessed in MCS at sweep 1. Respondents selected the category that captured their total annual family take-home income from all sources after tax and other deductions: “£0 to <£3,100” (coded 1), “£3,100 to <£10,400” (2), “£10,400 to <£20,800” (3), “£20,800 to <£31,200” (4), “£31,200 to <£52,000” (5), or “£52,000 and above” (6).
Family income was assessed in E-Risk at age 5 by asking how much total income the household received from all sources before tax in the previous 12 months. Respondents selected from 15 categories ranging from “<£3,999” (coded 0) to “>£41,000 or more” (15). These categories were re-coded as follows to correspond to the MCS income categories as closely as possible: “<£3,999” (coded 1), “£4,000-£9,999” (2), “£10,000-£19,999” (3), “£20,000-£31,999” (4), “£32,000-£40,999” (5), “£41,000 or more” (6).
Ethnicity of the MCS cohort members was reported at sweep 1 according to one of six census classes: White, Mixed, Indian, Pakistani and Bangladeshi, Black or Black British, or Other ethnic group.
The ethnicity of E-Risk participants was recorded at study baseline according to one of five categories: White, Asian, Black, Mixed race, or Other ethnic group.
Maternal psychological distress was included as an additional confounder in a planned MCS sensitivity test. It was assessed at sweep 2 using the 6-item Kessler (K6) Scale (Kessler et al., Reference Kessler, Andrews, Colpe, Hiripi, Mroczek, Normand, Walters and Zaslavsky2002). Items ask how often in the last 30 days the mothers felt depressed, hopeless, restless, worthless, nervous, or that everything was an effort. Responses were given on a 5-point scale from “all the time” (coded 1) to “none of the time” (5). Scores were reverse coded and rescaled from 0 to 4, then summed so that a higher total score reflected higher levels of psychological distress (α = 0.98).
Statistical analyses
Analyses were conducted using Stata 15 and proceeded in the following steps:
Step 1: Is high neighborhood deprivation in childhood associated with age-17 mental health problems in MCS?
We examined associations between high neighborhood deprivation in childhood and (i) total SDQ problems, (ii) internalizing problems, and (iii) externalizing problems self-reported at age 17. Because IMD is measured at LSOA level, MCS participants who live within the same LSOA may be more similar to one another than those who live in different LSOAs therefore this clustering must be taken into account. Traditional multilevel models are not appropriate as data are non-hierarchical (Figure 1) – that is, children do not necessarily belong to one and only one LSOA, rather this can change over time such that they are nested within a cross classification of up to six LSOAs (one at each study sweep). We accounted for this non-hierarchical data structure using cross-classified models. These were fitted using Markov Chain Monte Carlo (MCMC) estimation using the “runmlwin” command (Leckie & Charlton, Reference Leckie and Charlton2013); a simulation-based method involving multiple iterations (see Supplementary Material for further details). Parameter estimates were computed at each iteration from which means, standard deviations and credible intervals were calculated. We adjusted models for the potentially confounding effects of family incomeFootnote 3 and ethnicity, and accounted for MCS sample stratification by including dummy variables representing the nine strata of MCSFootnote 4.
Figure 1. Millennium Cohort Study (MCS) data are not “purely” nested (i.e., children do not necessarily belong to one and only one LSOA), instead participants are nested within a cross-classification of up to six LSOAs (one at each study sweep). Note. Circles represent MCS study sweeps in which address data was collected (age 9 months, 3 years, 5 years, 7 years, 11 years, and 14 years); squares represent lower-layer super output areas (LSOAs); participants 1 and 2 shown for illustrative purposes. (Figure adapted from Bakolis et al., Reference Bakolis, Murray, Hardy, Hatch and Richards2023).
Step 2: Which factors protect against age-17 mental health problems for children exposed to high neighborhood deprivation in MCS?
Focusing on the outcome(s) that were found to be significantly associated (at p < .05) with high neighborhood deprivation in Step 1, we tested each putative protective factor. Consistent with the concept of resilience (Luthar et al., Reference Luthar, Cicchetti and Becker2000; Rutter, Reference Rutter1985), for something to be protective, there must first have been exposure to a risk – here, high neighborhood deprivation. Therefore, we identified MCS participants who were exposed to high neighborhood deprivation and used cross-classified models to examine whether the following factors were associated with fewer mental health problems at age 17 in these participants: biological sex, infant temperament, cognitive skills, self-esteem, family income, parent-child relationship, parental knowledge of child’s whereabouts, green space, and social support. These models were adjusted for family income, ethnicity and MCS sample stratification. When higher family income was examined as a predictor of fewer adolescent mental health problems we adjusted for ethnicity and MCS sample stratification.
