Spatial distribution and temporal trends in social fragmentation in England, 2001−2011: a national study

Objective Social fragmentation is commonly examined in epidemiological studies of mental illness as high levels of social fragmentation are often found in areas with high prevalence of mental illness. In this study, we examine spatial and temporal patterns of social fragmentation and its underlying indicators in England over time. Setting Data for social fragmentation and its underlying indicators were analysed over the decennial Censuses (2001–2011) at a small area geographical level (mean of 1500 people). Degrees of social fragmentation and temporal changes were spatially visualised for the whole of England and its 10 administrative regions. Spatial clustering was quantified using Moran’s I; changes in correlations over time were quantified using Spearman’s ranking correlation. Results Between 2001 and 2011, we observed a strong persistence for social fragmentation nationally (Spearman’s r=0.93). At the regional level, modest changes were observed over time, but marked increases were observed for two of the four social fragmentation underlying indicators, namely single people and those in private renting. Results supported our hypothesis of increasing spatial clustering over time. Moderate regional variability was observed in social fragmentation, its underlying indicators and their clustering over time. Conclusion Patterns of social fragmentation and its underlying indicators persisted in England which seem to be driven by the large increases in single people and those in private renting. Policies to improve social cohesion may have an impact on the lives of persons who experience mental illness. The spatial aspect of social fragmentation can inform the targeting of health and social care interventions, particularly in areas with strong social fragmentation clustering.

Objective: Social fragmentation is commonly examined in epidemiological studies of mental illness as high levels of social fragmentation are often found in areas with high prevalence of mental illness. In this study we examine spatial and temporal patterns of the index of social fragmentation and its subdomains in England over time.
Setting: Data for social fragmentation and its subdomains were analysed over the decennial Censuses (2001Censuses ( -2011) at a small-area geographical level (mean of 1500 people). Degrees of social fragmentation and temporal changes were spatially visualised for the whole of England and its 10 administrative regions. Spatial clustering was quantified using Moran's I; changes in correlations over time were quantified using Spearman's ranking correlation.
Results: Between 2001 and 2011 we observed a strong persistence for social fragmentation nationally (Spearman's r=0.94). At the regional level, modest changes were observed over time, but marked increases were observed for two of the four social fragmentation subdomains, namely single people and those in private renting. Results supported our hypothesis of increasing spatial clustering over time. Moderate regional variability was observed in social fragmentation, its subdomains and their clustering over time.

Strengths and Limitations of the study
This study utilises longitudinal data for the whole of England and is the first to describe spatial and temporal patterns of social fragmentation. Spatial aspects of social fragmentation can inform organisation of mental health services and social interventions that aim to enhance social support, particularly in areas where social fragmentation is high and is spatially clustered Over the last two decennial Censuses (2001 and 2011), the index of social fragmentation in England strongly persisted at a small-area geographical level while its clustering and its subdomains increased over time. Over time social fragmentation remained relatively stable, but we found marked increases in the percentage of single people and the percentage of those living in private rented accommodation.

INTRODUCTION
The impact of local area characteristics, such as material deprivation, on health is well established. 1 2 However, similar to socioeconomic position, the social networks and the amount of social support in a neighbourhood may also partially determine an individual's health. [3][4][5][6] In this context, an area-level characteristic that has been the focus of much investigation in neighbourhood health research is social fragmentation. The index of social fragmentation measures lack of social cohesion, and it is used to define areas with a breakdown or absence of social capital, and was originally created to measure the noneconomic deprivation aspects of areas. 7 Social fragmentation is distinct conceptually from deprivation, since fragmentation is a feature of household composition and demographic structure and is not intrinsically linked to socioeconomic position. 8 However, the two measures tend to be positively correlated as they are both concentrated in large towns and cities. 8 9 Thus, many urban localities are both deprived and socially fragmented, although the two phenomena do not always co-exist. Several studies suggests that fragmented communities provide less stable social institutions and social bonds, although maintaining stable institutions and bonds within the communities can contribute substantially to the creation of shared identities, persistence of social relations, promotion of healthy behaviours and good physical and mental health for the population. [10][11][12][13] Social fragmentation has received less attention from researchers than deprivation, despite the fact that it can aid the understanding of mechanisms via which the wider social environment influences mental health. 14 Research on adults has shown that social fragmentation is more closely related to mental health than physical health outcomes, independent of material deprivation and individual-level risk factors. 7 15 In the context of mental health, social fragmentation has been linked with suicide and non-fatal self-harm, 7 16-18 mental disorders, 4 11 19 psychiatric health service use, 20 first admission for psychosis, [19][20][21] psychological distress, 3 11 and schizophrenia. 14 22 Nonetheless, physical illness outcomes do not appear to be as clearly linked to social fragmentation. 17 23 24 In the UK, an index of social fragmentation was developed to measure neighbourhood-level conditions which affect lack of opportunities for social integration. 7 The index was originally based on indicators of population turnover, percentage of married people, percentage of the population in private renting and percentage of one person households using data from the UK Census. The rationale for the selection of the underlying variables used for the calculation of the index was that several studies, both at ecological and individual level, have found raised mental illness prevalence in highly urbanised areas with high population turnover, many non-family households, and reduced social cohesion. 25 However, social trends towards increasing cohabitation, 10 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   5 urban areas 23 , suggest the use of proportion of single, widowed and divorced people instead of married people in the calculation of the index.

a phenomenon that occurs most commonly in
Social fragmentation in a population is likely to have profound effects on its health and wellbeing, and yet it is incompletely examined and reported in the existing literature, and a number of important research questions remain unanswered. First, how strongly clustered is social fragmentation? Second, how strongly does social fragmentation cluster and change longitudinally at regional level? Third, does social fragmentation persist over time? Answering all three questions is important to inform the organisation of healthcare services and to target social and healthcare interventions, especially for mental health for which social fragmentation appears to be a salient risk factor. In this study, we longitudinally and spatially describe the properties of social fragmentation in England and within each of the 10 administrative English regions and make comparisons across them. Finally, we quantify the persistence of social fragmentation over the study period.

Data sources
To derive the index of social fragmentation we accessed decennial Census data from the Nomis website 26 on: (1) One person households, % of all Census households; (2) single people (including widowed and divorced people), % all Census population; (3) Population turnover (percentage of the population that moved from other area or within the country), % Census population (4) Households in private renting, % of all Census households. Our analysis was conducted at a small-area geographical level, namely the Lower Layer Super Output Area (LSOA). LSOAs are administrative units of geography with a mean population of 1,500 people. All data were available at the LSOA level except for the population turnover data which were available at the output area level (OA). OAs provide the finest spatial scale at which the Census is enumerated. We assigned the OA level data to the LSOA level using lookup tables provided by the Office for National Statistics (ONS).
The social fragmentation index for each LSOA was calculated by adding the unweighted z scores for each of the four characteristics into a composite score. We used two time points for the index of social fragmentation, i.e. 2001 and 2011 over the two decennial Censuses. Across LSOAs the scores for the index ranged from -6.42 to 27.57 for 2001 and from -7.09 to 25.79 for 2011, with the upper boundaries indicating the most socially fragmented areas and the lower boundaries the most socially cohesive areas. Since the data used to derive the index in 2001 were reported using 2001 LSOAs, we used a weighted means algorithm to assign them to 2011 LSOAs using mid-year population estimates that were obtained from ONS. 27

Analyses
Our outcome of interest was the index of social fragmentation and its four subdomains measured separately at the 2001 and 2011 national Censuses. We visualised temporal changes in social fragmentation and its subdomains with the use of spatial maps for all of England and each of the 10 English regions (North East, North West, Yorkshire & the Humber, East Midlands, West Midlands, East of England, London, South East Coast, South Central and South West). To assess whether levels of spatial autocorrelation for social fragmentation changed over time we compared the values of Moran's I 28 in the two time points for England and each English region to allow within-England comparisons. The measure can identify the presence or absence of spatial clusters while accounting for the multi-dimensional and multi-directional nature of spatial autocorrelation. Under a random spatial pattern a higher than expected value of Moran's I for social fragmentation would indicate that areas with high levels of social fragmentation are clustered and also that LSOAs with high social fragmentation are bordered with LSOAs with similarly high levels of social fragmentation. We also quantified the persistency of social fragmentation over the two time points by calculating Spearman's correlation between 2001 and 2011. To visualise and compare temporal changes in the mean social fragmentation levels across the 10 English regions, we plotted population-weighted box plots for the main outcome and its subdomains. Finally, we used population-weighted box plots to visualise and compare temporal changes in the social fragmentation levels between the 10 English regions. Analyses were performed with Stata v.14.1 and R v.3.3.1. Due to the size of the dataset, effectively the whole of England, statistical significance is irrelevant; observed associations of minimal strength would be statistically significant, and thus we focussed on effect sizes wherever possible.

