FormalPara Key Points for Decision Makers

Informal care receipt has an asymmetric impact on health behaviours, increasing the probability of refraining from alcohol and cigarette consumption and decreasing the probability of eating healthily and being physically active.

The positive impact on refraining from alcohol and cigarette consumption is consistent with care recipients trying to minimise the burden of caregiving for friends and family.

The behaviour of the care recipient needs to be considered when measuring caregiving burden and evaluating the effectiveness of interventions impacting informal carers.

1 Introduction

Changes in an individual’s health, well-being or functioning can have profound and significant “spillover” effects on the health and well-being of the family and friends caring for them [1]. While informal care can improve the health outcomes of the recipient [2], it is also physically and emotionally straining for the caregiver and often entails them having to forgo opportunities for leisure and employment [3]. Therefore, although utility can be derived from being the one to provide care for a relative [1, 4], most studies find that caregiving overall negatively affects the health and well-being of the care provider [5,6,7].

Much of the literature examining caregiving and its burden examines these health and well-being spillover effects from the perspective of the caregiver [8]. Yet, caregiving is dyadic, and the literature considering caregiving from the perspective of the care recipient, in particular how the care recipient is affected by seeing a loved one burdened by the responsibilities of providing their care, is much less extensive. One study found that, when asked about their family’s involvement in care during semi-structured interviews, 46% of older adults spontaneously mentioned burden. Study participants raised concerns about complicating the lives of their adult children, feeling guilty about their health problems, and that their children were overly worried about their health [9]. Preferences for future long-term care support the existence of these concerns: older adults seem to favour the receipt of formal care only, suggesting a general aversion to burdening friends and family for extended periods [10]. Family members may continue to provide care despite the recipient’ concerns because they feel duty bound, are constrained financially and/or there is insufficient social support available [11].

As well as empathising with their informal carer’s burden, care recipients may alter their behaviour in response to caregiver efforts. However, it is not clear whether these behavioural changes are only motivated by a desire to alleviate carer burden. Fichera et al. [12] examined how recipients of formal healthcare responded to their providers’ efforts, and whether these efforts were complements or substitutes for patient effort. To do this, they exploited an exogenous increase in the size of financial incentives offered to general practitioners to reward their disease management as part of a pay-for-performance scheme. They found that patients responded to the increase in doctor effort induced by the incentive scheme by improving their own health behaviours. However, the responses of care recipients may differ depending on whether their care providers are formal or informal caregivers, and there is little evidence regarding the latter.

We address this gap in the literature by determining whether informal care recipients engage in behaviours to help alleviate their carer’s burden. We hypothesise that informal care recipients will reciprocate their informal caregivers’ effort and internalise carer burden by engaging in behaviours that improve or protect their own health. Improvements in care recipients’ health could reduce the onerousness of caregiving and the stress carers face from seeing a loved one in ill-health.

We tested our hypothesis by examining the response to informal care receipt in terms of health behaviours among older individuals using a nationally representative dataset of 3250 community-dwelling individuals aged 65 years and over with care needs. To explore whether the response depended on the type of health behaviour, we analysed both positive and negative health behaviours, including fruit and/or vegetable consumption and physical activity, and consumption of alcohol and cigarettes. To identify the impact of informal care on the behaviour of care recipients and minimise confounding bias, we used a doubly robust approach that combines regression adjustment and inverse probability weighting.

2 Methods

2.1 Data: UK Household Longitudinal Study (UKHLS)

We used data from the UK Household Longitudinal Study (UKHLS). The UKHLS is a nationally representative longitudinal study of people living in the UK. The UKHLS started in 2009 and comprised approximately 40,000 households. It contains rich information relating to almost all aspects of respondents’ lives. In the main study, each study wave is made up of a two-year fieldwork period. Individuals were required to be community dwelling in 2009–10 to be included in the initial sampling frame, but those who later moved to institutions (e.g. nursing homes, military bases) remained eligible for an interview in subsequent waves.

Wave 9 of the UKHLS (2017–19), a cross-section that contains data on 36,055 respondents, was used as our primary source of data. This is the most recent of the two study waves (waves 7 and 9) that include data on social care receipt and are unaffected by the COVID-19 pandemic [13]. We did not use data affected by COVID-19 restrictions because of concerns that the behaviours of informal care recipients and their carers would not be generalisable. There were 9544 older respondents (aged 65 years and over) interviewed face to face or via the web about their care needs. The 3265 individuals that reported having some care needs were asked about their receipt of both formal and informal care. Respondents interviewed via the telephone (0.4% of those aged 65 years and over) were not asked about their social care receipt and were hence excluded from our analyses. We further excluded individuals that reported living in institutions (0.08% of those aged 65 years and over) as their care experience is likely to have been very different to community-dwelling individuals.

2.2 Exposure: Informal Care Receipt

Basic activities of daily living (BADL) and instrumental activities of daily living (IADL) describe the functional ability of older adults. Basic activities of daily living are those activities necessary for self-care (e.g. bathing, dressing and feeding) and IADL are those activities necessary to adapt independently to the environment (e.g. shopping, transportation and housekeeping) [14].

To determine whether older adults require any assistance with BADL or IADL, UKHLS participants aged 65 years and over were asked to consider for 14 different tasks (ten BADL and four IADL), whether because of “long term physical or mental illness, disability or problems relating to old age”, they can manage the task “on their own”, “with help from someone else” or “not at all”. The BADL tasks included the ability to manage stairs, get around the house, get in/out of bed, cut toenails, bath/shower, use the toilet (including getting up and down), eat (including cutting up food), wash face and hands, get dressed and undressed, and take the right amount of medicines at the right times. The IADL tasks included the ability to walk down the road, do the shopping, do housework or laundry, and do paperwork or pay bills. Full details of the survey questions used to determine the ability of respondents to manage BADL/IADL are provided in Table A1 of the Electronic Supplementary Material (ESM1).

