Does Formal Employment Reduce Informal Caregiving?

HEALTH ECONOMICS Health Econ. 25: 829–843 (2016) Published online 28 May 2015 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/hec.3185

DOES FORMAL EMPLOYMENT REDUCE INFORMAL CAREGIVING? DAIFENG HE* and PETER MCHENRY** College of William and Mary, Williamsburg, VA, USA

ABSTRACT Using the Survey of Income and Program Participation, we examine the impact of formal employment on informal caregiving. We instrument for individual work hours with state unemployment rates. We find that, among women of prime caregiving ages (40–64 years), working 10% more hours per week reduces the probability of providing informal care by about 2 percentage points. The effects are stronger for more time-intensive caregiving and if care recipients are household members. Our results imply that work-promoting policies have the unintended consequence of reducing informal caregiving in an aging society. Copyright © 2015 John Wiley & Sons, Ltd. Received 20 March 2014; Revised 15 March 2015; Accepted 27 March 2015 KEY WORDS:

informal care; elderly care; formal employment; local business cycle; state unemployment rate

JEL Classifications:

I1; J14; J22

1. INTRODUCTION Along with many other developed countries, the USA faces twin headwinds in its attempts to take care of the elderly. On the one hand, demand for elderly care is increasing as the baby boomers hit retirement with rising longevity and high chronic disease prevalence (National Center for Health Statistics, 2012). Informal care— broadly defined as unpaid care provided by family members, friends, and charities—has been the majority source of elderly long-term care (McGarry, 1998; Arno et al., 1999; Gibson and Houser, 2007). As society continues to age, one can reasonably expect the demand for informal care to rise. On the other hand, certain demographic trends and various work-incentive policies may imply a dwindling supply of informal care if labor market activities crowd out informal caregiving. For example, one of the most prominent demographic trends in the USA in the past several decades is the dramatic increase in female labor force participation.1,2 How does this trend affect the informal caregiving decisions of women, who provide the majority of informal care (Johnson and Wiener, 2006; McGarry, 1998)? Meanwhile, the USA has implemented various policies that promote labor force participation (e.g., increasing the Social Security retirement age, adding work requirements to welfare programs, and implementing the Earned Income Tax Credit program). What are the implications of those policies for our ability to care for the elderly? A priori, it is not clear whether increased labor market activities necessarily reduce informal caregiving. Greater employment opportunities increase the opportunity cost of time for home production (e.g., informal

*Correspondence to: Department of Economics and the Thomas Jefferson Program in Public Policy at the College of William and Mary, Williamsburg, VA 23187, USA. E-mail: [email protected]; [email protected] 1

Women accounted for 46% of the US labor force in 2007, up from 34% in 1960, an increase of almost 48 million women (Gruber et al., 2009). 2 Other demographic trends that may negatively affect the supply of informal care include declining fertility and smaller family sizes.

Copyright © 2015 John Wiley & Sons, Ltd.

830

D. HE AND P. MCHENRY

caregiving) and might thereby reduce informal caregiving. However, if caregiving needs are inflexible, a person could reduce time spent in other activities (e.g., leisure such as relaxing or sleep, chores at home, hobbies, or exercise) in order to work more in the labor market, while maintaining a constant level of informal caregiving. In this case, the effect of employment on informal care would be zero. Also, increased earnings from the labor market might help the caregiver pay for formal care services to the care recipient rather than giving informal care; this would dampen the possibly negative effect of formal employment on informal caregiving toward zero. The majority of the literature on the relationship between formal employment and informal care assesses the effect of informal caregiving on caregivers’ labor market outcomes, the opposite of our focus.3 The results of those studies are somewhat mixed but on balance suggest that caregiving needs exert a modest negative pressure on the likelihood of formal employment and on work hours (e.g., Van Houtven et al., 2013). This evidence implies a trade-off between time spent in caregiving and labor market activities, but it does not necessarily show that work opportunities discourage informal caregiving. For example, work opportunities may induce caregivers to reduce their leisure instead of caregiving time, even if sudden caregiving needs such as an ailing parent tend to reduce work hours. In addition, the magnitude of the response to a shock in one activity (i.e., labor market work) may differ from the magnitude of the response to a shock in the other activity (i.e., informal caregiving). Among the smaller set of papers studying the effect of formal employment on informal caregiving, only a few try to correct for endogeneity of formal employment, and they yield mixed results. Stern (1995) uses previous employment status to instrument for current-period employment and finds no effect of employment on informal caregiving; the empirical identification relies on the strong assumption that individuals’ expectations about future caregiving activities do not influence their current employment decisions. Using a difference-indifferences approach comparing families with and without children (who were affected differently by welfare reform and Earned Income Tax Credit expansion in the 1990s), Golberstein (2008) finds that increased work incentives reduce the likelihood of a woman co-residing with a disabled parent, which is a key form of informal caregiving. Boaz (1996) and Doty et al. (1998) estimate simultaneous equation models of employment and caregiving, using somewhat implausible exclusion restrictions for employment (e.g., caregiver’s schooling, age, number of children, and disabilities). Mentzakis et al. (2009) and Carmichael et al. (2010) control for lagged employment status in a panel data context to deal with the endogeneity of contemporary employment. Michaud et al. (2010) address both directions of causation with estimates of reduced-form equations that model the dynamics of informal care and employment. Nizalova (2012) examines the effect of the wage on informal caregiving using cross-sectional state unemployment rates and industry structures as instruments. This paper aims to quantify the causal effect of formal employment on informal caregiving, using what we think is a more compelling instrumental variables method. The empirical challenge in identifying this causal effect is threefold. First, a negative correlation between formal employment and caregiving could reflect reverse causality: caregiving hurts individuals’ labor market prospects and thus reduces work hours (Ettner, 1995). Second, unobservable individual characteristics could influence both formal employment and caregiving simultaneously and thereby induce a spurious correlation between the two. Third, individual labor market activities such as employment and work hours are often measured with error, as most surveys use long recall periods such as 1 year or longer, which may further bias the relationship. Our instrumental variables strategy addresses all three of those concerns. Using data from three panels of the Survey of Income and Program Participation (SIPP 1996, 2001, and 2004), we instrument for individual work hours using within-state variation in the state unemployment rate. Our estimation strategy deals with the reverse causality issue because individual behavior is unlikely to affect the state unemployment rate. State-level unemployment rates probably also contain fewer measurement errors than individual work hours because the errors in individual responses could cancel each other in aggregation. More importantly, we argue that the state unemployment rate is a valid instrument that plausibly satisfies the exclusion restriction after we control for a

3

See, for example, Moen et al. (1994), Ettner (1995), Robison et al. (1995), Pavalko and Artis (1997), McGarry (2006), Heitmueller (2007), Leigh (2010), Coe et al. (2011), Ciani (2012), and Van Houtven et al. (2013).