For each factor found to be significantly associated (at p < .05) with fewer mental health problems among those exposed to high neighborhood deprivation, we then tested the statistical interaction with high neighborhood deprivation in the full MCS analysis sample. This tests whether they are associated with fewer mental health problems only in the context of exposure to high neighborhood deprivation in childhood (significant interaction) and thus have a protective effect or whether they are generally promotive for everyone regardless of deprivation exposure (no interaction and evidence of a main association). In other words, they exert a promotive effect in the whole sample. These models were adjusted for family income, ethnicity and MCS sample stratification.
Step 3: sensitivity analyses
To test the robustness of our findings we conducted sensitivity analyses (i) repeating Step 1 additionally controlling for potential confounding by maternal psychological distress (see Supplementary Material for detail); (ii) repeating Step 1 using only MCS participants who did not move residence between ages 14–17 (i.e. after the last IMD assessment); (iii) repeating Step 1 using alternative cut-points to define “high” neighborhood deprivation (i.e., below the median and using the full continuous scale); (iv) repeating Step 1 stratifying analyses by biological sex; and (v) repeating Steps 1 and 2 using age-17 parent-reported SDQ.
Step 4: Are the associations between high neighborhood deprivation and mental health problems found in MCS comparable in E-Risk, and do they extend to clinically relevant psychiatric diagnoses at age 18?
Using the E-Risk sample, we first used linear regression models to examine associations between high neighborhood deprivation at age 12 and (i) general psychopathology, (ii) internalizing symptoms, and (iii) externalizing symptoms factor scores at age 18. Next, we ran sensitivity analyses using logistic regression extending to (iv) any psychiatric disorder, (v) any internalizing disorder, and (vi) any externalizing disorder at age 18. There is no clustering of E-Risk participants at neighborhood level therefore we did not use cross-classified models. However, E-Risk twins are clustered within families, so we accounted for this using the Huber-White variance estimator (Rogers, Reference Rogers1994). Models were adjusted for the potential confounders of family incomeFootnote 5 and ethnicity.
Step 5: Are the putative protective factors found in MCS comparable in E-Risk?
For any significant associations (at p < .05) between neighborhood deprivation and mental health outcomes we then investigated whether the protective factors found in our MCS analyses were also associated with fewer mental health problems in E-Risk participants using linear regression models (logistic regression for categorical outcomes) and conceptually comparable variables.
Analytic samples
We analyzed complete cases (MCS: N = 5,422 and E-Risk: N = 1,920) because appropriately implementing multiple imputation in cross-classified data is extremely challenging (and understudied) due to the complex dependency between observations that is not captured by hierarchical models (Grund et al., Reference Grund, Lüdtke and Robitzsch2023) (see Supplementary Table S2 for details of missingness).
Results
Is high neighborhood deprivation in childhood associated with age-17 mental health problems in MCS?
A total of 1,498 MCS participants were exposed to high neighborhood deprivation during childhood (see Table 2 for descriptive statistics for all MCS study variables).
Table 2. Descriptive statistics of study variables in the whole Millennium Cohort Study (MCS) analysis sample and the subsample exposed to high levels of neighborhood deprivation in childhood
Note. SE = standard error. Tests for statistically significant differences between the whole MCS analysis sample and those exposed to high neighborhood deprivation could not be conducted as these groups are not independent of one another.
Exposure to high neighborhood deprivation was significantly associated with higher total problems and more internalizing problems (Table 3). No association was found for externalizing problems (Table 3).
Table 3. Association of high neighborhood deprivation in childhood with self-reported mental health problems at age 17 years in the Millennium Cohort Study (MCS)
Note. N = 5,422. CI = credible intervals; SD = standard deviation. Models were adjusted simultaneously for MCS sample stratification, family income and ethnicity. p is a Bayesian 1-sided p-value. Statistically significant estimates (p<0.05) are highlighted in bold.
Sensitivity analyses (i) additionally controlling for maternal psychological distress and (ii) using only MCS participants who did not move residence during adolescence replicated the above findings (Supplementary Tables S3 & S4).