Participants who stated they required help from someone else or were unable to manage at least one BADL or IADL task were asked to indicate which informal and formal care providers, if any, helped them with BADL/IADL tasks in the last month. We refer to these patients as having a BADL or IADL impairment. Informal care providers included family members, friends and neighbours. Formal care providers included a home care worker/home help/personal assistant, a member of the reablement/intermediate care staff team, occupational therapist/physiotherapist/nurse, voluntary helper, warden/sheltered housing manager, cleaner, council’s handyman or other. Full details of the survey questions used to determine who, if anyone, provided informal and formal care are provided in Table A2 of the ESM2.

Overall, 3258 community-dwelling individuals aged 65 years and over reported having at least one BADL or IADL impairment. These individuals were asked to detail, for both BADL and IADL impairments, whether they received any assistance, and if so, which informal and/or formal carer(s) provided it. There were 3250 (99.8%) individuals who provided this detail.

We defined an individual as receiving informal care if they indicated they received help with any of their BADL and/or IADL impairments from at least one informal care provider. Likewise, we defined an individual as receiving formal care if they indicated they received help with any of their BADL and/or IADL impairments from at least one formal care provider. These categories are not mutually exclusive, as some respondents received both informal and formal care. In Sect. 2.5.1, we address the potential compositional differences between individuals with both formal and informal care and those only receiving informal care.

It is plausible that a carer’s actions might directly influence some of the recipient’s health behaviours. For example, assistance with grocery shopping may affect the availability of fruit and vegetables, cigarettes and alcohol for the care recipient. In this case, the estimated effects of informal care receipt would not be capturing how care recipients attempted to alleviate their carer’s burden but rather how informal carers influenced health behaviours directly. To address this, from the 3250 community-dwelling individuals that indicated whether they received informal care, we dropped those that reported having BADL/IADL impairments that, if assisted with, could directly influence the respective health behaviours being examined.

Specifically, to get the subset of respondents used to inform the analysis sample of the fruit and/or vegetable consumption behaviour, we dropped 1510 individuals who reported an impairment with “eat[ing] (including cutting up food)” or “do[ing] the shopping”. Similarly, for the physical activity behaviour, we dropped 912 individuals who reported an impairment with “walking down the road” and for the smoking and drinking behaviours, we dropped 1502 individuals who reported an impairment “do[oing] the shopping”. These three non-mutually exclusive subsets of respondents inform the samples used to analyse the four health behaviours. Figure 1 illustrates the process through which the analysis samples were determined.

Fig. 1
figure 1

Data flow and sample derivation

Full size image

2.3 Dependent Variables: Health Behaviours

Study participants were also asked questions regarding their engagement in various health behaviours. We examined the responses to questions regarding four health behaviours that were likely observable to the caregiver: eating five fruits and/or vegetables per day, walking daily, quitting smoking and consuming a safe level of alcohol. The health behaviour or its impact on the care recipient’s health needs to be observable to ensure it can affect the onerousness of caregiving. Table A3 of the ESM3 provides a description of the health behaviours and the survey questions used to derive the variables. For the main analysis, we created four binary health behaviour variables to simplify the interpretation of regression output. We examined the sensitivity of our results to different classifications of these binary variables (detail in Sect. 2.5.5).

To study dietary behaviour, we constructed a binary variable indicating whether the respondent consumes at least five portions of fruit and/or vegetables per day. Five portions was chosen as the binary threshold because the World Health Organization recommends consuming five fruits and/or vegetables per day to reduce the risk of non-communicable diseases [15].

To study physical activity, we defined an individual as walking daily if they indicated that they walked for at least 10 min at a time 7 days per week. We chose a daily threshold as the UK Chief Medical Officer recommends that adults aged 65 years and over be physically active every day, even if it is just light activity [16].

To study smoking behaviour, we constructed a binary variable indicating whether the respondent has quit smoking. We restricted the analysis on smoking cessation to those who reported being a smoker in at least one of the previous four waves of the survey or reported having ever been a regular smoker when asked this question in wave 5 (2013–15). We constructed this variable to reflect a change in behaviour because smoking initiation usually occurs earlier in life, meaning the decision to smoke as an older adult is likely to reflect past choices and behaviours.

Regarding alcohol consumption, we classified individuals as drinking safely if they consumed alcohol fewer than two times per week and they typically consume three to four drinks or fewer on days when they drink. Assuming respondents always drink their typical amount, a maximum of four drinks per week can be consumed while still being classified as a safe drinker using our threshold. This is a slightly more conservative classification of the National Health Service’s recommendations to drink no more than 14 units of alcohol a week (around six pints of average strength beer), spread across 3 days or more [17]. We chose a conservative classification as it is often recommended that older adults drink less than the general population as they risk exacerbating health problems or having bad interactions with medications when they drink [18]. Those that typically drink more than three to four drinks were defined as not drinking safely, even if they did so infrequently because the National Health Service recommends avoiding drinking too much on any single occasion to avoid the risk of injury and misjudging risky situations [19]. We also tested the robustness of our results using a more lenient classification of drinking safely, under which respondents can drink a maximum of six drinks per week (see Sect. 2.5.5 for more detail). Where the questionnaire was administered face to face, questions relating to alcohol consumption were only asked to those that agreed to self-complete that section of the questionnaire. We excluded individuals who indicated they have never been a drinker from all analyses on alcohol consumption so as to focus on individuals for whom a change in behaviour is plausible.

