Health & Place 30 (2014) 226–233

Contents lists available at ScienceDirect

Health & Place journal homepage: www.elsevier.com/locate/healthplace

Rural–urban area of residence and trajectories of children's behavior in England Emily Midouhas a,n, Lucinda Platt b a b

Department of Psychology and Human Development, Institute of Education, 25 Woburn Square, WC1H 0AA, UK Department of Social Policy, London School of Economics and Political Science, Old Building, Houghton Street, WC2A 2AE, UK

art ic l e i nf o

a b s t r a c t

Article history: Received 17 April 2014 Received in revised form 24 August 2014 Accepted 4 September 2014

Despite extensive studies of neighborhood effects on children's outcomes, there is little evidence on rural–urban impacts on child mental health. We modeled trajectories of emotional–behavioral problems of white majority children at ages 3, 5, and 7 in England in areas with varying levels of rural and urban settlement, using the Millennium Cohort Study. After adjusting for area selection, children in less sparse rural areas had fewer conduct and peer problems, and children in areas with a mix of rural and urban settlement had fewer emotional symptoms, explained by the quality of their schools. Area differences remained in emotional problems. & 2014 Elsevier Ltd. All rights reserved.

Keywords: Child behavior Millennium Cohort Study Neighborhoods Rural Urban

1. Introduction Despite a wealth of literature on neighborhood effects on children's outcomes, there is little evidence on the role of rural– urban impacts on child mental health. The literature has instead tended to focus on deprivation in urban contexts (Leventhal and Brooks-Gunn, 2000). However, there are good theoretical reasons to expect that differences in the rural–urban makeup of an area will impact young children's behavioral development. First, the composition of those living in areas of different levels of rurality is likely to differ (Pateman, 2011), thus selection into particular sorts of areas by those with different family characteristics will result in indirect effects of an area's rurality. Secondly, the characteristics of the area may have effects over and above family-level characteristics, due to differences in social environment (Coleman, 1988; Putnam, 2000), access to quality education and services (Leventhal and Brooks-Gunn, 2000; Lupton, 2003), and the presence of others who can act as role models and help to enforce social control (Sampson et al., 1999). Third, it is possible that the intrinsic qualities of different areas, whether peaceful countryside, bustling town or densely populated inner city may directly impact children's wellbeing and behavior over and above the resources they offer (Lupton, 2003).

n

Corresponding author. Tel.: þ 44 02076126279; fax: þ44 02076126304. E-mail addresses: [email protected] (E. Midouhas), [email protected] (L. Platt). http://dx.doi.org/10.1016/j.healthplace.2014.09.003 1353-8292/& 2014 Elsevier Ltd. All rights reserved.

However, previous discussions of the rural ‘idyll’ have typically confounded the differential socio-economic status of those living in areas of varying levels of urbanity with desirable characteristics of rural areas themselves. Moreover, when exploring rural–urban effects it is important to identify the potential mechanisms through which differential effects operate. In this paper, we investigate, first, whether there are rural–urban differences in children's behavior using four emotional–behavioral measures; second, whether these are driven by selection of more or less advantaged families into particular areas; and third, whether any residual differences can be accounted for by two specific mechanisms, discussed further below: the presence of high status adults and school quality. Any remaining differences may then suggest some role played by intrinsic properties of different sorts of area. A small body of research has explored differences in adolescent and children's child cognitive ability in rural compared with urban areas of the US, UK, and Australia, producing mixed results (Gibbons and Silva, 2008; Midouhas and Flouri, 2013; National Centre for Social Research, 2009). One reason may be that the nature of ‘rural' and ‘urban' differs by country. For example, UK rural areas are closer to metropolitan areas than their US counterparts. Poverty is also more likely in US rural areas whereas it is disproportionately found in UK urban areas (Pateman, 2011). Another factor may be different rural–urban definitions. In the UK, some are based on population density whilst others capture occupational structure or the presence of services (Lupton, 2003). The UK Government Rural–Urban Definition for England and Wales focuses on settlement type of small areas

