Labour Market Detachment and Social Disconnection in Later Working Life: Evidence from the Australian Hidden Workforce

Abstract

Social disconnection, encompassing both loneliness and social isolation, is increasingly recognised as an important public health concern. While employment provides opportunities for social participation and role engagement, less is known about how different forms of labour market detachment relate to subjective and objective dimensions of social connection in later working life. This study examined the association between labour force attachment and both loneliness and social isolation among Australians aged 50–64 years using cross-sectional data from Wave 22 (2022) of the Household, Income and Labour Dynamics in Australia (HILDA) Survey (n = 3362). Participants were classified into labour force attachment groups including in work, underemployed hidden workers, unemployed hidden workers, discouraged workers, those not wanting work, and other. Survey-weighted logistic regression models were used to estimate adjusted predicted probabilities of loneliness and social isolation across labour force groups. After adjustment for sociodemographic and health characteristics, predicted probabilities of loneliness were elevated across hidden worker subtypes relative to those in paid employment, with point estimates 10–13 percentage points higher across categories. Differences in social isolation between hidden worker subtypes and those in paid work were small in magnitude. The highest adjusted predicted probability of social isolation was observed among individuals who reported not wanting work. These findings suggest that, in later working life, labour market marginalisation is associated more strongly with subjective experiences of social disconnection than with the structural availability of social contact. Interventions to reduce loneliness among older working-age adults may benefit from recognising the institutional functions of paid work alongside approaches targeting social contact.

1. Introduction

Social disconnection has emerged as a significant public health concern, associated with increased risk of morbidity and premature mortality across the life course (Holt-Lunstad et al. 2010; World Health Organization 2025). Conceptually, social disconnection is understood as an umbrella construct encompassing both objective and subjective dimensions of social relationships (Badcock et al. 2022). Social isolation refers to having ‘objectively few social relationships, social roles, group memberships, and infrequent social interaction’, whereas loneliness is ‘the subjective unpleasant or distressing feeling of a lack of connection to other people, along with a desire for more, or more satisfying, social relationships’ (Badcock et al. 2022). Although related, these constructs are distinct and may occur independently (Newall and Menec 2019); individuals may maintain frequent social contact while experiencing loneliness, or conversely report low levels of loneliness despite limited social interaction. Because loneliness and social isolation are distinct and may follow different patterns across populations, examining them jointly is particularly relevant when considering how broader structural determinants, such as labour market attachment, relate to social connection in later working life.

Historically, research on social disconnection has focused on interpersonal and individual-level determinants such as personality traits, relationship quality, or life events (Poscia et al. 2018). However, there is growing recognition that loneliness and social isolation are socially patterned outcomes that vary systematically with structural conditions and opportunities for social participation (Meehan et al. 2023). From a social ecological perspective, social connection is shaped by interacting determinants operating across multiple levels, including macro-level economic and policy contexts, organisational environments, community settings, and individual characteristics (Meehan et al. 2023, 2026). Within this social ecological framework, organisations that structure daily life, such as education systems and workplaces, play a critical role in facilitating access to social interactions (McLeroy et al. 1988). Workplaces are also important sites for the production of social capital, providing opportunities to develop relationships, reciprocal support, and shared norms through routine interaction with colleagues and organisational structures (Tsounis et al. 2023). These forms of social capital may extend beyond the workplace itself, and have been linked to broader patterns of social participation and perceived belonging.

Employment represents a key organisational context through which individuals engage in socially recognised roles and maintain routine interpersonal contact (Wang et al. 2025). Beyond the economic function, participation in paid work provides opportunities for social interaction, contributes to identity formation, and supports integration within broader social networks (Korber et al. 2024). Importantly, employment also confers institutional belonging: it situates individuals within socially normative roles that structure daily routines, create expectations of participation, and signal membership within collective economic and social life (Thissen et al. 2023). It is theorised that when labour market attachment is weakened or disrupted, individuals may experience reduced access to these institutionalised forms of belonging. This form of detachment has been associated with diminished perceived role legitimacy and social recognition, and may co-occur with feelings of loneliness even where interpersonal relationships remain intact. While previous research has examined associations between unemployment and loneliness or social isolation in older working-age, much of this research has focused on individuals entering retirement (Vigezzi et al. 2025). Less attention has been paid to those who are still marginally attached to the labour force but not in full-time employment.

Recent labour market research has drawn attention to the concept of the ‘hidden workforce’, encompassing individuals who are willing and able to work but remain excluded from paid employment due to structural barriers to participation (Fuller et al. 2021). This includes those who are underemployed, unemployed but seeking work, or discouraged from active job search despite a desire to work under different circumstances (Fuller et al. 2021). Unlike conventional measures of unemployment, which capture only those actively seeking employment, the hidden workforce reflects broader forms of labour underutilisation that remain largely invisible in official statistics (Fuller et al. 2021). Emerging Australian research has demonstrated that these groups can be identified within population-based surveys such as the Household, Income and Labour Dynamics in Australia (HILDA) Survey and represent a substantial proportion of older working-age adults who are not fully engaged in paid employment despite potential willingness to work (Lee et al. 2025a; Lee and Kang 2026). As such, hidden workforce participation may represent a form of institutional detachment from socially normative roles associated with paid employment, with potential implications extending beyond economic outcomes to include experiences of social integration and belonging.