Notably, we coded the quit smoking and safe drinking variables as positive health behaviours, so their interpretation is consistent with the other health behaviours. Accordingly, the respective variables were coded as 1 (Yes) if a respondent no longer smokes and 1 (Yes) if they consume alcohol fewer than 2 days per week and typically consume three to four drinks or fewer on days that they drink. Table A4 of the ESM3 provides further detail on the questions used to derive the health behaviours including what groups of individuals were asked what questions. Differing patterns of missing data across the questions used to derive the four health behaviours imply a variation in the sample size across our analyses (see Fig. 1).

2.4 Empirical Approach

Our analytical approach aims to estimate the response to informal care receipt across four health behaviours. We modelled the probability of engaging in each health behaviour as a function of informal care receipt. However, receipt of informal care is not randomly assigned and differences in health behaviours may simply reflect various sources of endogeneity. For example, omitted variables driving the receipt of informal care (e.g. intrinsic preference for informal care vs formal care, unobserved features of an individual’s health condition) may be correlated with the health behaviours, introducing bias into estimates. Whilst we cannot control for these unobservable characteristics, we can control for a comprehensive set of observable characteristics, thereby reducing the potential for bias from unobservable characteristics [20]. To identify the response of informal care receipt on our set of health behaviours, we assume mean independence of informal care conditional on a comprehensive set of individual-level control characteristics. This implies that, once these variables are controlled for, the expectation of the error term is uncorrelated with the receipt of informal care. To achieve this whilst simultaneously minimising the risk of model misspecification and improving estimation efficiency, we implemented a doubly robust approach, combining inverse probability weighting with regression adjustment. The advantage of using a doubly robust estimator is that it only requires one of either the treatment or the outcome model to be correctly specified to obtain a consistent estimator of average treatment effects [21].

We modelled the treatment equation using a logistic model. For the probability of informal care receipt \(P\left({\mathrm{IC}}_{i,t}=1\right)\), for individual \(i\), at study wave \(t,\) we estimated:

$$\mathrm{ln} \left[\frac{P\left({\mathrm{IC}}_{i,t}=1\right)}{1-P\left({\mathrm{IC}}_{i,t}=1\right)}\right]={\updelta }_{0}+{X}_{1i,t}{\updelta }_{1}+{X}_{2i,t-1}{\updelta }_{2}+{\upvarepsilon }_{i,t},$$
(1)

in which \(X\) are the covariates described below and \({\upvarepsilon }_{i,t}\) is the idiosyncratic error. Based on the estimates of (1), we obtained predicted treatment probability \({\widehat{p}}_{i}\) for individuals in both groups, which we used to define the inverse probability weights for the outcome model.

We modelled the outcome equations using linear probability models. The probability of carrying out the respective health behaviours \(Y\), for individual \(i\), in study wave \(t\), is defined as:

$$P\left({Y}_{i,t}=1\right)={\beta }_{0}+{X}_{1i,t}{\beta }_{1}+{X}_{2i,t-1}{\beta }_{2}+{\mu }_{it},$$
(2)

in which \({\mu }_{it}\) is an idiosyncratic error term assumed uncorrelated with \({\upvarepsilon }_{i,t}\).

In Eqs. (1) and (2), \({X}_{1i}\), is a vector of individual-level characteristics reported in survey wave 9 that could influence both informal care receipt and health behaviours. These individual-level characteristics include age (65–69, 70–74, 75–79, 80–84, 85+ years), gender, age-gender interaction, whether cohabits (yes or no), whether married (yes or no), the number of non-resident relatives, whether owns home outright (yes or no; used as an indicator for wealth [22, 23]), highest educational qualification (degree, other higher degree, advanced-level qualifications, general certificate of secondary education, other qualification, no qualification), whether has a long-standing illness or disability (yes or no) and whether receives formal care (yes or no).

\({X}_{2i}\) is a vector of individual-level characteristics that are lagged by one period (wave 8, 2016–18) and could determine informal care receipt and health behaviours. In order to account for differences in care need, ability to carry out health behaviours, and disposable income, these lagged individual-level characteristics include satisfaction with health (completely dissatisfied, mostly dissatisfied, somewhat dissatisfied, neither satisfied nor dissatisfied, somewhat satisfied, mostly satisfied, completely satisfied; included as a continuous variable), subjective well-being proxied by the GHQ-12 Likert scale from 0 (least distressed) to 36 (most distressed) and household income (log transformed and equivalised using the modified OECD equivalence scale [24]). We included these terms as lagged values to avoid incurring a bad control problem, with covariates being simultaneously affected by the treatment [25]. Data on individuals’ BADL/IADL were not included in the models because BADL/IADL status is used to define the exposure, i.e. only individuals reporting a BADL/IADL impairment are included in the sample and respondents were asked about social care receipt relating to their reported impairments only.

To obtain our doubly robust estimates for the average treatment effect of informal care receipt on the probability of engaging in health behaviour Y, we estimate the vector of coefficients \(\beta =\left[{{\beta }_{0},\beta }_{1},{\beta }_{2}\right]\) using the following weighted least-squares estimator, separately by treatment status (that is for \(\mathrm{IC}=0\) and \(\mathrm{IC}=1\)):

$${\mathrm{min}}_{{\upbeta }_{0},{\upbeta }_{1},{\upbeta }_{2}}\sum_{i=1}^{{N}^{IC}}\frac{{\left({Y}_{i,t}-{\upbeta }_{0}-{X}_{1i,t}{\upbeta }_{1}-{X}_{2i,t-1}{\upbeta }_{2}\right)}^{2}}{{\widehat{p}}_{i}},$$
(3)