RESULTS
We found strong correlation for social fragmentation at the two time points (rho=0.93), using Spearman's ranking correlation to account for the relative nature of the measure. We plotted spatial maps of social fragmentation for the whole of England and for each region for the two time points analysed. Maps of social fragmentation in 2001 and 2011 and their respective changes for the whole of England and London are presented in Figure 1 and Figure 2 respectively, whilst social fragmentation and its subdomains maps over time for each region are presented in supplementary Appendices 2-6. Temporal changes in social fragmentation for England and for all English regions are presented in Figure A in the online appendix. We identified a pattern of stability with modest increases over time for regions in the North, and small decreases in social fragmentation for regions in the South. However,  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 16.2% indicating a 62% increase. The largest increases in numbers of single people were found in the North East, North West and West Midlands while London had the highest levels of single people. Moreover, the largest increases for private renting were found in the North East, West Midlands and East Midlands while London had again the largest numbers of people living in private renting. For the other two social fragmentation subdomains, we found that the average number of one person households both nationally and regionally remained relatively stable. We also observed a modest decrease, both nationally and regionally (with the exception of London), in our measure of population turnover which reflects the levels of migration inside and outside areas. Finally, we observed great variability in levels of social fragmentation and its subdomains within regions at both time points. High levels of social fragmentation were concentrated in London, the North West and the South West. East England had the lowest levels of social fragmentation followed by West Midlands. There was a clear distinction between rural and urban areas, with the latter having far greater levels of social fragmentation.
Across the whole of England, Moran's I for social fragmentation was very low at 0.0449 in 2001 (95% CI 0.0447 to 0.0451) but by 2011 it had increased to 0.0794 (95% CI 0.0792 to 0.0797), indicating a small increase in spatial autocorrelation and clustering ( Figure 4). We also observed marked regional variability in social fragmentation spatial autocorrelation levels over time ( Figure 4). In both time points, the West Midlands had the lowest spatial autocorrelation in social fragmentation, followed by the East of England, while the South East and South West had the highest levels of spatial autocorrelation at both time points. In contrast to increases in spatial autocorrelation across the country between 2001 and 2011, the South East, North East and Yorkshire & the Humber had decreasing levels of spatial autocorrelation over the two time points.
Spatial clustering also appears to have increased over time for all fragmentation subdomains nationally, although its levels remained low ( Figures A1 and A2 in supplementary Appendix). For one person households we found very low regional clustering and small regional variation. For single persons and private renting regional variation in clustering was moderate and their levels were also low with the exception of the South East and South West. Finally, for population turnover we observed the greatest increases of  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y 8 spatial autocorrelation both nationally and regionally but levels of regional clustering were generally low with the exception of South West whilst regional variability was moderate.

DISCUSSION
Between 2001 and 2011, we found evidence of strong persistence of social fragmentation in England at small-area level. Regional changes over time were modest for social fragmentation, which is not surprising considering the large population denominators and the relative nature of the measure. Similarly, we observed modest increases over time for one person households and population turnover. However, changes over time for single persons and private renting were much greater, with substantial increases in both subdomains across regions, and these specific influences appear to have largely driven the heightened levels of social fragmentation overall. Furthermore, our findings provide evidence of increasing clustering for social fragmentation and its subdomains which, in conjunction with the regional variation observed between the two time points, indicates that areas in the South of England (including London) appear much more clustered than the North. Finally, we observed increased clustering for all subdomains of social fragmentation over the two time points.

Strengths and limitations of the study
The main strength of this study lies in its large longitudinal nature with data for the whole of England (53 million people in 2011), which we utilised to investigate spatial and longitudinal patterns of social fragmentation. To our knowledge, it is the first study of its kind to provide an insight into the temporal persistence of social fragmentation and its clustering.
However, some limitations exist. First, the extent to which the index captures social fragmentation may vary across the country as the demographic factors that we used may not in themselves be valid indicators of social fragmentation, or they may adequately measure social fragmentation in some areas but not in others. 21 However, the measure has been reported widely in the literature and on this basis social fragmentation has been identified as an important predictor of mental health outcomes such as suicide. Second, Census questions do not usually directly elicit the presence or quality of neighbourhood social ties or institutions, or people's relationship to their neighbourhood. Therefore, we might have failed to capture people's perception of social fragmentation through the Census data. Nevertheless, we used a measure capturing compositional aspects of neighbourhood which appear to be related to the capacity of a neighbourhood to act collectively, and is also important for the construction and maintenance of social ties and institutions. Third, whilst Census data is recognised to be of a high standard, some unobserved heterogeneity persists. 29 For example, LSOAs are defined from administrative  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   9 information and this may disguise local effects as contextual characteristics within administrative neighbourhoods are not homogeneous across LSOAs. 30 An alternative approach would require the use of non-administrative data to define levels, but the Census is currently the only source from which the data on the fragmentation subdomains is available. Finally, there were boundary changes following the 2011 Census that may have affected our findings, 31 but only 2.5% of LSOAs were affected and we developed an algorithm to make reasonable population weighted-based estimates for these localities, depending on whether they merged, split or some other change occurred.

Interpretation of findings
Our results suggest that urbanisation drives the high persistence of social fragmentation in England as we found that levels of social fragmentation and its subdomains were consistently higher in urban areas. This may be because urban areas are chosen by diverse social and cultural groups and living in these areas may lead to isolated spaces, lifestyles and problems of social integration. Moreover, even though the overall index of social fragmentation changed a little, the subdomains that drive the persistency of social fragmentation appear to be single people and households in private renting, a characteristic of urban environments, 32 as these were the subdomains with the largest increases ( Figures  3A and 3B).
In absolute terms, private renting was considerably increased in all regions and most notably London; a finding that pertains to unaffordable housing for first-time home buyers and less availability of social housing as a result of the 2008 financial crisis. 33 Similarly, all regions had higher levels of single people in 2011, indicating that numbers of young professionals, students and divorces have increased substantially over the decade. For the other two subdomains we did not observe any marked changes across regions, although levels of population turnover appeared to have decreased between the decennial Censuses. The increases we observed in the proportion of single people and those in private renting warrant implications for the prevalence of mental illness especially in urban centres. The private house market is characterised by poor housing conditions as opposed to social houses or owner occupation and this can have detrimental effects on mental health. 34 Furthermore, it is suggested that married individuals may have better mental health outcomes than single individuals. 35 We believe that these changes are certainly of concern as they may highlight the need for more effective social interventions to target socially fragmented areas in England.
Spatial clustering of neighbourhood social fragmentation appears to have increased over time in England while we observed moderate regional variations in levels of clustering despite the small changes in levels of social fragmentation across regions. Increased levels of spatial clustering for social fragmentation across and within regions can have implications  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   10 for the planning and organisation of health services, as this would imply needs that are not uniformly distributed across a region with the presence of hot spots of high levels of mental illness prevalence, especially in urban areas. To address such spatial health inequalities, services may need to be redesigned locally 36 while infrastructure and spending for mental health needs to be weighted towards social fragmentation hot spots rather than uniformly distributed.