E. Midouhas, L. Platt / Health & Place 30 (2014) 226–233

(leaving out social, economic and cultural elements which change more with time), thus allowing for a reasonably stable definition of rural–urban area. Using this definition, analysis of English National Pupil Database and other large-scale data on adolescent children's attainment shows that rural children on average have higher school attainment than their urban counterparts, though differences depend on settlement type and sparseness of area (Commission for Rural Communities, 2010; National Centre for Social Research, 2009). However, selection bias, caused by lack of independence between the selection mechanism into areas and child outcomes, may explain rural–urban area differences. National Centre for Social Research (2009) found that rural English children's higher attainment was primarily due to the higher social position of their parents. Differences in younger children's cognitive outcomes in England at higher geographies have also been identified. Midouhas and Flouri (2013) found that selective sorting of families into areas explained most rural–urban differences in ability according to a local authority classification. However, children in areas with a mix of rural and urban settlement had higher ability after accounting for selection, explained by the higher level of human capital in these areas. It remains possible that developmental differences also exist in rural and urban children's behavior. Research into neighborhood effects on behavioral outcomes has found that young children in deprived areas in the US (Duncan et al., 1994) and UK (McCulloch, 2006) have greater emotional and behavioral problems than their counterparts in less deprived areas, even after controlling for area selection. However, other UK research found that family characteristics explained the variation in children's behavioral problems according to area deprivation (e.g. Flouri et al., 2010). Notwithstanding the problem of area selection, theories of neighborhood effects point towards potential pathways from rural– urban areas to children's psychological difficulties (Leventhal and Brooks-Gunn, 2000). Jencks and Mayer's (1990) collective socialization/social control model and institutional model emphasize the potential benefits of ‘high status’ adults (those of higher socioeconomic position) in a neighborhood, who may act as positive role models, provide economic, social and educational resources, and help to maintain social control, thereby promoting opportunities and minimizing bad behavior. Children in rural areas are more likely to be exposed to such ‘high status’ adults than urban children (Commission for Rural Communities, 2010; Pateman, 2011). Another pathway may be the characteristics of local institutions, particularly schools, theorized to offer ways for parents to stimulate learning and healthy development in their children (Leventhal and Brooks-Gunn, 2000). School characteristics such as pupil socioeconomic composition, attainment levels and school climate, have been shown to differ by rural–urban area type (Commission for Rural Communities, 2010). Leventhal and Brooks-Gunn's (2000) norms/collective efficacy model also sees such resources as contributing to the supervision and monitoring of children. Additionally, school rather than neighborhood composition may explain ‘area effects’ (Owens, 2010). For example, attending school with high achieving students may expose less advantaged students to norms about both achievement and behavior (Gaviria and Raphael, 2001). Yet school achievement, often perceived as school ‘quality’, is related to parents' decisions to live in particular areas and might capture area selection rather than act as a pathway to child behavior (Browne and Goldstein, 2010). The present study examines the association between rural and urban area of residence and children's emotional and behavioral problem trajectories from ages 3 to 7. We attempt to account for selection bias caused by families' selective sorting into rural–urban areas by adjusting for mother's education level and social class, and income. In England, the percentage of people working in higher

227

managerial or professional occupations or qualified to at least degree level is higher than average in rural areas, but average or below average in sparse rural areas (Pateman, 2011). Similarly, people in urban areas and sparse rural areas are more likely than those in other areas to have no qualifications or to have low incomes (Rural Evidence Research Centre, 2004; Pateman, 2011). We further adjust for parental involvement (reading to the child), a potentially important influence on children's behavior (Flouri et al., 2010). In the presence of rural–urban effects that persist after reducing selection bias through these controls, we attempt to identify whether rural– urban effects are mediated by the achievement level of children's schools or the local presence of high status adults. Given previous work into area differences in young children's adjustment (Leventhal and Brooks-Gunn, 2000), and rural–urban differences in their cognitive ability (Midouhas and Flouri, 2013), we hypothesized that living in more rural, but not isolated, areas would be associated with more positive child behavioral outcomes. However, we also expected that accounting for selection of families into areas would attenuate the majority of these differences.

2. Method 2.1. Participants We used data from the first four sweeps (at children's ages 9 months, 3, 5 and 7 years) of the Millennium Cohort Study (MCS), a large-scale longitudinal study of children born in the four UK countries in 2000–2002 (Plewis, 2007). The MCS employed a stratified, clustered sample design, with oversampling from disadvantaged areas, areas with high ethnic minority populations and the smaller UK countries. Our analytic sample comprised singleton children of white majority ethnic background living in England at all four sweeps of MCS who did not change area type. We focus on families in England because the DEFRA classification of rural–urban we used applies to England only. We excluded children who changed area type at least once between the sweeps, as there is evidence that a true effect of an area's conditions on individuals' outcomes is not detected until they have been living in that area for several years (e.g. Sampson et al., 2007). We excluded ethnic minorities for two reasons. First, they tend to be relatively geographically concentrated, and predominantly resident in metropolitan areas (National Statistics, 2004). This renders comparison across different types of rural–urban area problematic, due to out of sample predictions. The second reason is that the relationships between area type and behavior may differ between groups, given distinctive patterns of behavior across ethnic groups, and that ethnic minorities who live in less urban areas are likely to be distinctive in possibly unobserved ways. Therefore, of 19,518 children who ever participated in MCS across all four countries, we focused on the 7414 in England in all four sweeps. Of these, 7224 were singletons. A further 825 were excluded because they made at least one move and 1556 were from an ethnic minority background resulting in our final analytic sample (n ¼4843). 2.2. Measures Emotional and behavioral difficulties were assessed from parental report on 20 items from the Strengths and Difficulties Questionnaire (SDQ; Goodman, 1997) at sweeps 2, 3, and 4 (ages 3, 5 and 7). The SDQ measures conduct problems, hyperactivity, emotional symptoms, and peer problems using five items for each domain. Responses ranged from 0 (not true) to 2 (certainly true) and were summed to provide a total score for each dimension ranging from zero to ten.