where \({N}^{IC}\) is the number of observations in the two groups defined by receipt of informal care, either \(\mathrm{IC}=1\) or \(\mathrm{IC}=0\). Given the two sets of estimated coefficients \({\widehat{\beta }}_{\mathrm{IC}=0}=\left[{{\widehat{\beta }}_{0},\widehat{\beta }}_{1},{\widehat{\beta }}_{2}\right]\) and \({\widehat{\beta }}_{\mathrm{IC}=1}=\left[{{\widehat{\beta }}_{0},\widehat{\beta }}_{1},{\widehat{\beta }}_{2}\right]\), we derive the average treatment effect as the average difference in potential outcomes (i.e. the fitted values based on estimated coefficients and observed covariates) for behaviour \({\widehat{Y}}_{i,\mathrm{IC}}\) between the treatment and control group:

$$\mathrm{ATE}=\frac{1}{{N}^{0}}\sum_{i=1}^{{N}^{0}}{\widehat{Y}}_{i,0}-\frac{1}{{N}^{1}}\sum_{j=1}^{{N}^{1}}{\widehat{Y}}_{j,1.}$$
(4)

In Sect. 2.5, we introduce our sensitivity analyses, which explore the stability of our results to these modelling choices. Statistical inference on our coefficients of interest across all models is based on standard errors obtained using the Huber–White heteroskedasticity-robust estimator for the variance-covariance matrix. In Tables A13–A15 of the ESM8, we also repeat our analysis using alternative treatment effect estimators. All analyses were conducted in Stata 14 using the teffect ipwra command.

2.5 Sensitivity Analyses

2.5.1 Sample Stratification

Receipt of formal care in addition to informal care may influence how informal care affects health behaviours or suggest compositional differences in our sample based on receipt of either or both types of care. For example, receipt of formal care may reflect different unobservable degrees of accessibility to formal care on the supply side (e.g. availability of formal carers) or demand side (e.g. financial support from family), which may, in turn, influence the receipt of informal care. To examine whether these unobserved factors correlated with formal care receipt introduced confounding in our main estimates, we stratified the sample by whether respondents received formal care, estimating the models for both groups independently and comparing the results.

We also examined whether the results varied across relevant population subgroups. First, we explored to what extent our results were driven by individuals particularly dependent on their carer. We proxy care dependency by the number of BADL/IADL impairments reported (irrespective of receipt of formal care) and stratify our analysis by individuals reporting multiple BADL/IADL impairments compared to those reporting one impairment. Second, we stratified the sample by gender, as demographic profiles and couple dynamics may determine asymmetric responses between men and women. Third, we checked whether our results differed for individuals aged under 74 years compared to those aged 75 years and over. Finally, we stratified our analysis by the relationship between the care recipient and caregiver, first, whether individuals received care from their children/grandchildren (yes or no) and second, whether individuals received care from a spouse (yes or no). Our assumption with this latter test is that reciprocity in behaviours may be different depending on the strength of family ties to the informal caregivers.

2.5.2 Restricting the Sample to Those Not Previously Receiving Informal Care

Our main analysis focuses on a cross-section of individuals, some of whom have been receiving informal care for a long time. The behavioural responses to receipt of informal care might differ for these individuals, compared with those newly exposed to informal care, mostly because of an adaptation mechanism. To rule out that our estimates of the behavioural response to informal care are diluted by adaptation and to examine the impact of change in treatment, we re-estimated our main models after restricting the samples to individuals who reported not receiving informal care in wave 7 (2015–17; i.e. received no care, formal care or reported not having a BADL/IADL impairment in wave 7).

2.5.3 Restricting the Sample to Remove Any Residual Direct Influence of the Carer on Health Behaviours

Whilst our main analysis sample is restricted to individuals without particular BADL/IADL impairments that, if assisted with, could directly influence the respective health behaviours being examined, some residual direct influence from the carer may remain. For example, if the informal carer undertakes the grocery shopping for the care recipient, this may reduce the recipients’ need to engage in physical activity if shopping requires walking. Furthermore, in our main analyses on quitting smoking and drinking safely, we attempted to remove the direct influence of carers’ actions on care recipients’ consumption behaviours through the exclusion of recipients who reported an impairment “do[ing] the shopping”. The reasoning being that if individuals are unable to do their shopping unaided, their carer may directly affect the availability of goods in the household. However, individuals who report an impairment doing the shopping may be still able to buy cigarettes or alcohol without the assistance of their carer if they can walk down the road because cigarettes and alcohol may be purchased separately from food shopping. As the samples we analyse for the effects on cigarette and alcohol consumption include some respondents with impairments walking down the road, it is conceivable that these care recipients could be reliant on the informal carer to buy them cigarettes and alcohol, and the carer may therefore influence the recipient’s behaviour directly.

To examine whether our results are robust to removing these examples of potential residual direct carer influence, we further restricted our analysis samples according to some additional BADL/IADL impairments that, if assisted with, could in theory directly influence health behaviours. We dropped a further 727 individuals with impairments “do[ing] the shopping” from the subset of respondents used to analyse the walks daily behaviour. We dropped a further 137 individuals with impairments “walking down the road” from the subset of respondents used to analyse the quit smoking and drinks safely health behaviours.

2.5.4 Adding Lagged Health Behaviours as Covariates

Exploiting the limited within-individual variation in health behaviours over time is also interesting, for at least two reasons. First, most health behaviours tend to be persistent over time [26,27,28], a dynamic relationship implying that current health behaviours will also depend on past health behaviours. Second, individuals who engaged in some health behaviours for a long time may be more or less likely to require informal care in the first place. Both points suggest that some unmeasured individual characteristics may drive both receipt of and behavioural response to informal care. This may bias our coefficients in directions that are a priori unclear. To partially address these concerns, we included, as covariates in the treatment and outcome models, the wave 7 (2015–17) versions of the non-binary response variables used to derive the health behaviours. The indicators for daily consumption of fruits and vegetables were included as continuous terms. The indicators for the number of days walked, frequency of alcohol consumption and number of drinks typically consumed on days when drinking were included as factor terms. We did not perform this analysis on the quit smoking sample because the dependent variable is already defined relative to past smoking behaviour.

