Inequalities in unpaid carer's health, employment status and social isolation

Abstract Providing higher‐intensity unpaid care (higher care hours or care within the household) is associated with negative impacts on people's paid employment, mental health and well‐being. The evidence of effects on physical health is mixed and carer's social and financial outcomes have been under‐researched. The biggest evidence gap, however, is on how outcomes vary by factors other than type or level of care provision, in particular socio‐demographic factors. Our study used two waves of data (2017/19 and 2018/2020) from the United Kingdom Household Longitudinal Study for people aged 16 and older. We investigated the effects of providing care for 10 or more hours a week or within the household in interaction with people's socio‐demographic characteristics. Outcomes included mental and physical health, social isolation, employment status and earnings. We found that caring responsibilities interacted with gender, ethnicity, socio‐economic status (as measured by highest educational qualification), or age to affect carers differentially in a number of areas of their lives leading to, and exacerbating, key disadvantages and inequalities.


| INTRODUC TI ON
How care needs are met, and disabled and older people enabled to live independent lives, is an important societal and social justice issue. It has had varied traction in the political and policy sphere, although has been higher on the agenda and public consciousness in recent years in many countries, the UK included (Her Majesty's Government, 2014;Scottish Government, 2014;Welsh Government, 2014). Currently, unpaid care comprises the majority of care provided and received (Verbeek-Oudijk et al., 2014). Discussions, and policy, about the optimal balance between formal and informal care varies across country, time and ideology. However, a consistent part of the equation is the impact on unpaid carers. There is now a substantial body of evidence showing that at higher care hours and/or for co-resident carers, providing care has significant negative impacts on carer's paid employment, mental health and well-being Kaschowitz & Brandt, 2017). There are also impacts on carer's physical health when caring for higher hours and/or providing co-resident care (e.g. Bauer & Sousa-Poza, 2015), although findings are mixed (Bom et al., 2019) and there is much less longitudinal evidence available. Social and financial outcomes have been much less researched (Spiers et al., 2021). The biggest evidence gap, however, is on how the experiences of carers vary by factors other than type or level of care provision, in particular socio-demographic factors (Young et al., 2020).
Few of these studies consider care hours but those that do suggest that gender differences in impacts on employment are not solely due to female carers providing higher hours of care than men (King & Pickard, 2013). Mental health was shown to be worse among women providing higher hours of care in a recent longitudinal study in Northern Ireland (Doebler et al., 2017). Cross-sectional studies show similar patterns (e.g. Office for National Statistics, 2013; Verbakel et al., 2017). Research on locus of care and gender suggests some interaction effect with regard to carer's employment (Arber & Ginn, 1995). In terms of other socio-demographic characteristics, evidence from a cross-sectional Swiss study in a very specific context (carers of partners with spinal cord injury) suggested that socioeconomic position was associated with greater perceived strain. This was not moderated by care hours (Tough et al., 2019). A European study using longitudinal data found that carers with higher wealth experienced greater life satisfaction (Brandt et al., 2021) although, in contrast a cross-sectional study in Japan found no interaction effect of care provision and income on depressive symptoms (Saito et al., 2018). However, in general and in the UK context, the evidence on factors other than gender -for example, age, ethnicity and socio-economic status (SES) -is scant (Spiers et al., 2021).
Carers are not a homogenous group and the gap in evidence on variations in outcomes matters because of the body of work showing that factors such as age, SES, gender and ethnicity are key determinants of outcomes in many domains (Dahlgren & Whitehead, 1991;Marmot et al., 2020;Solar & Irwin, 2010). In addition, in order to best support the most vulnerable carers, we need to first identify them. Our study investigated the interaction between provision of care and key sociodemographic factors. In doing so, we utilise a social determinants conceptual framework (Solar & Irwin, 2010). In this framework, socioeconomic position, which comprises social class and social stratifiers (e.g. age, gender, ethnicity, education), is a key structural determinant of outcomes. We also draw on Pearlin and colleagues' stress process model (Pearlin et al., 1990). This model postulates that impacts of care provision depend on both elements of the care itself (e.g. care hours, care type) but also on the context, including socio-economic position (gender, ethnicity, age, educational attainment) and resources. We know that providing care has impacts on people's lives in several domains and that gender, SES, ethnicity and age also impact people's outcomes. Our study aimed to add to the evidence base by exploring the effects of care provision and socio-demographic factors in combination to better understand who is most disadvantaged and how experiences differ. We focused on carers providing the most intense care (higher care hours or co-resident carers) because of evidence showing greater, or sometimes only any, impacts at these levels and types of care provision .