6
Region-level characteristics of the index of social fragmentation at both time points are 7 provided in Table 1. We found strong correlation for social fragmentation at the two time 8 points (rho=0.93), using Spearman's ranking correlation to account for the relative nature of 9 the measure. We plotted spatial maps of social fragmentation for the whole of England and for   15 appendix. We identified a pattern of stability with modest increases over time for regions in the 16 North, and small decreases in social fragmentation for regions in the South. However, temporal 17 changes of English regions for two of the four social fragmentation underlying indicators were 18 much larger and they are presented in figures 3A and 3B. 19 20 Large changes in absolute levels of single person households and private renting were found in 21 all regions. More specifically, in 2001 the average number of single persons of the total 22 population was 38.4% while in 2011 the figure was 52.6% indicating a 31% increase. Similarly, in 23 2001 the average number of people in private renting was 8.5% of all English households while 24 in 2011 this figure increased to 16.2% indicating a 62% increase. The largest increases in 25 numbers of single people were found in the North East, North West and West Midlands while 26 London had the highest levels of single people. Moreover, the largest increases for private 27 renting were found in the North East, West Midlands and East Midlands while London had 28 again the largest numbers of people living in private renting. For the other two social 29 fragmentation underlying indicators, we found that the average number of one person 30 households both nationally and regionally remained relatively stable. We also observed a 31 modest decrease, both nationally and regionally (with the exception of London), in our measure 32 of population turnover which reflects the levels of migration inside and outside areas. Finally, 33 we observed great variability in levels of social fragmentation and its underlying indicators 34 within regions at both time points. High levels of social fragmentation were concentrated in 35 London, the North West and the South West. East England had the lowest levels of social 36 fragmentation followed by West Midlands. There was a clear distinction between rural and 37 urban areas, with the latter having far greater levels of social fragmentation. Across the whole of England, Moran's I for social fragmentation was very low at 0.0449 in 2001 3 (95% CI 0.0447 to 0.0451) but by 2011 it had increased to 0.0794 (95% CI 0.0792 to 0.0797), 4 indicating a small increase in spatial autocorrelation and clustering ( Figure 4). We also observed 5 marked regional variability in social fragmentation spatial autocorrelation levels over time 6 ( Figure 4). In both time points, the West Midlands had the lowest spatial autocorrelation in 7 social fragmentation, followed by the East of England, while the South East and South West had 8 the highest levels of spatial autocorrelation at both time points. In contrast to increases in 9 spatial autocorrelation across the country between 2001 and 2011, the South East, North East 10 and Yorkshire & the Humber had decreasing levels of spatial autocorrelation over the two time 11 points. 12 13 Spatial clustering also appears to have increased over time for all fragmentation underlying 14 indicators nationally, although its levels remained low (Figures A1 and A2 in supplementary 15 Appendix). For one person households we found very low regional clustering and small regional 16 variation. For single persons and private renting regional variation in clustering was moderate 17 and their levels were also low with the exception of the South East and South West. Finally, for 18 population turnover we observed the greatest increases of spatial autocorrelation both 19 nationally and regionally but levels of regional clustering were generally low with the exception 20 of South West whilst regional variability was moderate. We provide graphs and discuss the 21 results from the LISA analysis for local spatial autocorrelation in the supplementary material 22 ( Figures B1 and B2). 23 25 26 Between 2001 and 2011, we found evidence of strong persistence of social fragmentation in 27 England at small-area level. Regional changes over time were modest for social fragmentation, 28 which is not surprising considering the large population denominators and the relative nature 29 of the measure. Similarly, we observed modest increases over time for one person households 30 and population turnover. However, changes over time for single persons and private renting 31 were much greater, with substantial increases in both underlying indicators across regions, and 32 these specific influences appear to have largely driven the heightened levels of social 33 fragmentation overall. Furthermore, our findings provide evidence of increasing clustering for 34 social fragmentation and its underlying indicators which, in conjunction with the regional 35 variation observed between the two time points, indicates that areas in the South of England 36 (including London) appear much more clustered than the North. Finally, we observed increased 37 clustering for all underlying indicators of social fragmentation over the two time points.  3 4 The main strength of this study lies in its large dataset for the whole of England (53 million 5 people in 2011) over two time points across 32,844 LSOAs, which we utilised to investigate 6 spatial and temporal patterns of social fragmentation. To our knowledge, it is the first study of 7 its kind to provide an insight into the temporal persistence of social fragmentation and its 8 clustering. 9 10 However, some limitations exist. First, the extent to which the index captures social 11 fragmentation may vary across the country as the demographic factors that we used may not in 12 themselves be valid indicators of social fragmentation, or they may adequately measure social 13 fragmentation in some areas but not in others. 22 However, the measure has been reported 14 widely in the literature and on this basis social fragmentation has been identified as an 15 important predictor of mental health outcomes such as suicide. Second, Census questions do 16 not usually directly elicit the presence or quality of neighbourhood social ties or institutions, or 17 people's relationship to their neighbourhood. Therefore, we might have failed to capture 18 people's perception of social fragmentation through the Census data. Nevertheless, we used a 19 measure capturing compositional aspects of neighbourhood which appear to be related to the 20 capacity of a neighbourhood to act collectively, and is also important for the construction and 21 maintenance of social ties and institutions. Third, whilst Census data is recognised to be of a 22 high standard, some unobserved heterogeneity persists. 33 For example, LSOAs are defined 23 from administrative information and this may disguise local effects as contextual characteristics 24 within administrative neighbourhoods are not homogeneous across LSOAs. 34 An alternative 25 approach would require the use of non-administrative data to define levels, but the Census is 26 currently the only source from which the data on the fragmentation underlying indicators is 27 available. Fourth, for the underlying indicator 'single people' we used data on legal marital 28 status in 2001 and 2011 and registered civil partnership status in 2011. This has implications for 29 the total numbers of single people, as those who cohabit will be counted as 'non-married' in 30 the calculation of the indicator. Fifth, there were boundary changes following the 2011 Census 31 that may have affected our findings, 35 but only 2.5% of LSOAs were affected and we developed 32 an algorithm to make reasonable population weighted-based estimates for these localities, 33 depending on whether they merged, split or some other change occurred. Finally, the index of 34 social fragmentation may serve as a suitable tool to inform allocation of mental health 35 resources in adults, although recent evidence suggests that social fragmentation may not have 36 an effect on children's mental health. 36 37  2 3 A previous study showed that between 1971 and 2001, levels of social fragmentation increased 4 steadily with large increases between 1981 and 1991 and modest increases between 1991 and 5 2001. Our results indicate that this trend was reversed in 2011 as we found a small decrease in 6 the overall index of social fragmentation between 2001 and 2011. 37 Furthermore, our results 7 suggest that urbanisation drives the high persistence of social fragmentation in England as we 8 found that levels of social fragmentation and its underlying indicators were consistently higher 9 in urban areas. This may be because urban areas are chosen by diverse social and cultural 10 groups and living in these areas may lead to isolated spaces, lifestyles and problems of social 11 integration. Furthermore, we found that the underlying indicators that drive the persistency of 12 social fragmentation appear to be single people and households in private renting, a 13 characteristic of urban environments, 38 as these were the underlying indicators with the 14 largest increases ( Figures 3A and 3B). 15 In absolute terms, private renting was considerably increased in all regions and most notably 16 London; a finding that pertains to unaffordable housing for first-time home buyers and less 17 availability of social housing as a result of the 2008 financial crisis. 39 Similarly, all regions had 18 higher levels of single people in 2011, indicating that numbers of young professionals, students 19 and divorces have increased substantially over the decade. For the other two underlying 20 indicators we did not observe any marked changes across regions, although levels of population 21 turnover appeared to have decreased between the decennial Censuses. The increases we 22 observed in the proportion of single people and those in private renting warrant implications 23 for the prevalence of mental illness especially in urban centres. The private house market is 24 characterised by poor housing conditions as opposed to social houses or owner occupation and 25 this can have detrimental effects on mental health. 40 Furthermore, it is suggested that married 26 individuals may have better mental health outcomes than single individuals. 41 We believe that 27 these changes are certainly of concern as they may highlight the need for more effective social 28 interventions to target socially fragmented areas in England. 29 Spatial clustering of neighbourhood social fragmentation appears to have increased over time 30 in England while we observed moderate regional variations in levels of clustering despite the 31 small changes in levels of social fragmentation across regions. Increased levels of spatial 32 clustering for social fragmentation across and within regions can have implications for the 33 planning and organisation of health services, as this would imply needs that are not uniformly 34 distributed across a region with the presence of hot spots of high levels of mental illness 35 prevalence, especially in urban areas. To address such spatial health inequalities, services may 36 need to be redesigned locally 42 while infrastructure and spending for mental health needs to 37 be weighted towards social fragmentation hot spots rather than uniformly distributed.  3 4 In the absence of individual-level data, area-level indicators are often the only source of 5 information to investigate how factors associated with mental illness may contribute to 6 population health, albeit when individual-level data are available and are fitted as covariates 7 these area-level associations may be attenuated substantially. Even though, social 8 fragmentation is increasingly considered a determinant of mental illness as more 9 epidemiological research emerges that investigates contextual effects on health. To improve 10 mental health in the population, policymakers need to address both relational and structural 11 issues, perhaps by investing in reorganisation of mental healthcare. Solving the problems of 12 cohesion and integration needs to be seen as a collective responsibility across central and local 13 government and all the local agencies. Policies needs to address social and educational 14 inequalities by strengthening the economic well-being of areas by lowering local 15 unemployment rates, encouraging populations to bring social support in the communities and 16 encouraging education of the population, which are key strategies for healthy communities. 17 Communities need to work towards tightening social networks and social bonds to help with 18 mental illness prevalence and suicide prevention, for example by promoting participation in 19 charity, voluntary or community groups and civic engagement. The spatial aspect of social 20 fragmentation is often overlooked, but it can provide vital information for the effective 21 organisation of health services and targeting of health or social interventions. The nature of 22 fragmentation and the mechanisms underlying its association with mental illness are of 23 increasing interest as the population gets older and the health care costs associated with 24 mental illness escalate in industrialised countries. 25  We would like to thank the Office of National Statistics for the wealth of information they have 3 collected and systematically organised, which made this study possible. 4 5 Contributors 6 EK, CG, DA and LM designed the study. CG extracted the data from all sources, performed the 7 analyses and drafted the manuscript. EK, RG, DA, LM, NK critically revised the manuscript. CG is 8 the guarantor of this work and, as such, had full access to all the data in the study and takes 9 responsibility for the integrity of the data and the accuracy of the data analysis.  Competing interests 20 None declared. 21 22 Provenance and peer review 23 Not commissioned; externally peer reviewed 24 25 Data sharing agreement 26 The data used in this study are freely available and the authors are happy to share in an 27 organised and cleaned final dataset. 28 29 Ethical approval and Patient and Public Involvement 30 Not applicable. The study uses freely reported data aggregates at low geography levels. No 31 patients or members of the public were involved in the study. 32 33 Open access 34 This is an Open Access article distributed in accordance with the terms of the Creative 35 Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and 36 build upon this work, for commercial use, provided the original work is properly cited. See: 37 http://creativecommons.org/licenses/by/4.0/ © Article author(s) (or their employer(s) unless         1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  One person households % E n g l a n d E a s t E n g l a n d E a s t M i d l a n d s