2.5.5 Re-defining Health Behaviours

The estimated impact of informal care receipt on health behaviours could be sensitive to the binary thresholds used to dichotomise the dependent variables. Hence, we tested the robustness of our results from the main analysis to different ways of classifying the survey responses into health behaviours.

For physical activity, we redefined someone as physically active if they completed at least 150 min of walking and/or moderate physical activity per week OR if they completed at least 75 min of vigorous activity per week. This is based on guidance from the UK Chief Medical Officer [16]. We incorporated minutes walked into our measure because the Chief Medical Officer’s definition of moderate activity includes brisk walking, something which is explicitly excluded in the UKHLS definition of moderate activity.

For alcohol consumption, we redefined drinking safely with a threshold that also allows safe drinkers to typically consume one to two drinks up to three times per week. Table A5 of the ESM3 shows how the questions on frequency and volume of alcohol consumption are combined to create our measures of drinking safely, illustrating which response options classify an individual as a safe drinker in our main analysis and under the redefined classification used in the sensitivity analysis.

For smoking, we expanded our definition of quitting smoking so it includes those who still smoke in wave 9, but infrequently. Amongst those who smoked previously (before wave 9), smoking infrequently was defined as smoking five cigarettes or fewer per day. We also examined whether the respondent was a non-smoker in wave 9 (2017–19), regardless of whether they had reported smoking in previous waves. Hence, this variable measures current smoking status, as opposed to quitting smoking.

Finally, we used, the original non-binary response categories for the eats five fruits and/or vegetables per day and walks daily health behaviours variables to explore whether our results are sensitive to the binary thresholds chosen, and in the case of eats five fruits and/or vegetables per day, to see whether there are differential effects for fruits versus vegetables.

Descriptions of the alternative specifications of the health behaviours and the questions used to derive them are presented in Table A6 of the ESM3. The smoking and drinking variables were again coded as positive health behaviours so that the interpretation of coefficients is consistent.

3 Results

3.1 Summary Statistics

Table 1 provides descriptive statistics for the subset of respondents that are used to inform the analysis sample for all physical activity variables. We chose to focus on this subset of respondents because it contains more individuals than the other two subsets. The table is categorised according to receipt of informal care to deal with reported BADL and/or IADL impairments. Those receiving no informal care were either in receipt of formal care or received no care. Those in receipt of informal care were either in receipt of informal care only or a combination of informal and formal care. In this subset of respondents, 1485 (63.5%) respondents reported receiving informal care.

Table 1 Summary statistics for individuals with at least one basic or instrumental activity of daily living impairment (subset of respondents used for the walking sample)
Full size table

Those receiving informal care differed substantially from those not receiving informal care across multiple characteristics. Informal care recipients were significantly more likely to be male, cohabit with another adult, be married, have more relatives outside the household, have no qualifications, have a long-standing illness or disability, have been unsatisfied with their health and have been distressed compared to those who did not receive informal care. Conversely, those that did not receive informal care were significantly more likely to own their home outright, have a higher household equivalised income, have a degree and receive formal care.

In both groups having one BADL impairment was the modal response, but the distribution in BADL impairment counts differed. The majority of those not receiving informal care reported having zero IADL impairments, whereas the majority receiving informal care reported having a positive number of IADL impairments. In these unadjusted comparisons, informal care recipients were significantly more likely to drink safely and those not receiving informal care were more likely to walk daily. We report a similar comparison for the subset of respondents used to analyse fruit and/or vegetable consumption (see Table A7 of the ESM4) as well as smoking and drinking (see Table A8 of the ESM4).

3.2 Main Analysis: Average Treatment Effects of Informal Care Receipt on Health Behaviours

In our four main samples (which differ depending upon the health behaviour being examined), the standardised difference in the weighted covariates is close to zero (see Table A9 of the ESM5). This suggests that the treated (informal care recipients) and untreated (those not receiving informal care) subjects are comparable and that our samples are balanced after our adjustment procedure [29]. We are also unable to reject the null hypothesis that covariates are balanced using an overidentification hypothesis test on each of the four samples [30].

Table 2 presents estimates of the average treatment effect of informal care receipt on health behaviours. Informal care receipt was found to reduce the probability of engaging in positive health behaviours (eating five fruits and/or vegetables per day and walking daily) but increase the probability of refraining from negative health behaviours (smoking and alcohol consumption). Informal care receipt reduced the probability of walking daily by 12.8% points. It also reduced the probability of eating five fruits and/or vegetables per day by 0.2% points, though this effect is not statistically significant. Conversely, receipt of informal care increased the probability of quitting smoking by 5.8% points and drinking safely by 8.4% points.

Table 2 Average treatment effects of informal care receipt on health behaviours
Full size table

The outputs for the treatment models used to derive the weights that were used to predict the inverse probability weighting with regression adjustment estimator are presented in Table A10 of the ESM6. Each of the treatment models correctly classified over 67% of respondents into the correct treatment level, suggesting that our treatment models were sufficiently accurate. Overlap plots are presented in Fig. A1 of the ESM6. For both treatment and control groups across all our analysis samples, the estimated conditional probability of not receiving informal care is strictly between zero and one. This suggests that the overlap assumption — crucial for a consistent estimation of the treatment effect with our chosen doubly robust approach — is not violated.