Health Econ. 25: 829–843 (2016) DOI: 10.1002/hec

FORMAL EMPLOYMENT REDUCES INFORMAL CAREGIVING

831

set of factors that may affect the supply of or demand for informal caregiving at the individual or state level (detailed in Section 2). We have two main findings. First, formal employment reduces the probability of caregiving. Among women of prime caregiving age (40–64 years old), working 10% more hours per week reduces the probability of providing informal care by about 2 percentage points. This is large relative to the sample average caregiving rate of 8%. Second, the effects of formal employment on caregiving are heterogeneous. The effects are stronger if the care recipients are household members, and we find some suggestive evidence that the effects are also stronger for more time-intensive caregiving. These results are consistent with stronger effects for more intensive caregiving (Lilly et al., 2007; Carmichael et al., 2010; Jacobs et al., 2014).

2. ESTIMATION STRATEGY To estimate the effect of formal employment on informal caregiving, we begin with the following equation: Prðcare_hoursist > 0Þ ¼ Φðδ0 þ δ1 lnðwork_hoursist þ 1Þ þ δ2 X ist þ δ3 W st þ vs þ ηt Þ

(1)

where care_hoursist is the number of weekly informal care hours given by individual i living in state s at time t and work_hoursist measures the number of weekly hours the individual works in the formal labor market, set to zero for those who do not work. In an alternative specification, we replace ln(work_hoursist + 1) with a dummy variable workist, indicating whether the person works at all. Xist is a vector of individual characteristics, Wst is a vector of residence state characteristics, νs is a residence state fixed effect, and ηt is a year fixed effect. Φ is the cumulative distribution function for the standard normal distribution, indicating that we will estimate probit models. We observe caregiving behavior in only one period per person, so we do not use panel methods (e.g., fixed or random effects) at the individual (i) level. δ1 is the parameter of interest, measuring how much formal labor market activities affect informal caregiving. Individuals make caregiving and employment decisions jointly, and other potentially confounding factors influence both decisions. For example, caregiving needs (e.g., the presence of an ailing parent) and the marginal utility of leisure relative to that of caregiving both directly affect the trade-off between formal employment and informal caregiving. Our empirical strategy is to focus on proxies for exogenously determined employment opportunities as shifters of individual formal employment decisions. Specifically, we instrument for individual employment, ln(work_hoursist + 1), with the unemployment rate in individual i’s residence state s in year t. This strategy identifies δ1 using variation in individual work hours that is induced by variation over time in employment opportunities in the individual’s residence state. We therefore estimate Equation (1) by maximum likelihood as an instrumental variables probit (IV probit) model. This allows us to isolate the causal effect of formal employment on informal caregiving, rather than the effect of informal caregiving on formal employment (reverse causality), because individual caregiving behavior, which may affect individual formal employment decisions, is unlikely to affect the unemployment rate at the state level. In addition, instrumenting with state unemployment rates likely reduces bias caused by measurement errors if aggregation of individuals’ unemployment statuses to the state level reduces the influence of recall bias and reporting errors (i.e., individual positive and negative errors cancel one another). Another potential concern is that unobserved individual or family traits affect informal care and are correlated with residual variation in our measures of employment (omitted variables). While it is possible that our controls are insufficient, we believe that our estimates are unlikely to reflect the effects of omitted variables. Our identifying assumption is that unobservable individual and family characteristics that influence both formal employment and informal caregiving are uncorrelated with the state’s unemployment rate after conditioning on a set of controls described in the following. We flexibly control for a comprehensive set of individual characteristics (Xist) including age (in years), age squared, age cubed, indicators for different educational levels, indicators for race and ethnicity, household income (excluding one’s own labor earnings), and household wealth. In addition, we include a set of year indicators (ηt) and a set of state fixed effects (vs). The year indicators control Copyright © 2015 John Wiley & Sons, Ltd.


832


for any factors that are common to everyone in a given year, such as nationwide economic conditions or time trends in caregiving. State fixed effects control for any time-invariant factors at the state level that can affect individual caregiving. For example, state culture and norms may play an important role when one decides whether to care for his or her elderly parents. Perhaps more importantly, for certain unobservable individual characteristics of the potential caregiver (e.g., individual ability at work and care, availability of siblings, and the presence of an ailing parent needing care), even if they are correlated with the local unemployment rate in a given year, they are unlikely to be correlated with within-state variation over time in the unemployment rate. This is because the distributions of such factors are plausibly time invariant in the short span of our sample period (about 10 years) and thereby are unlikely to be correlated with time variation in state unemployment rates. By the same argument, unobservable time-invariant individual characteristics of potential care recipients are also unlikely to be correlated with the within-state variation in unemployment rates in a short period. For example, potential care recipients in states with high unemployment rates may be particularly poor and unable to afford formal care, so they have to rely on informal care as a substitute. Because we do not observe potential care recipients’ characteristics in the SIPP data, this would be a problem if the potential care recipients’ wealth and income correlate with the time variation in state unemployment rates. But because most care recipients are elderly or disabled, their income and wealth, even if possibly correlated with state unemployment rates at a given point of time or nationwide business cycles, are unlikely to be substantively affected by local business cycles. For example, Social Security income does not fall when the state unemployment rate rises. The value of equity holdings, unless concentrated heavily on companies that are located in the person’s residence state, should not be affected by the local business cycle once we control for year fixed effects. We therefore believe this would introduce minimal bias, if any. Our specifications also control for several time-varying state-level variables (Wst) that might affect the supply of or demand for informal caregiving. In particular, we control for state Medicaid spending per enrollee to account for the possibility that formal care and informal care are substitutes (Van Houtven and Norton, 2004, 2008; Bolin et al., 2008) and that availability of Medicaid coverage likely is associated with state economic conditions. We also control for state Medicare expenditure per enrollee as a proxy for seniors’ health because a recent literature argues that seniors’ health worsens during recessions (McInerney and Mellor, 2012), which may call for more caregiving. Our coefficient estimates remain robust to the inclusion of those time-varying state-level controls. Note that we do not need to control for state-level time-varying variables such as income and wealth, education, racial composition, and age distribution even though those variables are likely correlated with the state unemployment rate, because we control for those variables at the individual level; for example, conditional on individual income, there is little reason to believe that the state average income level would affect an individual’s care decision. Our IV probit estimation strategy identifies a local average treatment effect. We estimate the average effect of formal employment on informal caregiving among people whose work hours are influenced by state-level changes in economic conditions (which influence the statewide unemployment rate). Our estimates may not generalize to the entire population. State-level economic conditions tend to have the largest effects on the employment of the less educated and members of minority groups (Hoynes, Miller, and Schaller, 2012). Our estimates are therefore likely relevant to women who have relatively few resources and whose potential care recipients may have considerable difficulty substituting expensive formal care for informal care, that is, women whose caregiving needs are relatively high.