Importantly, hidden workforce categories reflect different forms of labour market detachment that may carry distinct social and psychosocial implications. For some individuals, such as those who are underemployed, labour force participation may be on a part-time or casual basis; therefore work roles remain present but may be weakened or underutilised (Fuller et al. 2021). This form of role erosion may coincide with reduced occupational identity and recognition without removing individuals from workplace participation entirely. In contrast, individuals who are unemployed or discouraged from seeking work may experience more complete role loss, in which the socially recognised status associated with employment is removed altogether (Morrish and Medina-Lara 2021). Other individuals who report not wanting work may be outside the labour force due to health limitations, caregiving responsibilities, retirement transitions, or voluntary withdrawal from employment, reflecting broader participation constraints beyond the labour market itself (Fuller et al. 2021). These distinctions suggest that different forms of labour market attachment may relate to experiences of social connection through different pathways.

The relationship between labour force attachment and social disconnection may not operate uniformly across objective and subjective domains. Individuals who are marginally detached from paid employment may continue to maintain social relationships through family, community, or voluntary activities. Yet, they may also report diminished motivation to connect socially, alongside loss of socially recognised roles (Meehan et al. 2026). This potential discordance between objective social contact and perceived social connectedness highlights the importance of distinguishing between subjective and objective dimensions of social disconnection (Newall and Menec 2019). Examining both loneliness and social isolation is therefore critical for understanding how organisational contexts shape experiences of connection in later working life. Specifically, the loss of socially recognised roles associated with paid employment may be more closely linked to subjective experiences of loneliness, whereas changes in the structural availability of social contact may be more relevant to social isolation. This conceptual distinction motivates the joint examination of both outcomes in the present study.

Therefore, this study aims to examine the association between labour force attachment and both subjective (loneliness) and objective (social isolation) dimensions of social disconnection among Australians aged 50–64 years. Furthermore, we aim to assess whether labour force attachment is differentially associated with loneliness and social isolation in later working age. In doing this, we will conceptualise hidden workforce participation as an organisational determinant of social connection within a social ecological framework.

2. Methods

2.1. Study Design and Data Source

This study used cross-sectional data from Wave 22 (2022) of the HILDA Survey. HILDA is a nationally representative longitudinal household panel study that collects annual information on economic participation, labour market activity, health, and social wellbeing among Australians living in private dwellings. The initial HILDA sample was recruited in 2001 using a multi-stage stratified random sampling approach, with all household members aged 15 years and older invited to participate. A top-up sample was incorporated in 2011 to maintain national representativeness over time (Summerfield et al. 2021). A detailed description of the HILDA survey design has been published previously by Wooden and Watson (2007).

Data are collected annually through a combination of interviewer-administered questionnaires and a self-completion questionnaire (SCQ), which captures more sensitive information including measures of social wellbeing (Watson and Wooden 2012). While HILDA follows participants longitudinally, the present study adopts a cross-sectional analytic approach using Wave 22 data only in order to examine the contemporaneous association between labour force attachment and social disconnection among older working-age Australians. We have reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement (Von Elm et al. 2007), with the checklist available in Supplementary Table S1.

2.2. Study Sample

The analytic sample was restricted to adults aged 50–64 years to focus on later working life, which is a life-course stage characterised by increased labour market withdrawal, age-related employment barriers, and shifts in patterns of social participation that precede transitions out of the workforce. The lower bound of 50 years was selected to align with conceptualisations of older working-age adults used in Australian labour market research (Lee et al. 2025b), and reflects the age from which workforce participation rates begin to decline (Australian Bureau of Statistics 2025). Although narrower age cut-offs (e.g., 50–60 years) would centre on pre-retirement adults, this would exclude individuals approaching retirement transition points where labour market detachment is most pronounced and where the experiences of hidden workforce participation are most concentrated. The upper bound of 64 years reflects the period immediately preceding Age Pension eligibility, which currently commences at 67 years in Australia. To examine the robustness of findings to this threshold, a sensitivity analysis extended the sample to include adults aged 50–67 years, thereby capturing individuals approaching or recently reaching Age Pension eligibility.

2.3. Study Measures

Further details about the operationalisation of each variable can be found in the Supplementary File.

