Caregiving for family members is often described as a 36-hour day. Previous literature has suggested that family caregivers have little time to attend to their own health needs, such as participating in leisure-time physical activity. Using data from the Health and Retirement Study, we analyze whether time-allocation decisions reflect a conflict between time devoted to informal care and time devoted to self-health promotion through physical activity. The empirical model is a system of four correlated equations, wherein the dependent variables are hours spent caregiving, frequency of moderate and vigorous physical activity, and hours spent in paid work. Results from joint estimation of the four equations indicate limited evidence of a competition between time spent in caregiving and frequency of physical activity. Parental factors that increase allocation of care time to parents do not comprehensively induce reductions in the frequency of any type of physical activity, or in hours of work, among either men or women.
As the American population continues to age rapidly, the provision of long-term care (LTC) for older Americans becomes an increasingly urgent issue. The current system of LTC relies heavily on informal—that is, unpaid—caregivers. Informal caregivers are predominantly middle-aged females, a majority of whom are employed (Alecxih et al. 2002). The typical care recipient, most often a parent, is also female (Alecxih et al. 2002), mainly because women tend to live longer than men and often outlive their spouses. According to Donelan et al. (2002), family caregivers devote a substantial amount of time helping a relative: about one-third of family caregivers report providing 21 or more hours of help per week. In addition, many caregivers provide assistance for long periods of time, with more than 40 % having provided help for at least five years (Donelan et al. 2002).
Numerous studies have examined the adverse effects of caregiving on the physical and psychological health of informal caregivers. For instance, Pinquart and Sorensen (2003) conducted a meta-analysis of 228 studies and concluded that studies consistently report higher levels of depressive symptoms and mental health problems among caregivers than among noncaregivers. However, despite the existence of a large literature on “caregiving burden,” only a few studies have specifically analyzed the relationship between caregiving and leisure-time physical activity (Burton et al. 1997; Gallant and Connell 1997; King and Brassington 1997; Lim and Taylor 2005; Scharlach et al. 1997; Vitaliano et al. 2002).
Much of this work also suffers from methodological problems, including the use of nonrandom samples, control groups that are matched on the basis of only a few observables, and poorly specified empirical models. The literature also contains limited evidence on where the time spent in care comes from: it does not appear to have been diverted from home production time, and there is little agreement on whether it comes from time that was previously devoted to work.
In this article, we examine the relationship between parental caregiving and physical activity from a time-allocation perspective. Specifically, we ask whether a trade-off exists between hours spent in informal care and the frequency of leisure-time physical activity. An empirical model that specifies a jointly estimated system of four equations using pooled data from the Health and Retirement Study (HRS) is used to answer this question. Given the current level and projected growth of informal care provision, this is an important step toward designing effective caregiver support programs.
Physical activity relates to informal caregiving in at least two ways (Etkin et al. 2008). First, as a coping mechanism, it has the potential to buffer the impact of stressors (Howard et al. 1984). That is, if caregivers engage in physical activity, the resultant stresses associated with behavior and memory problems of care recipients might be better tolerated. Second, from a time-allocation perspective, more time spent in care provision might mean less time for physical activity (Castro et al. 2002).
Although the role of physical activity as a mediator of stress has been widely demonstrated (Boise et al. 2000; Castro et al. 2002), only a limited number of studies have examined the impact of caregiving on physical activity. A few of these studies have found that elderly caregivers to a spouse reported doing less exercise (Vitaliano et al. 2002), decreasing physical activity since becoming caregivers (Gallant and Connell 1997), and having less time for exercise than same-aged married noncaregivers (Burton et al. 1997; Vitaliano et al. 2002). Other studies have found no difference between caregivers and noncaregivers on rates of physical activity (Lim and Taylor 2005) or weekly exercise (King and Brassington 1997; Scharlach et al. 1997).
Not only is a consensus on this topic absent, but the methodology employed in many of these studies raises serious internal validity concerns. For instance, some studies have not included a control group for noncaregivers (Gallant and Connell 1997), whereas others have matched caregivers with noncaregivers only on the basis of a few observables—usually, age and gender (Burton et al. 1997; Vitaliano et al. 2002). Finally, studies (e.g., Lim and Taylor 2005; Scharlach et al. 1997) have also often failed to recognize that because caregiving and physical activity are two kinds of time uses decided by the same person, modeling one decision as a function of the other is inappropriate; instead, the two should be viewed as jointly chosen outcomes.