3. DATA We use the SIPP, a major ongoing survey conducted by the US Census Bureau. The SIPP is a series of nationally representative longitudinal samples of non-institutionalized civilians aged 15 and older in the USA. One advantage of the SIPP is its large sample sizes (more than 50,000 respondents per panel, including men and Copyright © 2015 John Wiley & Sons, Ltd.



833

women). We use three SIPP panels: 1996, 2001, and 2004. Each panel consists of multiple interview waves that were 4 months apart. The 1996 and 2004 panels have 12 waves each, and the 2001 panel has nine waves. We restrict our sample to working-age female respondents who are at prime caregiving age (between 40 and 64 years old). The SIPP core questionnaires, administered in every wave, asked respondents to recall their labor market participation (employment status and work hours) in each of the 4 months prior to the interview. We therefore can track each respondent’s labor market activities throughout the panel period. In addition, wave 7 of each of the three SIPP panels includes a topical module about informal caregiving activities. This module contains a rich set of information about the respondent’s informal caregiving behavior, including whether the respondent provided any informal care and the hours of care provided per week in the past month, whether providing care to household members or to non-household members and the respective care hours, and whether the care was given to a parent. To be consistent with the timing of the caregiving module, we define our main explanatory variable work_hoursist to be the average weekly work hours in the 12 months prior to the module. People who did not work at all receive a 0 for work_hoursist, and we add 1 to the hours value prior to taking logs. Because the caregiving module occurs only once for each respondent, our estimation sample is a repeated cross-sectional dataset even though the core SIPP surveys are longitudinal. We obtain state unemployment rates from the Census Local Area Unemployment Statistics. To be consistent with the timing of the main explanatory variable, we use the state unemployment rates 1 year preceding the informal care module. We obtain state-specific Medicare spending per enrollee (medicare_ perst) and Medicaid spending per enrollee (medicaid_perst) from the Centers for Medicare & Medicaid Services.4 We convert those spending data to constant 2005 dollars. For each household, we average household net worth available in waves 3 and 6 to reduce measurement errors (variable household_wealthist).5 We subtract a respondent’s own monthly earnings from his or her total monthly household income and average this ‘household income net of own earnings’ across the 12 months in the year prior to the informal care module (household_incomeist). Both variables are measured in constant 2005 dollars. Panel A of Table I displays summary statistics for our female SIPP samples. About 8% of women ages 40–64 years give informal care, and the average time devoted to caregiving is 2.3 hours weekly (including those with zero caregiving hours).6 Among caregivers, average weekly caregiving is 28.22 hours; this is comparable with the 24.5 average weekly hours of elder caregiving among female caregivers in the 2011 American Time Use Survey (Bureau of Labor Statistics, 2012). The mean employment rate in our sample of women is 66%. There is considerable variation in state unemployment rates both within and across years. The mean across states is 4.77 in 1997, rises to 5.45 in 2002, and falls back to 5.02 in 2005, and the standard deviation in each year is about 1 percentage point. Across the states, Medicaid and Medicare spending per enrollee increased over the period, and Medicaid long-term care spending per enrollee decreased.

Because the 1996 and 2001 SIPP panels combine Maine and Vermont into one ‘state’ identifier and North Dakota, South Dakota, and Wyoming into another ‘state’ identifier, we similarly group these states in the 2004 SIPP panel. We therefore have 45 ‘states’ plus the District of Columbia in our final working sample. To obtain unemployment rates and state spending data for the combined states, we calculate weighted averages with the weight being the state population. 5 The specific SIPP-created variable we use is called THHTNW, which is a sum of responses about asset and debt categories (e.g., equity in homes, vehicles, businesses, interest-earning assets at banks and other institutions, stocks and mutual fund shares, non-home real estate, other assets, individual retirement and Keogh accounts, and unsecured debt). The components of THHTNW were mostly top coded and include imputations. Matthew Marlay at the US Census Bureau kindly provided this information. 6 Survey of Income and Program Participation 1996 reported care hours using a continuous measure. SIPP 2001 and 2004 reported the care hours in brackets, so we impute the informal care hours in those panels by taking the midpoint of each closed bracket. We impute the midpoint of the lower end of the last open bracket and 112 hours, which is 7 days * (24 8 hours for sleep) per day. 4



834


Table I. Sample statistics Sample mean (standard deviation in parentheses)

Panel A: Individual-level variables Any care (care) Care to household members Care to non-household members Care to parents Care hours Care hours among caregivers Care hours more than 20 h Care houses more than 40 h Work hours Work hours among workers Any work Age Married Number of kids under 18 years High school Some college College degree Graduate degrees White Black Hispanic Household net assets (household_wealth, $1,000s) Household income net of self-earning (household_income, $1,000s) Panel B: State-level variables State unemployment rate State Medicaid spending per enrollee (medicaid_per, $1,000s) State Medicaid long-term care spending per enrollee (ltc_per, $1,000s) State Medicare spending per enrollee (medicare_per, $1,000s)

0.08 (0.28) 0.04 (0.20) 0.05 (0.22) 0.03 (0.16) 2.32 (11.80) 28.22 (31.03) 0.03 (0.18) 0.02 (0.14) 22.31 (19.20) 33.89 (12.94) 0.66 (0.47) 50.55 (6.95) 0.64 (0.48) 0.63 (1.01) 0.31 (0.46) 0.32 (0.47) 0.15 (0.36) 0.09 (0.29) 0.74 (0.44) 0.13 (0.34) 0.08 (0.28) 235.691 (1127.36) 3.93 (4.37) 1997

2002

2005

4.77 (1.18) 6.10 (1.84) 1.10 (0.62) 6.06 (0.99)

5.45 (0.96) 6.66 (1.86) 1.02 (0.62) 6.60 (0.89)

5.02 (1.04) 6.82 (1.87) 0.99 (0.55) 7.46 (0.96)

Sample statistics are taken across Survey of Income and Program Participation 1996, 2001, and 2004 panels over the female sample between ages 40–64 years. All dollar amount variables, that is, household_wealth, household_income, medicaid_per, ltc_per, and medicare_per, are in constant 2005 dollars.