Sensitivity analyses using average childhood IMD as a continuous scale (Supplementary Table S5) and using the median to define exposure to high deprivation (Supplementary Table S6) showed significant associations with all three mental health outcomes. Consistent with our findings above, the largest effects were found for total problems and internalizing problems.
Stratifying analyses by biological sex (Supplementary Table S7) showed that for males the largest effects were for total problems and internalizing problems, and there was no association with externalizing problems. For females, effect sizes were largest for total problems and were comparable for internalizing and externalizing problems (albeit statistically non-significant).
Sensitivity analyses using parent-reported adolescent mental health problems indicated significant associations between high neighborhood deprivation and more internalizing, externalizing, and total problems (Supplementary Table S8).
Which factors protect against age-17 total problems and internalizing problems for children exposed to high neighborhood deprivation in MCS?
Within MCS participants exposed to high neighborhood deprivation during childhood (N = 1,498), male sex, easier infant temperament, higher self-esteem, a more positive parent-child relationship, greater parental knowledge about their child’s whereabouts, and more social support were each associated with fewer total problems self-reported at age 17 (Table 4). None of the interactions with high neighborhood deprivation were found to be significant and all were robustly associated with fewer total problems in the full sample, with male sex showing the largest effects (Supplementary Table S9).
Table 4. Associations between putative protective factors and age-17 total problems among Millennium Cohort Study (MCS) participants exposed to high levels of neighborhood deprivation during childhood
Note. Putative protective factors were tested individually, one at a time. CI = credible intervals; SD = standard deviation. Models were adjusted simultaneously for MCS sample stratification, family income and ethnicity. The model testing high family income as a protective factor was only adjusted for MCS sample stratification and ethnicity. p is a Bayesian 1-sided p-value. Statistically significant estimates (p<0.05) are highlighted in bold.
Male sex, easier infant temperament, higher self-esteem, a more positive parent-child relationship, and more social support were also associated with fewer age-17 internalizing problems for those exposed to high neighborhood deprivation during childhood (Table 5). Conversely, more green space in their local area was associated with more adolescent internalizing problems for those exposed to high deprivation (Table 5).
Table 5. Associations between putative protective factors and age-17 internalizing problems among Millennium Cohort Study (MCS) participants exposed to high levels of neighborhood deprivation during childhood
Note. Putative protective factors were tested individually, one at a time. CI = credible intervals; SD = standard deviation. Models were adjusted simultaneously for MCS sample stratification, family income and ethnicity. The model testing high family income as a protective factor was only adjusted for MCS sample stratification and ethnicity. p is a Bayesian 1-sided p-value. Statistically significant estimates (p<0.05) are highlighted in bold.
The interaction between higher self-esteem and high neighborhood deprivation was significant in the full sample (Supplementary Table S10), with the association with fewer age-17 internalizing problems being slightly stronger for those who were exposed to high neighborhood deprivation (estimate = −0.35, SD = 0.04, p < .001, 95% CI = −0.43– −0.27) compared to those who were not (estimate = −0.29, SD = 0.03, p < .001, 95% CI = −0.34– −0.24). The interaction between green space and high neighborhood deprivation was also significant in the full sample (Supplementary Table S10) such that having more green space in the local area was associated with more internalizing problems for those exposed to high neighborhood deprivation (estimate = 0.12, SD = 0.06, p = .035, 95% CI = −0.01–0.24) but not for those who were unexposed (estimate = −0.21, SD = 0.03, p = .207, 95% CI = −0.07–0.03). No other interactions were found to be significant but male sex, a more positive parent-child relationship, and greater social support were all robustly associated with fewer internalizing problems in the full sample (Supplementary Table S10), with the largest effects found for male sex and social support.
Supplementary analyses using parent-reported age-17 mental health problems showed a consistent pattern across the three outcomes (Supplementary Table S11). An easier infant temperament, higher self-esteem, higher cognitive skills, greater parental knowledge of their child’s whereabouts, a more positive parent-child relationship, higher social support, and higher family income were associated with fewer internalizing, externalizing and total problems among those exposed to high neighborhood deprivation in childhood. Additionally, male sex was associated with fewer parent-reported internalizing problems whereas female sex was associated with fewer parent-reported externalizing problems. However, in the full sample, only easier infant temperament, greater parental knowledge, and higher social support were found to be more strongly associated with fewer parent-reported total and externalizing problems among those exposed to high neighborhood deprivation compared to those who were not (Supplementary Tables S12 and S13, respectively). Greater parental knowledge was found to be associated with fewer parent-reported internalizing problems in the high neighborhood deprivation group but not in the unexposed group (Supplementary Tables S14 and S15).