| ME THODS
Our methods strategy was as follows. Using data from the UK Household Longitudinal Study (UKHLS), we identified people aged 16 and older providing unpaid care at time 1 (wave 9; 2017/19) of (a) ten or more hours a week; (b) within the household. We then looked at how interaction of care provision and socio-demographic at time 1 was associated with a number of outcomes at time 2 (wave 10; 2018/20) (Figure 1).

UKHLS (University of Essex Institute for Social and Economic
Research, 2021). The UKHLS started in 2009 and collects data annually from a sample of household members aged 10 or older living in the UK. Sampling is based on a proportionately stratified, clustered sample of addresses selected by postcode, supplemented by specific additional samples added at subsequent waves (Knies, 2017).
Our sample comprised all panel members who took part in the study in both Wave 9 and Wave 10, who were aged 16 or older in Wave 9, and for whom data about caring responsibilities, hours and type were available. The resultant sample for carers caring for 10 or more hours a week was 25,935: 23,586 non-carers; 2349 carers and for co-resident carers N = 25,354: 23,586 non-carers; 1768 carers.
Attrition is an issue for any longitudinal data collection and whilst UKHLS takes a number of measures to minimise this, there is a degree of attrition between waves. However, whilst here is modest under-representation among the youngest age groups, men, ethnic minority respondents and those on the lowest incomes, the sample is still considered representative of the UK population (Lynn & Borkowska, 2018).

What Is Known about this Topic
• Providing unpaid care at higher hours or within the household impacts negatively on paid employment, mental health and well-being • Findings on association with physical health are mixed • There is evidence of gender differences for some outcomes but not much is known about inequalities in carer's experiences

What the Paper Adds
• Social determinants -identifying as female, ethnic minority or lower socio-economic group -interacts with care provision to result in greater negative effects on key life domains • Younger age in conjunction with care responsibilities is associated with poorer mental health and greater social isolation; older age with worse physical health • Higher-intensity caring is also associated with higher odds of social isolation or loneliness

| Caring responsibilities
The variables for caring responsibilities were derived from three questions asked of respondents at time 1: (i) 'Is there anyone living with you who is sick, disabled or elderly whom you look after or give special help to (for example, a sick, disabled or elderly relative/husband/wife/friend etc.)?' (ii) 'Do you provide some regular service or help for any sick, disabled or elderly person not living with you?' And the follow-up question: (i) 'Thinking about everyone who you look after or provide help for, both those living with you and not living with you -in total, how many hours do you spend each week looking after or helping them?' Non-carers were defined as people who answered no to both the first two questions. Higher care hour carers were defined as people who answered yes to either or both questions (i) and (ii) and indicated that they were providing care for 10 or more hours a week in question. Co-resident carers were defined as people who answered yes question (i) (either on its own or together with question (ii)).
Caring for ten or more hours a week was selected because previous research shows that this is the threshold at which impacts on carers are greatest; few or no impacts are observed at lower hours King & Pickard, 2013). In addition, preliminary analysis on the sample showed that providing care at 10 or more hours was the threshold at which negative outcomes in all domains were seen in our data. This effect was seen at the lower range (10-19 h). Within the 10 or more hours category, some higher care hour bands were associated with worse outcomes but this was not linear and no clear pattern emerged. Co-resident caring was selected because co-resident care provision is associated with greater impacts on carers than extra-resident (e.g. Bauer & Sousa-Poza, 2015;Kaschowitz & Brandt, 2017;Michaud et al., 2010;Nguyen & Connelly, 2014).