4. MAIN RESULTS 4.1. Strength of the instrument Table II reports the first-stage results from the IV probit estimation of Equation (1) by maximum likelihood, where ln(work_hoursist + 1), the natural log of average weekly hours worked in the previous 12 months, is regressed on the unemployment rate in the same period in the individual’s residence state. Columns 1–5 add controls gradually, and all columns also control for age, age squared, age cubed, state fixed effects, and year fixed effects. Table II consistently shows that higher unemployment rates at the state level predict fewer work hours at the individual level. The coefficient estimates of the unemployment rate are highly statistically significant. The last row of the table shows large first-stage partial F-statistics that indicate a strong instrument. The control variables are correlated with work hours largely as expected. More educated women work more hours. Married women work significantly less than single women (column 2), but conditional on wealth and income, married women work more than single women (columns 3–5). Household income net of own earnings and wealth are negatively correlated with work hours, consistent with a positive income effect on non-work time such as leisure. Having young children in the household significantly reduces work hours. Copyright © 2015 John Wiley & Sons, Ltd.


835


Table II. First-stage results from IV probit estimation (1)

(2)

Variables State unemployment rate

(3)

(4)

(5)

0.017*** (0.004) 0.299*** (0.017) 0.394*** (0.015) 0.483*** (0.018) 1.654*** (0.018) 0.048** (0.021) 0.072*** (0.025) 0.037 (0.032) 0.049*** (0.011) 0.064*** (0.004) 0.000** (0.000) 0.020*** (0.001) 0.004 (0.006) 0.013 (0.014) 38,489 16.08

First stage 0.020*** (0.005)

0.017*** (0.005) 0.288*** (0.017) 0.375*** (0.015) 0.441*** (0.017) 0.564*** (0.020) 0.046** (0.020) 0.090*** (0.024) 0.050 (0.031) 0.112*** (0.011) 0.070*** (0.004)

0.018*** (0.004) 0.299*** (0.017) 0.394*** (0.015) 0.483*** (0.018) 0.612*** (0.018) 0.048** (0.021) 0.072*** (0.025) 0.037 (0.032) 0.049*** (0.011) 0.064*** (0.004) 0.000** (0.000) 0.020*** (0.001)

0.018*** (0.004) 0.299*** (0.017) 0.394*** (0.015) 0.483*** (0.018) 1.654*** (0.018) 0.048** (0.021) 0.072*** (0.025) 0.037 (0.032) 0.049*** (0.011) 0.064*** (0.004) 0.000** (0.000) 0.020*** (0.001) 0.003 (0.006)

39,233 18.67

39,233 13.76

38,489 16.48

38,489 16.48

High school Some college College Graduate school White Black Hispanic Married Number of kids under 18 years household_wealth household_income medicaid_per medicare_per Observations F-statistic for instrumental variables strength

Dependent variable = log(work_hours + 1). Data from sample of women aged 40–64 years in Survey of Income and Program Participation 1996, 2001, and 2004 panels. Less than high school education is the omitted category for the education variables, and other races is the omitted category for the race/ethnicity variables. All regressions control for age, age squared, age cubed, state fixed effects, and year fixed effects. Robust standard errors clustered at the state-year level in parentheses. ***p < 0.01; **p < 0.05; *p < 0.1.

4.2. Results on providing informal care Table III reports our main results from estimating the effect of weekly work hours on the probability of providing any informal care. For ease of comparison, we present the probit marginal effect estimates in columns 1–5 and the IV probit marginal effect estimates in columns 6–10. Again, we gradually add individual-level and state-level controls, and all columns control for age, age squared, age cubed, state fixed effects, and year fixed effects. Columns 1–5 consistently show a statistically significant negative correlation between work hours and caregiving. The coefficient estimate of 0.015 implies that working 10% more hours per week is associated with a reduced probability of a woman providing informal care by 0.15 percentage points. Columns 6–10 show that the IV probit estimates of the marginal effects are substantially larger than the probit estimates. The coefficient of 0.206 in column 10—our preferred specification—implies that working 10% more hours per week on average reduces the likelihood of providing informal care by 2.06 percentage points. The much larger magnitude of our IV probit estimates in Table III implies that the upward bias in the previous probit estimation, perhaps due to omitted variable bias or measurement errors or a combination of both, dominates the downward bias of reverse causality. Unfortunately, we lose some precision in the IV probit estimation. The p-value of the Copyright © 2015 John Wiley & Sons, Ltd.



39,233

0.012*** (0.002)

0.016** (0.007) 0.013 (0.010) 0.0143 (0.010) 0.004 (0.003) 0.004*** (0.001)

39,233

(3)

38,489

0.016** (0.008) 0.014 (0.010) 0.013 (0.010) 0.001 (0.004) 0.004*** (0.001) 0.000 (0.000) 0.001** (0.000)

0.015*** (0.002) 0.007 (0.005) 0.023*** (0.006) 0.015*** (0.005) 0.021*** (0.008)

(4)

38,489

0.016** (0.008) 0.014 (0.010) 0.013 (0.010) 0.001 (0.004) 0.004*** (0.001) 0.000 (0.000) 0.001** (0.000) 0.002 (0.002)

0.015*** (0.002) 0.007 (0.005) 0.023*** (0.006) 0.016*** (0.006) 0.021*** (0.008)

Probit (marginal effects) 0.015*** (0.002) 0.006 (0.005) 0.022*** (0.006) 0.013** (0.006) 0.018** (0.008)

(2)

0.016** (0.008) 0.014 (0.010) 0.013 (0010) 0.001 (0.004) 0.004*** (0.001) 0.000 (0.000) 0.001*** (0.000) 0.002 (0.002) 0.013* (0.007) 38,489

0.015*** (0.002) 0.007 (0.005) 0.023*** (0.006) 0.015*** (0.006) 0.021*** (0.008)

(5)

39,233 18.67

0.217** (0.106)

(6)

0.028*** (0.010) 0.034*** (0.014) 0.025** (0.010) 0.030*** (0.012) 0.020*** (0.007)

39,233 13.76

(8)

38,489 16.48

(0.066) 0.029*** (0.011) 0.030** (0.013) 0.022** (0.010) 0.012** (0.007) 0.018*** (0.008) 0.43e06* (3.35e06) 0.005** (0.002)

0.230** (0.111) 0.071** (0.033) 0.110*** (0.041) 0.119*** (0.052) 0.153*

(9)

38,489 16.48

0.029*** (0.010) 0.030** (0.013) 0.022** (0.010) 0.012* (0.007) 0.018*** (0.007) 5.45e06* (3.33e06) 0.005*** (0.002) 0.003 (0.003)

0.237** (0.103) 0.074** (0.030) 0.114*** (0.037) 0.122** (0.048) 0.157*** (0.060)

IV probit (marginal effects) 0.241** (0.103) 0.071** (0.029) 0.108*** (0.035) 0.113*** (0.045)*** 0.146*** (0.056)

(7)

0.028*** (0.011) 0.028** (0.015) 0.021** (0.010) 0.010 (0.008) 0.016* (0.008) 5.18e06 (3.48e06) 0.005* (0.003) 0.003 (0.003) 0.010 (0.008) 38,489 16.08

0.206 (0.130) 0.065* (0.039) 0.100** (0.050) 0.110* (0.062) 0.138* (0.078)

(10)

Dependent variable = care, endogenous variable = log(work_hours + 1), instrumental variable = state unemployment rate. Data from sample of women aged 40–64 years in Survey of Income and Program Participation 1996, 2001, and 2004 panels. Less than high school education is the omitted category for the education variables, and other races is the omitted category for the race/ethnicity variables. All regressions control for age, age squared, age cubed, state fixed effects, and year fixed effects. Robust standard errors clustered at the state-year level in parentheses. ***p < 0.01; **p < 0.05; *p < 0.1.