Are the associations between high neighborhood deprivation and mental health problems found in MCS comparable in E-Risk and do they extend to clinically relevant psychiatric diagnoses at age 18?
A total of 463 (24.1%) E-Risk participants were exposed to high neighborhood deprivation at age 12 (see Table 6 for descriptive statistics of all E-Risk study variables).
Table 6. Descriptive statistics of study variables in the whole E-Risk study analysis sample and the subsample exposed to high levels of neighborhood deprivation in childhood
Note. E-Risk = Environmental Risk Longitudinal Twin Study; IQ = intelligence quotient; SD = standard deviation. Tests for statistically significant differences between the whole E-Risk analysis sample and those exposed to high neighborhood deprivation could not be conducted as these groups are not independent of one another.
There were no significant associations between high neighborhood deprivation in childhood and age-18 general psychopathology (estimate = −0.32, SE = 1.04, p = .758, 95% CI = −2.35–1.72), internalizing symptoms (estimate = −0.29, SE = 1.03, p = .779, 95% CI = −2.32–1.74) or externalizing symptoms (estimate = −0.61, SE = 1.02, p = .549, 95% CI = −2.61–1.39) in E-Risk. Similarly, for clinically relevant psychiatric diagnoses, there were no significant associations between age-12 high neighborhood deprivation and any psychiatric disorder (OR = 0.89, SE = 0.12, p = .372, 95% CI = 0.68–1.15), any internalizing disorder (OR = 0.98, SE = 0.14, p = .857, 95% CI = 0.74–1.28) or any externalizing disorder (OR = 0.82, SE = 0.12, p = .173, 95% CI = 0.62–1.09) at age 18.
Given the absence of a robust association between high neighborhood deprivation and mental health problems in E-Risk participants, we did not proceed to investigate putative protective factors as originally planned in this sample.
Discussion
Main findings
In MCS, exposure to high neighborhood deprivation in childhood was associated with more total mental health problems and internalizing (but not externalizing) difficulties in adolescence. Being male, having an easier infant temperament, higher self-esteem, more social support, a more positive parent-child relationship, and parents with more knowledge of their child’s whereabouts were associated with fewer total problems in individuals exposed to high neighborhood deprivation in childhood and those who were not. Similarly, being male, having higher self-esteem, a more positive parent-child relationship and more social support were also associated with fewer adolescent internalizing problems. These were promotive for all regardless of the level of neighborhood deprivation experienced, however higher self-esteem was linked more strongly with fewer internalizing problems for those exposed to high neighborhood deprivation. Statistically significant associations between neighborhood deprivation and poorer adolescent mental health were not found among E-Risk participants.
Importantly, the relationships between childhood neighborhood deprivation and adolescent internalizing and total problems in MCS were independent of family income. Since income likely influences where someone can live, accounting for this is crucial to demonstrate a true effect of the neighborhood environment over and above family-level disadvantage. The associations were also robust to several sensitivity checks including using the continuous measure of IMD and the median split.
Our finding that MCS participants who lived in relatively highly deprived neighborhoods during childhood reported more internalizing problems at age 17 is consistent with several studies and meta-analyses showing an association between deprived neighborhoods and depression and anxiety in children (Bakolis et al., Reference Bakolis, Murray, Hardy, Hatch and Richards2023) and in adults (Elovainio et al., Reference Elovainio, Vahtera, Pentti, Hakulinen, Pulkki-Raback, Lipsanen, Virtanen, Keltikangas-Jarvinen, Kivimaki, Kahonen, Viikari, Lehtimaki and Raitakari2020; Pearce et al., Reference Pearce, Cherrie, Shortt, Deary and Ward Thompson2018; Sui et al., Reference Sui, Ettema and Helbich2022). One possibility is that neighborhood deprivation exposes individuals to more stress alongside less resource to counteract it. Individuals’ feelings of safety may also be lower in more deprived areas (Mouratidis, Reference Mouratidis2020; Polling et al., Reference Polling, Khondoker, study team, Hatch and Hotopf2014) which has been linked with young people’s emotional but not conduct disorders (Meltzer et al., Reference Meltzer, Vostanis, Goodman and Ford2007) mirroring our finding of an association with internalizing but not externalizing problems.
