Does unemployment lead to greater levels of loneliness? A systematic review

There is evidence that loneliness and unemployment each have a negative impact on public health. Both are experienced across the life course and are of increasing concern in light of the COVID-19 pandemic. This review seeks to examine the strength and direction of the relationship between loneliness and unemployment in working age individuals, and in particular the potential for a self-reinforcing cycle with combined healthcare outcomes. A systematic search was undertaken in Medline, PubMed, PsycINFO, Embase and EconLit from inception to December 2020. PRISMA reporting guidelines were followed throughout this review, study quality was assessed using the Joanna Briggs Institute checklist and results were summarised in a narrative synthesis. English language studies evaluating the relationship between loneliness and unemployment in higher income western countries were included. Thirty-seven studies were identified; 30 cross-sectional and 7 longitudinal. Loneliness was measured by a direct question or loneliness scale while unemployment was self-reported or retrieved from a national register. A positive association between unemployment and increased loneliness was observed across all studies. Thus, across the life-course a clear yet complex relationship exists between unemployment and greater experience of loneliness. The magnitude of this relationship increases with the severity of loneliness and appears to peak at age 30–34 and 50–59. Logistic regression provided the greatest consistency at statistical significance revealing at least a 40% increase in the likelihood of reporting loneliness when unemployed. Recent longitudinal studies identified in this review found higher levels of loneliness following job loss, but also that loneliness was predictive of unemployment suggesting potential bi-directionality in the relationship. This bi-directionality may create a multiplier effect between loneliness and unemployment to form a self-reinforcing relationship and greater health concerns for those most at risk. Thus, review findings suggest the need for cross-sector awareness and intervention to tackle both loneliness and unemployment.


Introduction
Loneliness has been described as a 'public health epidemic' (Royal College of General Practitioners, 2018) and is of increasing concern in light of the current COVID-19 pandemic. Loneliness captures the deficit between desired and actual social relationships (Fulton and Jupp, 2015) and is increasingly acknowledged to be associated with health outcomes and unemployment (Leigh-Hunt et al., 2017;Matthews et al., 2018). While the literature on health outcomes related to loneliness has grown in recent years, particularly concerning older adults, understanding of the relationship between loneliness and wider socio-economic or economic factors, such as unemployment, remains limited (Leigh-Hunt et al., 2017). This remains the case despite recognition of greater loneliness in younger, working age adults (Barreto et al., 2021). While there is evidence that loneliness is detrimental to employee health (Michaelson et al., 2017), analysis has not been extended to investigate the resulting overarching relationship between loneliness and employment and does not consider the effect of loneliness on individuals that are unemployed. Thus, current understanding of loneliness in working age groups is incomplete. Furthermore, evidence of a relationship between loneliness and factors such as education and employment in adolescence and young adulthood neither extends across working age groups, nor pursues causal explanation (Matthews et al., 2018). More recently, throughout the COVID-19 pandemic, both rates of loneliness and unemployment have attracted increasing government scrutiny. A mapping of the UK illustrates how areas of higher unemployment have higher rates of loneliness during the coronavirus pandemic (Payne, 2021). Overall it is critical to understand the mechanisms and factors related to loneliness and unemployment in order to better understand the health and economic impacts of loneliness in general, as well as to help design interventions and policies to address the immediate impacts of the COVID-19 pandemic. This systematic review addresses these pressing policy questions by critically reviewing studies exploring the relationship between loneliness and employment outcomes.
Loneliness is commonly understood as the subjective or perceived experience of isolation or lack of social support. This perceived deficit can arise from reduced quantity and/or quality of social relationships Peplau, 1982). As loneliness incorporates not only quantity but also quality of social interaction and relationships, a person may experience loneliness without physical social isolation. Research into loneliness and its health related factors is largely concentrated in the older population (Holt-Lunstad et al., 2010;Leigh-Hunt et al., 2017), though the detrimental impact of loneliness can be seen across age groups. Health related consequences of loneliness have been considered in a large number of studies and review articles which observe consistently harmful effects (Heinrich and Gullone, 2006;Holt-Lunstad et al., 2010;Leigh-Hunt et al., 2017;Matthews et al., 2018;Ong et al., 2016;Petitte et al., 2015). These effects include worsened physical health through conditions such as cardiovascular disease, poorer mental health including depression and anxiety, lower wellbeing, negative health behaviours, increased incidence of suicide, and up to 50% increased likelihood of mortality in individuals experiencing loneliness. Overall loneliness has an impact on health comparable to that of smoking and greater than that of obesity and physical inactivity (Holt-Lunstad et al., 2010). There is increasing evidence of loneliness in younger populations with both young and middle-aged people reporting more loneliness than older people (Barreto et al., 2021). Loneliness in working age adults brings additional concerns, potentially impacting educational attainment, employment outcomes and earnings, alongside those effects established in health (Matthews et al., 2018).
It is conceivable that there could be not only an effect of unemployment on loneliness through aspects such as enforced isolation or diminished sense of belonging, but also an effect of loneliness on unemployment through mechanisms such as reduced motivation, performance, or productivity, in addition to poorer health and emotional recognition (Bruehlman-Senecal et al., 2020;Matthews et al., 2018;Michaelson et al., 2017;Ozcelik and Barsade, 2018). The impact of unemployment on health and wellbeing has been investigated in a large number of studies, including review articles, all of which evidence a detrimental effect of unemployment on health (Dooley et al., 1996;Herbig et al., 2013;Jin et al., 1997;Norström et al., 2014;Paul and Moser, 2009;Wanberg, 2012;Wilson and Walker, 1993). These consequences include mental health focused on depression and anxiety, overuse of healthcare resources through increased frequency of physician visits, suicide, and substance abuse including smoking and alcohol. There is also recent evidence that unemployment contributes to poorer quality of life (Norström et al., 2019). These health and socio-economic effects of prolonged unemployment are likely to be persisting (Wadsworth et al., 1999), while loneliness too has both short-and long-term effects on health and well-being (Cacioppo et al., 2003). These inter-related health and socio-economic factors bring complexity to the relationship between loneliness and unemployment. Aspects such as anxiety, depression, optimism, self-esteem, sense of belonging, life satisfaction, coping strategies, health and job-seeking behaviours, and educational attainment and qualifications are just a few of these potentially inter-related or mediating factors which may explain the connection between loneliness and unemployment (Matthews et al., 2018;Payne, 2021).
Given the persistence of both loneliness and unemployment, improved understanding of the mechanisms and factors related to each is key in improving public health across the life course. With increased incidence of both loneliness and unemployment as a result of COVID-19, and pre-existing but often overlooked evidence of a connection between the two experiences, greater understanding of the evidence for the relationship between loneliness and unemployment is required. To date no review has been conducted to untangle this complex, and possibly inter-related or even self-reinforcing (Bruehlman-Senecal et al., 2020) relationship between loneliness and employment. This review seeks to address this gap and develop an understanding of the evidence, strength and direction of the relationship between loneliness and unemployment. Where a relationship is suggested this review will better equip healthcare providers, public policy and employers to support individuals and instigate safeguards to mitigate the impact of loneliness and employment outcomes.