Observations First-stage F-statistic

medicare_per

medicaid_per

household_income

Number of kids under 18 years household_wealth

Married

Hispanic

Black

White

Graduate school

College

Some college

High school

Log(work_hours + 1)

Variables

(1)

Table III. The effect of employment on caregiving

836 D. HE AND P. MCHENRY


837


Table IV. Marginal effects of work and positive work hours on caregiving incidence (1) Variables work log(work_hours + 1) High school Some college College Graduate school White Black Hispanic Married Number of children Household wealth Household income State Medicaid per enrollee State Medicare per enrollee Observations First-stage F-statistic

(2) Marginal effects (standard error)

0.255 (0.310) 0.067 (0.066) 0.175 (0.081) 0.123 (0.092) 0.160 (0.114) 0.119 (0.059) 0.092 (0.068) 0.088 (0.062) 0.009 (0.026) 0.025 (0.013) 0.000 (0.000) 0.006 (0.004) 0.012 (0.134) 0.086 (0.047) 38,506 4.58

0.183 (0.283) 0.027 (0.027) 0.047 (0.036) 0.043 (0.044) 0.060 (0.058) 0.003 (0.009) 0.007 (0.011) 0.004 (0.012) 0.004 (0.006) 0.010 (0.012) 0.000 (0.000) 0.002 (0.003) 0.004 (0.004) 0.007 (0.009) 25,411 22.78

Dependent variable = care, instrumental variable = state unemployment rate. Data from sample of women aged 40– 64 years in Survey of Income and Program Participation 1996, 2001, and 2004 panels. In column 1, the sample includes those who work and those who do not. In column 2, the sample includes only workers (those with positive employment hours within the past 12 months). Marginal effects are reported. Specifications also control for age, age squared, age cubed, state fixed effects, and year fixed effects. Robust standard errors clustered at the state-year level in parentheses. ***p < 0.01; **p < 0.05; *p < 0.1.

marginal effect estimate for ln(work_hours + 1) is 0.11 in column 10, although it is less than 0.05 in the other specifications. However, the consistency across specifications of both the coefficient sign and magnitude suggests that working reduces informal caregiving considerably. The estimated effect size is large relative to the average informal caregiving rate in our sample (about 8%). However, our estimates are local average treatment effects that are relevant for women whose employment is sensitive to local economic conditions. Our conjecture is that these women may have an informal caregiving rate that is higher than that of an average woman in the population. While our main results in Table III pool workers and non-workers, Table IV shows results from specifications that separate the effect of working or not from the effect of work hours among workers. The two columns show results from IV probit specifications where the dependent variable is an indicator for informal caregiving (as in Table III). Column 1 uses an indicator for any work in the prior 12 months (workist) as the main independent variable. The estimated coefficient of 0.255 implies that being employed is associated with a lower incidence of informal caregiving, although the wide confidence interval includes zero. The first-stage partial F-statistic for the work indicator in this specification is quite low, so this is not a powerful test. Column 2 of Table IV shows the relationship between work hours and informal caregiving in the subsample of women who have positive work hours in the prior 12 months (workist = 1). The estimated effect is negative, although again not precise. It is similar in magnitude to the estimates based on the full sample of workers and nonworkers from Table III. Therefore, the results in Table IV are imprecise but still consistent with our finding that employment reduces informal caregiving among women. 4.3. Robustness checks The aforementioned findings are robust to alternative definitions of formal employment and informal caregiving. Table V presents the IV probit results of estimating Equation (1) with alternative definitions of formal Copyright © 2015 John Wiley & Sons, Ltd.


838


Table V. Robustness checks: varying the definition of employment (1)

(2)

(3)

Alternative independent variables

Marginal effects (standard error)

Observations

First-stage F-statistic

(1) Weekly work hours in previous 12 months (log(work_hours + 1)) (2) Work hours in the same month (log(work_hours_samemnth + 1)) (3) Work hours in previous 4 months (log(work_hours_4mnth + 1)) (4) Work hours in previous 8 months (log(work_hours_8mnth + 1)) (5) Work hours 1 year ago (log(work_hours_lag + 1)) (6) Any work in previous 12 months (work) (7) Any work in the same month (work_samemnth) (8) Any work in previous 4 months (work_4mnth) (9) Any work in previous 8 months (work_8mnth) (10) Any work 1 year ago (work_lag) (11) Weekly work hours in previous 12 months, among workers (log(work_hours + 1)) (12) Work hours in the same month, among workers (log(work_hours_samemnth + 1)) (13) Work hours in previous 4 months, among workers (log(work_hours_4mnth + 1)) (14) Work hours in previous 8 months, among workers (log(work_hours_8mnth + 1)) (15) Work hours 1 year ago among workers (log(work_hours_lag + 1))

0.206 (0.129) 0.226* (0.137) 0.205 (0.138) 0.182 (0.128) 0.155 (0.125) 0.255 (0.310) 0.516* (0.310) 0.539* (0.311) 0.246 (0.403) 0.049 (0.342) 0.183 (0.283) 0.175 (0.288) 0.157 (0.257) 0.165 (0.270) 0.121 (0.226)

38,489 38,413 38,434 38,412 37,621 38,506 36,427 38,478 38,412 37,637 25,411 24,041 24,067 24,072 24,836

16.08 5.52 7.51 10.63 12.82 4.58 5.66 5.71 6.50 9.80 22.78 3.84 5.79 7.19 12.11

Dependent variable = care, endogenous variable = 10 different variants of employment, instrumental variable = state unemployment rate. Data from sample of women aged 40–64 years in Survey of Income and Program Participation 1996, 2001, and 2004 panels. Each numbered row represents one instrumental variables probit regression except for rows 6–10, which are bivariate probit regressions. All specifications control for age, age squared, age cubed, state fixed effects, and year fixed effects, as well as the full set of controls in Table III. Robust standard errors clustered at the state-year level in parentheses. ***p < 0.01; **p < 0.05; *p < 0.1.

