Methods
Dataset
We address our research question using the English Longitudinal Study of Ageing (ELSA): a nationally representative survey of approximately 10,000 people aged 50+ years living in England, modelled on the well-established US Health and Retirement Survey (HRS;
http://hrsonline.isr.umich.edu). The wave 0 ELSA cohort consisted of individuals who participated in the 1998, 1999, and 2001 waves of Health Survey for England (HSE;
https://www.ucl.ac.uk/hssrg/studies/hse). Those who were aged 50+ in 2002 were invited to take part in the ELSA baseline (wave 1) which included an administered survey and a self-completion questionnaire. Subsequently data collection for ELSA has taken place at two-yearly intervals with the most recently available data for analysis being from wave 8 (2016). To compensate for attrition the ELSA sample has been refreshed with HSE participants aged 50 years in waves 3, 4, 5 and 6. Ethical approval was granted from the National Research and Ethics Committee. Further details about the design, sampling and methodology of ELSA are available elsewhere (
http://www.elsa-project.ac.uk/documentation).
Measures
Loneliness
We included two measures of loneliness in our analysis with dichotomisation to classify those who are lonely defined using the upper quartile of the distributions. Our main outcome variable is the revised UCLA loneliness scale [
35] consisting of three items: “How often do you feel you lack companionship?” “How often do you feel isolated from others?”, and “How often do you feel left out?” Participants selected their response from three options (hardly ever/never; some of the time; often) which are coded 1–2-3 and summed to give a total score ranging from 3 and 9. Scores are dichotomised, with scores 3–5 classified as not lonely and 6+ as lonely. Waves 3 and 7 of ELSA included a suite of questions focused on subjective neighbourhood evaluation [
36]. This included a question which asked participants to agree or disagree with the statement ‘I often feel lonely living in this area’ using a 7-point Likert scale. This single item has not, to our knowledge, been used elsewhere to evaluate feelings of loneliness within an explicit spatial context. The scale was recoded for consistency of directionality with the UCLA measure and dichotomised with a score of 4+ defining loneliness. As the use of this measure of loneliness has not has not previously been reported we undertook the evaluation of the measure in terms of its relationship with established loneliness predictors to determine its utility. This analysis was undertaken for both waves 3 and 7 to test stability of the relationships. Both measures were included in the self-completion questionnaire component of data collection and thus there are issues of missing data.
ELSA data routinely includes details of the 9 administrative regions of England. No more fine-grained data about the areas in which participants live are routinely provided to assure participants’ anonymity and confidentiality. Following the approval of a special data access request the National Centre for Social Research (NatCen) (
http://natcen.ac.uk/), the organisation that oversees such matters, provided us with a data set which included two area based measures (urban/rural classification; deprivation index) which were linked to individuals via their individual study number for wave 6 of ELSA.
The urban and rural area classification is produced by the Office for National Statistics (ONS) and areas are defined as Urban (includes urban areas, towns and urban fringes) and Rural (village, isolated dwellings/hamlets) based upon the size of the settlement and population density (
https://www.gov.uk/government/collections/rural-urban-classification). Our provided data set were grouped into 4 categories: urban, town and fringe, village, and hamlets and isolated dwellings. The Index of Multiple Deprivation (IMD) has been produced and validated by the Department of Communities and Local Government. It is area-level measure available for 32,844 small areas in England based upon 7 domains of deprivation (Income; employment; education, skills and training; health deprivation and disability; crime; barriers to housing and services; and living environment) (
https://data.gov.uk/dataset/imd_2004). Areas are ranked by score. As there is no absolute value that differentiates deprived from non-deprived areas analysis is usually undertaken using quintiles with 1st quintile characterising the least deprived area and 5th quintile the most deprived.
Covariates
The covariates included in our analysis are well-established predictors of loneliness: gender (coded as males/females), marital/partnership status (married/in partnership, always single, divorced/separated, widowed/partner died), social relationships (close relationships with 2+ family/friends and current participation in civic activities), employment status (out of job market, employed/self-employed, looking after home/family), health (self-rated health (excellent/very good/good, fair/poor; used from wave next to baseline, as in baseline it was not collected), depressive symptoms (CESD-8 scale dichotomised as 0-2 positive responses vs 3+ out of 8 questions), activities of daily living (ADL/IADL dichotomised as no difficulties with activities of daily living vs. at least one difficulty) and long-standing limiting illness (no, yes but not limiting, yes and limiting). We excluded co-variates such as income that were included in the IMD measure.