When discussing the impact of caregiving on physical activity, it is important to acknowledge recent evidence in support of the “healthy caregiver hypothesis”; this research suggests that as high-intensity care responsibilities increase, caregivers achieve better physical functioning because such responsibilities themselves may include a large physical component (Bertrand et al. 2006; Fredman et al. 2006). In other words, the physical activity inherent in performing certain care tasks may help to improve caregivers’ physical and cognitive health. Although potentially true, the healthy caregiver hypothesis does not preclude the need for leisure-time physical activity. Physical activity performed as a part of the caregiving process may not act as a mediator of caregiving stress; and to the extent that stress acts as a risk factor to a caregiver’s mental and physical health, leisure-time physical activity has an important role to play.
In recognition of the overall health benefits conferred by physical activity, the Centers for Disease Control and Prevention (CDC) (2008) has issued guidelines relating to minimum physical activity levels for children, adults, and older adults. Specifically, according to the CDC, to achieve important health benefit, adults need at least 30 minutes of moderate physical activity (i.e., brisk walking) five days per week as well as muscle-strengthening activities two or more days per week. Alternatively, adults could also do 20 minutes of vigorous physical activity (i.e., jogging or running) three days per week in addition to muscle-strengthening activities on two or more days per week to achieve the same benefits.
If increased care responsibilities are associated with a reduced frequency of physical activity such that the probability that individuals engage in physical activity “multiple days per week” is substantially reduced, existing caregiver support programs may need to be adjusted to include components that promote physical activity. Currently, the National Family Caregiver Support Program (NFCSP) is the key government program that provides assistance to family caregivers. With an overall budget of $154 million in FY 2010 (Administration on Aging 2013), the main components included in the NFCSP are provision of counseling, support groups, and respite care (Administration on Aging 2013). If caregivers’ physical activity levels are falling below the recommended guidelines, the NFCSP might usefully be extended to include strategies that promote physical activity among caregivers. For example, a telephone-supervised, home-based, physical activity regimen has been demonstrated in past research to be one such strategy for increasing physical activity in the caregiving population (King and Brassington 1997).
In addition to the policy context discussed earlier, a key motivation for this study relates to limitations of previous literature on caregiving and time use. Given the time demands associated with caregiving, researchers have attempted to address the trade-off between time spent in care activities and time spent in other activities. This work, however, has largely concentrated on estimating the causal relationship between caregiving and employment. Initial analyses were generally based on cross-sectional models of labor market participation and parental care (Muurinen 1986; Stone and Short 1990; White-Means 1992), but these studies failed to consider the endogeneity problem that arises when estimating the causal effect of informal care on work. Specifically, reduced-form estimates are prone to selection bias if adult children who have poor labor market prospects are also more likely to take on the caregiving role.
Newer research has tried to address this endogeneity problem in a variety of ways, but results from these studies have been mixed. Wolf and Soldo (1994) estimated a simultaneous equations model and found no reduction in married women’s employment or hours of work as a result of caregiving; in contrast, Ettner (1995) found that women’s labor supply is significantly reduced by coresidence with an elderly disabled parent, primarily because of withdrawal from the labor market. She used predictors of parents’ health status (education, age, and marital status) and the number of siblings as instrumental variables for coresidence in a two-stage estimation. Van Houtven et al. (2013) examined both intensive and extensive margins of labor market participation, controlling for time-invariant individual heterogeneity via fixed effects. They also controlled for any remaining endogeneity by using measures of parental health as instruments for informal caregiving. They found that care provision does not affect a woman’s probability of working and that caregiving is associated with relatively moderate reduction in work hours, with large decreases concentrated mostly among high-intensity caregivers.
The lack of consistent, conclusive evidence that informal care and labor supply decisions are negatively correlated has led some researchers to speculate that a time trade-off may come from activities other than work, such as housework or leisure (Wolf and Soldo 1994). Couch et al. (1999) considered four demands on household time and money resources: time spent working, time spent providing care to elderly parents, time spent performing household work, and monetary transfers to noncoresident elderly parents. Their findings indicate that parental factors associated with increased time transfers to parents do not induce reductions in either labor market or housework time.
The absence of negative correlations among competing time uses suggests that there is scope to further disaggregate time spent in leisure activities into other “productive” uses of time, such as physical activity. Because there are only 24 hours in a day, time for care must be found somewhere. This article attempts to determine whether care time is subtracted from time that would have otherwise been devoted to activities relating to investment in one’s own health—specifically, to exercise.