Spatial Maps
Digital vector boundaries for the 2011 LSOAs, generalised to 20 metres and clipped to the coastline to reduce size and improve visualisation, were obtained from the ONS open geography portal (2) . The vector boundaries were inputted in the Stata shp2dta command to calculate the centroid for each LSOA in the British National Grid format (3). These were then converted from British National Grid easting and northing to longitude and latitude in degrees (4).  Figure A2: Spatial autocorrelation with Moran's I for private renting (top) and population turnover (bottom), by region and over time       1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  illness as high levels of social fragmentation are often found in areas with high prevalence of mental 4 illness. In this study we examine spatial and temporal patterns of the index of social fragmentation 5 and its underlying indicators in England over time. nationally (Spearman's r=0.94). At the regional level, modest changes were observed over time, but 15 marked increases were observed for two of the four social fragmentation underlying indicators, 16 namely single people and those in private renting. Results supported our hypothesis of increasing 17 spatial clustering over time. Moderate regional variability was observed in social fragmentation, its 18 underlying indicators and their clustering over time. 19 20 Conclusion: Patterns of social fragmentation and its underlying indicators persisted in England 21 which seem to be driven by the large increases in single people and those in private renting. Policies 22 to improve social cohesion may have an impact the lives of persons who experience mental illness. 23 The spatial aspect of social fragmentation can inform the targeting of health and social care 24 interventions, particularly in areas with strong social fragmentation clustering. Strengths and Limitations of the study  This study utilises data over two time points for the whole of England and is the first to describe spatial and temporal patterns of social fragmentation.  Spatial aspects of social fragmentation can inform organisation of mental health services and social interventions that aim to enhance social support, particularly in areas where social fragmentation is high and is spatially clustered  Over the last two decennial Censuses (2001 and 2011), the index of social fragmentation in England strongly persisted at a small-area geographical level while its clustering and its underlying indicators increased over time.  Over time social fragmentation remained relatively stable, but we found marked increases in the percentage of single people and the percentage of those living in private rented accommodation.  Even though census data are considered of high quality, people's perception of social fragmentation may not be fully captured in the census while demographic factors may capture social fragmentation in some areas but not in others.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  The impact of local area characteristics, such as material deprivation, on health is well 4 established. 1 2 However, similar to socioeconomic position, the social networks and the 5 amount of social support in a neighbourhood may also partially determine an individual's 6 health. [3][4][5][6] In this context, an area-level characteristic that has been the focus of much 7 investigation in neighbourhood health research is social fragmentation. The index of social 8 fragmentation measures lack of social cohesion, and it is used to define areas with a 9 breakdown or absence of social capital, and was originally created to measure the non-10 economic deprivation aspects of areas. 7 Social fragmentation is usually measured using 11 census data which offer a great potential for the design and implementation of mental 12 health strategies when individual level data are not available. 13 14 Social fragmentation is distinct conceptually from deprivation, since fragmentation is a 15 feature of household composition and demographic structure and is not intrinsically linked 16 to socioeconomic position. 8 However, the two measures tend to be positively correlated as 17 they are both concentrated in large towns and cities. 8 9 Thus, many urban localities are both 18 deprived and socially fragmented, although the two phenomena do not always co-exist. 19 Several studies suggests that fragmented communities provide less stable social institutions 20 and social bonds, although maintaining stable institutions and bonds within the 21 communities can contribute substantially to the creation of shared identities, persistence of 22 social relations, promotion of healthy behaviours and good physical and mental health for 23 the population. 10-14 24 25 Social fragmentation has received less attention from researchers than deprivation, despite 26 the fact that it can aid the understanding of mechanisms via which the wider social 27 environment influences mental health. 15 Research on adults has shown that social 28 fragmentation is more closely related to mental health than physical health outcomes, 29 independent of material deprivation and individual-level risk factors. 7 16 In the context of 30 mental health, social fragmentation has been linked with suicide and non-fatal self-harm, 7 31 17-19 mental disorders, 4 11 20 psychiatric health service use, 21 first admission for psychosis, 20-32 22 psychological distress, 3 11 and schizophrenia. 15 23 Nonetheless, physical illness outcomes 33 do not appear to be as clearly linked to social fragmentation. 18 24 25 34 35 In the UK, an index of social fragmentation was developed to measure neighbourhood-level 36 conditions which affect lack of opportunities for social integration. 7 The index was originally 37 based on indicators of population turnover, percentage of married people, percentage of 38 the population in private renting and percentage of one person households using data from 39 the UK Census. The rationale for the selection of the underlying variables used for the 40 calculation of the index was that several studies, both at ecological and individual level, 41 have found raised mental illness prevalence in highly urbanised areas with high population turnover, many non-family households, and reduced social cohesion. 26 However, social 2 trends towards increasing cohabitation 10 27 suggest the use of proportion of single, 3 widowed and divorced people instead of married people in the calculation of the index. 4 5 Social fragmentation in a population is likely to have profound effects on its health and 6 wellbeing, and yet it is incompletely examined and reported in the existing literature, and a 7 number of important research questions remain unanswered. First, how strongly clustered 8 is social fragmentation? Second, how strongly does social fragmentation cluster and change 9 at regional level over two time points? Third, does social fragmentation persist over time? 10 Answering all three questions is important to inform the organisation of healthcare services 11 and to target social and healthcare interventions, especially for mental health for which 12 social fragmentation appears to be a salient risk factor. In this study, we temporally and 13 spatially describe the properties of social fragmentation in England and within each of the 14 10 administrative English regions and make comparisons across them. Finally, we quantify 15 the persistence of social fragmentation over the study period. 16

