Dec 28, 2010 - affordable. ... evidence points toward privacy being a normal good for young adults (Card and Lemieux, 1997; Ermish,. 1999 ... on daycare costs by providing babysitting services. .... of the SEO and SRC samples in PSID.
Living Arrangements of Mothers and Their Adult Children Over the Life Course* V. Joseph Hotz Duke University NBER and IZA Kathleen McGarry University of California, Los Angeles NBER Emily Wiemers University of Michigan Latest Draft: December 28, 2010
*This paper is based on an earlier version presented at the 2010 American Economic Association Annual Meeting, January 2-5. Hotz thanks the National Science Foundation (SES-0339051), McGarry gratefully acknowledges financial support from the National Institute on Aging (R01-AG016593) and Wiemers thanks the National Institute on Aging (AG000221-17).
Abstract Early in the last century, it was commonplace for elderly women to live with their adult children. Over time the prevalence of this type of living arrangement declined as incomes increased. However, the recent recession has brought with it numerous accounts of families responding to economic hardship by forming multigenerational households. In this paper we draw on the long panel of data available in the PSID to examine the living arrangements of older women and their adult children over the life course. We pay particular attention to the relationship between socioeconomic status and coresidence. Our results suggest that for much of the life course, coresidence serves to benefit primarily the adult children rather than the older parent. We also highlight a little known phenomenon, that of children who never leave the parental home and remain coresident well into their later adult years. KEYWORDS: Coresidence, living arrangements, family support, transfers
Introduction Over 16 percent individuals in the U.S. live in households containing at least two adult generations. This figure has increased from only 12 percent in 1980 and the rate of increase has sharpened during the current economic downturn (Pew Center, 2010). The popular press is almost daily reporting stories of families responding to the financial crisis by combining households. 1 It should come as no surprise to researchers studying the family that families provide support in times of need. There is a rich literature studying financial transfers from parents to their adult children, the transfer of time from children to their parents or vice versa in the case of grandparents providing care, and finally on the role of cohabitation in facilitating caregiving late in life. Indeed, throughout much of the last century, coresidence with children was the norm for elderly women (McGarry and Schoeni, 2000). Because the literature on coresidence has focused on these types of living arrangements as they relate to support for aging parents (typically widowed mothers) little is known about the evolution of these living arrangements over the life course, the prevalence of coresidence at earlier stages in life when the arrangement might serve more to benefit the adult child than the mother, or how these patterns vary with measures of socio-economic status much earlier in life. In this paper we address these issues to provide a more complete picture of living arrangements of adult children and their mothers. We draw on data from the Panel Study of Income Dynamics (PSID) which has followed its initial respondents and their progeny for nearly 40 years, with data thus extending from the time when children are young, until they start their own households, and through the time when their mother grows old. Despite the extremely long time frame covered by the PSID and its rich data on parents and children, its potential to help us understand the living arrangements of the elderly and other aspects of family behavior has not been fully exploited. 2 We focus our attention on differences in coresi1
For example, “Facing the Financial Pinch, and Moving In With Mom and Dad,” New York Times, March 2010; Cramped quarters: As children postpone their departure households get larger,” The Economist, September 2010; and “Grandparents Role Grows as the Economy Struggles,” New York Times, November 2010. 2
A notable exception is Ellwood and Kane (1990) that use the early waves of the PSID in an event history analysis of the marital status, disability, living arrangements and income of the elderly.
1
dence by age of the mother, her health, her income and the income of her children, and other characteristics of the child like unemployment and disability that may affect the demand for transfers of care or support from parents to children. Our paper is organized as follows. In the next section, we lay out the existing literature on the subject. In section 2 we describe the data available in the PSID and provide a discussion of how we constructed our analytic sample. Section 3 provides descriptive evidence of the choice of living arrangements and section 4 models the decision more formally in a multivariate context. Section 5 briefly summarizes our results and notes some of our plans for future research.
1. Background Although the data show that coresidence of the elderly with their (adult) children is not the norm today, less than a century ago it was quite common. In 1900, approximately 70 percent of elderly unmarried women lived with their children, but by 1990, this figure had fallen to below 20 percent (McGarry and Schoeni, 2000). A variety of hypotheses have been offered to explain this decline in the fraction of elderly women living with their children. 3 These include declines in fertility that reduced the availability of children with whom women could live and the rise in labor force participation of younger women over the 20th century which made it more difficult for daughters to care for their elderly parents (Kobrin, 1976; Ruggles, 1994). Also contributing to the change were improvements in health care, health status and life expectancy as well as programs such as Medicare and Medicaid that allowed individuals to maintain their independence to older ages (Wolf and Soldo, 1988). 4 And finally, changes in attitudes and norms, including the rise in individualism and the decline in traditional values. However, the explanation that has received the lion share of support, particularly among economists, is the increase in the incomes of the el3
See McGarry and Schoeni (2000) and Ruggles (2007) for reviews of the literature on the hypotheses and evidence for explaining the long run trends in the living arrangements of the elderly.