228

E. Midouhas, L. Platt / Health & Place 30 (2014) 226–233

Rurality/urbanity (‘rural–urban area type’): Rural and urban is defined using the UK government's classification for Local Authorities (LA) in England (Rural Evidence Research Centre, 2009). The six-fold classification applies to lower tier authorities (‘districts’). At MCS sweep 1, the 354 lower tier LAs in England had 2148– 138,977 residents, with an average of 138,819. This large geographical unit can capture the pathways of more remote rural areas where neighbors and schools are more spread out. The classification measures the amount of rural settlement LAs contain, based on aggregate information on 2001 Census population statistics and settlement type for census Output Areas (OAs), the basic building blocks of UK Census geography. The six areas (Table 1) follow a continuum of urban to rural. Major Urban LAs are the most urban, situated in big cities likes London and Liverpool. Rural 80% LAs – the most rural – are found in sparse areas including North Cornwall and the Cotswolds. The Other Urban and the Significant Rural areas are the most ‘mixed’ with the latter areas being more rural. The child-level covariates were age in years at sweeps 2–4, and sex. Family-level covariates: We tested the association between family-level covariates that we expected to be linked to selection into areas and rural–urban area type, retaining only those that were significantly associated with area: maternal socio-economic status and academic qualifications, weighted household income (in quintiles), and parental involvement based on the frequency the mother reads to the child, all measured at the beginning of the observation window (i.e. around age 3). Socio-economic status was based on the 5-level National Statistics Socio-economic Classification (NS-SEC). Highest academic qualification was classified to six groups ranging from higher degree to no qualification. We measured achievement of schools attended by children at age 5 with Key Stage 1 (KS1) average point scores from the January 2006 school census. Key stages are stages of the English state education system. At the end of KS1 (age 7), children are assessed in English, Maths, and Science. These KS1 data are for state-maintained schools

only and therefore are missing for pupils attending fee-paying schools (around 7% of the sample). The KS1 data were banded into quintile groups based on all schools. The presence of high status adults in the area was measured with three variables using Office for National Statistics neighborhood statistics for each LA assessing: (1) Income deprivation based on the Indices of Multiple Deprivation (IMD) Income score (quintiles) (Noble et al., 2006), (2) Percentage of residents with a qualification equivalent to first degree (quintiles), and (3) Percentage of residents with higher socio-economic status, namely Higher/ Lower Managerial, Administrative or Professional Occupations (quintiles). Descriptives of the four dependent variables are shown in Table 2, broken down by rural–urban area. This shows the extent to which unadjusted levels of problem behavior vary by area type. Appendix A shows the full set of descriptive statistics for all variables by rural–urban area, revealing cross-area variation in family characteristics and parental behaviors that we would expect to be linked to problem behavior. On the whole, rural areas have residents with more favorable characteristics than urban areas do, with Rural 50% areas being the most advantaged and Large Urban areas the least. Compared to all MCS families, those in our analytic sample are slightly overrepresented in Large Urban and Significant Rural areas and slightly under-represented in Major Urban areas, but overall the distribution is comparable.

2.3. Analytic approach We conducted 2-level multilevel growth curve models (Snijders and Bosker, 1999) to model individual and average trajectories of children's problems. The fixed effects parameters identify the average effect on the dependent variable, while the random effects parameters identify the extent to which the variance is accounted for by variation between observations. The measurement occasion

Table 1 Rural–urban classification and % of MCS families in analytic sample by area type. Area type

Description

% MCS families

Major urban (e.g. LAs in London, Liverpool) Large urban (e.g. LAs in Sheffield, Bristol) Other urban (e.g. LAs in Canterbury, Norwich) Significant rural (e.g. LAs in Colchester, Harrogate) Rural 50% (e.g. LAs in Shropshire, Central Bedfordshire) Rural 80% (e.g. LAs in North Cornwall, the Cotswolds)

Either minimum of 100,000 people or minimum of 50% of total population is resident within a major urban area (i. e. area with at least 750,000 residents). Either minimum of 50,000 people or minimum of 50% of total population is resident within a large urban urea (i.e. area of 250,000–750,000 residents). Less than 26% of population lives in rural settlements (including larger market towns, i.e. urban areas of 10,000– 30,000 residents). More than 26% and less than 50% of their population lives in rural settlements and larger market towns.

35.3

At least 50% but less than 80% of population lives in rural settlements (including larger market towns).

11.6

At least 80% of population lives in rural settlements (including larger market towns).

16.0 13.4 14.3

9.5

Note: proportions are weighted. Families in our sample lived in a total of 200 LAs.