19
Data sources 20 21 To derive the index of social fragmentation we accessed decennial Census data from the 22 Nomis website 28 on: (1) One person households, % of all Census households; (2) single 23 people (including widowed, divorced and separated people), % all usual residents aged 16 24 or over; (3) Population turnover (percentage of the population that moved into the area 25 from within the UK and from outside the UK), % Census population in the year preceding the 26 census (4) Households in private renting, % of all Census households. Our analysis was 27 conducted at a small-area geographical level, namely the Lower Layer Super Output Area 28 (LSOA). LSOAs are geographical areas with a mean population of 1,500 people. All data were 29 available at the LSOA level except for the population turnover data which were available at 30 the output area level (OA). OAs provide the finest spatial scale at which the Census is 31 enumerated. We assigned the OA level data to the LSOA level using lookup tables provided 32 by the Office for National Statistics (ONS). More details on the data used to derive each 33 social fragmentation underlying indicator are provided in the Supplementary appendix. 34 35 The social fragmentation index for each LSOA was calculated by adding the unweighted  areas. Since the data used to derive the index in 2001 were reported using 2001 LSOAs, we 2 used a weighted means algorithm to assign the data for each underlying indicator to 2011 3 LSOAs using mid-year population estimates that were obtained from ONS. 29  Midlands, East of England, London, South East Coast, South Central and South West), which 17 were the highest level of organisation available in NHS England during the study period. To 18 assess whether levels of spatial autocorrelation for social fragmentation changed over time 19 we compared the values of Moran's I 31 in the two time points for England and each English 20 region to allow within-England comparisons. The measure can identify the presence or 21 absence of spatial clusters while accounting for the multi-dimensional and multi-directional 22 nature of spatial autocorrelation. Under a random spatial pattern a higher than expected 23 value of Moran's I for social fragmentation would indicate that areas with high levels of 24 social fragmentation are clustered and also that LSOAs with high social fragmentation are 25 bordered with LSOAs with similarly high levels of social fragmentation. We also investigated 26 local indicators for spatial autocorrelation (LISA) 32 to identify local patterns of spatial 27 associations. We also quantified the persistency of social fragmentation over the two time 28 points by calculating Spearman's correlation between 2001 and 2011. To visualise and 29 compare temporal changes in the mean social fragmentation levels across the 10 English 30 regions, we plotted population-weighted box plots for the main outcome and its underlying 31 indicators. Finally, we used population-weighted box plots to visualise and compare 32 temporal changes in the social fragmentation levels between the 10 English regions. 33 Analyses were performed with Stata v.14.1 and R v.3.3.1. Due to the size of the dataset, 34 effectively the whole of England, statistical significance is irrelevant; observed associations 35 of minimal strength would be statistically significant, and thus we focussed on effect sizes 36 wherever possible. 37 Table 1. We found strong correlation for social fragmentation at the two time 3 points (rho=0.93), using Spearman's ranking correlation to account for the relative nature of 4 the measure. We plotted spatial maps of social fragmentation for the whole of England and 5 for each region for the two time points analysed. We created maps of social fragmentation 6 from a pooled sample in 2001 and 2011 for the whole of England and London in order to 7 facilitate comparisons between areas over time and we present those in Figure 1 and Figure   8 2 respectively. Furthermore, we present maps of social fragmentation and its underlying 9 indicators at each time point for every region in supplementary Appendix 2. Temporal 10 changes in social fragmentation for England and for all English regions are presented in 11 Figure A in the online appendix. We identified a pattern of stability with modest increases 12 over time for regions in the North, and small decreases in social fragmentation for regions in 13 the South. However, temporal changes of English regions for two of the four social 14 fragmentation underlying indicators were much larger and they are presented in Figure 3.  27 We also observed a modest decrease, both nationally and regionally (with the exception of 28 London), in our measure of population turnover which reflects the levels of migration inside 29 and outside areas. Finally, we observed great variability in levels of social fragmentation and 30 its underlying indicators within regions at both time points. High levels of social 31 fragmentation were concentrated in London, the North West and the South West. East 32 England had the lowest levels of social fragmentation followed by West Midlands. There 33 was a clear distinction between rural and urban areas, with the latter having far greater 34 levels of social fragmentation. 35 36 Across the whole of England, Moran's I for social fragmentation was very low at 0.0449 in 37 2001 (95% CI 0.0447 to 0.0451) but by 2011 it had increased to 0.0794 (95% CI 0.0792 to 38 0.0797), indicating a small increase in spatial autocorrelation and clustering (Figure 4). We 39 also observed marked regional variability in social fragmentation spatial autocorrelation 40 levels over time (Figure 4). In both time points, the West Midlands had the lowest spatial 41 autocorrelation in social fragmentation, followed by the East of England, while the South East and South West had the highest levels of spatial autocorrelation at both time points. In 2 contrast to increases in spatial autocorrelation across the country between 2001 and 2011, 3 the South East, North East and Yorkshire & the Humber had decreasing levels of spatial 4 autocorrelation over the two time points. 5 6 Spatial clustering also appears to have increased over time for all fragmentation underlying 7 indicators nationally, although its levels remained low ( Figures A1 and A2 in supplementary   8 Appendix). For one person households we found very low regional clustering and small 9 regional variation. For single persons and private renting regional variation in clustering was 10 moderate and their levels were also low with the exception of the South East and South 11 West. Finally, for population turnover we observed the greatest increases of spatial 12 autocorrelation both nationally and regionally but levels of regional clustering were 13 generally low with the exception of South West whilst regional variability was moderate. We 14 provide graphs and discuss the results from the LISA analysis for local spatial autocorrelation 15 in the supplementary material ( Figures B1 and B2). 16 18 19 Between 2001 and 2011, we found evidence of strong persistence of social fragmentation in 20 England at small-area level. Regional changes over time were modest for social 21 fragmentation, which is not surprising considering the large population denominators and 22 the relative nature of the measure. Similarly, we observed modest increases over time for 23 one person households and population turnover. However, changes over time for single 24 persons and private renting were much greater, with substantial increases in both 25 underlying indicators across regions, and these specific influences appear to have largely 26 driven the heightened levels of social fragmentation overall. Furthermore, our findings 27 provide evidence of increasing clustering for social fragmentation and its underlying 28 indicators which, in conjunction with the regional variation observed between the two time 29 points, indicates that areas in the South of England (including London) appear much more 30 clustered than the North. Finally, we observed increased clustering for all underlying 31 indicators of social fragmentation over the two time points. 32 33 Strengths and limitations of the study 34 35 The main strength of this study lies in its large dataset for the whole of England (53 million 36 people in 2011) over two time points across 32,844 LSOAs, which we utilised to investigate 37 spatial and temporal patterns of social fragmentation. To our knowledge, it is the first study 38 of its kind to provide an insight into the temporal persistence of social fragmentation and its 39 clustering. However, some limitations exist. First, the extent to which the index captures social 2 fragmentation may vary across the country as the demographic factors that we used may 3 not in themselves be valid indicators of social fragmentation, or they may adequately 4 measure social fragmentation in some areas but not in others. 22 However, the measure has 5 been reported widely in the literature and on this basis social fragmentation has been 6 identified as an important predictor of mental health outcomes such as suicide. Second, 7 Census questions do not usually directly elicit the presence or quality of neighbourhood 8 social ties or institutions, or people's relationship to their neighbourhood. Therefore, we 9 might have failed to capture people's perception of social fragmentation through the 10 Census data. Nevertheless, we used a measure capturing compositional aspects of 11 neighbourhood which appear to be related to the capacity of a neighbourhood to act 12 collectively, and is also important for the construction and maintenance of social ties and 13 institutions. Third, whilst Census data is recognised to be of a high standard, some 14 unobserved heterogeneity persists. 33 For example, LSOAs are defined from administrative 15 information and this may disguise local effects as contextual characteristics within 16 administrative neighbourhoods are not homogeneous across LSOAs. 34 An alternative 17 approach would require the use of non-administrative data to define levels, but the Census 18 is currently the only source from which the data on the fragmentation underlying indicators 19 is available. Fourth, for the underlying indicator 'single people' we used data on legal marital 20 status in 2001 and 2011 and registered civil partnership status in 2011. However, the 21 population base for this indicator is not restricted to those households for which legal 22 marital status information is available and this has implications for the total numbers of 23 single people. For example, some of those who are single, widowed or divorces could be 24 living in institutions as well as cohabiting unions which will result in counting cohabitants as 25 'single, widowed or divorced' in the calculation of the indicator. Fifth, there were boundary 26 changes following the 2011 Census that may have affected our findings, 35 but only 2.5% of 27 LSOAs were affected and we developed an algorithm to make reasonable population 28 weighted-based estimates for these localities, depending on whether they merged, split or 29 some other change occurred. Finally, the index of social fragmentation may serve as a 30 suitable tool to inform allocation of mental health resources in adults, although recent 31 evidence suggests that social fragmentation may not have an effect on children's mental 32 health. 36 33 34 Interpretation of findings 35 36 A previous study showed that between 1971 and 2001, levels of social fragmentation 37 increased steadily with large increases between 1981 and 1991 and modest increases 38 between 1991 and 2001. Our results indicate that this trend was reversed in 2011 as we 39 found a small decrease in the overall index of social fragmentation between 2001 and 2011. 40 37 Furthermore, our results suggest that urbanisation drives the high persistence of social fragmentation in England as we found that levels of social fragmentation and its underlying 2 indicators were consistently higher in urban areas. This may be because urban areas are 3 chosen by diverse social and cultural groups and living in these areas may lead to isolated 4 spaces, lifestyles and problems of social integration. Furthermore, we found that the 5 underlying indicators that drive the persistency of social fragmentation appear to be single 6 people and households in private renting, a characteristic of urban environments, 38 as 7 these were the underlying indicators with the largest increases ( Figure 3). 8 In absolute terms, private renting was considerably increased in all regions and most 9 notably London; a finding that pertains to unaffordable housing for first-time home buyers 10 and less availability of social housing as a result of the 2008 financial crisis. 39 Similarly, all 11 regions had higher levels of single people in 2011, indicating that numbers of young 12 professionals, students and divorces have increased substantially over the decade. For the 13 other two underlying indicators we did not observe any marked changes across regions, 14 although levels of population turnover appeared to have decreased between the decennial 15 Censuses. The increases we observed in the proportion of single people and those in private 16 renting warrant implications for the prevalence of mental illness especially in urban centres. 17 The private house market is characterised by poor housing conditions as opposed to social 18 houses or owner occupation and this can have detrimental effects on mental health. 40 19 Furthermore, it is suggested that married individuals may have better mental health 20 outcomes than single individuals. 41 We believe that these changes are certainly of concern 21 as they may highlight the need for more effective social interventions to target socially 22 fragmented areas in England. 23 Spatial clustering of neighbourhood social fragmentation appears to have increased over 24 time in England while we observed moderate regional variations in levels of clustering 25 despite the small changes in levels of social fragmentation across regions. Increased levels 26 of spatial clustering for social fragmentation across and within regions can have implications 27 for the planning and organisation of health services, as this would imply needs that are not 28 uniformly distributed across a region with the presence of hot spots of high levels of mental 29 illness prevalence, especially in urban areas. To address such spatial health inequalities, 30 services may need to be redesigned locally 42 while infrastructure and spending for mental 31 health needs to be weighted towards social fragmentation hot spots rather than uniformly 32 distributed. 33 34 Conclusion 35 36 In the absence of individual-level data, area-level indicators are often the only source of 37 information to investigate how factors associated with mental illness may contribute to 38 population health, albeit when individual-level data are available and are fitted as covariates these area-level associations may be attenuated substantially. Even though, social 2 fragmentation is increasingly considered a determinant of mental illness as more 3 epidemiological research emerges that investigates contextual effects on health. To improve 4 mental health in the population, policymakers need to address both relational and 5 structural issues, perhaps by investing in reorganisation of mental healthcare. Solving the 6 problems of cohesion and integration needs to be seen as a collective responsibility across 7 central and local government and all the local agencies. Policies needs to address social and 8 educational inequalities by strengthening the economic well-being of areas by lowering local 9 unemployment rates, encouraging populations to bring social support in the communities 10 and encouraging education of the population, which are key strategies for healthy 11 communities. Communities need to work towards tightening social networks and social 12 bonds to help with mental illness prevalence and suicide prevention, for example by 13 promoting participation in charity, voluntary or community groups and civic engagement. 14 The spatial aspect of social fragmentation is often overlooked, but it can provide vital 15 information for the effective organisation of health services and targeting of health or social 16 interventions. The nature of fragmentation and the mechanisms underlying its association 17 with mental illness are of increasing interest as the population gets older and the health 18 care costs associated with mental illness escalate in industrialised countries. 19  We would like to thank the Office of National Statistics for the wealth of information they 3 have collected and systematically organised, which made this study possible.  18 Research, the Department of Health, or the MRC. 19 20 Competing interests 21 None declared. 22 23 Provenance and peer review 24 Not commissioned; externally peer reviewed 25 26 Data sharing agreement 27 The data used in this study are freely available and the authors are happy to share in an 28 organised and cleaned final dataset. 29 30 Ethical approval and Patient and Public Involvement 31 Not applicable. The study uses freely reported data aggregates at low geography levels. No 32 patients or members of the public were involved in the study. 33 34 Open access 35 This is an Open Access article distributed in accordance with the terms of the Creative 36 Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt 37 and build upon this work, for commercial use, provided the original work is properly cited. 38 See: http://creativecommons.org/licenses/by/4.0/ © Article author(s) (or their employer(s) 39 unless otherwise stated in the text of the article) 2018. All rights reserved. No commercial 40 use is permitted unless otherwise expressly granted.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59