Time-allocation studies are grounded in the New Home Economics models of the 1960s, which view families as engaging in production of goods, much like a firm (Becker 1965). Families convert time, material resource inputs, and purchased services into abstract household goods. In particular, Becker’s approach recognized the importance of time allocation in the production of household goods.
In simple models, two activities fully accounted for time available: time in the labor market and time in leisure. Gronau (1977) expanded the traditional two-dimensional time-allocation model to three uses of time: market work, leisure, and home production. Because these categories are mutually exclusive and exhaustive, when two of the three categories are determined, the third is implied. The optimal time allocation by each individual depends on the value of time in each activity (or, the opportunity cost of that time) as well as the preferences of the individual. Thus, an individual’s time-allocation decisions are not only simultaneously determined but are also dependent on tastes and other unobserved variables that may be correlated across outcomes (Couch et al. 1999).
Following Gronau’s work, Kooreman and Kapteyn (1987) modeled the allocation of time by couples between market work and a variety of nonmarket activities, including home production, child care, hobbies, and personal care. Although their model takes account of multiple time uses, it does not allow for cross-equation correlations between them.
More recently, in studying mothers’ time use, Kimmel and Connelly (2007) further expanded the Gronau triad into five aggregated uses of time: (paid) market work, unpaid housework, child care, leisure, and other (including sleep, personal care time, education, and so on). Their results show that mothers’ caregiving time increases with the number of children, decreases with age of the child, and increases with price of child care.
Model Specification and Estimation
As discussed earlier, we assume that households are productive units whose primary resources are time and money. Because caregiving and physical activity are two kinds of time use, it is theoretically indefensible to model one decision as a function of the other. Such an equation would not have a precise ceteris paribus interpretation because the amount of time devoted to both care and physical activity is decided by the same individual (Wooldridge 2002). The interrelationships among time uses suggest that all factors associated with the value of time spent in any one activity influence decisions about all other time allocations. In this case, one cannot hold time spent in caregiving fixed because anything that influences the decision on how much time to spend on physical activity also simultaneously influences the decision regarding care hours and all other uses of time. This also rules out the existence of an exclusion variable that influences one category of time allocation without affecting all other types of time uses.
Here, X is an array of explanatory variables common to all equations.
These four equations are estimated jointly through a mixed-process model that included an “ordered probit with known thresholds” (Lillard and Panis 2003) for care hours,1 two probits for the physical activity outcomes, and a Tobit for work hours. The joint estimation was done using the statistical software aML (Lillard and Panis 2003). To account for clustering in the pooled data, a time-invariant, person-level random effect was included in each equation with correlations at the person-level across equations.
The analysis uses data from four waves (2004–2010) of the Health and Retirement Study (HRS), which is a nationally representative, biannual survey of the near-elderly in the United States (Juster and Suzman 1995). Persons aged 51–61 entered the sample initially, thus making their parents prime candidates to be care recipients. The HRS collects not only detailed information about the respondents and their spouses but also important information about their parents and siblings. We use only four waves because questions on the frequency of various types of physical activity were asked for the first time in 2004.
Although the HRS employs the recall method to collect data on various time uses, time diaries (such as those collected by the American Time Use Survey, ATUS) are generally considered the gold standard for measurement of time use mainly because they collect time-allocation data in a structured way and involve a relatively short recall period (Van den Berg and Spauwen 2006). This gold standard is, however, not universally accepted. According to Bittman et al. (2004), time diaries fail to capture the real burden of caregiving and therefore can potentially underestimate time inputs. Self-reported hours of care may reflect supervisory time—that is, a need to be “on call.” Diaries, on the other hand, are designed primarily to record activities, and being on call seldom shows up as an activity (Budig and Folbre 2004).
An additional complication is that care activity may be embedded in and absorbed into normal domestic activity (Wolf 2004). For example, in time diaries, joint production of a meal to be consumed by both the caregiver and the care recipient is already assigned a domestic activity code and does not register as a specific activity associated with caregiving. In evaluating the accuracy of the recall method for measuring time spent in informal care, Van den Berg and Spauwen (2006) concluded that if one assumes that respondents take into account joint production when completing the recall questionnaire, the recall method is a valid instrument to measure time spent on the provision of informal care.
For these reasons and because the ATUS does not collect data on parental need factors (which tend to provide key exogenous variation for time spent in care tasks), we use the HRS for this analysis. Any measurement error in reporting time use attributable to recall or social desirability bias will be present in the dependent variables, which has relatively less serious consequences (Bound et al. 2001).