employment, using the full set of controls as in column 10 of Table III. For reference, row 1 repeats the main marginal effect estimate from column 10 of Table III. Rows 2–5 show results from alternative formal employment measures with different reference periods: instead of average weekly work hours in the previous 12 months prior to the informal caregiving questions, we use average weekly work hours in the previous month (work_hours_samemnthist), in the previous 4 months (work_hours_4mnthist), in the previous 8 months (work_hours_8mnthist), or in the year prior to the previous year (from 24 to 12 months ago: work_hours_lagist). Rows 1–5 show that the marginal effect estimates are generally robust to these alternative definitions. We also replace these continuous measures of work hours with a set of dichotomous measures: an indicator for any work hours in the previous 12 months (workist) and a parallel set of variants with different reference periods. They are an indicator for any work in the same month (work_samemnthist), an indicator for the previous 4 months (work_4mnthist), an indicator for the previous 8 months (work_8mnthist), or an indicator for any work between 24 and 12 months prior to the informal care module (work_lagist). Row 6 repeats the specification in column 1 of Table IV. Rows 6–10 show that the coefficients are mostly robust across these various alternative definitions; working reduces the probability of providing informal care by 0.049–0.539. The instrument, however, is less powerful in predicting the work indicators as shown by the smaller first-stage F-statistics, so we caution that some of the estimates here may be biased. Rows 11–15 in Table V vary the reference period for the work hours measure with the workers-only sample. Point estimates are negative and fairly consistent. In another robustness check, we replace state Medicaid spending per enrollee by state Medicaid long-term care spending per enrollee because not all Medicaid spending is for care that is potentially a substitute for informal care. The marginal effect estimate in this specification is qualitatively similar to those reported in Table III (results upon request). In yet another robustness check, we control for state-specific linear time trends, and the effect of employment is again qualitatively similar, although the strength of the instrument is considerably reduced (results upon request). Copyright © 2015 John Wiley & Sons, Ltd.


839


4.4. Heterogeneous effects The effect of employment on caregiving may vary across types of caregiving activities. We first explore whether the employment effect differs by the time intensity of caregiving. It is possible that care provision has some fixed costs (e.g., time spent traveling to the recipient’s house) so that there is a threshold of care effort past which additional informal caregiving is not very costly. In such a case, working would have a smaller effect on the high end of the care hours distribution. On the other hand, some prior research has suggested that the effect of intensive informal caregiving (i.e., more than 20 hours/week) has a larger effect on employment than less-intensive informal caregiving has (Carmichael and Charles, 1998; Heitmueller, 2007; Lilly et al., 2007; Carmichael et al., 2010; Jacobs et al., 2014). Intensity of care may be relevant for the effect of employment on informal caregiving as well. For example, it may be easier to continue modest caregiving tasks while holding a job, while very time-consuming informal care is more difficult to continue while working in the labor market. Table VI reports results from instrumental variables ordered probit specifications that allow more flexibility in our measure of informal caregiving than a single caregiving indicator. The dependent variable takes on separate discrete values associated with different ranges of informal caregiving hours. As previously discussed, we instrument for the individual’s employment with his or her state’s unemployment rate. The three panels show results for different categorizations of informal caregiving intensity. Panel A of Table VI shows estimated marginal effects of work hours on predicted probabilities of different levels of informal caregiving: none, between 0 and 20 hours, and more than 20 hours. Consistent with our main results, greater work hours is associated with lower informal caregiving. The estimated effect of work hours on the likelihood of more than 20 caregiving hours (0.171) is greater than the estimated effect on the likelihood of 0–20 hours (0.052). While the difference is not statistically distinguishable from zero, it is suggestive of a greater effect on more intensive care. This is consistent with prior findings of a stronger relationship between employment and more intensive informal caregiving. Panels B and C of Table VI show results from alternative categorizations of informal caregiving intensities. These results are also consistent with a negative effect of employment on informal caregiving, with a possibly stronger effect on the probability of more time-intensive caregiving. We also examine heterogeneous effects of formal employment by characteristics of the care recipient. Table VII reports the IV probit results from estimating Equation (1) with the column heading indicating alternative dependent variables. Column 1 reports the estimated marginal effect of work hours on the likelihood of Table VI. Ordered probit estimation of work hours on caregiving (marginal effects on the predicted probabilities)

Panel A: log (work_hours + 1) Panel B: log (work_hours + 1) Panel C: log (work_hours + 1)

Care hours = 0

0 < care hours ≤ 20

Care hours > 20

0.223* (0.117) 0.220* (0.123) 0.225 (0.111)

0.052** (0.017)

0.171 (0.134)

0.052** (0.017)


0.08*** (0.019)


0.028** (0.003)

Care hours > 40

0.14 (0.142) 0.145 (0.131)

Number of observations

First-stage F-statistic

38,506

16.16

38,506

16.16

38,506

16.40

Dependent variable = care hours categories (see row headings for the categories), endogenous variable = log(work_hours + 1), instrumental variable = state unemployment rate. Data from sample of women aged 40–64 years in Survey of Income and Program Participation 1996, 2001, and 2004 panels. All regressions control for age, age squared, age cubed, state fixed effects, and year fixed effects, as well as the full set of controls in Table III. All regressions are instrumental variables ordered probit regressions. The panels only differ in the cutoffs used to define the categories of the dependent variable. Panel A dependent variable care hours has three categories, using 0 and 20 h as cutoffs (i.e., care hours being zero, care hours positive but less than or equal to 20 h, and care hours greater than 20 h). Panel B dependent variable care hours also has three categories but uses 0 and 40 h as the cutoffs. Panel C dependent variable care hours has four categories, using 0, 20, and 40 h as cutoffs. Robust standard errors clustered at the state-year level in parentheses. ***p < 0.01; **p < 0.05; *p < 0.1. Copyright © 2015 John Wiley & Sons, Ltd.