2011
illness. In this study we examine spatial and temporal patterns of the index of social fragmentation 5 and its underlying indicators in England over time. nationally (Spearman's r=0.94). At the regional level, modest changes were observed over time, but 15 marked increases were observed for two of the four social fragmentation underlying indicators, 16 namely single people and those in private renting. Results supported our hypothesis of increasing 17 spatial clustering over time. Moderate regional variability was observed in social fragmentation, its 18 underlying indicators and their clustering over time. 19 20 Conclusion: Patterns of social fragmentation and its underlying indicators persisted in England 21 which seem to be driven by the large increases in single people and those in private renting. Policies 22 to improve social cohesion may have an impact the lives of persons who experience mental illness. 23 The spatial aspect of social fragmentation can inform the targeting of health and social care 24 interventions, particularly in areas with strong social fragmentation clustering. Strengths and Limitations of the study  This study utilises data over two time points for the whole of England and is the first to describe spatial and temporal patterns of social fragmentation.  Spatial aspects of social fragmentation can inform organisation of mental health services and social interventions that aim to enhance social support, particularly in areas where social fragmentation is high and is spatially clustered  Over the last two decennial Censuses (2001 and 2011), the index of social fragmentation in England strongly persisted at a small-area geographical level while its clustering and its underlying indicators increased over time.  Over time social fragmentation remained relatively stable, but we found marked increases in the percentage of single people and the percentage of those living in private rented accommodation.  Even though census data are considered of high quality, people's perception of social fragmentation may not be fully captured in the census while demographic factors may capture social fragmentation in some areas but not in others.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  The impact of local area characteristics, such as material deprivation, on health is well 4 established. 1 2 However, similar to socioeconomic position, the social networks and the 5 amount of social support in a neighbourhood may also partially determine an individual's 6 health. [3][4][5][6] In this context, an area-level characteristic that has been the focus of much 7 investigation in neighbourhood health research is social fragmentation. The index of social 8 fragmentation measures lack of social cohesion, and it is used to define areas with a 9 breakdown or absence of social capital, and was originally created to measure the non-10 economic deprivation aspects of areas. 7 Social fragmentation is usually measured using 11 census data which offer a great potential for the design and implementation of mental 12 health strategies when individual level data are not available. 13 14 Social fragmentation is distinct conceptually from deprivation, since fragmentation is a 15 feature of household composition and demographic structure and is not intrinsically linked 16 to socioeconomic position. 8 However, the two measures tend to be positively correlated as 17 they are both concentrated in large towns and cities. 8 9 Thus, many urban localities are both 18 deprived and socially fragmented, although the two phenomena do not always co-exist. 19 Several studies suggests that fragmented communities provide less stable social institutions 20 and social bonds, although maintaining stable institutions and bonds within the 21 communities can contribute substantially to the creation of shared identities, persistence of 22 social relations, promotion of healthy behaviours and good physical and mental health for 23 the population. 10-14 24 25 Social fragmentation has received less attention from researchers than deprivation, despite 26 the fact that it can aid the understanding of mechanisms via which the wider social 27 environment influences mental health. 15 Research on adults has shown that social 28 fragmentation is more closely related to mental health than physical health outcomes, 29 independent of material deprivation and individual-level risk factors. 7 16 In the context of 30 mental health, social fragmentation has been linked with suicide and non-fatal self-harm, 7 31 17-19 mental disorders, 4 11 20 psychiatric health service use, 21 first admission for psychosis, 20-32 22 psychological distress, 3 11 and schizophrenia. 15 23 Nonetheless, physical illness outcomes 33 do not appear to be as clearly linked to social fragmentation. 18 24 25 34 35 In the UK, an index of social fragmentation was developed to measure neighbourhood-level 36 conditions which affect lack of opportunities for social integration. 7 The index was originally 37 based on indicators of population turnover, percentage of married people, percentage of 38 the population in private renting and percentage of one person households using data from 39 the UK Census. The rationale for the selection of the underlying variables used for the 40 calculation of the index was that several studies, both at ecological and individual level, have found raised mental illness prevalence in highly urbanised areas with high population 2 turnover, many non-family households, and reduced social cohesion. 26 3 4 Social fragmentation in a population is likely to have profound effects on its health and 5 wellbeing, and yet it is incompletely examined and reported in the existing literature, and a 6 number of important research questions remain unanswered. First, how strongly clustered 7 is social fragmentation? Second, how strongly does social fragmentation cluster and change 8 at regional level over two time points? Third, does social fragmentation persist over time? 9 Answering all three questions is important to inform the organisation of healthcare services 10 and to target social and healthcare interventions, especially for mental health for which 11 social fragmentation appears to be a salient risk factor. In this study, we temporally and 12 spatially describe the properties of social fragmentation in England and within each of the 13 10 administrative English regions and make comparisons across them. Finally, we quantify 14 the persistence of social fragmentation over the study period. 15