2.3.1. Independent Variable

Hidden workforce status was derived from Wave 22 labour force participation indicators, following the classification scheme developed in prior Australian research using HILDA data, to which members of the present research team have contributed (Lee and Kang 2026; Lee et al. 2025a). Participants were classified into one of six mutually exclusive categories based on combinations of current labour force status, stated desire for work, and reasons for part-time hours. ‘In work’ comprised respondents who were employed full-time, employed part-time, or employed with unknown usual hours. ‘Underemployed hidden workers’ were respondents working part-time (fewer than 35 h per week) whose stated main reason for part-time hours reflected constrained labour market participation, including caring for children, caring for disabled or elderly relatives, other personal or family responsibilities, inability to find full-time work, or potential reductions to welfare payments or pension entitlements from working full time. ‘Unemployed hidden workers’ were respondents who were not employed but actively looking for full-time or part-time work. ‘Discouraged workers’ were respondents who were not in the labour force (either marginally attached or not marginally attached) and who reported that they wanted, or might want, to work. ‘Does not want work’ comprised respondents not in the labour force who reported not wanting to work. A residual ‘other’ category captured respondents not in the labour force for whom the relevant labour force preference items were missing. Detailed operationalisation, including HILDA variable names and decision rules, is provided in Supplementary Table S2.

2.3.2. Outcome Variables

Social disconnection was operationalised using two complementary constructs measured in the Wave 22 self-completion questionnaire.

Loneliness (Subjective Disconnection)

Loneliness was assessed using three items from the Wave 22 self-completion questionnaire, drawn from a short-form measure of perceived social connectedness previously validated for use in HILDA (Lim et al. 2023; Manera et al. 2022): ‘People don’t come to visit me as often as I would like’, ‘I often need help from other people but can’t get it’, and ‘I often feel very lonely’. Each item was scored on a 7-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree); higher scores indicated greater loneliness. A continuous loneliness score was calculated as the mean of available item responses (range: 1–7). For regression analyses, respondents were dichotomised as lonely if their median item score was greater than or equal to 4, consistent with the threshold applied in previous Australian population-based research using HILDA (Lim et al. 2023; Manera et al. 2022). This threshold corresponds to respondents who did not disagree with statements indicating loneliness, and was selected because it has been shown to discriminate between chronic, episodic, and non-lonely groups in HILDA (Lim et al. 2023). Psychometric properties of the scale, including internal consistency, have been reported by Manera et al. (2022).

Social Isolation (Objective Disconnection)

Social isolation was assessed using four items from the Wave 22 self-completion questionnaire, drawn from a short-form measure of perceived social support previously validated for use in HILDA (Manera et al. 2022): ‘There is someone who can always cheer me up when I’m down’, ‘I enjoy the time I spend with the people who are important to me’, ‘When something’s on my mind, just talking with the people I know can make me feel better’, and ‘When I need someone to help me out, I can usually find someone’. Each item was scored on a 7-point Likert scale from 1 (strongly disagree) to 7 (strongly agree); because items were positively worded, lower scores indicated greater social isolation. A continuous score was calculated as the mean of available item responses (range: 1–7). For regression analyses, respondents were dichotomised as socially isolated if their median item score was less than or equal to 4, consistent with previous Australian population-based research (Lim et al. 2023; Manera et al. 2022). Psychometric properties of the scale have been reported by Manera et al. (2022).

Loneliness and social isolation were treated as co-primary outcomes to capture both subjective and objective dimensions of social disconnection.

2.3.3. Covariates

Analyses adjusted for socio-demographic and health characteristics identified in prior literature as being associated with both labour force participation and loneliness, including:

• Age (continuous);
• Gender (male/female);
• Marital status (married/de facto, separated/divorced, never married and not de facto);
• Self-assessed health status (5-point scale from poor–excellent);
• Educational attainment (tertiary, trade certificate, high school certificate, did not complete high school).

All covariates were measured at Wave 22.

2.4. Statistical Analysis

All analyses were conducted using Stata (version 19.5) (StataCorp 2025). To account for the complex survey design of HILDA, analyses incorporated person-level self-completion questionnaire weights alongside stratification and primary sampling unit identifiers. Analyses were restricted to the relevant analytic subpopulation using the ‘subpop()’ command to avoid bias associated with dropping non-eligible cases.

Descriptive statistics were used to characterise the distribution of loneliness and covariates across hidden workforce categories. Survey-weighted logistic regression models were used to estimate the association between hidden workforce status and dichotomised measures of loneliness and social isolation. A series of models were fitted for each outcome variable:

• Model 1: unadjusted;
• Model 2: adjusted for age and gender;
• Model 3: additionally adjusted for marital status;
• Model 4: fully adjusted for age, gender, marital status, self-assessed health, and educational attainment.

Models were fitted sequentially in a pre-specified, hierarchical manner with confounders entered in conceptually grouped blocks (demographic, then social, then health and education), rather than through data-driven stepwise variable selection. All variables were retained in subsequent models regardless of statistical significance.