| Carer characteristics
The carer characteristics used as predictors in the models were gender (male = 0; female = 1); ethnicity (White = 0; Asian = 1; Black = 2); highest educational qualification (degree or higher degree = 0; primary, secondary or no formal qualifications = 1) as a measure of socio-economic status; and age group (16-24; 25-44; 45-65; 66-74; 75+). Highest educational qualification is a well-used measure of socio-economic status and was used in our study because it has a good response rate, is easy to measure and includes people who are unemployed. The social class variables available in UKHLS use information from current job. As a result, over 40% of responses are 'inapplicable'.

| Outcomes
We considered five outcomes separately at time 2: employment status; annual earnings from paid employment; physical health; mental health; and social participation (loneliness or isolation). Employment status was recoded into two categories: 0 = in paid-or selfemployment; 1 = not in paid employment. The continuous variable for earnings was based on a question about monthly earnings from paid employment and thus excluded earnings from self-employment.

F I G U R E 1 Analysis framework
C Ca ar re e p pr ro ov vi is si io on n a. For more than 10 hours a week b. Within the household S So oc ci io o--d de em mo og gr ra ap ph hi ic c c ch ha ar ra ac ct te er ri is st ti ic cs s We multiplied the monthly figure by 12 to give annual earnings. For employment and earnings, we excluded sample members who were in full-or part-time education or training at the same time as paid employment and sample members who were over the UK state pension age at the time (66 and older). For health outcomes, the variables used were the Physical and Mental Components of the Short-Form 12 Health Survey (SF12, PCS and MCS) which measure physical and mental health, respectively; they have been validated for use in the general population (Ware et al., 1996). Lower scores indicate poorer physical/mental health. Social participation was derived from two questions asked at Wave 10 about how often the person felt lonely or felt isolated from others, recoded into 0 for hardly ever or never and 1 for some of the time or often.

| Analysis
First, we used descriptive statistics to report the characteristics of the sample. We then investigated (a) the effect of care provision; (b) the effect of people's socio-demographic characteristics; and (c) the combined, interactive, effect. We used two-step multivariate regression models which used the factors on their own and with interaction terms. We looked at care provision and carer characteristics at time 1 (Wave 9) and outcomes at time 2 (Wave 10). Multivariate logistic regression was used for categorical outcome variables: employment status and social participation. Multivariate linear regression was used for continuous outcome variables: physical health score (PCS) and mental health score (MCS). For analysis of earnings, as this variable has a skewed distribution with a substantial number of zeros, we used two-part Generalised Linear Models (Mullahy, 1998).
We used a modified Park test (Manning & Mullahy, 2001)  Covariates varied by outcome but included the carer characteristics listed above excluding the one used in the interaction term as the predictor in each model. In addition, for employment outcomes (employment status, earnings), we included carer's partnership status (single or in a partnership) and health (presence or absence of a 'long-standing physical or mental impairment, illness or disability;' LLTI). In analyses of health outcomes, additional covariates were partnership status and housing tenure (coded as owner-occupied, social-rented or privately rented). Social-rented housing (or 'public housing') in the UK is provided at more affordable rents, usually by local government or non-profit sector housing associations. We used the unweighted sample for the regression analysis. Where sampling weights are solely related to the independent variables, as they are in our models, unweighted estimates are preferred (Winship & Radbill, 1994).

All tests of statistical significance used robust standard errors.
A significance level of 0.05 was used as the criterion to determine statistical significance and 0.10 to determine marginal significance.
We conducted analyses using Stata 14.2 (StataCorp, 2015). Table 1 shows that 66% of carers caring for 10 or more hours week were female and 34% were male. Women also made up a higher proportion of co-resident carers than men. The majority of higher care hour carers and co-resident carers identified as White. This was 85% for higher care hours, similar to the proportion in the sample overall (84%) and slightly lower (81%) for co-resident caring. Asian carers comprised 11% of higher care hour carers and 15% of co-resident carers. Two-thirds of carers had primary, secondary or no qualifications, higher than their proportion in the sample overall (57%).