Methods
This systematic review follows the PRISMA reporting guidelines (Moher et al., 2009). Given the distinct nature of loneliness from social isolation (Heinrich and Gullone, 2006;Peplau, 1982) social isolation was not considered simultaneously in this review.

Identification of studies
Following a brief scoping of existing reviews in loneliness (Boss et al., 2015;Deckx et al., 2018;Dyal and Valente, 2015;Lim et al., 2018;Petitte et al., 2015) and in unemployment (Milner et al., 2013;Norström et al., 2014;Sumner and Gallagher, 2017) we developed a search strategy to identify studies considering any relationship between loneliness and unemployment. The term lonel* was combined with employ* OR unemploy* OR job OR work OR labor/labour to search in title and abstract. We searched the following electronic databases from their inception to December 07, 2020: Embase(Ovid), PsycINFO(Ovid), EconLit(EBSCOhost), MEDLINE(EBSCOhost) and PubMed(NCBI). Titles and abstracts obtained from searches were screened by one reviewer (NM) and potential includes screened by a second reviewer (AML). Full articles were screened for eligibility by one reviewer (NM) and agreed by the second reviewer (AML). Forward and backward citation chasing was conducted on studies identified as included at full-text. These citations were screened following the same process outlined for database searches. Any disagreement was resolved by consensus. Screening was conducted using EndNote X9.

Inclusion criteria
Studies had to evaluate the relationship between loneliness and unemployment. Loneliness was required to be explicitly identified and measured in the analysis. Related concepts such as social isolation or other objective measures were considered insufficient. Studies were to be published in the English language; consider the working age population (aged 16 to 65); and be set in higher income western countries in Europe, North America and Oceania. These were defined by the World Bank as those with a Gross National Income per capita above $12,375 (The World Bank, 2020), and included Europe, Canada, USA, Australia and New Zealand. Included studies were restricted to higher income western countries, in Europe, North America and Oceania, in order to ensure greater consistency in demographics, societal structures, and health and social care systems. Studies were included even if they used data from the same source or sample of respondents. All types of quantitative and qualitative study design were considered.

Exclusion criteria
Studies considering specific populations, such as immigrants or twins, specific occupations, such as dentists, or participants with existing medical conditions, were excluded. Studies were omitted if one or more of the eligibility criteria were not met or if the studies were: methodological, commentaries, letters, editorials or abstracts.

Data extraction
Standardised data forms were used to extract the relevant information from each of the included studies. The data extracted included: author, year, country of data, sample size, age, gender, data source, study design, measurement of loneliness and unemployment, methodology, and key statistical findings. One reviewer (NM) extracted the data and the information extracted was checked by a second reviewer (AML). Results were described in a narrative synthesis and the possibility to perform meta-analysis was explored taking into account the heterogeneity of the studies included with respect to study design and the measures of unemployment and loneliness. Critical appraisal was conducted in line with the Joanna Briggs Institute (JBI) Checklists for Cross-Sectional Studies and for Cohort Studies (Moola et al., 2020). Studies were appraised by one reviewer (NM) and checked by a second (AML). Critical appraisal highlighted any shortcomings or areas of concern within the included papers to ensure balance in the interpretation of findings and highlight areas for improvement in future study on this topic. Consideration was given to any paper with a large number of clear shortcomings before being included in the synthesis.