Cross-cohort comparisons
The links between childhood neighborhood deprivation and poorer adolescent mental health demonstrated in MCS were not found in our second cohort, the E-Risk study. This could reflect generational social change such that neighborhood deprivation has become more important for mental health in recent years. Alternatively, it could reflect methodological differences between the cohorts. E-Risk used only a single assessment of IMD at age 12 compared to MCS’s multiple measures across childhood with the latter capturing more information about deprivation exposure.
The mental health measures also differ between the cohorts with MCS using a population mental health metric compared to the more clinical interview measure used in E-Risk. Furthermore, there are important differences in the “internalizing” and “externalizing” problems and “total problems” and “general psychopathology” constructs derived from these measures between the two cohorts. Specifically, MCS uses observed scores on the SDQ, the subscales of which can be correlated, whereas E-Risk uses bifactor modeling to obtain independent (uncorrelated) factors from which factor scores are analyzed. Thus, there will be positive intercorrelations between the “total problems,” “internalizing” and “externalizing” problems scores in MCS whereas the internalizing and externalizing factor scores in E-Risk represent the constructs of externalizing and internalizing symptoms over and above the latent “general psychopathology” factor and are specified to be uncorrelated with each other (see Schaefer et al., Reference Schaefer, Moffitt, Arseneault, Danese, Fisher, Houts, Sheridan, Wertz and Caspi2018). These differences in the measurement of mental health may also contribute to the different findings we report across the MCS and E-Risk cohorts.
Parent- versus self-reported mental health
Finally, we revealed a different pattern of associations for self- versus parent-reported adolescent mental health problems in MCS with an association between high levels of neighborhood deprivation and externalizing problems evident only when parent-report was utilized. Discrepancies between adolescent and parent reports of mental health are well-documented (Booth et al., Reference Booth, Moreno-Agostino and Fitzsimons2023; De Los Reyes et al., Reference De Los Reyes, Augenstein, Wang, Thomas, Drabick, Burgers and Rabinowitz2015). Whilst adolescents are assumed to have better insight into their own internalizing problems because feelings and emotions may be unseen by parents (Sourander et al., Reference Sourander, Helstelä and Helenius1999), parent-report may be more suitable for identifying externalizing problems, particularly oppositional behaviors and those related to conduct disorder and ADHD (Aebi et al., Reference Aebi, Kuhn, Banaschewski, Grimmer, Poustka, Steinhausen and Goodman2017). There were also some differences in the pattern of associations found with the putative protective factors and their interactions with neighborhood deprivation. Most notably, higher cognitive skills were associated with fewer parent-reported mental health problems across all three outcomes for those exposed to high neighborhood deprivation and those who were not. However, despite some differences, social support and self-esteem continued to yield the largest effect sizes of the modifiable factors (alongside higher family income), consistent with our findings using self-report.
Implications
Our study adds to an emerging literature suggesting that tackling the root causes of inequality to prevent neighborhood deprivation may help prevent some mental health issues (Elovainio et al., Reference Elovainio, Vahtera, Pentti, Hakulinen, Pulkki-Raback, Lipsanen, Virtanen, Keltikangas-Jarvinen, Kivimaki, Kahonen, Viikari, Lehtimaki and Raitakari2020; Leventhal & Brooks-Gunn, Reference Leventhal and Brooks-Gunn2000; Pearce et al., Reference Pearce, Cherrie, Shortt, Deary and Ward Thompson2018; Visser et al., Reference Visser, Bolt, Finkenauer, Jonker, Weinberg and Stevens2021). Structural-level interventions to improve equity in this way have the potential benefit of impacting many people at the same time with cascading effects over the life course. However, our findings also identify several individual-level modifiable factors that could be the focus of interventions to reduce adolescent mental health problems among those exposed to neighborhood deprivation as well as those who are not.
For example, we found a role for parenting (positive parent-child relationship and greater knowledge of child’s whereabouts) in promoting better adolescent mental health. This was the case for individuals exposed to high levels of childhood neighborhood deprivation and those who were not and is consistent with research showing the wider benefits of secure attachment (Brumariu, & Kerns, Reference Brumariu and Kerns2010) and parental monitoring (Cadman et al., Reference Cadman, Paul, Culpin, Sallis, Bould and Pearson2022) for adolescent mental health. Parents who are more involved and knowledgeable about their offspring’s whereabouts may be better able to support and help them. Greater knowledge and a more positive parent-child relationship may also reflect open communication allowing the child/adolescent to disclose any difficulties and seek help thus potentially reducing the likelihood of mental health problems.