In addition to reporting regression coefficients, adjusted predicted probabilities were derived from fully adjusted survey-weighted logistic regression models using predictive margins (Graubard and Korn [1999] 2004; Williams 2012). Predicted probabilities represent the model-estimated likelihood of loneliness or social isolation for individuals within each labour force attachment group, averaged over the observed distribution of covariates in the analytic sample. This approach was used to facilitate interpretation of absolute differences in outcome prevalence across labour force groups, as odds ratios may overestimate relative differences when outcomes are common in the study population (Graubard and Korn [1999] 2004). Pairwise differences in adjusted predicted probabilities between labour force attachment categories were computed using the ‘pwcompare’ option in Stata, generating Wald-based 95% confidence intervals around each contrast. Inference focused on the magnitude and precision of estimated differences, characterised by point estimates and confidence intervals, rather than on hypothesis testing of individual contrasts. This approach was adopted because cross-sectional observational data with modest cell sizes in some categories provide stronger support for descriptive characterisation of patterns of association than for confirmatory tests of individual contrasts (Amrhein et al. 2019; Wasserstein et al. 2019). For transparency, the full matrix of pairwise contrasts between all six labour force attachment categories (15 contrasts per outcome), with unadjusted 95% confidence intervals, is presented in Supplementary Table S3. Adjusted odds ratios from the logistic regression models are presented in Supplementary Table S4. Coefficients from the sensitivity linear regression models, fitted using the continuous loneliness and social isolation scales as outcomes, are presented in Supplementary Table S5.

As a sensitivity analysis, survey-weighted linear regression models were fitted using the continuous loneliness and social isolation scales as the outcome. Additional sensitivity analyses were conducted using an expanded analytic sample including individuals aged 50–67 years. Missing data on covariates were handled through casewise deletion, consistent with the default behaviour of survey-weighted regression in Stata when using the ‘subpop()’ command. Respondents with missing outcome data on the self-completion questionnaire were excluded at the eligibility stage rather than dropped during analysis.

3. Results

3.1. Sample Characteristics

The final analytic sample comprised 3362 respondents aged 50–64 years from Wave 22 of the HILDA Survey, drawn from 3822 respondents in scope (50–64 years), of whom 460 did not complete the self-completion questionnaire and were therefore ineligible. Detailed sample characteristics can be seen in

Table 1. Sample characteristics by labour force attachment among Australians aged 50–64 years.

Characteristic	Labour Force Attachment
	Total	In Work	Under Employed Hidden	Unemployed Hidden	Discouraged	Don’t Want Work	Other
Mean age (years)	56.7	56.1	55.6	56.3	57.2	59.1	57.1
Gender (%)
Male	51.3	54.4	17.0	62.1	38.6	38.5	26.5
Female	48.7	45.6	83.0	37.9	61.4	61.5	73.5
Marital status (%)
Married/de facto	71.8	74.5	78.1	70.1	54.3	68.3	68.6
Separated/divorced	16.3	16.0	11.1	14.0	24.9	14.8	25.1
Widowed	2.3	1.6	4.3	1.5	4.0	3.9	1.4
Never married	9.6	7.9	6.6	14.4	16.8	13.0	4.8
Educational attainment (%)
Tertiary	32.4	37.5	24.1	28.5	22.2	21.5	16.1
Trade certificate	39.3	39.5	42.3	44.8	42.5	37.0	28.5
High school certificate	9.5	8.3	17.2	9.7	11.8	11.3	7.9
Did not complete high school	18.8	14.7	16.5	17.0	23.6	30.2	47.4
Self-rated health (%)
Excellent	6.2	7.2	7.1	2.7	2.5	4.8	0.0
Very good	32.6	35.9	49.1	11.5	19.2	25.2	40.2
Good	40.1	42.9	30.3	55.0	37.5	31.7	33.1
Fair	16.7	12.6	13.4	26.7	27.4	26.8	14.4
Poor	4.4	1.5	0.0	4.2	13.5	11.5	12.4
Loneliness (%)
Not lonely	67.8	72.1	63.7	52.4	49.1	62.4	67.3
Lonely	32.2	27.9	36.3	47.6	50.9	37.6	32.7
Social isolation (%)
Not socially isolated	86.6	88.9	94.3	81.6	81.5	79.3	89.1
Socially isolated	13.4	11.1	5.7	18.4	18.5	20.7	10.9

Note: Values are weighted proportions derived from survey-adjusted analyses of HILDA Wave 22 data (2022) among adults aged 50–64 years.

.

The majority of respondents were in paid work (67.1%), while 12.5% were classified as hidden workers, including those who were underemployed (3.3%), unemployed (1.7%), or discouraged from seeking work (7.5%). A further 19.5% reported not wanting work.

Mean age varied across labour force attachment groups, ranging from 55.6 years among underemployed hidden workers to 59.1 years among those who did not want work. Hidden worker groups were also more likely to report adverse socioeconomic and health characteristics compared to those in paid employment. For example, discouraged workers were more likely to report fair or poor self-rated health (40.9% vs. 14.1%), not being partnered (45.7% vs. 25.5%), and not completing high school (23.6% vs. 14.7%) compared with those in work.

The prevalence of loneliness also varied across labour force attachment groups, with 27.9% of those in paid work reporting loneliness compared with 36.3% of underemployed hidden workers, 47.6% of unemployed hidden workers, and 50.9% of discouraged workers.

3.2. Labour Force Attachment and Loneliness

In unadjusted analyses, the prevalence of loneliness was highest among discouraged workers (50.9%) and unemployed hidden workers (47.6%), and lowest among those in paid employment (27.9%).