We focus exclusively on care provided by adult children for parents because parental caregiving is the most common care scenario and is also the most relevant given the middle-age profile of the HRS respondents. In contrast, spousal caregiving is a commonly reported care situation for those older than 75 years.
We restrict the sample to those respondents who have at least one parent alive or a parent-in-law alive, or those who have experienced the recent death of either a parent or a parent-in-law. Respondents who have experienced the death of a parent or a parent-in-law since the time of the last interview are included because a substantial amount of care is provided at the end-of-life stage. Further, to ensure that respondents were of working age—that is, those for whom labor market participation is most relevant—we restricted the sample to individuals younger than age 65.
These sample selection criteria led to a total of 8,998 observations. After deleting observations with missing values (3.8 %),2 a final sample size of 8,658 was achieved. Of these, 3,892 (45 %) are women, and 4,766 (55 %) are men. Individuals appear in the pooled analysis file from one to four times, depending on how often they meet the inclusion criteria.
The four dependent variables (Table 1) used in this analysis are hours of care, frequency of moderate physical activity, hours of vigorous physical activity, and hours spent in paid employment. Hours of care is asked only if the respondent provided at least 100 hours of care in assisting parents with activities of daily living (ADLs) or instrumental activities of daily living (IADLs). ADLs include self-care activities, such as eating, bathing, dressing, transfers, and walking. The survey question is worded as follows: “Did you spend a total of 100 or more hours [since last interview, in the last two years] helping your (deceased) [parents/in-laws] with basic personal activities like dressing, eating, and bathing?” Unlike ADLs, IADLs are not necessary for fundamental functioning, but they facilitate independent living in a community. Similar to the question on ADLs, the HRS asks for IADLs, “Did you spend a total of 100 or more hours [since last interview] helping your (deceased) [parents/in-laws] with other things such as household chores, errands, transportation, etc.?” If the answer either of these questions is “yes,” the respondent is asked, “Roughly how many hours did you spend [since last interview, in the last two years] giving such assistance to your parents/in-laws?” This question is repeated, separately, for the spouse.
We used a combined measure of caregiving (personal care and chores) as the dependent variable in the first equation. That is, the effective time (over two years) spent caring for parents is defined as the sum of time spent helping parents with basic personal needs and time spent helping parents with household chores. Although some studies separate ADL and IADL care to account for heterogeneity of response based on the type of care, it is appropriate from a time-allocation perspective to combine the two because ADL care is frequently accompanied by IADL care, thus providing a more comprehensive measure of the total amount of time spent in informal care.
As discussed in the Background section, both ADL and IADL caregiving tasks may involve physical activity. It is important to acknowledge that there is likely to be a tremendous amount of heterogeneity in the degree of physical exertion involved in care tasks across individuals. However, from an empirical perspective, we are unable to take this heterogeneity into account because our data do not provide this level of detail on the nature of individual-care scenarios.
A sizable proportion of the respondents who responded “yes” to providing more than 100 hours of ADL also answered “don’t know” to the subsequent question on the actual number of care hours. This may reflect the difficulty of recalling the intensity of care efforts as much as two years in the past. A follow-up question then asked these respondents to choose from three possible ranges of care hours: 0–199 hours; 200–499 hours; and 500–5000 hours.3
Table 1 presents summary statistics for “hours of care” for women and men. Interval-coded values appear as lower and upper bounds, respectively.4 For those who specified a value for their care hours, the same value appears as a lower and upper bound. The unconditional mean (i.e., including zeros) for the lower bound of care hours for women is approximately 198 hours over a two-year period, while the same for the upper bound is 684 hours. Not surprisingly, these numbers are much higher for women than men, suggesting that in general, women provide more hours of care. Also, as expected, for caregivers5 (42 % among women and 36 % among men), the mean values for both the lower and upper bounds are considerably higher than those for the entire sample of women and men (unconditional means).
With regard to the frequency of different types of physical activity, the HRS asks respondents how often they engage in three kinds of physical activity: mild, moderate, and vigorous. Because the CDC guidelines are limited to only moderate and vigorous physical activity, we don’t include mild physical activity in our analysis.
For vigorous physical activity, the HRS asks, “How often do you take part in sports or activities that are vigorous, such as running or jogging, swimming, cycling, aerobic or a gym workout?” Similarly, for moderate physical activity, the HRS asks respondents, “How often do you take part in sports or activities that are moderately energetic, such as brisk walking, gardening, cleaning the car, dancing, floor or stretching exercises?” The response categories for both questions are “hardly ever or never,” “one to three times a month,” “once a week,” and “more than once a week/every day.” Given the inherently physical nature of many care tasks, there might be concern that some respondents think of caregiving as physical activity. However, because the questions contain explicit examples, it is unlikely that individuals included care tasks in their responses on physical activity participation.