Search results
The electronic searches yielded 4740 hits. When duplicates were removed 2349 unique hits remained. Of these 2281 were excluded through title and abstract screening. Full texts were obtained for 68 studies to check for eligibility. After reviewing full texts 22 studies met the inclusion criteria. Screening of the reference lists added 15 additional papers with a total of 37 studies were included in this review of which 30 were cross-sectional and 7 longitudinal studies. Further details of the study selection process are presented in the PRISMA diagram (Fig. 1).
The opportunity for meta-analysis was explored, however not deemed possible based on the level of heterogeneity between studies in measurement of both unemployment and loneliness (exposures and outcomes) and in study design. While 75% (28/37) studies measured loneliness by a multi-item scale and so could have been combined in meta-analysis, only 30% (11/37) of studies were comparable in terms of employment status measure, being binary employed vs unemployed. The remaining 70% (26/37) of studies had varying definitions and categories defining employment status such that they could not be combined. Furthermore, given the variety in study design, statistical methods, confounders and reporting of results, studies were also not able to be combined by study design. Thus, overall, narrative synthesis was better suited to both manage measurement and methodological heterogeneity, and to capture the complexity of the relationship under consideration.

Critical appraisal
Quality varied across included studies. Of the cross-sectional studies (n = 30) only three fully met the quality criteria with at least one N. Morrish and A. Medina-Lara Social Science & Medicine 287 (2021) 114339 criterion unclear or not fulfilled in the remaining 27 studies. Most crosssectional studies (29/30) clearly defined their inclusion criteria, though often this was dictated by the chosen data source. The most common area lacking clarity for cross-sectional studies was the identification of confounding factors and the stated strategies to deal with these. Five studies failed to describe study subjects in detail while a further three were unclear. Overall, ten cross-sectional studies had at least one category in which they clearly fell short of the desired quality. Of those, eight studies reported percentages, means or correlation coefficients to indicate a relationship between loneliness and unemployment. Throughout the review, less weight is given to these studies in understanding the relationship such that minor shortfalls in quality did not indicate reason for exclusion. The remaining two cross-sectional studies with a clear 'no' in response to the JBI checklist utilised regression analysis and only fell short in description of study subjects thus did not warrant exclusion. Authors described the location and year of the study but demographic details were not provided.
None of the seven cohort studies fully met the quality criteria. Strategies to deal with confounding variables and whether participants were free from exposure at the start of the study were often unclear. Four studies had clear shortcomings in quality, largely surrounding incomplete follow-up without explanation or exploration of this fact. One of these studies  also did not employ any strategy to address the incomplete follow up. This poor reporting was not considered grounds for exclusion, however caution is taken throughout this review in presenting a balanced overview of causality and the possibility of bi-directionality providing rationale for further targeted research in this area. Overall, no studies were excluded based on quality appraisal as no shortcomings were considered of critical concern to this review. A summary of study quality for cross-sectional and cohort studies is provided in Tables 1 and 2 respectively.

Study characteristics
Almost 90% (33/37) of included studies were undertaken in a single country in Europe (n = 22), USA (n = 7), Australia (n = 3) or New Zealand (n = 1). The remaining four studies had a multi-country perspective. Sample size ranged from 114 to 60,341 with seven studies using fewer than 1000 participants. All studies focused on the working age population. One study (von Soest et al., 2020) also included adolescents and many included individuals aged over 65. In these cases mean age remained between 16 and 65. Where reported (20/37), mean age ranged from 21 to 63 years old. Over 65% (25/37) of studies had between 40% and 55% male populations, six studies did not provide descriptive statistics by gender, and six studies were over represented by women with at least 60% of participants identifying as female. The majority (27/37) of studies were carried out using existing national or international databases while 10 studies collected their own data. A large proportion (38%, 14/37) of the studies, included in this review were published in 2020 of which five considered the impact of COVID-19.