Analytic sample
Our analysis is based on 4663 ELSA core members that satisfied two criteria: (a) they participated in waves 3 and 7 (collected in 2006 and 2014; which both included the UCLA scale and also the area-based loneliness measure) and (b) they reported the same address in wave 7 as in wave 6 (as area characteristics data was only available for wave 6). Only 5% of those who participated in waves 3 and 7 changed address between waves 6 and 7, therefore our analytical sample represents 95% of those who took part in both waves 3 and 7.
Analysis plan
Our analysis plan consisted of two phases. First, we used descriptive statistics and bivariate regression analysis to profile our analytic sample and to evaluate the crude association between independent (loneliness) and dependent variables separately for both wave 3 and wave 7 and to evaluate the utility of the area-based loneliness measure. We then undertook a series of multivariable regression analyses to evaluate the role of area characteristics in reported loneliness. We used the complete case sample as we considered it inappropriate to use imputation for missing socio-demographic data. Our three regression models were as follows: model A, loneliness measured by the 3-item UCLA scale adjusted for age and gender; model B, further adjusted for social network characteristics (marital status, evaluation of close feelings to at least two members of the family or friends, civic participation, job market participation, UCLA-measured loneliness in baseline (wave 3), and mutually adjusted for area loneliness (when analysing individual loneliness, we have adjusted also for area-based loneliness and vice versa)); model C, further adjusted for health status characteristics (depressive symptoms, long-standing limiting illness, self-rated health, and difficulties with activities of daily living). We repeated this analysis using the area-based loneliness measure as our outcome. Analyses were carried using STATA MP Version 13.0 with p-value < 0.05 signifying statistical significance.
Discussion
Research examining the antecedents of loneliness in older adults have predominantly focused upon individual characteristics. In our study we moved the focus away from individuals to the types of area in which they live as community/meso-level factors are neglected in loneliness research. We aimed to add to the existing evidence base by focusing upon the importance of place and the environment in which people live as potential loneliness vulnerability factors. We investigated the importance of three geographical categories in relation to loneliness: area typology (urban/rural), geographical region, and deprivation. We used two measures of loneliness: the 3-item UCLA scale (measuring self-reported personal loneliness status), and a measure focused upon ‘loneliness based on the area of residence’. We show that there are no relationships with region or area type or deprivation for the UCLA scale. However there was, after adjustment for confounding factors, a statistically significant relationship between area-based loneliness and deprivation.
Existing research focused upon understanding the prevalence of, and risk factors for, loneliness has largely concentrated on seeking explanation at the individual level. Sullivan et al. (2016) have discussed many of the limitation of this approach to studying loneliness including the presumption of shared understanding and the dynamic nature of the experience combined with the complexity and difficulty people may have in describing the experience of loneliness [
38]. There is also an increasing acceptance that loneliness is not a static experience but one which may fluctuate during a day, a week or a year [
39] and that the population characterised as ‘lonely’ is not homogeneous but includes those for whom loneliness is an enduring part of their life whilst for others loneliness may increase or decrease as they age.
We used two waves (3 and 7) of the English Longitudinal Study of Ageing (ELSA) to consider the relationship between individual loneliness and area-based loneliness as this latter question was included in these waves of data collection. In terms of our two loneliness outcome measures the revised UCLA scale is well established. The question on evaluating loneliness in the area where participants live was not and has not, to our knowledge, been used elsewhere. However, it does offer a novel insight into the experience of loneliness by locating it in the area in which people live.
Our three measures of area characteristics are all designed for administrative rather than research purposes. The measure of deprivation is designed for use as a means of targeting resources to areas in need. The urban-rural classification is limited in that it does not distinguish large conurbations such as London from smaller urban areas. Furthermore, our data show that England is a predominantly urban society and thus we may have had too few participants from rural areas making our study insufficiently powered to identify any differences. The third area characteristics was the geographical regions classification which gives only broad information from which part of England the participant comes but could help with distinguishing London area from the others regions. These caveats frame the confidence that we can have in our overall findings and highlight some issues to be addressed in further research.