In a further effort to keep the analysis relevant to CDC guidelines, we created dichotomous variables expressing physical activity frequency as either “multiple times per week” or “less than multiple times per week” for both moderate and vigorous physical activity. Table 1 demonstrates that approximately 56 % of women and 60 % of men report doing moderate physical activity multiple times per week. In contrast, only 25 % of women and 33 % of men report doing vigorous physical activity multiple times per week. It is not surprising that a higher percentage of men and women frequently perform moderate physical activity. In addition to being physically more strenuous, vigorous activity also entails a higher investment on the part of an individual. For example, one can do stretching exercises (an example of moderate physical activity) in the confines of his/her residence, but an aerobic workout or a swim (examples of vigorous physical activity) will entail a gym membership, special attire, and other sports gear. Thus, even from a nonphysical perspective, engaging in moderate physical activity is relatively more feasible. In general, the descriptive statistics also show that men are likely to engage in both types of physical activity more frequently than women.
The hours of work variable is taken from the RAND HRS (2011) data files. It is the sum of the typical number of hours per week that the respondent works at the main job and at a secondary job, if any. If the respondent is not working, the hours of work are coded to zero. Table 1 shows substantial differences between the unconditional and conditional means for employment hours because 39 % of women and 29 % of men do not work. When everyone is included, the mean hours spent in paid employment are about 23 hours per week for women and 32 hours per week for men; when considering only those who are employed, the mean hours spent in paid employment are approximately 38 hours per week for women and 45 hours per week for men.
The explanatory variables (summarized in Table 1) include characteristics of parental health and need, Pi; the respondent’s individual and demographic characteristics, Zi; characteristics of respondent’s health, Hi; and characteristics of the respondent’s household, Di.
The vector Pi represents the key explanatory variables—those that measure parental health status. Separate indicators for mother, father, mother in-law, and father in-law are included in the regression analyses. As parental health declines, hours of care are hypothesized to increase, and frequency of physical activity and hours of work are expected to decrease. These variables include a dummy variable for whether the parent can be left alone for an hour (coded as 1 if the parent cannot be left alone for an hour) and another indicating whether the parent has ADL needs. Almost all respondents who indicated that the parent cannot be left alone for an hour also reported that the parent has ADL needs. To avoid overlap, the measure for whether the parent has ADL needs excludes those who indicated that the parent cannot be left alone for an hour; thus, the “ADL needs” variable represents needs beyond any revealed by the “cannot be left alone” variable.
We also control for parents’ marital status. If parents are married (to each other or otherwise), the adult child is less likely to provide care because of the availability of an alternate caregiver (the parent’s spouse). The three “need” indicators discussed earlier (including parent’s marital status) pertain only to parents who were alive at the time of the interview.
Finally, for those individuals whose parents died recently (i.e., since the last interview), we include a variable indicating whether the parent died as the result of an illness lasting three months or more, and another variable indicating whether the parent died without an illness lasting three months or more. Because the period before death is likely to characterize a significant need for care regardless of cause, we expect both variables to positively predict care hours. That said, we hypothesize that death attributable to an illness would lead to a higher increase in care hours compared with death without an illness. Similarly, we expect that death following an illness would lead to a larger decrease in the frequency of physical activity and hours of work (if any) compared with death not resulting from an illness. Disaggregating parent’s recent death by cause allows us to examine how different levels of parental need influence various time-allocation decisions.
As depicted in Table 1, almost 60 % of the women have a living mother, but only 24 % have a living father. Although the numbers are smaller for men, the trend is similar. Not only are more mothers alive than fathers, but mothers are also more likely to require care: 7 % of the women report that their mothers cannot be left alone for an hour, but only 2 % of the women report that their fathers cannot be left alone for an hour. Similarly, even among men, 7 % report their mothers have ADL needs, but only 3 % report that their fathers have ADL needs.
The vector Zi includes the respondent’s age, education, race, marital status, number of siblings (number of sisters and number of brothers separately), the natural log of hourly wage, and wealth. The mean age for women and men is 58 and 59 years, respectively. Approximately 78 % of women and 84 % of men are white. Further, a large majority of both men and women are married or living together.