Measurement of loneliness
Measurement of loneliness is explicitly defined in all but one study (Bak et al., 2012). Twenty-six studies measured loneliness through an established loneliness scale, one through a scale designed for the survey (Hawkins-Elder et al., 2018), and twelve through a single-item question. Of these, three studies (Bu et al., 2020b;von Soest et al., 2020;von Soest et al., 2018) utilised both a single-item question and a loneliness scale, which is considered the 'gold standard' by the UK Office for National Statistics (Osborne et al., 2018). Loneliness scales are considered an indirect measure as they do not explicitly use the term 'loneliness' but rather ask about certain feelings that implicitly capture loneliness, such as how often an individual may feel left out, miss the company of others or whether they have people they can trust (Shiovitz-Ezra and Ayalon, 2011). As such, multi-item scales do not require the respondent to openly admit feeling lonely. Meanwhile single-item questions involve a direct question about the respondents' experience or feeling of loneliness and as such require personal recognition of the experience. In this review 13 studies opted for a form of UCLA Loneliness scale (Russell et al., 1978), 9 studies for a form of De Jong Gierveld scale (de Jong-Gierveld and Kamphuls, 1985), 2 studies used the SELSA (Social and Emotional Loneliness Scale for Adults) (DiTommaso and Spinner, 1993), and 2 the Hughes short scale for measuring loneliness (Hughes et al., 2004). The UCLA loneliness scales varied from 20-item to 3-item versions. The UCLA scale evaluates individual's feelings of loneliness and isolation, and traditionally uses four response categories: never; rarely; sometimes; often (Russell et al., 1978). Studies utilising the De Jong Gierveld scale varied from the more extended 11-item to the 3-item version, and in one case the short scales which evaluate emotional and social loneliness separately were used. The De Jong Gierveld has three response options: no; more or less; yes (de Jong-Gierveld and Kamphuls, 1985). Of those using direct questions to measure loneliness, most (11/12) provided at least 3 response options, most commonly including a selection from; never, rarely, sometimes, often, always. Of these, three studies recoded into a binary format for analysis (Algren et al., 2020;Emerson et al., 2020;Page and Cole, 1991). The remaining study (1/12) provided binary response to feeling of loneliness (Stankunas et al., 2009).

Measurement of employment status
Employment status was either self-reported in a survey (n = 31) or assessed by national register (n = 2). Additionally, one study used both register and survey data (Lasgaard et al., 2016), one included only unemployed individuals (Stankunas et al., 2009), and two studies did not report how data on employment status was collected (Beutel et al., 2017;Tong et al., 2019). Over half of the studies (19/37) included further data collection categories alongside a measurement for being employed or unemployed. These additional categories included being in education; sick/disability; retired; homemaker; economically inactive; benefit claimant; part-time work; military volunteer; rehabilitation; incapacitated.

Study design
Over 80% (30/37) were cross-sectional studies while the remaining 7 used longitudinal data. A third of the studies (13/37) applied weights to account for missing data and/or ensure a representative population. Loneliness was predominantly (80%, 30/37) used as the dependent variable in analysis. Unemployment acted as dependent variable in only 2 studies (Morris, 2019;von Soest et al., 2020). Of the 34 studies utilising regression methodology only 24 used regression methods to evaluate the relationship between loneliness and unemployment. Where conducted, regression standardisation or adjustment changed neither the direction nor statistical significance of the coefficient and resulting relationship identified. The remaining studies (10/34) reported descriptive statistics or correlations for this relationship and used regression for analysis elsewhere. Of the longitudinal studies included, two did not seek to understand directionality (Kalil et al., 2010;Pagan, 2020), three considered the causal impact of employment status on loneliness (Bu et al., 2020a;Buecker et al., 2020;von Soest et al., 2018) and two the reverse of loneliness on employment outcomes (Morris, 2019;von Soest et al., 2020). Further details of study characteristics, data collection and analysis ordered by date of publication can be found in Table 3.

Table 2
Critical appraisal of cohort studies.

Key findings
The papers can be broadly categorised into exploring the: 1) descriptive statistics on loneliness and unemployment; 2) relationship between loneliness and employment; 3) relationship between loneliness and unemployment; 4) potential for bi-directional relationship; 5) differential effects of impacts by age and gender, and 6) impact of COVID-19 on loneliness and unemployment. Descriptive statistics provide an outline of underlying associations expanded in later sections. Synthesis of the relationship between loneliness and employment is separated from that of unemployment to reflect the differences in experience or onset of loneliness across individuals in and out of work. Where papers consider more than one research category they are included within both categories.

Descriptive statistics on loneliness and unemployment
Most studies (22/37) reported descriptive statistics to indicate experience of loneliness or employment status. These included either the percentage of individuals experiencing loneliness and unemployment or employment (n = 11), mean values (n = 6), or correlation coefficients (n = 5). Of these 22 studies, nine simultaneously reported a linear or logistic regression coefficient. The remaining studies (15/37) presented only regression coefficients (linear, logistic or probit) without this descriptive information.
Across the 11 studies reporting percentages, feelings of loneliness were reported in between 11.1% and 70.2% of unemployed individuals with lower percentages in those who were employed (4.4%-43.3%). Two studies found rates of unemployment rose with severity of loneliness. Unemployment rose from 36.1% in those feeling slight loneliness to 45.5% when severe (Beutel et al., 2017); or from 22.1% with low loneliness to 29.2% with high loneliness (Hawkins-Elder et al., 2018). Four studies (Algren et al., 2020;Emerson et al., 2020;Kearns et al., 2015;Li and Wang, 2020) distinguished individuals who were 'often' lonely. Here between 4.4% and 10.2% of employed individuals experienced loneliness 'often', compared to between 8.18% and 15.8% of those unemployed. By extending the definition of loneliness to include 'fairly often' and 'sometimes' the prevalence of loneliness, as measured by percentage, increased up to fourfold. In the 5 studies evaluating the correlation between loneliness and unemployment statistical significance was achieved in 3 studies with correlation coefficients identified between 0.06 (p ≤ 0.001) (von Soest et al., 2018) and 0.49 (p ≤ 0.001) (Tong et al., 2019), indicating an increase in loneliness corresponds to an increase in unemployment. However, this relationship is only weak in almost all studies.