The full output from the outcome models used to estimate the average treatment effect is provided in Table A11 of the ESM7. Table A12 of the ESM7 provides estimates of goodness of fit for alternative specifications of the outcome models, using the Akaike Information Criterion. Stepwise models are fitted for each health behaviour by adding our 13 covariates (see Sect. 2.4.) one at a time. The Akaike Information Criterion scores suggest we need most of or all the 13 covariates in our models. Considering the doubly robust framework, we err on the side of more variables is preferable for the sake of getting the best of the two parts of the model in case both the treatment and outcome models are slightly misspecified [31].

3.3 Sensitivity Analyses

3.3.1 Sample Stratification

Table 3 presents estimates of the average treatment effect of informal care receipt on health behaviours after stratifying the sample by receipt of formal care, reporting more than one BADL/IADL impairment, gender, being aged 75 years and over, receiving care from their child or grandchild, and receiving care from a spouse. The estimated effects in this analysis are measured less precisely than those in the main analysis, particularly for those receiving formal care and those with multiple BADL/IADL impairments as sample sizes are smaller. Some caution, therefore, needs to be exercised when interpreting the results.

Table 3 Average treatment effects of informal care receipt on health behaviours, for stratified samples
Full size table

Amongst those not receiving formal care, the impact of receiving informal care (informal care only vs no care) is concordant with the main analysis: receipt of informal care decreased the likelihood of engaging in positive health behaviours and increased the likelihood of refraining from negative health behaviours. However, amongst those receiving formal care, we no longer observe statistically significant effects for the quit smoking and safe drinker health behaviours.

For both the least dependent individuals (i.e. single BADL/IADL impairment) and the most dependent individuals (multiple BADL/IADL impairments), we generally find that informal care receipt decreased the likelihood of engaging in positive behaviours and increased the likelihood of refraining from negative behaviours. However, amongst the least dependent individuals, informal care had a positive impact on eating five fruits and/or vegetables per day, though the effect is still not statistically significant; and amongst the most dependent, the effect on drinks safely is now negative, though not statistically significant. The effects on the probability of walking daily and quitting smoking are more pronounced for the most dependent individuals compared with the least dependent individuals.

For female individuals, the magnitude of the negative effect on walking daily is larger than for male individuals. However, male individuals appeared to be more likely to quit smoking in response to informal care.

For individuals aged 74 years or below, the magnitude of the positive effect on quitting smoking is larger than for the older population. However, the magnitude of the negative effect on walking daily is larger for the older population. The effects are very similar for those that did not receive informal care from a child or grandchild and those that did, although the positive effect on quitting smoking is slightly larger for those with a child carer.

Finally, amongst those not receiving informal care from a spouse, we observe a large positive effect on the drinking safely behaviour, whereas the effect among those with a spousal carer is negative. The magnitude of the negative effect on eating five fruits and/or vegetables per day is much larger for those receiving informal care from a spouse. The magnitudes of the negative effect on walking daily and the positive effect on quitting smoking are slightly larger for those receiving care from a spouse.

3.3.2 Restricting the Sample to Those Not Previously Receiving Informal Care

The findings that informal care receipt reduces the probability of walking daily and increases the probability of drinking safely are robust to restricting the sample to those who did not receive informal care in 2015–17 (see Table 4), suggesting that our main results for these behaviours are not being diluted by those adapting to informal care receipt over time. The effect on quitting smoking remains positive but diluted in magnitude and is no longer statistically significant, suggesting that the result in our main analysis is primarily driven by those who have been exposed to informal care for longer periods. This could reflect that quitting smoking is rarely instantaneous and often takes multiple attempts [32]. Unlike in the main analysis, the effect on eating five fruits and/or vegetables per day is positive for those who have been more recently exposed to informal care. This could suggest that individuals eat healthily upon receipt of informal care to try and alleviate their carer’s burden but that they do not sustain this over time.

Table 4 Average treatment effects of informal care receipt on health behaviours, restricted to those not receiving informal care in 2015–17
Full size table

3.3.3 Restricting the Sample to Remove Any Residual Direct Influence of the Carer on Health Behaviours

The finding that informal care receipt reduces the probability of walking daily is robust to dropping those with impairments doing the shopping from the analysis sample, although the effect is diluted in magnitude by 5.1% points (see Table 5). This suggests that the possibility of informal carers removing the need for the care recipients to walk in order to undertake essential food shopping could have some influence but is not the sole driver of the result in the main analysis. The drop in magnitude could also be partially attributable to the exclusion of a large number of individuals with high levels of BADL/IADL impairment. The findings that informal care receipt increases the probability of quitting smoking and drinking safely is robust to dropping those with impairments walking down the road from the respective samples, suggesting that the main results are not being driven by the informal care recipient being reliant on their carer to walk to the shop to purchase cigarettes and alcohol.

Table 5 Average treatment effects of informal care receipt on health behaviours, further excluding those with particular basic or instrumental activities of daily living impairments
Full size table

3.3.4 Adding Lagged Health Behaviours as Covariates

The finding that informal care receipt reduced the probability of engaging in positive health behaviours and increased the probability of refraining from negative health behaviours is robust to including lagged components of the dependent variables as covariates (see Table 6). However, these effects are diluted in magnitude by 0.1, 1.4 and 4.9% points for eating five fruits/veg per day, walking daily and drinking safely, respectively.