Additionally, consistent with a wealth of evidence highlighting the benefits of social support (Gariepy et al., Reference Gariepy, Honkaniemi and Quesnel-Vallee2016; Latham et al., Reference Latham, Arseneault, Alexandrescu, Baldoza, Carter, Moffitt, Newbury and Fisher2022; Wang et al., Reference Wang, Mann, Lloyd-Evans, Ma and Johnson2018) and self-esteem (Orth & Robins, Reference Orth and Robins2022) for mental health, we also found evidence that these helped reduce the level of total problems and internalizing problems reported at age 17. These effects were the largest (aside from biological sex) and thus may be viable targets for interventions to increase young people’s self-esteem and the social support they have available to them to help prevent mental health difficulties. For example, cognitive behavior therapy and competitive memory training targeting self-esteem have been shown to be effective (Kolubinski et al., Reference Kolubinski, Frings, Nikčević, Lawrence and Spada2018; Korrelboom et al., Reference Korrelboom, IJdema, Karreman and van der Gaag2022) with beneficial effects on depression and anxiety also reported (Korrelboom et al., Reference Korrelboom, IJdema, Karreman and van der Gaag2022; Staring et al., Reference Staring, Van den Berg, Cath, Schoorl, Engelhard and Korrelboom2016). Whilst our findings suggest that this would be beneficial for everyone, regardless of exposure to neighborhood deprivation, they also indicate that improvements to self-esteem may be especially beneficial for those in deprived neighborhoods. When resources are limited, targeting those in more deprived areas first or with greater intensity (Marmot, Reference Marmot2010) is recommended given their higher risk for developing mental health problems.
Our study also adds to literature highlighting the need to better understand the societal conditions that drive gender inequalities in adolescent mental health problems (Campbell et al., Reference Campbell, Bann and Patalay2021; McElroy et al., Reference McElroy, Tibber, Fearon, Patalay and Ploubidis2023). Whilst biological sex at birth is not in itself a modifiable target for intervention, identifying what underlies the benefits of being male for having fewer total problems and internalizing difficulties may help inform interventions to promote adolescent mental health.
Strengths and limitations
To our knowledge, this is the first UK study to examine the association between childhood neighborhood deprivation and adolescent internalizing and externalizing problems, plus investigate factors that are protective of mental health in this risky context. Strengths include the use of two large nationally representative samples, cross-classified models to appropriately model complex data structures, repeated measures of neighborhood deprivation, longitudinal prospective measures of mental health and key confounding variables, and sensitivity analyses to test the robustness of findings.
We also acknowledge limitations. By averaging the repeated measures of neighborhood deprivation in MCS we did not model whether changes reflected improvements or worsening of deprivation exposure over childhood, and it is possible that mental health outcomes may differ accordingly. Averaging the measures also meant that we were blind to any cumulative or sensitive period effects. Furthermore, in MCS changes in IMD during childhood only capture changes resulting from participants moving neighborhoods, changes in the relative deprivation of neighborhoods themselves could not be captured as only the 2004 IMD has been applied to participants’ addresses. However, whilst some change in the relative rank of neighborhoods is expected over time, evidence suggests that among the most and least deprived areas there is considerable continuity (Office for National Statistics, 2011, 2015).
Although we use IMD to measure neighborhood deprivation in both MCS and E-Risk, the samples consist of different individuals who may be disadvantaged in different ways. This within-group variation is not captured by our study as we investigated group-level differences rather than individual-level prediction. Consistent with the majority of research in this field, we utilized the overall IMD rank and therefore did not separately examine the different domains of neighborhood deprivation. These domains may be differentially associated with mental health problems (Flouri et al., Reference Flouri, Midouhas and Francesconi2020) and they may have different protective factors.