Relationship between loneliness and employment
Twelve studies used regression analysis to consider the relationship between loneliness and being employed or in paid work. Of these, only one study attempted casual analysis through use of a longitudinal dataset . In evaluating the relationship between loneliness and being actively employed, linear regression analysis indicated a negative relationship. Four studies yielded negative linear regression coefficients at 5-percent statistical significance. These linear regression coefficients ranged between b = − 0.03 (p ≤ 0.05) (Fokkema and Naderi, 2013) and b = − 0.218 (p ≤ 0.001) (Lykes and Kemmelmeier, 2014). A further study achieved statistical significance only for a specific subgroup (romantic loneliness) with linear regression coefficient b = − 0.14 (p ≤ 0.001) (Neto, 2015). Conversely, only one study identified positive linear regression coefficients (Luhmann and Hawkley, 2016). For both positive and negative linear regression coefficients the magnitude of the relationship between loneliness and paid work was greater in full-time than part-time employment (Luhmann and Hawkley, 2016;Pagan, 2020). Five linear regressions presented negative coefficients but did not achieve statistical significance, which was also the case in causal analysis . Logistic regression also suggested reduced experience of loneliness amongst the employed. One study evaluated loneliness and being in employment through multivariate logistic regression and found middle-aged adults in paid work were less likely to be lonely (AOR = 0.71, 95%CI[0.56-0.90]) than their unemployed counterparts. This was reinforced by logistic regression on a sample with mean age 45 which yielded odds ratio OR = 0.73 (95%CI [0.56, 0.94]).

Relationship between loneliness and unemployment
The relationship between loneliness and unemployment was explored through regression analysis in 14 studies. Most studies used cross-sectional data (9/14) while five attempted causal analysis on longitudinal data. Evaluating the relationship between being unemployed and experiencing loneliness, linear regression revealed a statistically significant relationship across the whole sample in only one study (Bu et al., 2020b). Otherwise statistical significance was achieved only in specific subgroups with linear regression coefficients: b = 0.04 (p ≤ 0.01) when considering initial participant status with a loneliness scale as an indirect measure (von Soest et al., 2018); b = 0.31 (p ≤ 0.05) using a direct question for loneliness (von Soest et al., 2020); b = 0.11 (p ≤ 0.001) for only the social loneliness subscale (Neto, 2015); b = 0.338 (p ≤ 0.05) for recent unemployment in a non-USA subsample (Wei, 2020). Meanwhile most studies (4/5) using logistic regression found statistically significant results (Bjelajac et al., 2019;Lasgaard et al., 2016;Li and Wang, 2020;Morris, 2019). One of these studies (Morris, 2019) considered unemployment as the outcome variable in longitudinal data revealing odds ratio OR = 1.252 (p ≤ 0.001) suggesting a 25.2% increase in the likelihood of reporting unemployment, or as the authors describe 'work disability', when experiencing loneliness as measured by the UCLA 3-item scale. The remaining studies with statistically significant odds ratios used loneliness as the dependent variable (Bjelajac et al., 2019;Lasgaard et al., 2016;Li and Wang, 2020). They found at least 40% greater odds of reporting loneliness when unemployed based on odds ratios between OR = 1.40 (p ≤ 0.001) and OR = 2.81 (p ≤ 0.05), reaching to OR = 6.93 (p ≤ 0.05) when restricted to experience of severe loneliness. This finding of a greater relationship between loneliness and unemployment in the presence of severe loneliness is reinforced by growth mixture modelling (Bu et al., 2020a).

Potential for bi-directional relationship
Seven studies utilised a longitudinal dataset. Of these, five studies conducted causal analysis in the evaluation of loneliness and unemployment suggesting potential bi-directionality in the relationship. Two studies found loneliness to predict later life unemployment. Morris identified loneliness as predictive of work disability through logistic regression with odds ratio OR = 1.252 (SE = 0.062) (Morris, 2019) while probit regression found loneliness in adolescence and young adulthood predicted higher midlife unemployment with probit regression coefficient b = 0.31, p ≤ 0.05 when using a direct question to measure loneliness (von Soest et al., 2020). Conversely, propensity score matching with a generalized additive model  revealed higher reported loneliness following a job loss with linear regression coefficient b = 0.314, p ≤ 0.05, 95%CI[0.111, 0.516]. Unemployment was also observed to positively predict change (von Soest et al., 2018) and growth trajectory (Bu et al., 2020a) of loneliness over time, though without statistical significance. Overall, studies exploring causal inference found higher levels of loneliness following a job loss , but also that loneliness was predictive of unemployment in later life (Morris, 2019;von Soest et al., 2020), suggesting a bi-directional relationship.        Unemployed with no paid work purported significantly higher levels (p < 0.01) than the unemployed with regular paid work. Occupation made significant contribution in predicting social loneliness: F (3,144) = 3.83, p < 0.05 and accounted for 7% of the variance.  Wei (