After adjustment for age, gender, marital status, self-rated health, and educational attainment, hidden worker subtypes continued to show elevated adjusted predicted probabilities of loneliness relative to those in paid work (

Table 2. Adjusted predicted probability of loneliness and social isolation by labour force attachment among Australians aged 50–64 years.

Labour Force Attachment	Loneliness Predicted Probability (%)	95% CI	Social Isolation Predicted Probability (%)	95% CI
In work	30.0	27.1–33.0	11.8	10.1–13.4
Underemployed hidden	40.7	29.9–51.5	8.5	1.7–15.3
Unemployed hidden	42.6	28.0–57.2	14.7	6.2–23.2
Discouraged	41.8	34.2–49.3	14.9	10.4–19.4
Does not want work	33.4	28.6–38.2	18.3	13.8–22.9
Other	29.4	8.8–50.0	10.7	−2.0–23.4

Note: Estimates derived from survey-weighted logistic regression models adjusted for age, gender, marital status, self-rated health, and educational attainment.

). The adjusted probability of loneliness among those in work was 30.0% (95% CI: 27.1–33.0), compared with 40.7% (95% CI: 29.9–51.5) among underemployed hidden workers, 42.6% (95% CI: 28.0–57.2) among unemployed hidden workers, and 41.8% (95% CI: 34.2–49.3) among discouraged workers. Across all three hidden worker subtypes, point estimates were elevated by between 10.7 and 12.6 percentage points relative to those in paid employment, with the magnitude of difference being broadly similar across categories despite varying precision.

The pairwise difference between discouraged workers and those in paid work was 11.7 percentage points (95% CI: 3.6–19.8), with a confidence interval excluding zero. The pairwise differences for underemployed and unemployed hidden workers were similar in magnitude (10.7 and 12.6 percentage points, respectively) but estimated with less precision due to smaller analytic cell sizes (underemployed n = 110; unemployed n = 57), with confidence intervals that included zero. The full matrix of pairwise contrasts is presented in Supplementary Table S3.

3.3. Labour Force Attachment and Social Isolation

Overall, 13.4% (95% CI: 11.8–15.1) of respondents were classified as socially isolated. In fully adjusted models (Table 2), the adjusted predicted probability of social isolation among those in paid work was 11.8% (95% CI: 10.1–13.4), 8.5% (95% CI: 1.7–15.3) among underemployed hidden workers, 14.7% (95% CI: 6.2–23.2) among unemployed hidden workers, and 14.9% (95% CI: 10.4–19.4) among discouraged workers. The highest adjusted predicted probability of social isolation was observed among those not in the labour force who did not want work at 18.3% (95% CI: 13.8–22.9).

The pairwise difference between those who did not want work and those in paid employment was 6.6 percentage points (95% CI: 1.8–11.3), with a confidence interval excluding zero. The pairwise difference between those who did not want work and underemployed hidden workers was 9.8 percentage points (95% CI: 2.0–17.6), also with a confidence interval excluding zero. Differences between hidden worker subtypes and those in paid employment were small in magnitude and estimated with confidence intervals that included zero. The full matrix of pairwise contrasts is presented in Supplementary Table S3.

3.4. Sensitivity Analysis

Two sensitivity analyses were conducted to assess the robustness of the main findings. First, the analytic sample was extended to include adults aged 50–67 years, capturing individuals approaching or recently reaching Age Pension eligibility (n = 3916). Survey-weighted logistic regression models were re-fitted with this extended sample. Results were directionally consistent with the main analysis (Supplementary Table S4). The adjusted predicted probability of loneliness among those in paid work was 29.3% (95% CI: 26.5–32.2), with elevated probabilities across hidden worker subtypes: 38.7% (95% CI: 28.4–49.1) among underemployed hidden workers, 42.4% (95% CI: 28.4–56.4) among unemployed hidden workers, and 39.8% (95% CI: 33.1–46.5) among discouraged workers. For social isolation, the adjusted predicted probability among those in paid work was 11.1% (95% CI: 9.2–13.0), with the highest probability again observed among those who did not want work (20.7%; 95% CI: 14.9–26.5). Predicted probabilities for hidden worker subtypes were 5.7% (95% CI: 1.4–10.0) among underemployed hidden workers, 18.4% (95% CI: 7.9–28.9) among unemployed hidden workers, and 18.5% (95% CI: 12.1–24.9) among discouraged workers.

Second, survey-weighted linear regression models were fitted within the main 50–64 sample, using the continuous loneliness and social isolation scales as outcomes in place of the dichotomised indicators (Supplementary Table S5). Adjusted differences in continuous loneliness scores were positive across hidden worker subtypes relative to those in paid work (underemployed β = 0.56, SE = 0.20; unemployed β = 0.76, SE = 0.29; discouraged β = 0.51, SE = 0.17), with confidence intervals excluding zero. For social isolation, where lower scores indicate greater social isolation, coefficients for hidden worker subtypes were small in magnitude with confidence intervals including zero (underemployed β = −0.14, SE = 0.13; unemployed β = −0.43, SE = 0.22; discouraged β = −0.13, SE = 0.10). Adjusted social isolation scores were lower among those who did not want work than among those in paid employment (β = −0.19, SE = 0.08; CI excluding zero), consistent with the direction of association observed in the main dichotomised analysis. Across both sensitivity analyses, the overall pattern of associations was consistent with the main results.