4
The establishment of the Medicare and Medicaid programs in 1965 made medical care and home health care more affordable. The Medicare program provides health insurance for nearly all those 65 or older. In addition to acute care needs, it currently provides limited coverage of long-term care needs. Medicaid is a means tested transfer program that provides health insurance to the poor. Both programs also provide support for care for the elderly in the form of nursing homes. For example, Medicare covers some nursing home stays while Medicaid funds a larger fraction of nursing home expenditures than any other source.
2
derly through the substantial expansion of the Social Security System and work-related retirement programs, and general economic growth (e.g. Michael, Fuchs and Scott, 1980; Schwartz, Danziger and Smolensky, 1984; Costa, 1999; McGarry and Schoeni, 2000). With privacy viewed as a normal good, individuals used the increased resources to maintain their independence. In this paper we therefore pay particular attention to the role of incomes of both the parent and the child. Coresidence is an important issue for the elderly not only because of the potential financial benefits, but because of the informal care that children can provide. In the United States, the vast majority of individuals with functional disabilities live in the community with only 25 percent of such needy elderly in nursing homes. For the remaining 75 percent who are community-dwelling individuals, informal care is by far the dominant means of assistance with over 90 percent of those who receive assistance receiving some form of informal support (Health Policy Institute). 5 While a spouse is typically the primary caregiver for married elderly, for unmarried individuals, children play the leading role with large differences in the amount of care provided by living arrangement. For instance, among those caring for individuals 70 years old or older in the Asset and Health Dynamics Survey (AHEAD), non-coresident children averaged 8.5 hours of care per week while coresident children helped nearly full-time, averaging 38.5 hours (McGarry, 1998). For an elderly individual who needs a great deal of care or constant supervision, an alternative to shared living arrangements is formal nursing home care. Formal care can be provided in one’s own home or in a nursing home but both types of assistance are expensive. Home health care costs average $19 an hour or nearly $40,000 a year for 40 hours of care a week (and several times that for around the clock care). Skilled nursing facilities average $75,000 a year (MetLife, 2007). Despite what appears to be a strong dislike of institutional care and significant expense, nearly 40 percent of individuals over the age of 65 will eventually spend some time in a nursing home. The typical stay however, is short; two-thirds of
5
78 percent rely exclusively on informal care, 14 percent receive some formal and some informal care and 8 percent receive formal care only.
3
all stays last three months or less.6 The decision of whether to living in a nursing home, with children, or independently, is thus central in our understanding of the receipt of care and has important financial implications for all generations involved. Although much of the work in economics has looked at coresidence as a situation benefiting the elderly parent, in response to recent changes in the transition to adulthood, there is now a small but growing literature on coresidence between younger adults and their parents. Like work on the elderly, the focus is on identifying the effect of income and local market factors on the decision to coreside with parents. Most studies focus on the effect of the income of the child in determining living arrangements and the evidence points toward privacy being a normal good for young adults (Card and Lemieux, 1997; Ermish, 1999; Matsudaira, 2010). Several studies suggest that increases in parental income are associated with increases in coresidence (Manacorda and Moretti, 2006; Ermisch, 1999). Notably Manacorda and Moretti (2006) use a change in the mandated retirement age for certain cohorts of Italian men as an exogenous shock to parental income to identify the effect of increases in parental income on home leaving behavior of young Italian men. They argue that it is parents’ tastes for coresidence along with children’s indifference to living with parents that explains a surprising stylized fact that over 80 percent of Italian men aged 18-30 live with their parents. Card and Lemieux (1997) and Matsudaira (2010) take a more market based approach, using aggregate data from the US and Canada, to estimate the effect of labor market conditions on living arrangements, school enrollment and work effort of young adults. Both studies show that that improving local demand conditions lowers the probability of living at home but that higher costs of housing raise these probabilities. Unlike the decision of elderly parents to coreside with a child which usually necessitates a move by one or both parties, young adults can live with parents either because they have not yet left the parental home, the “failure to launch”, or they can live with parents because they have moved back in after a pe6
The three month time frame is not coincidental. Medicare will cover stays of up to 100 days if they follow a hospital stay and the individual requires medical rather than, or in addition to, custodial care. Garber and MaCurdy (1993) document a spike in nursing home discharges when Medicare coverage ends.
