Loneliness is a significant public health and wellbeing concern associated with adverse health outcomes including cardiovascular disease, cognitive decline, depression, and mortality [1–4], resulting in substantial societal and economic costs [5, 6]. Loneliness is a subjective state of perceived social deficit typified by feelings of isolation; social isolation is an objective lack of social contacts and relationships [2, 3, 7]. Due to the subjective nature of loneliness it is captured using various measurement tools and outcome classifications, such as emotional and social loneliness [8, 9]. Prevalence of loneliness can vary according to measurement tools, loneliness types, and across multiple factors including individual characteristics and life circumstances, such as gender, and age, which may influence social aspirations across the life course. Markers of social isolation include reduced social networks, fewer social contacts, and living alone. International research has associated living alone with both loneliness and adverse health outcomes including elevated mortality and suicide risk, particularly among middle-aged men [3, 10–13].

This dual aspect of living alone, as a correlate of loneliness and an independent health determinant, underscores potential value of including living alone measures in population research. Increased loneliness is also associated with living arrangements and partner bereavement [14, 15]. However, few studies have examined loneliness using population level administrative data including the inferred reason for why someone lives alone.

Additional loneliness risk factors include age, with mixed evidence; large coordinated analyses suggest a U-shaped curve (higher in younger and older adults) [16], whereas recent UK monitoring reports highest prevalence in younger adults [17], likely reflecting measure and context differences. Gender gaps are generally small across the life course [18]. Evidence from Wales and England reports increased loneliness for those from an ethnic minority background, those who live in material deprivation, and for those living in social or privately rented properties [14, 19]. Broader UK analyses report lower education and lower socioeconomic status being linked to higher loneliness and social isolation [20]. Neighbourhood greenness, walkability, and housing type have been linked with greater social connection and lower loneliness in certain contexts [21–23]. However, systematic review evidence emphasises variation across populations and setting [24–26].

Loneliness is likely to arise through the interaction of multiple individual, social and environmental factors. Living arrangements, health status and place-based characteristics may each influence opportunities for social connection and experiences of loneliness. Understanding these factors together may therefore provide a more comprehensive picture of loneliness risk within populations. Measurement of loneliness and living alone can be challenging - large surveys are costly, may include non-representative or modest samples, and repeated measures are needed to capture longitudinal trends, with measures often limited to binary classifications [27]. With advances in access to population level linked administrative data (generated through individuals’ interactions with public services [28]) it is possible to improve our understanding of these circumstances by including place-based factors which may impact on, or shape experiences, of loneliness. With a statutory loneliness national indicator [14, 29], Wales provides a useful setting for this research using the mature Secure Anonymised Information Linkage (SAIL) Databank which supports privacy-protecting linkage of person-based health, social care and demographic data at population level [30]. Wales has an ageing population [31], has distinct small-area deprivation across diverse urban, rural, and post-industrial communities [32].

The SAIL Databank enables linkage of longitudinal administrative and survey data alongside demographic, health, and built environment measures at population-scale. This study used administrative data to develop richer, longitudinally consistent indicators of living alone as an exposure variable along with other risk factors and investigated associations with loneliness. By linking longitudinal administrative and survey data, we (1) explored the scale, temporal trends, and demographic characteristics of living alone in Wales, United Kingdom (UK), from 2008–2024 (Phase 1); (2) developed novel living alone reason and duration indicators (Phase 1); and (3) assessed associations between emotional, social, and overall loneliness and living alone as well as other potential risk factors including health, demographic, and built environment factors (Phase 2).

Figure 1 shows trends in living alone households in Wales from 2008 to 2024. The proportion of households classified as living alone increased from 27.0% (314,806) in 2008 to 29.6% (363,133) in 2024, representing a 2.6 percentage-point increase (9.6% relative increase). Living alone was most common among those aged 76 years and over throughout the study period, although prevalence declined modestly over time. In contrast, the largest increase was observed among those aged 16–30 years, with smaller increases across the remaining age groups. Females consistently had a higher prevalence of living alone than males, although the proportion of males living alone increased slightly across the study period. Across the study period, the mean number of living alone households was 336,646 (SD=13,722), compared with 859,801 (SD=5,660) multiple-person households.