Impact of COVID-19 on loneliness and unemployment
Five studies, three in the UK (Bu et al., 2020a, b;Li and Wang, 2020), one in Norway (Hoffart et al., 2020), one across multiple countries (Wei, 2020), were conducted during the COVID-19 pandemic. In Norway, the relationship between loneliness and paid work during the COVID-19 pandemic was indicated by linear regression coefficient b = − 1.36, but did not achieve statistical significance at the 5% level (Hoffart et al., 2020). One UK based COVID-19 study (Bu et al., 2020b) used linear regression on loneliness and unemployment data from two different surveys. While unemployment remained a persistent risk factor, analysis revealed a drop in the association between loneliness and unemployment from linear regression coefficient b = 0.65 (95%CI [0.5-0.8]) before COVID-19 to b = 0.40 (95%CI[0.25-0.55]) during. Through hierarchical regression, the multi-country study (Wei, 2020) observed a positive and statistically significant relationship between loneliness and recent unemployment due to COVID-19 in non-USA countries (b = 0.338, p ≤ 0.05), but not in the USA. Further details of all key study findings are provided in Table 4.

Discussion
Interest in the relationship between loneliness and unemployment has increased in recent years with over a third of the included studies published in 2020. There is also a trend towards the use of existing national databases, which in general provides a larger and more varied sample, though at the expense of choice in the measurement of loneliness and employment status. Most studies used self-report employment status and all used a self-report loneliness measure. While loneliness is inherently subjective, and so a self-report measure appropriate for its assessment, there are limitations to self-labelled loneliness, as respondents must personally identify loneliness in themselves. This can be mitigated by using multi-item loneliness scales that capture feelings of loneliness through a wider set of indirect questions. As such, established loneliness scales are preferred over direct questioning, with the De Jong Gierveld and UCLA scales most popular. Only three studies utilised both a direct question and an established multi-item scale to measure loneliness, which is considered the 'gold standard' by the UK Office for National Statistics (Osborne et al., 2018). This review identifies a lack of targeted research into the relationship between loneliness and unemployment. Prior to 2018 inferences were limited to correlations by cross-sectional data and underuse of longitudinal datasets. The evaluation of loneliness in relation to unemployment often arises from the inclusion of covariates or controls, and so findings are frequently limited to the descriptive statistics or arise as a by-product in regression analysis. As a result studies are often limited in their evaluation of the nuance and complexity in the relationship between loneliness and unemployment. As expected, there is evidence that being in paid work may have some protection against experience of loneliness. The greatest benefit is observed in full-time employment given the magnitude of the relationship between loneliness and being employed is consistently greater for those in full-time than part-time work. This suggests the volume of time spent in the workplace can influence loneliness, perhaps due to a sense of community and belonging not always achieved outside of work, and indicates value in promoting workplace engagement for those at greater risk of feeling lonely. Furthermore, one should also consider the impact of workplace environment, including changing working patterns, working from home or frequent changes in jobs, all of which have been increasingly prominent in the COVID-19 pandemic.
Studies evaluating the connection between loneliness and unemployment are consistent in their observation that being unemployed is associated with increased loneliness, particularly with more severe experience of loneliness. Logistic regression, while not determining causality, provided statistically significant findings for the whole sample to reveal at least 40% greater likelihood of unemployed individuals experiencing loneliness (Bjelajac et al., 2019;Bu et al., 2020a;Lasgaard et al., 2016;Li and Wang, 2020) increasing to 693% for those reporting an outcome of severe loneliness (Lasgaard et al., 2016). A greater detrimental effect of more persistent and severe experience of loneliness is also observed in existing research in mental health and executive functioning (Ong et al., 2016;Peplau et al., 1985). This suggests particular importance in providing support for individuals with more severe experience of loneliness and the need to improve understanding of loneliness in relation to employment status could impact both health and economic progression.
This review also identified evidence of at least 25% higher odds of individuals reporting unemployment when lonely (Morris, 2019). However, overall, few studies analysed causal inference and a number while confirming directionality did not achieve statistical significance in evaluating the relationship between loneliness and employment status. This is likely due to the analytical methods and at times sample size available. It could also suggest the need for more focussed analysis controlling for confounders or mediators in the relationship. Employment status is again often included as a sociodemographic control variable rather than a key factor related to loneliness resulting in the relationship being observed as a by-product. This however does reduce concern over publication bias with evidence of a statistically significant relationship arising in the majority of studies. It also promotes the need for future work specifically exploring the complex relationship between loneliness and employment status.
Where statistical significance is achieved in causal inference, studies found higher levels of loneliness following a job loss , but also that loneliness was predictive of unemployment (Morris, 2019;von Soest et al., 2020). This suggests potential for a bi-directional relationship between loneliness and unemployment and implies possible self-reinforcing or self-fulfilling relationships where pressures pushing on both factors create a multiplier effect between the two concepts, alongside scope for joint prevention initiatives. This idea of a self-reinforcing loneliness cycle is consistent with existing studies in mental health and behaviour (Bruehlman-Senecal et al., 2020) and is of particular note given the increased rates of both loneliness and unemployment resulting from the COVID-19 pandemic. Due to the limited number of causal inference studies, and the variation in statistical significance which is not achieved by two studies (Bu et al., 2020a;von Soest et al., 2018), these early results should be interpreted with caution with further research needed in this area. While there is potential for direct self-reinforcement between loneliness and unemployment, it is also likely that the cycle could arise from inter-related factors. The nuance and intricacy in the relationship between loneliness and unemployment remains largely unexplored, however some included studies evidenced possible mediation effects surrounding income (Luhmann and Hawkley, 2016) and depression (Morris, 2019). Contributing or mediating factors could include a wide range of inter-related health, social and economic factors linked to both loneliness and unemployment such as anxiety, depression, wellbeing, life satisfaction, executive functioning, and health and job seeking behaviours (Matthews et al., 2018;Ong et al., 2016). Given evidence of a relationship between loneliness and unemployment highlighted in this review, future research should look to further understanding of the complex relationship between loneliness, unemployment and related factors.
This review also indicates that the differential impacts of loneliness on employment have considered age effects, with underreporting of the impacts across gender and ethnicity. The relationship between loneliness and employment status remains prevalent throughout working life. By combining the most at risk identified by Lasgaard and Franssen the greatest association is seen to occur at age 30-34 and age 50-59, consistent with Luhmann's two loneliness peaks, at age 30 and age 60. However, greater effect is observed in the youngest, age 19-34, and the oldest, age 60-75, upon model adjustment. In longitudinal analysis Buecker finds older individuals reported a lower than average increase in loneliness following change in employment status. Findings illustrate the importance of age classification, in particular the oldest (age 65-75) and youngest (age 16-21) where factors such as education and retirement could impact the amount and intensity of paid work. This could also influence the process of defining working age subsamples from the entire population.
As expected, the relationship between loneliness and employment status prevails throughout COVID-19. Evidence from Norway (Hoffart et al., 2020) shows potential for a larger magnitude in the relationship between loneliness and paid work following the onset of the pandemic. However, this study alone, particularly given its lack of statistical significance, is not sufficient for a conclusion to be drawn and as such further research is required. Perhaps unexpected are findings from Bu who did not achieve statistical significance in causal analysis (Bu et al., 2020a) and observed a reduction in the association with the onset of the pandemic (Bu et al., 2020b). There is however no consideration of the impact of the 'furlough' scheme, a temporary paid absence scheme introduced by the UK government in response to COVID-19 to prevent mass unemployment, on this finding. This is a key omission given its direct link to unemployment rates. Furthermore, data is restricted to the first three months of UK lockdown (March-May 2020), such that longer-term effects are not considered. On balance the impact of COVID-19, which may persist for a number of years, needs additional consideration. It could be expected that the impact of increased pandemic loneliness on unemployment would become increasingly evident as individuals are encouraged to go back to normalcy and job protection schemes, such as UK furlough, end. With increased incidence of both loneliness and unemployment as a result of the pandemic any association between the two remains of concern.