The E-Risk study applied a bifactor model to its measurement of adolescent mental health symptoms which is a popular approach to capture common and unique elements of psychopathology (e.g., Caspi et al., Reference Caspi, Houts, Belsky, Goldman-Mellor, Harrington, Israel, Meier, Ramrakha, Shalev, Poulton and Moffitt2014; Laceulle et al., Reference Laceulle, Vollebergh and Ormel2015; Sharp et al., Reference Sharp, Wright, Fowler, Frueh, Allen, Oldham and Clark2015). However, concerns have recently been raised about this approach including how bifactor models are evaluated – e.g., that the model has a tendency to overfit such that global fit statistics favor the model even when it is a poor description of the data – and questions about how the “general psychopathology” (or “p factor”) and “internalizing and “externalizing” factors are interpreted (Bornovalova et al., Reference Bornovalova, Choate, Fatimah, Petersen and Wiernik2020; Watts et al., Reference Watts, Greene, Bonifay and Fried2024). For example, to what extent internalizing and externalizing factors can be interpreted as traits that are independent of the p factor, and exactly what the causal mechanisms involved in the p factor are (Bonifay et al., Reference Bonifay, Lane and Reise2017; Watts et al., Reference Watts, Greene, Bonifay and Fried2024).
Our analyses controlled for family income measured at study baseline. Repeated measures of family income were available in MCS such that an alternative strategy would have been to calculate an average family income variable. However, because family income was only measured at study baseline in E-Risk, we opted to maximize comparability between the studies.
We examined a large number of associations and although the significance level at 5% is purely nominal and the likelihood of type I error is inflated, the majority of our findings in MCS were consistent in terms of the direction and size of the association. In addition, we analyzed complete cases which may have introduced some bias into our MCS results where missing data was higher. As in any observational study, unmeasured confounding is a source of potential bias.
Finally, the E-Risk sample was comprised of twins and the extent to which findings can be generalized to non-twin samples is sometimes questioned. However, the prevalence of mental health problems has been shown to be comparable between twins and non-twins (Kendler et al., Reference Kendler, Martin, Heath and Eaves1995), and the E-Risk families closely match the deciles of the UK government IMD (Supplementary Figure S1) suggesting neighborhood deprivation was also comparable.
Conclusion
Children who grow up in more deprived UK neighborhoods may be at risk of developing more mental health problems, including more internalizing problems, in adolescence. Tackling the root causes of inequality may help improve mental health whilst interventions that focus on improving self-esteem and the availability of social support may also be effective in promoting better adolescent mental health in the whole population.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0954579425000203.
Acknowledgements
We are grateful to the Centre for Longitudinal Studies (CLS), UCL Social Research Institute, for the use of the Millennium Cohort Data and to the UK Data Service for making them available. However, neither CLS nor the UK Data Service bear any responsibility for the analysis or interpretation of these data. We are grateful to the E-Risk study mothers and fathers, the twins, and the twins’ teachers and neighbors for their participation. Our thanks to Professors Terrie Moffitt and Avshalom Caspi, the founders of the E-Risk study, and to the E-Risk team for their dedication, hard work, and insights. We also thank Dr Joanne Newbury and Professor Candice Odgers for their work on the IMD measure within the E-Risk Study.
Funding statement
The Millennium Cohort Study is supported by the Centre for Longitudinal Studies, Resource Centre 2015–20 grant [ES/M001660/1], 2022-25 grant [ES/W013142/1] and a host of other co-funders. The E-Risk Study is funded by the Medical Research Council (MRC) [G1002190, MR/X010791/1]. Additional support was provided by the National Institute of Child Health and Human Development [HD077482] and the Jacobs Foundation. Drs Latham and Moreno-Agostino and Profs Das-Munshi, Ploubidis and Fisher are supported by the Economic and Social Research Council (ESRC) Centre for Society and Mental Health at King’s College London [ES/S012567/1]. Prof Ploubidis is also supported by the ESRC Centre for Longitudinal Studies Resource Centre grant at University College London [ES/W013142/1]. Profs Bakolis and Das-Munshi are part supported by the National Institute for Health and Care Research (NIHR) Applied Research Collaboration South London (NIHR ARC South London) at King’s College Hospital NHS Foundation Trust and by the NIHR Maudsley Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London. Prof Das-Munshi is supported by UK Research and Innovation funding for the Population Mental Health consortium [MR/Y030788/1], part of Population Health Improvement UK (PHI-UK). For the purposes of Open Access, the author has applied a CC BY public copyright license to any Author Accepted Manuscript version arising from this submission. The views expressed are those of the authors and not necessarily those of the ESRC, MRC, NIHR, the Department of Health and Social Care, or King’s College London. These funders played no role in study design; in the collection, analysis, and interpretation of data; in the writing of the report; nor in the decision to submit this article for publication.
Competing interests
All authors declare that they have no conflicts of interest.
Open access