840


Table VII. Heterogeneous effects by care recipients (marginal effects reported) (1)

(2)

(3)

IV Probit Variables

Whether any care to household members

Whether any care to non-household members

Whether any care to parents

log(work_hours + 1) Observations First-stage F-statistic

0.304*** (0.063) 33,413 9.55

0.039 (0.115) 38,489 16.16

0.204 (0.148) 38,489 16.16

dependent variable = (see the column heading), endogenous variable = log(work_hours + 1), instrumental variable = state unemployment rate. Data from sample of women aged 40–64 years in Survey of Income and Program Participation 1996, 2001, and 2004 panels. All regressions control for age, age squared, age cubed, state fixed effects, and year fixed effects, as well as the full set of controls in Table III. Robust standard errors clustered at the state-year level in parentheses. ***p < 0.01; **p < 0.05; *p < 0.1.

providing care to a household member. The negative marginal effect is somewhat larger than our preferred estimate in Table III for overall caregiving and statistically significant. It is large relative to the sample mean of caregiving to household members (4%). This large local average treatment effect is plausible if women whose employment is sensitive to state economic conditions have a high informal caregiving rate. The dependent variable in column 2 is an indicator for caregiving to a non-household member; there is no evidence that work hours reduce non-household caregiving. Relatedly, research on the effect of informal caregiving on employment has found a stronger trade-off when the caregiving is among household members than among nonhousehold members (e.g., Heitmueller, 2007; Carmichael et al., 2010). In particular, Carmichael et al. (2010) note that potential caregivers who are not working may feel extraordinary pressure to care for family members more than to care for someone who is not in the same household. Cohabitation with care recipients is itself a decision that could correlate with local business cycles and could also result in more caregiving (implying an endogeneity issue). State-level recessions may reduce household income and wealth and increase the cohabiting rate; thus, our main estimates in Table III could reflect a recession effect on cohabiting (which incidentally encourages informal caregiving) rather than a true effect of employment on informal caregiving. While we cannot completely rule out this interpretation, we think it is not as compelling as our interpretation. Our specifications control for the main channels through which this might operate: family income and household wealth. In addition, cohabitation due to care recipient’s financial situation is unlikely once we control for year effects; elderly people’s incomes are likely sheltered from local shocks once we control for year effects, because they rely more on Social Security and returns to financial investments (presumably diversified enough to be independent of local shocks). Table VII column 3 investigates caregiving to parents. Potential caregivers probably place more weight on their parents’ well-being than on other people’s, so working more hours might affect caregiving to parents less than it affects caregiving to other people. On the other hand, a woman who exits the labor market may be more willing to use her time to care for parents than for others. Our column 3 estimate of 0.204 is very similar in magnitude to our main marginal effect estimate of 0.206 in Table III, indicating that work hours influence caregiving to parents and others similarly.

5. CONCLUSION We use a large nationally representative dataset covering multiple years to estimate the effect of formal employment on informal caregiving in the USA. To address the endogeneity issue, we instrument for individual formal employment using state-level unemployment rates. We find that among women of prime caregiving age Copyright © 2015 John Wiley & Sons, Ltd.



841

(40–64 years old), working 10% more hours per week reduces the probability of providing any informal care by about 2 percentage points. The effect of formal employment on caregiving to household members is stronger than that on caregiving to non-household members, and we find suggestive evidence that the effect is also stronger on more time-intensive caregiving. As with all instrumental variable estimators, our IV probit estimation strategy estimates a local average treatment effect, in this case, for those individuals whose formal employment is responsive to local labor market fluctuations. Our estimates therefore may not necessarily generalize to the entire population. Because statelevel economic conditions tend to have the largest effects on the employment of the less educated and members of minority groups (Hoynes et al., 2012), our estimates are most relevant to women with relatively scarce financial resources. Also note that the effect we estimate is a partial equilibrium effect because we investigate the influence of relatively high-frequency changes in state unemployment rates on individuals’ work hours. This partial equilibrium effect could potentially be different from the general equilibrium effect of large-scale and permanent movements toward employment, such as the long-term increase in women’s labor force participation occurring over the past several decades. Nevertheless, our results are consistent with increased employment opportunities putting downward pressure on informal caregiving supply by women. This implies that there will be continuing strains on the longterm care infrastructure in the USA. Increasing demand for elderly care brought on by demographic trends only adds to the strains. Unfortunately, the formal caregiving sector provides no panacea. Nursing home care is often prohibitively expensive to families, the private insurance market is very small, and the government insurance program Medicaid offers incomplete coverage with copays and coinsurance almost equal to one’s wealth and income (Norton, 2000). A government-run compulsory long-term care insurance program—such as the one enacted in Germany in 1995—could directly increase the provision of elderly care (Mellor, 2000), but the likelihood of increased social spending for a new entitlement program in the USA is minimal. As a result, many elderly still largely rely on informal care, the majority of which is provided by women. Policymakers should consider the trade-off between informal care needs and formal employment when regulating health care and labor markets. Some labor policies may help, such as mandatory flexible work schedules, the ability to work remotely, and other ‘family-friendly’ policies. The US federal policy, primarily through the Family and Medical Leave Act of 1993, however, mandates less coverage for paid leave and work flexibility than most other countries (Heymann et al., 2008). On the other hand, companies acting in their own best interests have been increasingly adopting more flexible work environments. For example, significantly more Americans worked at home in 2010 than in 2000 (Shah, 2013). Such changes may relieve some of the tension between informal care needs and formal employment, even without explicit public policy changes. However, existing research points to limitations of the ability of flexible work policies to increase informal caregiving (Bryan, 2012). More research on the relationship between flexible work and informal caregiving would be timely. Ultimately, the nation needs to strike a balance between encouraging more people to work and satisfying the growing needs of elderly care in an aging society.

CONFLICT OF INTEREST The authors state that there is no potential conflict of interest between themselves and others that might bias this work.

STATEMENT OF ORIGINAL PUBLICATION The authors state that no part of this work has been published anywhere except as a working paper. The work is not being submitted for publication elsewhere at the same time. Copyright © 2015 John Wiley & Sons, Ltd.


842


ETHICS STATEMENT The authors state that there is no ethical violation of any sort involved in this work.