As discussed earlier, for a large percentage of individuals in the sample—those not presently working—the hourly wage rate is missing. Because this is suggestive of selection bias, we predict wages for the entire sample (men and women together) using a two-step Heckman estimation procedure (results available on request). Variables included in the reduced-form employment equation that are not included in the log-wage equation include measures of parental need, spousal need, respondent health, and marital status. Age, education, experience, a second-order term for experience, and gender are included in the wage equation.6
In the first-stage estimation, several of the identifying variables, including spousal need and some measures of parental health, are significant predictors of employment. In addition, the F test of the wage equation estimation is also significant at the .01 level, indicating that all the included variables contribute to the prediction of wage levels. After the second stage, selectivity-corrected predicted log wage is imputed for all respondents. As expected, the mean predicted log wage is higher for men than women.
We also control for nonhousing financial wealth. This includes the sum of appropriate wealth components (such as savings and checking accounts, mutual funds, stocks, bonds, and Treasury Bills) minus debt. This measure is taken from the RAND version of the HRS. Because the RAND version of the HRS imputes values for those who either gave no response or an interval response, there were no missing observations.
The vector Hi includes plausibly exogenous variables that determine a respondent’s health status. Health status is likely to have an effect on how the respondent decides to allocate his/her time. Hi, consists of two indicator variables: (1) whether the respondent ever smoked (specifically, smoked 100 cigarettes or more in his/her lifetime), and (2) a lagged measure (from the previous wave of the HRS) of the respondent’s body mass index (BMI). Other health status variables in the HRS—for example, self-reported physical and mental health status—are not included because they may be simultaneously determined along with the dependent variables and thus may be subject to endogeneity bias.
Finally, household characteristics, Di, include indicator variables for spouse’s health, the number of coresident children younger than age 18, spouse’s age, number of spouse’s siblings, and spouse’s work experience. Spousal health characteristics and the number of coresident children are intended to capture intrahousehold demands on an individual’s time. With deteriorating spousal health, we expect that an individual’s time allocation in parent care, physical activity, and work will compete with his/her time allocation in spousal care. We use spouse’s ADL needs (as measured by the sum of five possible ADL limitations) to control for spousal health. Similarly, time allocation in any activity not related to childcare may also decrease as the number of children younger than age 18 in the household increase.
It is important to recognize that among married individuals, the decision regarding time transfers to parents or parents-in-law is taken in a household context. For example, spousal time allocation in parental care is a key predictor of the respondent’s time allocation in the same task. From a modeling perspective, however, explicitly controlling for the amount of time the spouse spends in various activities creates an endogeneity problem. Therefore, in order to represent this intrahousehold decision-making context, we include certain exogenous variables that are likely to directly influence spouse’s time-allocation decisions. These variables include spouse’s age, spouse’ work experience, and spouse’s siblings (number of sisters and number of brothers separately). The control for spousal health described earlier is an additional control for this intrahousehold decision-making context. Note that for those individuals who don’t have a spouse, these variables are coded to zero.
Tables 2 and 3 (for women and men, respectively) present the estimated marginal effects on the unconditional expected value of hours spent in caregiving, predicted probabilities of engaging in moderate and vigorous physical activity multiple times a week, and unconditional expected value of hours spent in employment. Marginal effects are slopes for continuous explanatory variables (e.g., wage rates) or discrete changes in the outcome variable for dummy-coded explanatory variables (e.g., mom has ADL needs). All standard errors are Huber-corrected.
As expected, care hours respond positively to almost all parental need characteristics. Having a mother or father who cannot be left alone for one hour or who requires help in completing ADL tasks leads to significant increases in care hours. For instance, among women, having a mother who cannot be left alone for an hour increases average care hours by 214 hours over a two-year period. Among men, having a father who cannot be left alone for an hour increases average care hours by 114 hours over a two-year period. Further, as expected, experiencing a recent death of a parent resulting from an illness leads to a greater increase in care hours compared with a recent death of a parent without an illness. For example, among women, having had a mother who died recently with an illness increases care hours by 265 hours over a two-year period; in contrast, having had a mother who died without an illness increases care hours by a lesser amount: 187 hours. The results for men are similar.
The results suggest that women’s increases in time allocated to caregiving in response to parental needs are larger than men’s. In addition, as demonstrated in the literature previously (Lee et al. 1993), there is evidence of same-gender preferences: women are likely to increase care hours more in response to their mother’s physical needs than to their father’s. Similarly, men are likely to increase their care hours more in respect to father’s health needs as opposed to the mother’s.