4. Discussion

This study examined the association between labour force attachment and both subjective (loneliness) and objective (social isolation) dimensions of social disconnection among Australians aged 50–64 years. The findings indicate that labour market detachment in later working life is more strongly associated with subjective experiences of disconnection than with objective social isolation. Adjusted predicted probabilities of loneliness were consistently elevated across hidden worker subtypes—including underemployed, unemployed, and discouraged workers—relative to those in paid employment, with point estimates approximately 10–13 percentage points higher after adjustment for sociodemographic and health characteristics. In contrast, differences in social isolation between these groups and those in paid work were small in magnitude and estimated with imprecision. These findings are consistent with the interpretation that labour market marginalisation may be more closely linked to perceived social belonging and role legitimacy than to the structural availability of social contact. Together, the results highlight the importance of understanding labour market participation not only as an economic determinant of wellbeing but also as an institutional context within which experiences of social connection take shape.

The differentiated pattern of associations across labour force attachment subtypes suggests that the relationship between marginalisation from paid work may not operate uniformly across subjective and objective dimensions. Underemployed workers had elevated loneliness point estimates without corresponding elevation in social isolation. Unemployed hidden workers showed similarly elevated loneliness, alongside modest point-estimate increases in isolation. These patterns are descriptive of the present cross-sectional data and are open to multiple interpretations, which we consider in turn. One interpretive possibility, consistent with social capital frameworks, is that employment environments provide opportunities for bonding and bridging social capital through routine interactions with colleagues, shared activities, and participation in organisational life (Tsounis et al. 2023). Within this framework, partial or full detachment from these institutional contexts may coincide with diminished perceived belonging and social recognition even when interpersonal networks outside work remain relatively stable. Alternative interpretations are also possible. Selection effects, in which individuals experiencing higher loneliness self-select into reduced labour force participation, cannot be ruled out with cross-sectional data. Unmeasured confounding by mental health status, prior occupational history, financial strain, or other factors influencing both labour force engagement and perceived social connection may also contribute to the observed pattern. The data analysed here do not adjudicate between these accounts.

Discouraged workers—those who have ceased active job search despite conditional willingness to work—showed the most precisely estimated elevation in loneliness relative to those in paid employment. One interpretation, drawing on cumulative disadvantage frameworks in life-course epidemiology (Dannefer 2003; Ferraro and Shippee 2009), is that individuals in this group may have experienced repeated barriers to labour market participation across the later working life course, including age-related employment barriers, skills or credential mismatches, health limitations, or unsuccessful job search experiences. Such barriers have been theorised to accumulate over time, reshaping individuals’ relationships to labour market institutions and the social roles they confer (Phillipson 2019). This interpretation is consistent with the elevated loneliness observed in this group but cannot be adjudicated with the present data; longitudinal designs would be needed to examine whether the relationship between accumulated labour market barriers and loneliness reflects such cumulative processes.

In contrast, individuals classified as not wanting work exhibited the highest adjusted probability of social isolation, but comparatively lower predicted probabilities of loneliness. The ‘does not want work’ category is heterogenous and likely includes individuals who are retired, experiencing health limitations, undertaking caring responsibilities, or otherwise outside the labour force due to functional or contextual constraints. For many individuals in this group, reduced labour force participation may co-occur with broader limitations in mobility, health or time availability that also restrict participation in community and social activities. Evidence from retirement research suggests that transitions out of paid work are complex and have been associated with changes in social connection mediated by shifts in daily routine, identity, and social participation (Vigezzi et al. 2025), with associations varying across contexts and subgroups (Hagani et al. 2024). On this interpretation, the elevated social isolation observed in this group may reflect reduced opportunities for participation across multiple domains, rather than solely the absence of an occupational role. The pattern observed for hidden worker subtypes, where loneliness was elevated but isolation was not, is consistent with a different configuration of disconnection—one where institutional exclusion from paid employment despite the potential desire or capacity to participate, may co-occur with subjective experiences of disconnection more than with reductions in social contact. Notably, the contrast in adjusted predicted social isolation between those who did not want work and underemployed hidden workers (difference: 9.8 percentage points, 95% CI: 2.0–17.6; Supplementary Table S3) was even more pronounced than the contrast against those in paid work, reinforcing the interpretation that the does-not-want-work group sits qualitatively apart from the other groups on this outcome.