Table 2 Descriptive sample statistics. Behavioral problem

N

All areas (N ¼ 4843) M(SD)

MU (N ¼1574) M(CIs)

LU (N¼ 853) M(CIs)

OU (N ¼705) M(CIs)

SR (N ¼728) M(CIs)

R50 (N ¼ 512) M(CIs)

R80 (N ¼471) M(CIs)

Conduct problems Hyperactivity Peer problems Emotional symptoms

4812 4795 4802 4806

1.50(1.47) 3.23(2.33) 1.04(1.36) 1.30(1.51)

1.46(1.38,1.53) 3.23(3.12,3.35) 1.07(1.00,1.14) 1.27(1.19,1.35)

1.63(1.52,1.73) 3.39(3.22,3.55) 1.22(1.12,1.33) 1.46(1.35,1.58)

1.53(1.42,1.64) 3.37(3.20,3.54) 1.06(0.97,1.16) 1.27(1.16,1.37)

1.41(1.30,1.52) 3.25(3.09,3.41) 0.93(0.94,1.03) 1.24(1.14,1.33)

1.20(1.09,1.32) 2.89(2.70,3.08) 0.85(0.75,0.95) 1.23(1.14,1.33)

1.32(1.19,1.44) 3.09(2.88,3.29) 0.99(0.87,1.11) 1.38(1.24,1.51)

Note: CIs¼ 95% confidence intervals.

E. Midouhas, L. Platt / Health & Place 30 (2014) 226–233

(i.e. sweeps 2, 3 and 4) was at Level 1. At Level 2 was the child. Level 1 estimates the growth trajectory of each child's behaviors across time. Level 2 estimates individual variations in behaviors and trajectories between children. Linear and quadratic fixed terms for age in years (grand mean centered at 5.20 years) were included in all models. Additionally, a random linear slope was included for age to allow changes in difficulties to vary by child. The full sequence of models is in Table 3. The MCS stratified sampling design (Plewis, 2007) was accounted for by including the strata for England: England-Advantaged, England-Disadvantaged, and England-Ethnic. The inclusion of these variables will attenuate the rural–urban associations since they capture some characteristics associated with rural–urban differences. Model 1 investigated the average levels and growth of problems by regressing them on centered age in years and its square. Model 2 added the rural– urban indicator, specified to be related to the intercept and slopes (linear only) of problems to examine whether levels of problems at

229

around age 5 and the rate of change in problems over time shifted with rural–urban area. Originally we specified our rural–urban indicator to be related to the non-linear slope of problems. We dropped these terms from the models as they were not statistically significant. Model 3 added the child sex and family/parent Table 3 Model summary. Model

Variables

1 2 3 4 5 6 7

Ageþ age2 þ strata variables Model 1þ rural–urban areaþrural–urban area  age Model 2þ child sex þ parent/family factors Model 3þ school achievement Model 3þ LA income deprivation þLA income deprivation  age Model 3þ LA human capital þ LA human capital  age Model 3þ LA social class þLA social class  age

Table 4 Fixed and random effects estimates for Models 2–4: conduct problems (standard errors in parentheses). Parameters

Model 2 b(se)

Constant Stratum (Ref: England-advantaged) England-disadvantaged England-ethnic Age Age2 Rural–urban area (Ref: large urban) Major urban Other urban Significant rural Rural 50% Rural 80% Major urban  age Other urban  age Significant rural  age Rural 50%  age Rural 80%  age Girl

Model 3 b(se)

Model 4 b(se)

1.40(0.05)nnn

2.49(0.11)nnn

2.19(0.12)nnn

0.42(0.04)nnn 0.50(0.11)nnn  0.37(0.02)nnn 0.14(0.004)nnn

0.11(0.04)nn 0.12(0.11)  0.37(0.02)nnn 0.14(0.004)nnn

0.10(0.05)n 0.11(0.11)  0.36(0.02)nnn 0.14(0.004)nnn

 0.16(0.06)n  0.05(0.07)  0.14(0.07)n  0.24(0.08)nn  0.14(0.08) 0.01(0.02) 0.01(0.02) 0.05(0.02)n 0.06(0.03)n 0.03(0.03)

 0.08(0.06)  0.02(0.07)  0.12(0.07)  0.18(0.08)n  0.15(0.08) 0.01(0.02) 0.01(0.02) 0.05(0.02)n 0.06(0.03)n 0.03(0.03)  0.24(0.04)nnn

 0.04(0.06)  0.01(0.07)  0.02(0.07)  0.13(0.08)  0.08(0.08) 0.01(0.02)  0.002(0.03) 0.04(0.03) 0.05(0.03) 0.04(0.03)  0.22(0.04)nnn

 0.44(0.12)nnn  0.60(0.09)nnn  0.46(0.08)nnn  0.44(0.07)nnn  0.22(0.08)nnn  0.15(0.18)  0.17(0.02)nnn

 0.53(0.12)nnn  0.62(0.09)nnn  0.46(0.08)nnn  0.43(0.07)nnn  0.24(0.08)nn  0.15(0.18)  0.08(0.02)nnn

 0.13(0.05)nn  0.13(0.06)nn  0.07(0.09) 0.05(0.08) 0.15(0.02)nnn

 0.13(0.06)n  0.14(0.06)n  0.06(0.09) 0.09(0.08) 0.06(0.01)nnn  0.04(0.02)n

Maternal qualification (Ref: none) Higher degree First degree A levels or HE diploma GCSE a-c GCSE d-g Other qualification Household income Social class (Ref: routine) Professional/managerial Intermediate Small business or self-employed Lower supervisory or technical Less frequent reading to the child School-level achievement Random effects Level 2 (child) Intercept variance Slope variance Covariance