Limitations
In quantitative studies, which were the most common in this review, reporting guidelines profess the need to explicitly state covariates, such as loneliness and unemployment. However, lack of clarity and consensus in terminology related to both loneliness and unemployment brought limitation to this review. A number of studies may have been inadvertently missed, particularly where unemployment was not stated as a key covariate or sociodemographic control. While comprehensiveness could have been increased by inclusion of additional terms from which to infer employment, such as financial satisfaction, financial resource, work disability, deprivation, life events and sociodemographic, this comes at the expense of precision. We believe our search strategy struck a good balance between pragmatism, comprehensiveness and precision. Furthermore, the electronic search was supplemented by both forward and backward citation chasing. Although, inclusion criteria remained wide, looking for evidence of both loneliness on unemployment and unemployment on loneliness, this provided a more comprehensive overview of the existing literature and greater insight into the complexity of the relationship.
While the potential for meta-analysis was considered in this review, it was concluded that the included studies had insufficient methodological homogeneity to synthesise the results in a meta-analysis. Thus, narrative synthesis was utilised to best capture the nuance of the existing literature on loneliness and unemployment. Included studies contained large variation in study design and reporting of results for standardisation, possibly a result of the lack of targeted research in this area to date. Existing meta-analyses on unemployment and mental health report the need for consistent categories of unemployment, with preference for binary classification to avoid misrepresentation of non-employed persons not seeking work such as students, homemakers or retired individuals (Paul and Moser, 2009). However, studies included in this review contained a wide range of categories for employment status. Given the uncertainty surrounding the impact of using a direct question rather than multi-item loneliness scale in comparing outcomes Masi et al., 2011), it was concluded that while meta-analysis would be possible across different multi-item loneliness scales, studies using a direct question could not be combined with those using a multi-item scale.