Within Phase 1 the study evidenced living alone patterns, with a slow continuously increasing trend in the scale by number and proportion of living alone households in Wales. Within modelled results, distinct living alone patterns by age and gender were observed across the life course; younger males showed the highest prevalence of living alone, while among females prevalence increased from midlife and remained elevated in older age groups. The largest adjusted association was observed for those living in flats compared with other housing types (detached, semi-detached and terraced). We also introduced a novel method for inferring and tracking longitudinal living alone measures using administrative data at population level.

Phase 2 of the study found that loneliness in Wales is strongly associated with factors related to living alone, poor health, and certain demographic factors, with the highest risk amongst those who have recently lost a cohabiting partner to death or entry to a care home. A linear negative association between age and overall loneliness was observed with younger adults reporting highest loneliness. Important age-gender differences were evident with contrasting patterns between emotional and social loneliness subscales. Males reported consistently higher social loneliness than females, especially around middle age (31-60), whereas females reported higher levels of emotional loneliness than males with a widening gap with age. Built environment factors showed little evidence of association.

The significantly increased odds of emotional loneliness for those recently bereaved aligns with previous findings indicating relationship loss is a key determinant of loneliness [15]. We quantified differences in measures of loneliness between living alone reasons: partner loss, separation, and unassigned cases, demonstrating that context matters beyond a simple binary indicator. This supports arguments that living alone is an important correlate of loneliness and health outcomes [3, 10, 11]. Our results confirm previous research linking living alone factors to loneliness and demonstrate the ability to apply such methods using administrative data [15, 49, 52].

We observed that modelled overall loneliness declined with age, with younger adults reporting the highest levels, and clear age-gender differences across the emotional and social subscales—a pattern consistent with large-scale survey evidence [53] and ONS monitoring [17]. This partly contrasts with large-scale studies reporting a U-shaped age pattern [16, 54]. Notably, NSW analysis found no independent associations of overall loneliness with either age or limiting long-term health conditions, alongside strong associations with general health and mental illness [14]. As noted previously [55], inconsistencies may reflect differences in measurement and modelling approaches; our examination of emotional and social subscales separately, and to include an age-gender interaction, may explain why we identified clearer declines with age alongside gender specific differences in line with previous research [17].

Age-gender differences between emotional and social subscales underline the value of analysing these dimensions separately, as determinants can differ by subscale [9, 51]. The UCLA-3 is widely used as an overall loneliness indicator and correlates with both De Jong Gierveld subscales [56].

Place-based effects were small in contrast to international studies reporting lower loneliness in more favourable neighbourhood and social environments [21–23, 57]. Our null findings are consistent with systematic reviews emphasising contextual variation and mixed results [24, 25]. This may reflect limitations of the available measures, or that, in Wales, individual level factors such as health and living arrangements dominate over place effects.

Our results highlight several groups at elevated risk of loneliness. Young adults reported the highest levels of emotional loneliness, while recently bereaved adults showed particularly high risks of emotional loneliness persisting years after partner loss. Poor self-reported health and long-term illness were consistently associated with loneliness across subscales, suggesting potential implications for health and social care service use. We also observed higher social loneliness among midlife men, aligning with evidence of increased social isolation and suicide risk in this group [12, 13]. Taken together, these findings identify several groups that may benefit from further investigation and consideration within intervention and service-planning approaches, including bereavement support, community outreach and health-linked social prescribing.

Findings also suggest that loneliness may be particularly relevant for groups already in contact with health and social care services, such as those in poor health, living in social housing, and those recently bereaved; highlighting potential opportunities for identification and support.