1.52(0.04) 0.09(0.01)  0.15(0.01)

1.31(0.04) 0.09(0.01)  0.14(0.01)

1.17(0.03) 0.09(0.01)  0.12(0.01)

Level 1 (occasion) Intercept variance

1.05(0.02)

1.05(0.02)

0.98(0.02)

2

Model 2¼ age þage þ strata variables þrural–urban areaþrural–urban area  age; Model 3 ¼Model 2þ child sexþ parent/family factors; Model 4¼ Model 3þ School achievement. n

po 0.05. p o0.01. p o 0.001.

nn

nnn

230

E. Midouhas, L. Platt / Health & Place 30 (2014) 226–233

covariates. If rural–urban differences remained, Models 4–7 introduced the potential mechanisms in turn. Model 4 added schoollevel achievement; and Models 5–7 entered presence of highstatus adults, measured by the three LA characteristics, specified to be related to the intercept and linear slope of problems. All analyses were conducted in Stata Version 12.

3. Results Conduct problems, hyperactivity and peer problems decreased at an average annual linear rate of 0.35, 0.15 and 0.08 on the SDQ scale from ages 3 to 7 (Model 1, all outcomes, not illustrated but available from authors on request). The positive age2 terms for conduct problems (b¼0.14), hyperactivity (b¼ 0.08), and peer problems (b¼0.05) demonstrated an additional slight upward curve between ages 5 and 7. Emotional symptoms, however, increased at a linear rate of 0.04 points annually, and the positive age2 term (b¼0.02) demonstrated an acceleration of problems. In relation to the random effects, for each behavioral problem there was substantial variation between children in their age 5 scores. There was also considerable within-child variation in scores between measurement points. However, the overall trajectories of behavioral problems varied much less between children. There were no differences in children's hyperactivity according to the rural–urban area where they lived (Model 2). These results remained after controlling for key child and family/parent covariates in Model 3, and are not further discussed. (Tables available from authors on request.) However, children in Major Urban areas, Significant Rural areas and Rural 50% areas had 0.16, 0.14, and 0.24 fewer points on the conduct scale, at central age (Model 3, Table 4). Children in Significant Rural and Rural 50% areas had a small linear increase in their problems over time (b¼ 0.05 and b¼0.06, respectively). The inclusion of child and family/parent background factors in Model 3, Table 4 explained only the effects of living in a Major Urban area and a Significant Rural area at central age. The predicted mean trajectory of conduct problems for a child living in each rural–urban area based on Model 3 is illustrated in Fig. 1. The trajectories are nearly parallel, starting out with more problems at age 3, making substantial improvements until age 6, and then slightly increasing in their problems between ages 6 and 7. At age 5, the child in a Rural 50% area has the lowest scores and the

child in the Large Urban area has the highest scores. Also, the child in the Significant Rural area slightly surpasses the child in the Large Urban area in the level of problems between ages 6 and 7. Adding a measure of school-level achievement fully attenuated the main effect of living in a Rural 50% area, and the effects of Significant Rural area  age, and Rural 50% area  age on conduct problems (Table 4, Model 4). Furthermore, greater school achievement was associated with fewer problems at age 5 (b¼  0.04). LA income deprivation (Model 5, not shown) also explained both the tiny effects of Significant Rural  age and Rural 50%  age, and LA social class (Model 6, not shown) fully attenuated the similarly small effect of Significant Rural area  age. However, LA human capital (Model 7, not shown) did not alter any remaining effects. Moreover, the main effects of these area composition factors were not significantly associated with conduct problems. As with conduct problems, following adjustments for selection, children in Rural 50% areas had fewer peer problems at central age (b¼  0.16). This difference was again explained by school achievement (b ¼ 0.04) (Model 4, not shown). Area composition factors did not similarly attenuate this effect (Models 5–7, not shown). Turning to emotional difficulties, children in Major Urban, Other Urban, Significant Rural and Rural 50% areas had significantly fewer emotional symptoms at age 5 (b¼  0.22, b¼  0.15, b¼  0.14, and b¼  0.15, respectively; Table 5, Model 2). Furthermore, living in a Significant Rural area was associated with a drop in 0.05 points annually; whereas living in a Rural 50% area was related to an annual increase of 0.05 points. Adjusting for selection (Model 3) explained the effect of living in a Rural 50% area at central age, but not the other area main effects, or the interactive effects of rural areas and child's age. In Model 4, school achievement explained the main effect of Significant Rural area, and each LA characteristic independently explained the small Rural 50%  age effect (b¼0.05; Models 5–7; not shown). However, the effects of a Major Urban and Other Urban area at central age, and the effect of a Significant Rural area on change in problems were unexplained by the two pathways in Models 4–7. (Tables for Models 5-7 available on request.) Fig. 2 plots the predicted rural–urban trajectories of emotional symptoms based on Model 3. By contrast with conduct problems, most of these trajectories appear to be linear, steadily increasing over time at varying rates. The children in the Major Urban and Other Urban areas follow a similar trajectory and have lower levels of emotional symptoms than the child in the Large Urban area, at any given timepoint, with the gap between these two trajectories

Fig. 1. Predicted mean trajectories of conduct problems by rural–urban area (Model 3). Notes: predictions are plotted for children whose mothers have a professional/ managerial occupation, a GCSE a–c qualification, and otherwise the reference group for each categorical variable, and at the mean of each continuous variable.