18
Data sources 19 20 To derive the index of social fragmentation we accessed decennial Census data from the   used a weighted means algorithm to assign the data for each underlying indicator to 2011 2 LSOAs using mid-year population estimates that were obtained from ONS. 28  Midlands, East of England, London, South East Coast, South Central and South West), which 16 were the highest level of organisation available in NHS England during the study period. To 17 assess whether levels of spatial autocorrelation for social fragmentation changed over time 18 we compared the values of Moran's I 30 in the two time points for England and each English 19 region to allow within-England comparisons. The measure can identify the presence or 20 absence of spatial clusters while accounting for the multi-dimensional and multi-directional 21 nature of spatial autocorrelation. Under a random spatial pattern a higher than expected 22 value of Moran's I for social fragmentation would indicate that areas with high levels of 23 social fragmentation are clustered and also that LSOAs with high social fragmentation are 24 bordered with LSOAs with similarly high levels of social fragmentation. We also investigated 25 local indicators for spatial autocorrelation (LISA) 31 to identify local patterns of spatial 26 associations. We also quantified the persistency of social fragmentation over the two time 27 points by calculating Spearman's correlation between 2001 and 2011. To visualise and 28 compare temporal changes in the mean social fragmentation levels across the 10 English 29 regions, we plotted population-weighted box plots for the main outcome and its underlying 30 indicators. Finally, we used population-weighted box plots to visualise and compare 31 temporal changes in the social fragmentation levels between the 10 English regions. 32 Analyses were performed with Stata v.14.1 and R v.3.3.1. Due to the size of the dataset, 33 effectively the whole of England, statistical significance is irrelevant; observed associations 34 of minimal strength would be statistically significant, and thus we focussed on effect sizes 35 wherever possible. 36