Policy implications and future research
This review highlights the need for earlier and more effective recognition of loneliness in both the workplace and outside of work. To achieve these objectives, the relationship between loneliness and unemployment must first be raised amongst employers (Michaelson et al., 2017). As a response, employers should look to ensure good relationship and community, preventing workplace loneliness and associated unemployment. As suggested in recent UK guidance to employers this could be achieved by tackling culture and infrastructure; management; people and networks; work and workplace design; and action in the wider community (Department for Digital et al., 2021). Such measures could also help prevent negative employment outcomes, such as poor employee health, identified in previous literature (Michaelson et al., 2017). Policy objectives should not however be restricted to those in paid work. Increased awareness of unemployment as a risk factor for loneliness or conversely loneliness for unemployment, alongside improved support for newly unemployed and long-term unemployed, could help prevent loneliness and the many detrimental health, social and economic outcomes associated. Providing support through agencies, set up to assist individuals looking or work, could be one such method to reduce or prevent loneliness and/or increase employability to break any self-fulfilling cycles between unemployment and increased loneliness.
The COVID-19 pandemic has increased both unemployment and loneliness (Mental Health Foundation, 2020) highlighting the critical importance of these health limiting issues. COVID-19 has affected both our working lives (economic uncertainty and in some cases the introduction of furlough schemes), and our social lives (lockdowns and social distancing measures). This review raises the finding of a bi-directional relationship between loneliness and unemployment, implying a potential multiplier effect and challenges of possible self-reinforcing or self-fulfilling relationships, alongside the opportunity for joint prevention initiatives. This connection suggests policy makers should not consider loneliness and unemployment in isolation, but focus on both factors simultaneously as they 'build back better' following the COVID-19 pandemic. COVID-19 studies included in this review are restricted to early experience of the pandemic where an association of unclear magnitude is observed. Further research into the ongoing experience and aftermath of the pandemic is required with deeper and more focussed attention on the relationship between loneliness and unemployment. In particular, with unemployment set to rise as schemes such as furlough end, effort should be put into understanding and assisting loneliness in those entering unemployment. As the pandemic persists and experience of loneliness and unemployment increases findings suggest a strengthening relationship between the two variables. Furthermore, as loneliness is subjective and not solely due to physical isolation, it will not necessarily be 'solved' by mixing after lockdown measures are lifted. Thus, policies should seek to help individuals readjust to a more open society and find ways to reduce any ongoing experience of loneliness.
Further research is also needed to reinforce and confirm the bidirectional relationship between loneliness and unemployment. This is particularly important in the presence of a self-fulfilling multiplier effect, and given the detrimental impact of both loneliness and unemployment on health (Dooley et al., 1996;Heinrich and Gullone, 2006;Herbig et al., 2013;Holt-Lunstad et al., 2010;Leigh-Hunt et al., 2017;Matthews et al., 2018;Norström et al., 2014;Ong et al., 2016;Paul and Moser, 2009). Understanding could be achieved through additional analysis of longitudinal datasets or utilisation of causal methods, such as propensity score matching. This study illustrates how the relationship between loneliness and employment status is often identified as a by-product. Purposeful consideration is required to enable comparison across severity and duration of both loneliness and unemployment, highlighting those at the greatest risk of being caught in a negative cycle. Direct analysis should also consider the inter-related factors or potential mediators in the relationship, such as depression (Morris, 2019) and income (Luhmann and Hawkley, 2016), in order to improve understanding of the complex and nuanced relationship between loneliness and unemployment. Additionally, more research is needed on the longer-term impact of COVID-19 including exploration of government policies, such as the UK furlough scheme, in mitigating the impact of loneliness and employment outcomes. Overall, this review is the first step in understanding the relationship between loneliness and unemployment, presenting evidence of an association and potential bi-directionality. Further research should seek to unpick the complexity of the relationship and improve understanding of bi-directionality.

Conclusion
Studies show a clear relationship between loneliness and unemployment which extends across the life-course. Individuals who experience loneliness are also more likely to be unemployed. Causality in this relationship is largely under-researched. There are however causal studies suggesting a bi-directional relationship between these outcomes that job loss leads to loneliness, but also that experiencing loneliness leads to subsequent unemployment. Further targeted research is required to better understand the relationship between loneliness and employment status, particularly around this notion of bi-directionality which can in turn create a self-reinforcing relationship, or multiplier effect, for those most at risk. Future research should also consider interrelated factors in this relationship to better understand its complexity and nuance, particularly regarding the area of mental health. There is a definite trend towards the use of national datasets and inclusion of indirect loneliness measures, or loneliness scales, to assess experience of loneliness. Care should be taken in choice of measure and dichotomisation of loneliness, and also in age group classification as these can influence findings. Research in the area of loneliness and unemployment is of particular importance given the current COVID-19 pandemic. Both loneliness and unemployment are set to rise and, based on the findings of this review, potentially exacerbate and reinforce each other.

Funding
This work was supported by the National Institute for Health Research through the Pre-Doctoral Fellowship programme. The funding source had no involvement in study design, writing of the article, or decision to submit for publication. The views expressed in this publication are those of the author(s) and not necessarily those of the National Institute for Health Research or the Department of Health and Social Care.

Declaration of competing interest
The authors declare they have no conflicts of interest.