E. Midouhas, L. Platt / Health & Place 30 (2014) 226–233

231

Table 5 Fixed and random effects estimates for Models 2–4: emotional symptoms (standard error in parentheses). Parameter

Model 2 b(se)

Constant Stratum (Ref: England-advantaged) England-disadvantaged England-ethnic Age Age2 Rural–urban area (Ref: large urban) Major urban Other urban Significant rural Rural 50% Rural 80% Major urban  age Other urban  age Significant rural  age Rural 50%  age Rural 80%  age Girl

Model 3 b(se)

1.36(0.05)nnn

2.05(0.10)nnn

0.20(0.04)nnn 0.42(0.10)nnn 0.06(0.02)nnn 0.01(0.004)nn

0.02(0.04) 0.22(0.10)n 0.10(0.02)nnn 0.01(0.004)nn

 0.22(0.05)nnn  0.15(0.06)n  0.14(0.06)n  0.15(0.07)n  0.10(0.07)  0.03(0.02)  0.003(0.02)  0.05(0.02)n 0.05(0.02)n  0.01(0.03)

Maternal qualification (Ref: none) Higher degree First degree A levels or HE diploma GCSE a–c GCSE d–g Other qualification Household income Social class (Ref: routine occupation) Professional/managerial Intermediate Small business or self-employed Lower supervisory or technical Less frequent reading to the child School-level achievement

 0.18(0.05)nn  0.14(0.06)n  0.13(0.06)n  0.12(0.07)  0.11(0.07)  0.03(0.02)  0.004(0.02)  0.05(0.02)n 0.05(0.03n)  0.01(0.03) 0.04(0.03)

Model 4 b(se) 2.13(0.11)nnn 0.004(0.04) 0.22(0.10)n 0.05(0.02)nn 0.01(0.01)nn  0.17(0.05)nn  0.14(0.06)n  0.11(0.06)  0.09(0.07)  0.12(0.07)  0.02(0.02) 0.004(0.02)  0.05(0.02)n 0.05(0.03)n 0.01(0.03) 0.04(0.04)

 0.19(0.11)  0.36(0.08)nnn  0.30(0.07)nnn  0.30(0.06)nnn  0.13(0.07)  0.41(0.17)n  0.10(0.02)nnn

 0.14(0.11)  0.33(0.08)nnn  0.26(0.07)nnn  0.27(0.06)nnn  0.13(0.07)  0.34(0.18)  0.08(0.02)nnn

 0.18(0.05)nnn  0.19(0.05)nnn  0.20(0.08)n  0.10(0.08) 0.05(0.02)nn

 0.18(0.05)nn  0.19(0.06)nnn  0.19(0.09)n  0.08(0.08) 0.04(0.02)n  0.04(0.01)nn

Random effects Level 2 (child) Intercept variance Slope variance Covariance

1.09(0.03) 0.06(0.01) 0.13(0.01)

1.03(0.03) 0.06(0.01) 0.13(0.01)

1.03(0.03) 0.06(0.01) 0.13(0.01)

Level 1 (occasion) Intercept variance

1.16(0.02)

1.15(0.02)

1.16(0.03)

2

Model 2¼ age þage þ strata variables þrural–urban areaþrural–urban area  age; Model 3 ¼Model 2þ child sexþ parent/family factors; Model 4¼ Model 3þ School achievement. n

po 0.05. p o0.01. nnn p o 0.001. nn

and that of the child in the Large Urban area widening slightly over time. Emotional problems increase less for children in Significant Rural areas, following a U-shaped, non-linear trajectory.

4. Discussion This study set out to explore whether trajectories of children's emotional and behavioral problems differed depending on the conduct problems level of rurality/urbanity, even after accounting for families' selective sorting into particular areas. In unadjusted models, there were no rural–urban differences in child hyperactivity. Children's conduct and peer problems and emotional symptoms were, however, patterned by area rurality, even in models that accounted for selection. Living in a Rural 50% area at around age 5 was related to fewer conduct and peer problems. And children in Rural 50% and Significant Rural areas showed an increase – albeit

slight – in conduct problems each year compared to those in Large Urban areas. This may be due to a greater relative reduction in these problems for urban children who began with more problems around age 3 and therefore had more room to improve. Differences in emotional symptoms between Major Urban, Other Urban and Significant Rural areas compared to Large Urban areas were not explained by selection, and neither were the differential trajectories of those children living in Significant Rural and Rural 50% areas. Overall, it was clear that selection played a major role in accounting for urban–rural differences in behavior problems, and residual differences where they were found were relatively small. However, there did seem to be a relatively consistent pattern of modest rural– urban differences across three out of the four problem domains, in particular with more positive outcomes for those in Rural 50% and Significant Rural areas. These differences were themselves largely explicable in terms of the achievement levels of the primary schools attended by rural children. Although achievement is considered a