Table 6 Average treatment effect of informal care receipt on health behaviours, including lagged dependent variables as additional covariates
Full size table

3.3.5 Re-defining Health Behaviours

Table 7 presents the results when estimated using alternative ways of classifying the survey responses into measures of health behaviours. The effect of informal care receipt remains negative for most alternative ways of classifying the positive health behaviour variables and positive for all ways of classifying refraining from negative health behaviour variables. The estimated effect on the number of vegetable portions eaten per day becomes positive but is still not statistically significant. Notably, the estimated effect of informal care receipt on the probability of drinking safely when using the alternative, more lenient threshold (which includes those that drink one to two drinks three times per week) is 4.0% points lower than when using the stricter interpretation in the main analysis. This is suggestive of a gradient in the effect of informal care receipt on drinking behaviour, with a stronger influence on those with less frequent drinking habits. The estimated effect is also lower by 4.0% points for whether respondents complete 150 min of walking/moderate physical activity or 75 min of vigorous physical activity per week when compared with whether walks daily. This suggests that the negative effect of informal care receipt could be stronger for less intense forms of physical activity. Aside from the intensity of the effect and the effect on vegetable consumption, the results are broadly consistent with our main estimates.

Table 7 Average treatment effects of informal care receipt on alternative definitions of health behaviours
Full size table

4 Discussion

4.1 Main Findings

Providing informal care has been shown to have a negative effect on the caregiver’s health and well-being, but little is known about how recipients respond to the receipt of care from an informal provider. To capture whether informal care recipients internalise their carer’s burden, we examined the average treatment effect of informal care receipt on positive and negative health behaviours amongst older adults with care needs. We did so after excluding those with BADL/IADL impairments that suggest carers may have a direct influence on health behaviours. We examined health behaviours that were likely observable to the caregiver using a doubly robust approach involving inverse probability weighting and regression adjustment. We controlled for numerous individual-level characteristics across multiple waves of data that could have affected informal care receipt and health behaviours. We found the direction of the estimated effect to differ depending on whether the health behaviour in question was positive or negative. On the one hand, informal care receipt was found to reduce the probability of engaging in positive health behaviours, reducing care recipients’ likelihood of consuming five fruits and/or vegetables per day and walking daily. Conversely, informal care receipt was found to increase the probability of care recipients refraining from engaging in negative health behaviours, reducing recipients’ likelihood of consuming cigarettes and alcohol. These effects were strongest for walking daily, quitting smoking and safe alcohol consumption.

These results are broadly consistent across our battery of sensitivity checks. However, the type and intensity of care required seem to exacerbate some behavioural responses.

We found that the most care-dependent individuals responded more strongly to the receipt of informal care for all health behaviours except drinking safely. Likewise, we found that those receiving care from a spouse responded more strongly for all health behaviours except drinking safely, where, in contrast to those not receiving care from a spouse, informal care receipt had a negative impact on the probability of drinking safely. The degree of behavioural response also depended on gender and age, but how these characteristics interacted with informal care receipt was not consistent across health behaviours.

Whilst our main analysis controls for the receipt of formal care, our sensitivity check suggests that the recipients’ responses are different for those receiving and not receiving formal care. This might point towards a confounding of formal care on the perceived experience of delivering or receiving informal care. However, if anything, the results are more pronounced when focusing on individuals not in receipt of formal care, relaxing concerns about bias in our main estimates.

Restricting the sample to those who had not previously received informal care diluted the positive response on the quitting smoking variable, suggesting that the result is primarily driven by those exposed to informal care for a longer period. We may therefore be under-estimating the long-term impact of informal care receipt on smoking cessation in the main analysis.

When accounting for behaviour persistence by controlling for lagged behaviour, our coefficients were concordant in sign but diluted, which could suggest some omitted variable bias in the main analysis. Hence, this sensitivity analysis provides a plausible lower-bound estimate for our effect of interest.

4.2 Reconciling Asymmetrical Effects on Positive and Negative Health Behaviours

The finding that informal care receipt increased the probability of refraining from negative health behaviours appears consistent with our initial hypothesis, that informal care recipients will reciprocate their caregiver’s efforts and attempt to alleviate carer burden by not harming their own health. To explain why informal care recipients might choose to alleviate carer burden this way, it is useful to further explore the concept of reciprocity. To do so, we refer to the social exchange theoretical framework. Where individuals depend on one another for the things they value and maximise benefits and minimise costs, exchanges in familial relations operate on the generalised rules of reciprocity [33]. In a reciprocal relationship, familial exchanges should be relatively balanced over the long-term. If the exchange of resources is persistently unidirectional, as is sometimes the case in a caregiving dyad [34], the relationship may become intolerably burdensome. Evidence suggests that balance in caregiving exchanges (e.g. sharing household chores, or where both parties show warmth and regard) reduces carer burden [35,36,37]. A systematic review by Park and Schumacher [36] found growing evidence that mutuality in a caregiving relationship, where both parties share the same meaning, attitudes and orientation towards an illness [38], was associated with improvements in the caregiver’s emotional health outcomes. Given that older adults are averse to burdening family and friends with caring responsibilities [9], and that reciprocity in a caregiving relationship can reduce carer burden, it seems logical that informal care recipients would be motivated to reciprocate their carer’s efforts by avoiding alcohol and cigarette consumption.

After removing the direct influence of carers on the basket of goods available to care recipients, we found receipt of informal care to increase the probability of quitting smoking by 5.8% points. This is larger than the effect of financial incentives for smoking cessation, which have been estimated to increase the probability of quitting by an average of 3.5% points [39]. Our findings demonstrate the strength of social incentives routed in reciprocity, suggesting that they have the potential to be more powerful than financial incentives in influencing health behaviours.