Levels of loneliness, as assessed by our two measures, were broadly stable at 18% for the individual-based loneliness and 25% for the area-based one in both waves (Additional file
1: Table S1). Inviting participants to evaluate loneliness in the context of the area where they lived generated a higher level of loneliness and this finding is, to our knowledge, novel in the literature. The congruence between the measures was good for those reporting that they were not lonely but was under 50% of those reporting loneliness (Additional file
1: Table S6 a, b). This suggests that the area-based measure is extending into a domain of loneliness not embraced by the social/emotional relationship focus of the items included in the UCLA scale. Clearly the potential of the area-based measure and characteristics of those who report individual-based loneliness in one but not both measures merits further investigation.
Drawing comparisons with previous research is complex because of differences in how area typologies are defined, and loneliness is measured. Once other factors were considered, loneliness was not associated with region or area classification in terms of urban/rural. This lack of an association with urban/rural area classification aligns with the studies from Ireland [
22], Canada [
32] and Poland [
32]. Our study reported increased prevalence of loneliness in deprived areas which, whilst not reaching the levels reported by Scharf and de Jong Gierveld [
28] of 57%, are significantly higher than those in the least deprived areas (range 36–80% depending on measure and data collection wave). Once socio-demographic, social and health characteristics were considered the significant relationship between area-based loneliness and high levels of deprivation remained robust. This may reflect the features of the specific environment such as terrain or amenities, demographic characteristics, housing conditions, high crime rates, potential opportunities for engagement or issues of trust and neighbourliness and population turn-over. However as with the related concept of resilience there is a need to embrace the role of macro- (societal) and meso- (community/neighbourhood) factors in the emergence of vulnerability to loneliness. Further research with older people living in these types of areas is required to understand what is driving this relationship and what interventions might ameliorate it and to understand how micro-, meso-, and macro- level factors combine to protect or render older people vulnerable to loneliness.
The strength and limitations
The strength of our study is rooted in two key areas: our research questions and the use of the ELSA data set. ELSA is currently the largest, most representative and longest established longitudinal study of older people in the community within the UK and, as such, is the best UK-based data set. The development of longitudinal studies of ageing, using the same model as ELSA, in Northern Ireland (The Northern Ireland Longitudinal Study of Ageing-NICOLA study established in 2012) and Scotland (Healthy Ageing in Scotland-HAGIS study completed a pilot study in 2017) include more rural areas than England and thus offer the potential to develop our analysis in the future. Our research questions attempt to extend our understanding of loneliness beyond the individual and extend it to the area in which they live. To the best of our knowledge, the question on area-based loneliness has not been used in other research and, as such, adds to our understanding of the complexity of the experience of loneliness. By including both wave 3 and wave 7 in our analysis we have been able to establish the utility of the measure as a complementary method to the individual measure. In addition, the construction of our analytic cohort enables us to include previous loneliness experiences from wave 3 in our predictive models based on wave 7. This is important as much loneliness modelling does not take into account past experiences.
Nevertheless, there are limitations to our study which relate to the conduct of ELSA including attrition, missing data because the loneliness questions are included in the self-completion rather direct interview element of ELSA and the exclusion of those older people living in care homes. More specifically the area-based loneliness question did not offer guidance to respondents in terms of the size of the area to which the question refers. The interpretation of results regarding deprivation are preliminary given that the data provided relate to 2004 and our data collection to 2006 (wave 3) and 2014 (wave 7) and there might have been changes in deprivation profile for some areas. Given the positive relationship between loneliness and deprivation it would have been useful to conduct a longitudinal analysis of how changes in deprivation linked to changes in loneliness at individual and area level. We were unable to obtain details of area classification for both time points thereby precluding a cohort study. However, this is a potential area for future research as is qualitative a finer grained quantitative research to see which elements of the deprivation measure are important in explaining the link with loneliness.
Open AccessThis article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit
http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (
http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.