Proportions of both higher care hours carers and co-resident carers
were highest in mid-life (45-64). Table 2 shows the association between providing care for ten or more hours a week and each outcome; the association between outcomes and people's socio-demographic characteristics; and Lower-qualified carers were a particularly disadvantaged group (Table 2). The combination of lower qualifications and being a carer for ten or more hours a week resulted in being six times less likely to be in paid employment, an earnings penalty of £12,000 a year, and having significantly lower mental and physical health scores, indicating worse health. The intersection of age and care provision had differential effects on health. The combination of being younger -aged 16-24 or 25-44 -and being a carer was associated with significantly worse mental health, whereas being older and a carer was associated with significantly worse physical health. Being younger and a carer in combination was also associated with higher odds of being lonely or isolated. The interaction of being a co-resident carer and carer's gender, ethnicity, qualification and age showed similar results to the interaction of providing higher care hours and those characteristics (Table 3). This includes for experiencing loneliness and social isolation.

| DISCUSS ION
The effects of providing unpaid care at higher care hours or within the household on carer's employment, earnings and mental health are well-established in the literature and our study findings concur. However, our study adds to that body of evidence by showing that gender, ethnicity, SES and age interact with care provision to amplify these effects. That is that care and context contribute to outcomes (as postulated by Pearlin's stress process model (Pearlin et al., 1990)) and that social determinants also influence carer's outcomes. A further contribution of our paper is the finding that care provision at higher hours or within the household was associated with poorer physical health or greater social isolation; outcomes which are much less studied. Interaction effects were seen here as well leading to greater impacts for some population sub-groups. the UK context, which found differential earnings effects by gender (Keating et al., 2014). The effect of higher hours of care provision and co-resident caring on earnings are likely due to the fact that these types of care are associated with a lower likelihood of being in paid employment and a higher likelihood of working reduced hours and/or in lower-paid jobs Keating et al., 2014). This is exacerbated for female carers by the gender pay gap whereby women earn on average 15% less than men in the UK (Office for National Statistics, 2020a). The amplifying effect of the interaction may be due to female carers being more likely to reduce their paid employment hours and/or to take a less well-paid job (Keating et al., 2014). There is evidence that strategies with regards to managing employment and care differ by gender with men more likely to organise care round their work and women their work around care (Auth et al., 2019).
The types of care provided also may vary by gender. Women provided higher hours of care in our study and may also provide more personal care. Whilst this is part of the caring context per se, it is also related to the unequal gendered nature of caring and to gender roles and is therefore currently inextricably, but not irrevocably, linked with gender. We also observed an interaction effect of being female and providing care on mental and physical health and on social isolation. The care effect is likely to be due to the mental and physical stresses and strains of providing high-level or within-household care and reduced time to spend on social participation. The interaction effect may be due to gender roles, differential access to resources (Solar & Irwin, 2010), and to the complex relationship between gender roles and coping strategies, agency and gender differences in self-care (Zygouri et al., 2021). A recent review found that female carers found it harder to maintain a sense of self-agency and 'felt socially restricted in pursuing their interests, personal needs and career ambitions' (Zygouri et al., 2021). As with employment, differences in the type of care may also play a part, in particular women providing more emotional care as well as higher hours.
For ethnicity, the picture was more complex. Being Asian and Abbreviation: ns, non-significant.