Additionally, it is plausible that informal care receipt reduced the probability of engaging in negative health behaviours if informal care recipients are disproportionally admonished for these by their carer compared to formal care recipients and individuals receiving no care. If we assume that carers are responsible for the care recipient’s overall health rather than just specific impairments, carers actions might incentivise certain health behaviours and disincentivise others [12, 40]. They could praise care recipients for being healthy (e.g., eating healthily and being physically active) and admonish them for being unhealthy (e.g. smoking and drinking). The cost of admonishing care recipients may differ between carer types, with formal care providers likely to have higher costs, potentially viewing doing so as beyond their professional remit or feeling uncomfortable admonishing somebody outside of their social network. If this is the case, informal care recipients would have a greater incentive to avoid smoking and drinking compared to the incentive to engage in positive health behaviours. At the same time, informal carers (especially those living with the care recipient) are more likely to stringently tackle negative behaviours (e.g. permanent monitoring, low-level hostile sanctioning) than formal carers who likely spend less time with care recipients.

Yet, these mechanisms alone do not explain the coexistence of asymmetrical behavioural responses. Informal care recipients refraining from negative behaviours but reducing positive health behaviours could be explained by risk and effort compensation. Individuals could be compensating risk, as it has been suggested that individuals have a desirable level of risk and are therefore constantly adjusting their behaviour to maintain that level [41]. In our study, respondents could have been avoiding cigarette and alcohol consumption to help alleviate carer burden but have offset this reduced risk to their personal health by eating less healthily and being more sedentary. Similarly, the negative effects on diet and physical activity health behaviours could have been respondents attempting to maintain some desirable level of effort expenditure by offsetting the additional effort they put into avoiding cigarette and alcohol consumption.

Individuals may have specifically chosen to compensate their risk and efforts by reducing positive health behaviours in favour of avoiding negative health behaviours because they perceive the consumption of alcohol and cigarettes to more severely damage the well-being of friends and family [42, 43]. Similarly, because these behaviours can be so harmful to oneself and others, informal care recipients may anticipate being more severely admonished by their carer for smoking and drinking. Therefore, to compensate risk and effort, the care recipient could rationally decide to give up negative behaviours and put less effort into pursuing positive behaviours.

4.3 Strengths and Limitations

To our knowledge, this is the first study to explore whether informal care recipients engage in behaviours designed to alleviate their carer’s burden using a doubly robust approach, which requires only one of either the treatment or outcome model to be correctly specified to produce unbiased estimates. The behaviour of the care recipient is rarely considered in studies evaluating informal care provision.

Our study used a large sample of individuals with care needs. Data on care needs, social care receipt and health behaviours were determined using participants’ responses to a household survey. The survey questions used to derive the health behaviours in our analyses did not reference carer burden nor any other mechanisms that could have influenced health behaviour, thus minimising the potential that the framing of the survey questions influenced our results and their interpretation.

To test the robustness of our results, we performed multiple sensitivity analyses including stratifying the sample by various individual-level characteristics and the receipt of formal care, restricting the sample to those that had not previously received informal care, restricting the sample to remove any residual direct influence of the carer on health behaviours, exploiting the panel structure of the data by controlling for lagged health behaviours and using alternative specifications of the health behaviours.

Despite these findings, our study has some limitations. First, one potential source of endogeneity in the main model is simultaneous causality bias. While informal care receipt can determine health behaviours, it is also plausible that health behaviours can determine whether somebody receives informal care. For example, somebody may volunteer to provide care for their family member because they are eating unhealthily. We partly address this issue by dropping individuals with particular BADL/IADL impairments relevant to the specific health behaviours (e.g. impairments doing the shopping or eating when studying fruit and/or vegetable consumption) and controlling for past health behaviours. Despite this, we cannot be sure that some simultaneous causality bias did not persist.

Second, whilst we take measures to remove the direct influence that carers may have on health behaviours, some influence may remain. For example, informal care recipients may reduce their physical activity because their needs for social interaction are partially met by their carers visiting them in their homes, meaning they may be less inclined to walk to visit friends and family. However, if this were the case, we would expect the negative effect to be less pronounced for those receiving care from a spouse, who will be cohabiting in the vast majority of cases. If anything, we find the opposite to be true in our stratified analysis (see Table 3).

Third, despite using a doubly robust approach to reduce threats of model misspecification and testing robustness to additional covariate terms, we cannot fully rule out the possibility that coefficients were biased either because of endogeneity in the model or because of unobserved individual heterogeneity (e.g. some residual differences in care need), hence excluding any strictly causal interpretation of our results. Finally, there were insufficient data available on our exposure of interest (receipt of informal care) in other waves to use a formal longitudinal model, such as a fixed-effect model. However, we did utilise data from previous waves where possible, using five waves of UKHLS data in our analyses to provide clarity on the mechanisms driving the effects.

4.4 Future Research

In a setting where data permits, future research could implement an instrumental variable approach to address the potential for simultaneous causality bias and unobserved heterogeneity when examining the impact of informal care receipt [44, 45]. To better understand the mechanisms underpinning a change in health behaviours and differential changes among population subgroups, it would also be beneficial for future research to incorporate longer longitudinal series and qualitative research components in a mixed-methods approach. Informal care recipients detailing their motivation for any changes in health behaviours over time could provide valuable insight into the strategic interplay between the informal care recipient and their carer.

5 Conclusions

We provide evidence that informal care recipients were less likely to engage in positive health behaviours, but also more likely to refrain from negative health behaviours, compared with individuals receiving either formal or no care for the same care needs. This asymmetry suggests that the mechanisms driving positive and negative health behaviours may be different. In the discussion, we outline some potential explanations for this dual response, including risk and effort compensation. Failure to account for the behavioural responses of informal care recipients may lead to under-estimation or over-estimation of the extent of caregiving burden and the effectiveness of interventions impacting informal carers.