39
Region-level characteristics of the index of social fragmentation at both time points are 40 provided in Table 1. We found strong correlation for social fragmentation at the two time  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 26 We also observed a modest decrease, both nationally and regionally (with the exception of 27 London), in our measure of population turnover which reflects the levels of migration inside 28 and outside areas. Finally, we observed great variability in levels of social fragmentation and 29 its underlying indicators within regions at both time points. High levels of social 30 fragmentation were concentrated in London, the North West and the South West. East 31 England had the lowest levels of social fragmentation followed by West Midlands. There 32 was a clear distinction between rural and urban areas, with the latter having far greater 33 levels of social fragmentation. 34 35 Across the whole of England, Moran's I for social fragmentation was very low at 0.0449 in 36 2001 (95% CI 0.0447 to 0.0451) but by 2011 it had increased to 0.0794 (95% CI 0.0792 to 37 0.0797), indicating a small increase in spatial autocorrelation and clustering (Figure 4). We 38 also observed marked regional variability in social fragmentation spatial autocorrelation 39 levels over time (Figure 4). In both time points, the West Midlands had the lowest spatial 40 autocorrelation in social fragmentation, followed by the East of England, while the South 41 East and South West had the highest levels of spatial autocorrelation at both time points. In  3 autocorrelation over the two time points. 4 5 Spatial clustering also appears to have increased over time for all fragmentation underlying 6 indicators nationally, although its levels remained low ( Figures A1 and A2 in supplementary   7 Appendix 1). For one person households we found very low regional clustering and small 8 regional variation. For single persons and private renting regional variation in clustering was 9 moderate and their levels were also low with the exception of the South East and South 10 West. Finally, for population turnover we observed the greatest increases of spatial 11 autocorrelation both nationally and regionally but levels of regional clustering were 12 generally low with the exception of South West whilst regional variability was moderate. We 13 provide graphs and discuss the results from the LISA analysis for local spatial autocorrelation 14 in the supplementary appendix 1 (Figures B1 and B2). 15 17 18 Between 2001 and 2011, we found evidence of strong persistence of social fragmentation in 19 England at small-area level. Regional changes over time were modest for social 20 fragmentation, which is not surprising considering the large population denominators and 21 the relative nature of the measure. Similarly, we observed modest increases over time for 22 one person households and population turnover. However, changes over time for single 23 persons and private renting were much greater, with substantial increases in both 24 underlying indicators across regions, and these specific influences appear to have largely 25 driven the heightened levels of social fragmentation overall. Furthermore, our findings 26 provide evidence of increasing clustering for social fragmentation and its underlying 27 indicators which, in conjunction with the regional variation observed between the two time 28 points, indicates that areas in the South of England (including London) appear much more 29 clustered than the North. Finally, we observed increased clustering for all underlying 30 indicators of social fragmentation over the two time points. 31 32 Strengths and limitations of the study 33 34 The main strength of this study lies in its large dataset for the whole of England (53 million 35 people in 2011) over two time points across 32,844 LSOAs, which we utilised to investigate 36 spatial and temporal patterns of social fragmentation. To our knowledge, it is the first study 37 of its kind to provide an insight into the temporal persistence of social fragmentation and its 38 clustering. However, some limitations exist. First, the extent to which the index captures social 2 fragmentation may vary across the country as the demographic factors that we used may 3 not in themselves be valid indicators of social fragmentation, or they may adequately 4 measure social fragmentation in some areas but not in others. 22 However, the measure has 5 been reported widely in the literature and on this basis social fragmentation has been 6 identified as an important predictor of mental health outcomes such as suicide. Second, 7 Census questions do not usually directly elicit the presence or quality of neighbourhood 8 social ties or institutions, or people's relationship to their neighbourhood. Therefore, we 9 might have failed to capture people's perception of social fragmentation through the 10 Census data. Nevertheless, we used a measure capturing compositional aspects of 11 neighbourhood which appear to be related to the capacity of a neighbourhood to act 12 collectively, and is also important for the construction and maintenance of social ties and 13 institutions. Third, whilst Census data is recognised to be of a high standard, some 14 unobserved heterogeneity persists. 32 For example, LSOAs are defined from administrative 15 information and this may disguise local effects as contextual characteristics within 16 administrative neighbourhoods are not homogeneous across LSOAs. 33 An alternative 17 approach would require the use of non-administrative data to define levels, but the Census 18 is currently the only source from which the data on the fragmentation underlying indicators 19 is available. Fourth, for the underlying indicator 'single people' we used data on legal marital 20 status in 2001 and 2011 and registered civil partnership status in 2011. However, the 21 population base for this indicator is not restricted to the household population and this has 22 implications for the total numbers of single people. For example, some of those who are 23 single, widowed or divorces could be living in institutions as well as cohabiting unions which 24 will result in counting cohabitants as 'single, widowed or divorced' in the calculation of the 25 indicator. Fifth, there were boundary changes following the 2011 Census that may have 26 affected our findings, 34 but only 2.5% of LSOAs were affected and we developed an 27 algorithm to make reasonable population weighted-based estimates for these localities, 28 depending on whether they merged, split or some other change occurred. Finally, the index 29 of social fragmentation may serve as a suitable tool to inform allocation of mental health 30 resources in adults, although recent evidence suggests that social fragmentation may not 31 have an effect on children's mental health. 35 32 33 Interpretation of findings 34 35 A previous study showed that between 1971 and 2001, levels of social fragmentation 36 increased steadily with large increases between 1981 and 1991 and modest increases 37 between 1991 and 2001. Our results indicate that this trend was reversed in 2011 as we 38 found a small decrease in the overall index of social fragmentation between 2001 and 2011. 39 36 Furthermore, our results suggest that urbanisation drives the high persistence of social 40 fragmentation in England as we found that levels of social fragmentation and its underlying indicators were consistently higher in urban areas. This may be because urban areas are 2 chosen by diverse social and cultural groups and living in these areas may lead to isolated 3 spaces, lifestyles and problems of social integration. Furthermore, we found that the 4 underlying indicators that drive the persistency of social fragmentation appear to be single 5 people and households in private renting, a characteristic of urban environments, 37 as 6 these were the underlying indicators with the largest increases ( Figure 3). 7 In absolute terms, private renting was considerably increased in all regions and most 8 notably London; a finding that pertains to unaffordable housing for first-time home buyers 9 and less availability of social housing as a result of the 2008 financial crisis. 38 Similarly, all 10 regions had higher levels of single people in 2011, indicating that numbers of young 11 professionals, students and divorces have increased substantially over the decade. For the 12 other two underlying indicators we did not observe any marked changes across regions, 13 although levels of population turnover appeared to have decreased between the decennial 14 Censuses. The increases we observed in the proportion of single people and those in private 15 renting warrant implications for the prevalence of mental illness especially in urban centres. 16 The private house market is characterised by poor housing conditions as opposed to social 17 houses or owner occupation and this can have detrimental effects on mental health. 39 18 Furthermore, it is suggested that married individuals may have better mental health 19 outcomes than single individuals. 40 We believe that these changes are certainly of concern 20 as they may highlight the need for more effective social interventions to target socially 21 fragmented areas in England. 22 Spatial clustering of neighbourhood social fragmentation appears to have increased over 23 time in England while we observed moderate regional variations in levels of clustering 24 despite the small changes in levels of social fragmentation across regions. Increased levels 25 of spatial clustering for social fragmentation across and within regions can have implications 26 for the planning and organisation of health services, as this would imply needs that are not 27 uniformly distributed across a region with the presence of hot spots of high levels of mental 28 illness prevalence, especially in urban areas. To address such spatial health inequalities, 29 services may need to be redesigned locally 41 while infrastructure and spending for mental 30 health needs to be weighted towards social fragmentation hot spots rather than uniformly 31 distributed. 32 33 Conclusion 34 35 In the absence of individual-level data, area-level indicators are often the only source of 36 information to investigate how factors associated with mental illness may contribute to 37 population health, albeit when individual-level data are available and are fitted as covariates 38 these area-level associations may be attenuated substantially. Even though, social  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y  1 fragmentation is increasingly considered a determinant of mental illness as more 2 epidemiological research emerges that investigates contextual effects on health. To improve 3 mental health in the population, policymakers need to address both relational and 4 structural issues, perhaps by investing in reorganisation of mental healthcare. Solving the 5 problems of cohesion and integration needs to be seen as a collective responsibility across 6 central and local government and all the local agencies. Policies needs to address social and 7 educational inequalities by strengthening the economic well-being of areas by lowering local 8 unemployment rates, encouraging populations to bring social support in the communities 9 and encouraging education of the population, which are key strategies for healthy 10 communities. Communities need to work towards tightening social networks and social 11 bonds to help with mental illness prevalence and suicide prevention, for example by 12 promoting participation in charity, voluntary or community groups and civic engagement. 13 The spatial aspect of social fragmentation is often overlooked, but it can provide vital 14 information for the effective organisation of health services and targeting of health or social 15 interventions. The nature of fragmentation and the mechanisms underlying its association 16 with mental illness are of increasing interest as the population gets older and the health 17 care costs associated with mental illness escalate in industrialised countries.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  We would like to thank the Office of National Statistics for the wealth of information they 3 have collected and systematically organised, which made this study possible. 4 5 Contributors 6 EK, CG, DA and LM designed the study. CG extracted the data from all sources, performed 7 the analyses and drafted the manuscript. EK, RW, DA, LM, NK critically revised the 8 manuscript. CG is the guarantor of this work and, as such, had full access to all the data in 9 the study and takes responsibility for the integrity of the data and the accuracy of the data 10 analysis. 11 12 Funding 13 This study is funded by the National Institute for Health Research School for Primary Care 14 Research (NIHR SPCR), through CG's PhD. This report is independent research by the 15 National Institute for Health Research. LM acknowledges financial support from the MRC 16 Skills Development Fellowship (MR/N015126/1). The views expressed in this publication are 17 those of the authors and necessarily those of the NHS, the National Institute for Health 18 Research, the Department of Health, or the MRC. 19 20 Competing interests 21 None declared. 22 23 Provenance and peer review 24 Not commissioned; externally peer reviewed 25 26 Data sharing agreement 27 The data used in this study are freely available and the authors are happy to share in an 28 organised and cleaned final dataset. 29 30 Ethical approval and Patient and Public Involvement 31 Not applicable. The study uses freely reported data aggregates at low geography levels. No 32 patients or members of the public were involved in the study. 33 34 Open access 35 This is an Open Access article distributed in accordance with the terms of the Creative 36 Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt 37 and build upon this work, for commercial use, provided the original work is properly cited. 38 See: http://creativecommons.org/licenses/by/4.0/ © Article author(s) (or their employer(s) 39 unless otherwise stated in the text of the article) 2018. All rights reserved. No commercial 40 use is permitted unless otherwise expressly granted.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59

Discussion
Key results 18 Summarise key results with reference to study objectives p.5-p.6 Limitations 19 Discuss limitations of the study, taking into account sources of potential bias or imprecision.
Discuss both direction and magnitude of any potential bias p.7 Interpretation 20 Give a cautious overall interpretation of results considering objectives, limitations, multiplicity of analyses, results from similar studies, and other relevant evidence p.8 Generalisability 21 Discuss the generalisability (external validity) of the study results p.9 Other information Funding 22 Give the source of funding and the role of the funders for the present study and, if applicable, for the original study on which the present article is based p.11 *Give information separately for cases and controls in case-control studies and, if applicable, for exposed and unexposed groups in cohort and cross-sectional studies.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60