Overall, there is limited support for the existence of a time conflict between hours of care and frequency of physical exercise. The results demonstrate that even though most parental-need characteristics positively predict hours of care, they do not comprehensively predict compensatory decreases in the frequency of physical activity.
Among women, having a father who needs help with ADL tasks increased the number of care hours over a two-year period by 161 hours and reduced the frequency of engaging in vigorous physical activity multiple times a week by 10 %. There is some indication that having a father who died without an illness—a type of parental need that is relatively less severe because of its presumably shorter length—led to an increase in the probability of frequent participation in physical activity. However, it is not entirely clear how having had a father who died without an illness affects care hours: for both men and women, this variable increased care hours by only a very small and statistically insignificant number. Finally, among women, having a mother who is married increased the probability of engaging in moderate physical activity multiple times per week by 10 %. The mother’s marital status negatively predicts care hours, although again, the estimate is not statistically significant. None of the other parental need variables predicted a statically significant trade-off between care and physical activity among either men or women.
Interestingly, among men, the estimated correlations of unobservables between time spent in caregiving and the two types of physical activity are positive (see the Rho: Care row, at the bottom of Table 3). This suggests that among men, after parental need and other variables are controlled for, unobservable factors influence time allocation in caregiving and physical activity in the same direction. Some of these unobservables may include self-efficacy, beliefs regarding quality of life and health, and so on. Further, it is also likely that better mental health, which we do not include in the regressions given that it is almost certainly endogenous, positively predicts both the amount of care time and frequency of physical activity. The estimated correlations between care and physical activity are positive but are small and statistically insignificant among women.
Focusing on another type of time conflict—that between care hours and work hours—the results demonstrate that among men, having a mother who cannot be left alone for an hour negatively influences hours of work. Specifically, having a mother with intensive care needs reduces average work hours by close to four hours per week. The corresponding result for women is also negative but statistically insignificant. Similarly, facing a non-illness-related death of a mother in the last two years is associated with an increase in work hours for both men and women. As discussed earlier in the article, the literature has yet to reach a general agreement on the effect of informal care provision on labor force participation. Our results are somewhat consistent with recent evidence from the United States, suggesting that personal care assistance reduces the probability of working among men and leads women who are working to reduce work hours (Van Houtven et al. 2013). It is possible that our estimates are imprecise for women because almost 40 % of the women in our sample do not work.
With regard to error correlations, among both women and men, the correlation between care hours and work hours is negative and statistically significant. In this case, the unobservable factors provide a clear indication of a conflict between two the competing uses of time.
Theory predicts that higher wages should lead to an increase in hours of work. In addition, as the opportunity cost of time increases, hours of care should decline. The results show that an increase in predicted log wage leads to only small decreases in care hours for both men and women. Further, the results are statistically insignificant in both cases. Conclusions from recent literature help explain this result. According to Nizalova (2012), wage elasticity estimates of informal care are subject to an omitted variable bias and are thus biased upward. For example, some people may be more productive in everything they do, which is difficult to control for with a conventional set of variables available to researchers. Therefore, these people would provide more care but also would be rewarded in the market with higher wages. In addition, the price of formal care is likely to be higher for people living in high-wage areas, and a high price of formal care might mean more hours of informal care (if formal and informal care are substitutes for each other). Thus, failure to control for price of formal care would further result in an upwardly biased estimate of wage effect on informal caregiver time.
Although some previous research has suggested that blacks express stronger kinship support than whites (Dilworth-Anderson et al. 2005), the overall evidence on racial differences in caregiving has been mixed. For instance, Janevic and Connell (2001) conclude that minority groups may not have more available support than whites. In addition, prior research has also found that black elders receive more informal help mainly because blacks are more disabled (Li and Fries 2005). Nevertheless, we find that even after we control for parental need, being black (as opposed to white) increases care hours among both men and women (although the result is statistically significant only for men).
Among women, being black decreases the frequency of moderate physical activity by about 9 %. Each additional sister reduces average care hours for both men (by about nine hours) and women (by about 24 hours). Similarly, among women, each additional sister-in-law reduces women’s average care hours by about 14 hours. These variables are most likely picking up the presence of substitute caregivers.
Interestingly, indicators of health risk factors (smoking history and lagged BMI) do not significantly predict care hours for either men or women. However, as expected, each additional unit of lagged BMI is associated with decreases in the probability of engaging in both types of physical activity multiple times a week for both men and women. BMI is also associated with decreased work hours among women.