An interpretive framework that aligns with the observed pattern is one in which loneliness among hidden worker subtypes is understood as reflecting diminished institutional integration rather than diminished interpersonal opportunity. In Berkman and Glass’s influential conceptual model, macro-social structures are theorised to condition the formation of social networks, which in turn relate to health via psychosocial, behavioural, and physiological pathways (Berkman et al. 2000). Within this framing, employment functions not merely as an economic exposure, but as a socially organising institution that confers normative roles, daily structure, and routine participation in collective life. The latent deprivation model, drawn from work and unemployment theory, proposes that work provides time structure, social contact, collective purpose, status, and activity, and that the withdrawal of these ‘latent functions’ may be associated with poorer wellbeing independently of income loss (Paul et al. 2023). The patterns observed in the present analysis are consistent with predictions from these frameworks but do not test them directly; cross-sectional, observational data cannot establish that institutional detachment generates loneliness, only that the two co-occur in the present sample in ways the frameworks would anticipate.

From this perspective, it is plausible that many hidden workers retain family ties and some degree of community contact, yet experience loneliness alongside the loss of role-based belonging. The cross-sectional pattern reported here is broadly compatible with this interpretive account, though we emphasise again that the temporal ordering of these associations cannot be established with the present data. From a policy standpoint, the framework suggests that labour market participation may function as one mechanism through which institutional belonging is sustained in later working life; when access to the labour market is disrupted, either through poor job matching, barriers to hiring, credentialism, discrimination, or health constraints, subjective experiences of disconnection may be the proximate domain in which this disruption is registered.

This observed divergence between loneliness and social isolation across labour force attachment categories highlights the importance of examining these two related but distinct concepts together. In fully adjusted models, adjusted predicted probabilities of loneliness were elevated across hidden worker subtypes, while differences in social isolation relative to those in paid work were small in magnitude. This pattern is consistent with the possibility that, in later working life, labour market marginalisation is more closely related to the meaning and perceived legitimacy of social roles than to the structural availability of social contact. Individuals who are marginally detached from employment may continue to maintain family relationships, friendships, and community ties, while still reporting loneliness, particularly if they perceive a loss of social recognition, purpose, or institutional belonging. On this view, loneliness may be particularly sensitive to disruptions in socially valued roles, whereas social isolation more directly reflects the quantity or frequency of social interaction.

The hidden workforce construct, prominently articulated by Fuller et al. (2021) as individuals who are willing and able to work but excluded by structural barriers, helps clarify why such divergence might occur. Hidden workers, by definition, are not necessarily socially withdrawn; rather, they are institutionally sidelined. In the Australian context, Lee et al. (2025a) demonstrate that hidden worker groups can be identified in HILDA and differ meaningfully from standard ‘unemployed’ categories, underscoring how conventional labour statistics can understate the scale and heterogeneity of labour underutilisation. The present findings extend this work by suggesting that institutional sidelining may be linked to loneliness through perceived status loss, reduced identity continuity, or weakened sense of collective purpose. These are interpretive pathways that do not require sufficient parallel reductions in contact frequency or participation to register on a social isolation index. If loneliness is responsive to perceived belonging, recognition, and role legitimacy, then interventions focused only on increasing social contact may not address the institutional mechanism hypothesised here.

4.1. Strengths and Limitations

This study has several strengths relevant to both research and policy. The use of nationally representative HILDA data supports generalisability to older working-age Australians and enables a nuanced categorisation of labour force attachment beyond the common employed/unemployed dichotomy. This is especially important given evidence that hidden worker groups are socio-demographically distinct and not well captured in standard labour force reporting (Lee et al. 2025a). The presentation of adjusted predicted probabilities improves interpretability for policy and practice audiences, particularly when outcomes are common and odds ratios can exaggerate perceived differences (Graubard and Korn [1999] 2004).

Several limitations should also be acknowledged. First, the cross-sectional design precludes causal inference and cannot resolve directionality. The associations reported here are open to multiple temporal interpretations: loneliness may contribute to labour market detachment via reduced job search efficacy, health-related withdrawal or related pathways; labour market detachment may co-occur with loneliness; or both may be jointly shaped by underlying factors not captured in the present analysis. Selection effects, in which individuals experiencing higher loneliness self-select into reduced labour force participation, are a particular concern for cross-sectional analyses of this kind and cannot be ruled out with the present data.

Second, residual confounding remains a possibility despite adjustment for key sociodemographic and health covariates. Unmeasured factors including mental health status, functional limitations, caregiving intensity, prior occupational history, job quality, financial strain, and place-based labour market context could plausibly influence both labour force attachment and social disconnection, and the present design cannot exclude their contribution to the observed associations.

Third, loneliness and social isolation were operationalised using items available within HILDA rather than the most widely used standalone scales (Maes et al. 2022), such as the UCLA Loneliness Scale (Russell et al. 1980). Although the HILDA-derived measures have been psychometrically evaluated in Australian population-based research (Manera et al. 2022) and applied in previous longitudinal work (Lim et al. 2023), the use of non-standard scales may limit cross-study comparability and introduce some measurement error. Choice of dichotomisation thresholds may also influence the magnitude of estimated differences across labour force categories; threshold sensitivity was partially addressed through the use of both continuous and dichotomised outcome measures (Supplementary Table S5).