REFERENCES

Arno PS, Levine C, Memmott MM. 1999. The economic value of informal caregiving. Health Affairs 18(2): 182–188. Boaz RF. 1996. Full-time employment and informal caregiving in the 1980s. Medical Care 34(6): 524–536. Bolin K, Lindgren B, Lundborg P. 2008. Informal and formal care among single-living elderly in Europe. Health Economics 17(3): 393–409. Bryan ML. 2012. Access to flexible working and informal care. Scottish Journal of Political Economy 59(4): 361–389. Bureau of Labor Statistics. 2012. American time use survey. Charts by topic: eldercare. United States Department of Labor. http://www.bls.gov/tus/charts/eldercare.htm. Accessed May 15, 2013. Carmichael F, Charles S. 1998. The labour market costs of community care. Journal of Health Economics 17(6): 747–765. Carmichael F, Charles S, Hulme C. 2010. Who will care? Employment participation and willingness to supply informal care. Journal of Health Economics 29(1): 182–190. Ciani E. 2012. Informal adult care and caregivers’ employment in Europe. Labour Economics 19(2): 155–164. Coe NB, Skira M, Van Houtven CH. 2011. Do couples self-insure? The effect of informal care on a couple’s labor supply. Center for Retirement Research at Boston College Working Paper 2011–2016. Doty P, Jackson ME, Crown W. 1998. The impact of female caregivers’ employment status on patterns of formal and informal eldercare. The Gerontologist 38(3): 331–341. Ettner SL. 1995. The impact of ‘parent care’ on female labor supply decisions. Demography 32(1): 63–80. Gibson MJ, Houser AN. 2007. Valuing the invaluable: a new look at the economic value of family caregiving. AARP Public Policy Institute Issue Brief. http://assets.aarp.org/rgcenter/il/ib82_caregiving.pdf. Accessed June 3, 2013. Golberstein E. 2008. The effects of women’s employment incentives on their co-residence with older, disabled parents. In Essays on long-term care and aging. PhD diss, University of Michigan. Gruber J, Milligan K, Wise DA. 2009. Social security programs and retirement around the world: the relationship to youth employment, introduction and summary. National Bureau of Economic Research Working Paper 14647. Heitmueller A. 2007. The chicken or the egg? Endogeneity in labour market participation of informal carers in England. Journal of Health Economics 26(3): 536–559. Heymann J, Earle A, Hayes J. 2008. The work, family, and equity index: how does the United States measure up? Institute for Health and Social Policy at McGill University. Hoynes H, Miller DL, Schaller J. 2012. Who suffers during recessions? Journal of Economic Perspectives 26(3): 27–47. Jacobs JC, Laporte A, Van Houtven CH, Coyte PC. 2014. Caregiving intensity and retirement status in Canada. Social Science & Medicine 102: 74–82. Johnson RW, Wiener JM. 2006. A Profile of Frail Older Americans and Their Caregivers, The Urban Institute: Washington, DC. Leigh A. 2010. Informal care and labor market participation. Labour Economics 17(1): 140–149. Lilly MB, Laporte A, Coyte PC. 2007. Labor market work and home care’s unpaid caregivers: a systematic review of labor force participation rates, predictors of labor market withdrawal, and hours of work. The Milbank Quarterly 85(4): 641–690. McInerney M, Mellor JM. 2012. Recessions and seniors’ health, health behaviors, and healthcare use: analysis of the Medicare current beneficiary survey. Journal of Health Economics 31(5): 744–751. McGarry K. 1998. In Inquiries in the Economics of Aging, Wise DA (ed.), Caring for the Elderly: The Role of Adult Children, University of Chicago Press: Chicago. McGarry K. 2006. In Health Care Issues in the United States and Japan, Wise DA (ed.), Does Caregiving Affect Work? Evidence Based on Prior Labor Force Experience, University of Chicago Press: Chicago. Mentzakis E, McNamee P, Ryan M. 2009. Who cares and how much: exploring the determinants of co-residential informal care. Review of Economics of the Household 7(3): 283–303. Mellor J. 2000. In Care Work: Gender, Labor, and Welfare Studies, Meyer MH (ed.), Filling in the Gaps in Long-term Care Insurance: Policy Implications for Care Workers, Routledge Press: London. Michaud PC, Heitmueller A, Nazarov Z. 2010. A dynamic analysis of informal care and employment in England. Labour Economics 17(3): 455–465. Moen P, Robison J, Fields V. 1994. Women’s work and caregiving roles: a life course approach. Journal of Gerontology: Social Sciences 49(4): S176–S186. Copyright © 2015 John Wiley & Sons, Ltd.



843

National Center for Health Statistics. 2012. Health, United States, 2011: with special feature on socioeconomic status and health. http://www.cdc.gov/nchs/data/hus/hus11.pdf. Accessed June 19, 2013. Nizalova O. 2012. The wage elasticity of informal care supply: evidence from the health and retirement study. Southern Economic Journal 79(2): 350–366. Norton EC. 2000. In Handbook of Health Economics, Culyer AJ, Newhouse JP (eds.), Long-term Care, Elsevier: Amsterdam. Pavalko EK, Artis JE. 1997. Women’s caregiving and paid work: casual relationships in late midlife. Journal of Gerontology: Social Sciences 52B(4): S170–S179. Robison J, Moen P, Dempster-McClain D. 1995. Women’s caregiving: changing profiles and pathways. Journal of Gerontology: Social Sciences 50B(6): S362–S373. Shah N. 2013. More Americans working remotely. Wall Street Journal. http://online.wsj.com/article/ SB10001424127887324539404578342503214110478.html Stern S. 1995. Estimating family long-term care decisions in the presence of endogenous child characteristics. Journal of Human Resources 30(3): 551–580. Van Houtven CH, Coe NB, Skira MM. 2013. The effect of informal care on work and wages. Journal of Health Economics 32(1): 240–252. Van Houtven CH, Norton EC. 2004. Informal care and health care use of older adults. Journal of Health Economics 23(6): 1159–1180. Van Houtven CH, Norton EC. 2008. Informal care and medicare expenditures: testing for heterogeneous treatment effects. Journal of Health Economics 27(1): 134–156.



Combining Formal and Informal Caregiving Roles: The Psychosocial Implications of Double- and Triple-Duty Care.

Appraisals, burnout and outcomes in informal caregiving.

Health risks and informal employment in South Africa: does formality protect health?

Cost of informal caregiving for patients with heart failure.

Diabetes self-management support using mHealth and enhanced informal caregiving.

Dementia care: intersecting informal family care and formal care systems.

Informal and formal care of older people in Iceland.

Residence differences in formal and informal long-term care.

Rural dementia carers: formal and informal sources of support.

Formal and informal classroom settings: effects on hyperactivity.

Informal employment and health status in Central America.

Learning in later life: participation in formal, non-formal and informal activities in a nationally representative Spanish sample.

Feeling the squeeze? The effects of combining work and informal caregiving on psychological well-being.

Predictors of caregiving satisfaction in informal caregivers of people with dementia.

Quantifying the burden of informal caregiving for patients with cancer in Europe.

Informal caregiving and its impact on health: a reappraisal from population-based studies.

Research and commentary. Impact of informal caregiving roles on children and young people.

The effect of free personal care for the elderly on informal caregiving.

Cost and burden of informal caregiving of dependent older people in a rural Indian community.

Improving caregiving competence, stress coping, and mental well-being in informal dementia carers.

Socioeconomic and demographic factors modify the association between informal caregiving and health in the Sandwich Generation.

Cost of informal caregiving associated with stroke among the elderly in the United States.

Socio-demographic determinants of informal caregiving: co-resident versus extra-resident care.

positive home-based caregiving appraisals by informal carers of the elderly in Japan.