232

E. Midouhas, L. Platt / Health & Place 30 (2014) 226–233

Fig. 2. Predicted mean trajectories of emotional symptoms by rural–urban area (Model 3). Note: See note for Fig. 1.

proxy for ‘quality’ by many parents and policymakers, the achievement of pupils may also relate to socioeconomic composition of the school which has been theorized to influence children's well-being indirectly through peer group, instructional, organizational and/or leadership processes (Thrupp et al., 2002). Higher SES schools may have superior management, resources and other supports available to address children's difficulties before they get out of hand and interfere with teaching and learning. Our results add to the growing body of research pointing towards the role of schools in explaining why children's developmental outcomes differ by neighborhood (e.g. Heilmann et al., 2013). Hence, they align with theoretical perspectives highlighting the role of institutional resources and norms/collective efficacy (as proxied by school ‘quality’) in accounting for neighborhood effects (Leventhal and Brooks-Gunn, 2000). We should note, however, that not all children attended schools in their immediate neighborhoods, due to school choice policies in England. Therefore, school achievement may partly have captured selection rather than proved mediation, since perceived school ‘quality’ is a key reason that families are drawn to particular neighborhoods. An area's level of high status adults mediated the effects of children living in the less sparse rural areas on their trajectories of emotional and behavioral problems. However, these effects were small to begin with; this might indicate that they or their attenuation was due to random noise. Furthermore, the main effects of the LA composition factors were not significant, giving little support to their role as pathways from area to child outcome. Neither schools nor resident characteristics explained why children in Major Urban and Other Urban areas had fewer emotional symptoms at central age, or why living in a Significant Rural area was related to a reduction in emotional symptoms over time. This provides some prima facie evidence for the relevance of area type to children's emotional development. It is possible that unobserved characteristics of families associated with area selection could explain these remaining differences. Alternatively, they may be due to ‘parental mediation’ (Galster, 2012): parenting approaches (e.g. how much they monitor their child's behavior) may be influenced by the conditions of their neighborhood through, for example, the extent to which parents perceive threats in their community (Leventhal and Brooks-Gunn, 2000). Parents' hardships resulting from poor conditions in Large Urban areas or through collective ‘cultures’ of parenting may influence their parenting practices as well. Our results suggest that there is a need for the neighborhood effects literature to more explicitly examine such mechanisms (Small

and Feldman, 2012), and to take seriously rural–urban as well as within-urban factors contributing to children's wellbeing. Nevertheless, these findings should be interpreted with some caution. First, they are generalizable only to white majority children living continuously in their area types across early-mid childhood. Second, we assumed that area characteristics were fixed over time. Change over time in resident composition, employment opportunities, cultures of poverty, social capital, collective action, or local policies (Lupton and Power, 2004) might obscure our ability to assess rural–urban area differences. Third, we did not take into account potential spillover effects of surrounding areas, particularly for families living near boundaries. In addition, LAs are relatively large areas, and even for those not living near the borders, the overall composition of the LA may not fully reflect their ‘neighborhood’ experience. Despite these limitations we do provide indicative evidence on the ways by which rural–urban area may be linked to children's behavior. Our findings suggest that school quality linked to area differences, has an impact on the development of conduct and peer problems, problems that are by and large relational and thus more sensitive to school context. We additionally provide some suggestive evidence that the intrinsic characteristics of rural– urban area may matter for emotional symptom trajectories, which may be most susceptible to area environmental influences. Hence, our findings are preliminary; yet provide a basis for future lines of inquiry and the further illumination of rural–urban effects. Future research could valuably expand on these results by exploring the extent to which they persist across childhood and over more data points, and contribute to their theoretically informed elucidation. Acknowledgments The funding for this research was provided by a Bloomsbury Colleges Consortium studentship to the first author. We are grateful to Shirley Dex, John Shepherd, Kirstine Hansen, and Heather Joshi for their useful comments on this research. Special license data for area units were obtained through the UK Data Archive (reference 42291).

Appendix A. Supporting information Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.healthplace.2014. 09.003.