TA B L E 3 (Continued)
provision on health. Ethnicity was not associated with differences in care hours provided but there may be other differences in type of care provision for Asian carers that may contribute to the outcomes seen. The findings for mental health and Black ethnicity were unexpected. Despite a wealth of existing evidence on ethnicity and health leading us to expect poorer outcomes for Black participants compared to White (Bignall et al., 2019), the findings for Black carers and mental health score were non-significant in our study. Black participants in the UKHLS sample overall had higher mental health scores than White participants, indicating better mental health. This may be due to methodological issues. There is higher attrition among minority ethnic participants, for example, although attrition in the sample is unrelated to health status (Lynn & Borkowska, 2018), and a higher proportion of missingness for the mental health variable Black participants in our sample were younger on average than White participants and a higher proportion were female. However, neither of these help explain our finding because women, and younger people, had poorer mental health in our study. Black participants also had higher educational qualifications, which is associated with better mental health.
SES, as measured by the highest educational qualification, was on its own and in intersection with care provision associated with negative impacts in every domain. Lower qualifications are associated with lower earnings and employment rates (Office for National Statistics, 2020b) and are a major determinant of health (Marmot & Bell, 2012;Solar & Irwin, 2010). The interaction effect of care provision and qualification on employment and earnings may also be because higher qualified carers are less likely to have flexible work practices or be able to negotiate them (Spiess & Schneider, 2003). Lower qualified people may have less resources available to them and therefore less alternatives to providing that care themselves.
The stress process model also sets out how the resources available to carers can increase or decrease the impacts of caring on wellbeing (Pearlin et al., 1990). The choices available to lower qualified carers are not just due to their lower financial and other resources.
Choice is also delineated by cultural and familial expectations and these may vary by SES. Expectations about who provides care also vary by ethnicity and gender (della Giusta et al., 2009;Parveen et al., 2011). Greater role captivity may thus also be a contributor to the interaction effect for female and Asian carers both because of the pressure of societal expectations and because women and ethnic minorities are less likely to seek or receive care services and thus to have fewer alternatives to providing that care themselves (Greenwood et al., 2014;Zygouri et al., 2021).
The effects of age and the interplay between age and care provision were striking. The combination of age and caring responsibilities mean that younger carers had much worse mental health than older carers whereas older carers had poorer physical health.
The interaction effect may be linked to a combination of younger people's and carer's poorer mental in general. However, it may also result in part from care provision among younger carers being particularly linked to lack of alternatives (Olsen, 2000) and from their fewer emotional, financial and other resources to mediate the effects of providing care (Aldridge, 2018). Carers aged 25-44 had similarly poorer mental health. This may also be due to role strain and need to juggle competing commitments of work and childcare; such factors exacerbate the stresses of care provision Pearlin et al., 1990). That caring exacerbates mental ill health among younger carers and physical ill health among older carers is a cause for concern and for action. In our study, the odds of younger people expressing being lonely was higher than older people and the combined effect of care and age was seen most in younger age groups for both higher care hour carers and co-resident carers. Care provision at higher hours will reduce the time available for social participation. Stigma and fear of being judged, particularly among younger carers, may cause concerns about bringing people home and/or disconnection from their non-carer peers (Becker & Becker, 2008;Joseph et al., 2020).

| CON CLUS ION
In conclusion, we found that caring responsibilities interact with socio-demographic factors to affect carers differentially in a number of life domains leading to, and exacerbating, key disadvantages and inequalities. Our findings reinforce the need for differentiation of carer support. One clear example is the need for mental health support and prevention for younger carers and physical health support and prevention for older carers. One of the pathways by which social factors determine health and other outcomes is by the ability to access health, long-term care and other services (Solar & Irwin, 2010) and there is evidence of differential access to care services among carers and the people they support (Floridi et al., 2021;García-Gómez et al., 2015;Ilinca et al., 2017). Thus a further implication is a need to reduce or remove barriers to support for sub-groups of carers, examples being through targeting and/or changes to charging regimes and other barriers. Because caring responsibilities are a contributory factor to poorer outcomes, good and targeted support for carers including services for the person they care for has an important role to play. However, support for carers is just one part of what is needed. For example, for female or ethnic minority carers, the gender or ethnic pay gap may be as much of an issue as their care responsibilities. Future qualitative research might fruitfully explore in depth some of the reasons for the sub-group differences seen in our study.

ACK N OWLED G EM ENTS
We are very grateful to the UK Department of Health and Social Care for funding this study, to the Understanding Society (UKHLS) data collection team at the Institute for Social and Economic Research, University of Essex for a great resource, and to the UKHLS team and the UK Data Service for making the data available to us.

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the UK Data Service. Restrictions apply to the availability of these data, which were used under licence for this study. Data are available from the UK Data Service and only be accessed with the permission of the UK Data Service.

E TH I C S
Ethical approval for the UKHLS data used in this study was obtained by the University of Essex Ethics Committee which has approved all data collection on the UKHLS main study and innovation panel waves.