Fourth, gender composition varied substantially across labour force attachment categories. Underemployed hidden workers were predominantly female (83.0%), consistent with the concentration of part-time work, caregiving responsibilities, and welfare-related constraints among women in later working life. By contrast, unemployed hidden workers were predominantly male (62.1%), potentially reflecting different pathways into unemployment among older men, including job loss in declining industries. Although models adjusted for gender, this adjustment cannot capture potential effect modification, and pooled estimates may obscure gender-specific pathways into and experiences of hidden workforce participation. Future research using stratified or intersectional approaches would be valuable to clarify how gender shapes these associations.

Finally, because outcomes were derived from the self-completion questionnaire, selection bias may arise if socially disconnected individuals are less likely to complete this component, potentially leading to underestimation of prevalence and associations. The HILDA self-completion questionnaire response rate among in-scope respondents in the present analysis was approximately 88%, suggesting this potential bias is unlikely to be substantial but cannot be excluded entirely.

4.2. Implications for Policy and Practice

The findings have implications for both intervention design and labour market policy that should be interpreted in light of the limitations outlined above. If loneliness among hidden workers is linked to loss of socially recognised roles and institutional integration, then interventions focused solely on network-building may be insufficient. Social prescribing is often positioned as a promising response to loneliness in primary care and community settings, although systematic review evidence indicates mixed impacts and emphasises heterogeneity across programme models and outcome measurement (Bickerdike et al. 2017; Reinhardt et al. 2021). Where social prescribing has been associated with reductions in loneliness, mechanisms appear to extend beyond increased social contact to the restoration of meaningful participation, including structured activities that provide purpose (Foster et al. 2021). This is consistent with the present pattern, in which loneliness was elevated among hidden workers despite the likely retention of family and community contacts: interventions matched to perceived belonging and meaningful role engagement may be more relevant for this group than those focused on contact frequency alone. The present results also support an established argument in gerontology and social epidemiology that loneliness and social isolation should be examined together rather than as interchangeable indicators of social disconnection given their distinct correlates and potentially distinct intervention needs (Newall and Menec 2019; Hawkley and Cacioppo 2010). Policies aimed at reducing loneliness through increased contact may be poorly matched to populations whose primary deficit is perceived belonging; conversely, individuals experiencing objective isolation may benefit from structural supports that increase opportunities for participation across multiple life domains.

Workplace and labour market policy may function as a lever for social connection among older working-age adults. Hidden workers are, by definition, an underutilised labour pool constrained by structural barriers including age-related employment barriers, skills or qualification mismatches, and health constraints (Lee et al. 2025a; Fuller et al. 2021). For this population, policies supporting flexible work, job redesign, and supportive organisational climates may simultaneously sustain employment participation and protect social integration during a life stage where health limitations and caring responsibilities become more common (Vanajan et al. 2020; Pak et al. 2019). Evidence from studies of older workers with chronic health conditions indicates that flexible working arrangements and psychologically safe organisational climates that enable workers to negotiate accommodations have been associated with sustained workforce participation and improved health outcomes (McGonagle et al. 2015; Vanajan et al. 2020). From a public health perspective, the present findings strengthen the case that ‘good work’ and inclusive employment practices may function not only as economic goals but as potential upstream supports for social connection in later working life.

4.3. Future Research Directions

Future research should examine longitudinal transitions into and out of hidden workforce status to clarify the temporal ordering of associations between labour force attachment and social disconnection in later working life, and to address the questions of selection and reverse causation that cross-sectional designs cannot resolve. Given the substantial gender heterogeneity across labour force attachment categories observed in the present analysis, stratified and intersectional analyses examining how gender, socioeconomic status, marital status, and health jointly shape pathways into hidden workforce participation would be particularly informative. Further investigation of candidate mediating constructs, including perceived role identity, opportunities for meaningful participation, and quality of social ties, would help to test the interpretive frameworks discussed above and to inform the design of interventions aimed at supporting social connection among those marginally attached to the labour market. Qualitative work exploring the lived experiences of older adults in hidden worker categories would also be valuable in elucidating how institutional belonging and social connection are experienced and negotiated in this group.

5. Conclusions

This study examined the association between labour force attachment and social disconnection among Australians aged 50–64 years, distinguishing between subjective loneliness and objective social isolation. Adjusted predicted probabilities of loneliness were elevated across hidden worker subtypes, including underemployed, unemployed, and discouraged workers, relative to those in paid employment after adjustment for sociodemographic and health characteristics. In contrast, differences in social isolation were less pronounced, with the highest predicted probability of isolation observed among those who reported not wanting work. These findings are consistent with the interpretation that, in later working life, labour market marginalisation is more closely associated with perceived belonging and role legitimacy than the structural availability of social contact.

The results underscore the importance of understanding employment not only as an economic determinant of wellbeing but also as an institutional context within which opportunities for social integration are organised. Policies and interventions aimed at addressing loneliness in later working life may benefit from recognising the role of labour market participation and meaningful social roles in supporting social connection, in addition to those targeting social contact and network size. Efforts to reduce social disconnection among older working-age adults should consider both the structural barriers that limit access to employment and the broader social functions that work provides within everyday life.