E. Midouhas, L. Platt / Health & Place 30 (2014) 226–233

References Browne, W., Goldstein, H., 2010. MCMC sampling for a multilevel model with nonindependent residuals within and between cluster units. J. Educ. Behav. Stat. 35, 353–373. http://dx.doi.org/10.3102/1076998609359788. Coleman, J., 1988. Social capital in the creation of human capital. Am. J. Sociol. 94, S95–S120. http://dx.doi.org/10.2307/2780243. Commission for Rural Communities, 2010. The state of the countryside update: Children and educational services. Commission for Rural Communities, Cheltenham. Duncan, G.J., Brooks-Gunn, J., Klebanov, P.K., 1994. Economic deprivation and early childhood development. Child Dev. 65, 296–318. http://dx.doi.org/10.1111/ j.1467-8624.1994.tb00752.x. Flouri, E., Mavroveli, S., Tzavidis, N., 2010. Modeling risks: effects of area deprivation, family socio-economic disadvantage, and adverse life events on young children's psychopathology. Soc. Psychiatry Psychiatric Epidemiol. 45, 611–619. http://dx.doi.org/10.1007/s00127-009-0101-x. Galster, G.C., 2012. The mechanism(s) of neighbourhood effects: theory, evidence, and policy implications. In: Van Ham, M., Manley, D., Bailey, N., Simpson, L., Maclennan, D. (Eds.), Neighbourhood Effects Research: New Perspectives. Springer, Netherlands, Nieuwerkerk, pp. 23–56. Gaviria, A., Raphael, S., 2001. School-based peer effects and juvenile behavior. Rev. Econ. Stat. 83, 257–268. http://dx.doi.org/10.1162/00346530151143798. Gibbons, S., Silva, O., 2008. Urban density and pupil attainment. J. Urban Econ. 63, 631. http://dx.doi.org/10.1016/j.jue.2007.04.006. Goodman, R., 1997. The Strengths and Difficulties Questionnaire: a research note. J. Child Psychol. Psychiatry 38, 581–586. http://dx.doi.org/10.1111/j.14697610.1997.tb01545.x. Heilmann, A., Kelly, Y., Stafford, M., Watt, R.G., 2013. OP57 The contribution of neighbourhoods and schools to cognitive test performance at age seven: Findings from the UK Millennium Cohort Study. J. Epidemiol. Commun. Health 67, A28. http://dx.doi.org/10.1136/jech-2013-203126.57. Jencks, C., Mayer, S.E., 1990. The social consequences of growing up in a poor neighbourhood. In: Lynn, L.E., McGeary, M.G.H. (Eds.), Inner-City Poverty in the United States. National Academy Press, Washington. Leventhal, T., Brooks-Gunn, J., 2000. The neighborhoods they live in: the effects of neighborhood residence upon child and adolescent outcomes. Psychol. Bull. 126, 309–337. http://dx.doi.org/10.1037/0033-2909.126.2.309. Lupton, R., 2003. ‘Neighbourhood Effects’: can we measure them and does it matter? (CASEPaper 73). Centre for Analysis of Social Exclusion, London. Lupton, R., Power, A., 2004. The growth and decline of cities and regions. CASE– Brookings Census Brief 1. CASE, London.

233

McCulloch, A., 2006. Variation in children's cognitive and behavioural adjustment between different types of place in the British National Child Development Study. Soc. Sci. Med. 62, 1865–1879. http://dx.doi.org/10.1016/j. socscimed.2005.08.048. Midouhas, E., Flouri, E., 2013. Rural/urban area differences in the cognitive abilities of primary school children in England. Popul. Space Place , http://dx.doi.org/ 10.1002/psp.1815 (advance online publication). National Centre for Social Research, 2009. Educational attainment in rural areas. National StatisticsFocus on Ethnicity and Identity. Office for National Statistics, London. Noble, M., Wright, G., Smith, G., Dibben, C., 2006. Measuring multiple deprivation at the small-area level. Environment and Planning A 38, 169–185. Owens A., 2010. Neighborhoods and schools as competing and reinforcing contexts for educational attainment, Sociol. Educ. 83 287–311, http://dx.doi.org/10.1177/ 0038040710383519. Pateman T., 2011, Rural and urban areas: Comparing lives using rural/urban classifications. Regional Trends 43, 11–86, http://dx.doi.org/10.1057/rt.2011.2. Plewis, I., 2007. Millennium Cohort Study First Survey: Technical Report on Sampling, 4th Ed. Centre for Longitudinal Studies, Institute of Education, London. Putnam R.D., Bowling alone: The collapse and revival of American community, 2006, Simon Schuster, New York. Rural Evidence Research CentreSocial and Economic Change and Diversity in Rural England. Department for Environment Food and Rural Affairs, London. Rural Evidence Research Centre, 2009. Defra classification of Local Authorities in England: An updated technical guide. Available at 〈http://www.rerc.ac.uk/ findings/documents_rural/LA_Class_Technical_Guide_April_2009.pdf〉. Sampson, R.J., Morenoff, J.D., Earls, F., 1999. Beyond social capital: spatial dynamics of collective efficacy for children. Am. Sociol. Rev. 64, 633–660. Sampson, R., Sharkey, R., Raudenbush, S., 2007. Durable effects of concentrated disadvantage on verbal ability among African–American children. Proc. Natl. Acad. Sci. 105, 845–852. Small, M.L., Feldman, J., 2012. Ethnographic evidence, heterogeneity, and neighbourhood effects after moving to opportunity. In: van Ham, M., Manley, D., Bailey, N., Simpson, L., Maclennan, D. (Eds.), Neighbourhood Effects Research: New Perspectives. Springer, Netherlands, pp. 57–77. Snijders, T.A.B., Bosker, R.J., 1999. Multilevel analysis: an introduction to basic and advanced multilevel modeling. Sage Publishers, London. Thrupp, M., Lauder, H., Robinson, T., 2002. School composition and peer effects. Int. J. Educ. Res. 37, 483–504. http://dx.doi.org/10.1016/S0883-0355(03)00016-8.