Among women, each additional ADL need of the spouse leads to a 22-hour reduction in care hours to parents. This suggests that time allocated to caring for spouse competes with time allocated to caring for parents. Another control for intrahousehold demands on the respondent’s time is the number of children in the household. Among women, there is no effect of each additional child in the household on time allocation to parents. Interestingly, among men, each additional child leads to a small but statistically significant decrease in care hours to parents. Previous literature on the phenomenon of “sandwich generation” has found very limited evidence of a competition for assistance between children and older parents (Grundy and Henretta 2006). It is possible that men are responding to their spouse’s time allocation to parents by taking on more responsibility for child care. However, this is only a speculation given that empirical research on this topic has focused almost exclusively on women (DeRigne and Ferrante 2012).
Caregiving for family members has often been described as a 36-hour day. This notion has motivated researchers to ask the following question: If family members are allocating their time to provide care, what other productive uses of their time are they giving up? In this article, we examined whether, from a time-allocation perspective, a conflict exists between care hours and frequency of leisure-time physical activity. If such a trade-off is present and if it induces a decrease in physical activity to levels below minimum guidelines set forth by the CDC, the promotion of physical activity might become a goal of publicly supported programs, such as the NFCSP.
In our joint model of time use, parental factors associated with increased allocation of time to parents do not appear to strongly induce corresponding reductions in the frequency of physical activity. Further, unobserved factors influencing time transfers to parents and frequency of physical activity—factors collectively represented by the regression error terms—were positively correlated across equations, at least among men. This positive correlation indicates that net of measured covariates, individuals who are motivated to provide care also have a taste for engaging in physical activity. Put another way, these two types of time-allocation decisions appear to be complements rather than substitutes. It may be the case that busy, active individuals are those most likely to take on caregiver tasks, but that they simply add those tasks to an already-busy schedule, contributing to the “36-hour day” image.
In the previous section, we provided a few examples of unobservable factors that might explain this positive correlation. In addition to those factors, it is possible that the variation in the time units (within the HRS) for measuring each type of activity—biannually for care hours, and weekly for exercise and work hours—led to this result. Because care hours are measured over such a long interval, estimates could potentially be biased downward. That is, over a long period (such as two years), individuals are likely to make adjustments to accommodate potential time-allocation conflicts. Alternatively, it is probable that if one had daily measures on time spent in various activities (such as from time diary data), the results would demonstrate large negative trade-offs. In a very short time interval, using one’s time in any given way tends (almost mechanically) to crowd out other potential time uses. However, such results might overstate the effects if these time conflicts are smoothed over a two-year period. Therefore, the period over which to examine these conflicts remains an open question—one with key implications for any policy intervention in this area.
The empirical results of this article are suggestive of a trade-off between caregiving and work hours. Even after controlling for parental need, predicted log wage, and other variables, the error correlations show a statistically significant, negative correlation between care time and paid work for both men and women, indicating that time spent in caring competes with time in the labor market. Also, among men, having a mother who cannot be left alone for an hour (one of the more severe need categories) reduces weekly work hours by approximately four hours.
To conclude, we found that increases in caregiving hours primarily appear to be in response to the severity of parental health needs. We found limited evidence that parental need factors were also associated with a decreasing frequency of physical activity. An extension of this work would be to further disaggregate leisure hours (into sleep, recreation, and family time) to more comprehensively answer the question, Where do care hours come from?
Earlier versions of this article were presented at the 2013 annual meeting of the Population Association of America and at the Cornell-Syracuse University Encore Conference (2012). We also received helpful comments from John Cawley (Cornell University) and from three anonymous reviewers.
“Ordered probit with known thresholds” is a generalized censored regression also known as “interval regression.”
A comparison of means between the original sample (N = 8,998) and the final sample (N = 8,658) revealed no evidence of any systematic bias resulting from missing data.
The RAND HRS family data files do provide an imputed scalar value for respondents who reported an interval measure for care hours. However, concerns regarding an imputed value’s accuracy become key given the wide range and two-year period of the interval. For this reason, we model the interval range as the dependent variable in the care-hours equation. Compared with imputed data, using intervals leads to larger variances, thus yielding tests of significance that are generally more conservative.
Observations reporting zero care hours are coded as falling into the –∞, 0 interval because “ordered probit with known thresholds” is a generalized Tobit estimation.
Those with positive ADL or IADL care hours.
To check for wage penalties, we included parental need variables in one specification of the wage equation. We found no statistically significant evidence of reduced wages as a result of poor parental health.