This study has several strengths. In Phase 1, we created longitudinal population level snapshots for Wales, enabling analysis of living alone trends over time. We also developed novel living alone reason indicators which may prove useful for wider research using similar administrative data. In Phase 2, we drew on relatively large numbers by linking two survey sources to administrative data, applied validated multiscale loneliness scores, and demonstrated the practical use of novel living alone factors in modelling risk of loneliness. Inclusion of age-gender interactions across multiple loneliness scales provided more nuanced understanding than combining gender or limiting analyses to an overall loneliness score, avoiding the risk of masking potentially modifiable associations relevant for policy.

There are also several study limitations. As is standard with data linkage, some loss of cases occurred, for example when individuals were not assigned a linkage key (around 9%); or when NSW survey data were not successfully linked (around 6%) (Supplementary Table 3); this study therefore remains broadly representative of residents in Wales over 2016-2022. Use of annual snapshots may introduce temporal error due to factors such as discrepancies between survey dates, events such as death, and mid-year linkage points. The built environment ALE transit measure was available at LSOA level, which may be less sensitive when exploring associations. The NSW methodology also changed across the study period; not all variables were collected annually, and the COVID-19 pandemic may have influenced loneliness scores. Given the observational design and partial reliance on cross-sectional survey measures, associations should not be interpreted as causal. Relationships between loneliness, health status, housing circumstances and living arrangements are likely to be complex and potentially bidirectional.

Potential selection bias may have arisen through participant consent for linkage, successful creation and matching of anonymised linkage identifiers, and study-specific exclusions due to missing covariate or outcome information. However, comparisons between the initial eligible survey sample and final analytic cohort demonstrated broadly similar age distributions and only modest differences in sex distributions (Supplementary Table 6). Sensitivity analyses restricted to pre-pandemic and post-pandemic survey waves produced broadly consistent findings.

Finally, while our novel living alone reason measure provided additional insight, a simple binary indicator of living alone also showed clear associations with loneliness. For monitoring purposes, such binary indicators may therefore be sufficient, offering reduced complexity and easier implementation across large datasets. The more detailed measure highlights heterogeneity, especially for emotional loneliness, but comes at the cost of added derivation and interpretive complexity. Accepting this trade-off may be appropriate where increased simplicity and comparability are prioritised, though our findings suggest nuanced indicators add value in research contexts. Ability to derive and assign reason indicators varies over time as a natural data artefact related to address registration data; for individuals already living alone prior to the APS 2008 start date, reasons could not be inferred. These inferred categories represent administrative approximations of household transition circumstances rather than precise social or relationship histories and may therefore be subject to some misclassification. Population level living alone reason trends (Supplementary Figures 2,3) indicate more recent snapshots provide a clearer basis for understanding living alone dynamics.

We present longitudinal living alone trends and introduce a novel method to derive living alone reasons at the population level using large-scale, routinely collected data. Living alone, particularly when characterised by more granular inferred reasons, was strongly associated with emotional, social, and overall loneliness after adjustment for demographic and health factors, with the strongest associations for emotional loneliness. Including an age-gender interaction revealed nuanced differences across the life course. These findings extend the loneliness evidence base and offer scalable indicators for monitoring living alone dynamics. They also provide tools for future health and social care research using linked data and complement national statistics on loneliness and wellbeing.

RDJ: Conceptualisation, methodology, formal analysis, data curation, visualisation, writing – original draft, approval of final submission.

RB, SB: Conceptualisation, supervision, methodology, writing – review & editing, approval of final submission.

RF: Conceptualisation, supervision, writing – review & editing, approval of final submission.

LG: Conceptualisation, supervision, methodology, writing – review & editing, approval of final submission.

This work is supported by Administrative Data Research (ADR) Wales (Grant ref: ES/W012227/1), part of the ADR UK investment, uniting research expertise from Swansea University Medical School and WISERD (Wales Institute of Social and Economic Research and Data) at Cardiff University with analysts from Welsh Government. ADR UK is funded by the Economic and Social Research Council (ESRC), part of UK Research and Innovation.