4
riod of independence, the phenomenon of “boomerang children”. While some studies use individual panel data to separate these two phenomena (Ermish, 1999 on “failure to launch” and Kaplan, 2010 on “boomerang children”), it is more common to use aggregate data in which those who never moved out and those who moved back in are indistinguishable from one another (Card and Lemieux, 1997; Manacorda and Moretti, 2006; Matsudaira, 2010). Kaplan (2010) provides some of the first evidence on “boomerang children” with low levels of education. He uses individual employment and living arrangement transitions to show that unemployment is important in explaining the returns to the parental home for the less welleducated. As is the case with the frail elderly, young adult children may benefit not just financially, but in terms of care as well. Not so much with respect to care for the child himself (although we note that we find that coresidence is likely when the child is disabled) but with respect to the provision of grandchild care. A grandparent who lives with her child and grandchild may help the child remain employed or save on daycare costs by providing babysitting services. Our work builds on the work on elderly women coresiding with adult children and the work on younger adults living in the parental home. We are able to use the long panel in the PSID to distinguish between the “boomerang” and “failure to launch” reasons for coresidence, even among the elderly and to control for the age and health of the elderly parent to proxy for needed care. Furthermore, we can examine these phenomenon across SES, providing some of the first evidence on how the evolution of living arrangements of the elderly differ according to the economic well-being of mothers and their children.
2. Data Our study is based on a sample of older women 7 drawn from the Panel Study of Income Dynamics (PSID). 8 In this section, we describe the basic structure of the PSID and the selection criteria we use to con7
Although our focus is on the coresidence decisions of mothers and their children, there are numerous interesting questions with regard to caring for fathers which we ignore. Unfortunately, identifying non-custodial fathers in most data sets – particularly those fathers who were never married to the child’s mother – is much more difficult than identifying mothers and is likely to lead to severe selection biases in who is included in our sample. 8
An alternative to the PSID is the Health and Retirement Study (HRS). The HRS provides a much larger sample of
5
struct our analytic sample. We then provide some descriptive information on the characteristics of our sample and illustrate some of the basic patterns of living arrangements for these older women. 2.1
The PSID
The PSID is a household based panel survey first fielded in 1968 at which time it was representative of the population of households in the United States. Interviews were conducted annually until 1997 and have been administered in alternate years ever since. Because of data limitations in the earlier waves, we limit our analyses to outcomes measured from 1984 to 2007 but draw on information in previous waves to explain these later choices. 9 The PSID has followed not just the original respondents but has added to its sample newly born (or adopted) children. Importantly, these children are followed when they leave the parental home and establish their own household, allowing us to link data from mothers to that of their grown children over an extended period of time. 10 With these continual additions to the original sample through marriages and births the PSID sample is now much larger than the original 18,230 respondents. Even with deaths, attrition, and a budget related reduction in sample size in 1997, there are currently over 23,500 respondents. This policy of following PSID children as they form their own households allows us to obtain self-reported information on measures such as the income and employment of adult children. This data availability gives the PSID an advantage over other data sources such as the Health and Retirement Survey (HRS) and National Long Term Care Survey (NLTCS), wherein the information that is available for children is reported by a parent. Proxy reports from parents about their adult children likely produce reli-
elderly women with information reported on each of their children in each wave of the survey. However, the HRS does not have interviews of the children or cover as long a time span. 9
Because the PSID sample was drawn in 1968, Hispanics, Asians, and other immigrant groups arriving in large numbers after that time are under-represented relative to today’s population. (The PSID attempted to correct this under-sampling in 1997 with the addition of an immigrant sample and new sample weights, but for most of our sample period we have few observations on these groups.) Thus, although past analyses suggest significantly different patterns of living arrangements and care giving for elderly Hispanics, we are unable to address this issue. We do examine differences between blacks and non-blacks. 10 New spouses of sample persons are added to the survey as well and are interviewed as long as they remain with the sample person; they are not followed in the event that the couple splits. In a similar vein, stepchildren who were not originally in the sample are not followed if they leave the PSID respondent’s home.
6
able data on such things as a child’s gender and the number grandchildren, but the accuracy of such reports on the income or wealth of children is clearly more suspect. Moreover, the accuracy of proxy reports may vary systematically with how close a mother is to her children either geographically or emotionally, leading to biased estimates. The drawback of obtaining data directly from the children is that if the child cannot be located or refuses to be interviewed, we have no information on that child. Again, this missing data may be related to the closeness of a parent and child. Perhaps most importantly, the PSID exceeds other panel surveys in the length of time for which respondents are followed. The long time frame allows us to observe living arrangements over the life cycle as well as to examine how resources early in life are related to later outcomes. 2.2
Analytic Sample
The primary focus of our study is on the living arrangements of mothers and their adult children. The sample we draw from the PSID therefore consists of female PSID respondents who have at least one child who themselves is a PSID sample member. Because we want to examine living arrangements when a woman’s children are young adults and long before she needs care, we limit our sample to women who are observed starting at age 58. 11 We also require that we observe these women when they are age 65 or older so that we can follow them for an extended period of time and to an age at which they are likely to be retired. Finally, we require that the final observation be in the year 1984 or later because 1984 was the first year in which the PSID collected information on health status—a likely crucial determinant of living arrangements. 12 This selection criterion has the effect of excluding from our sample those women from the very earliest birth cohorts who die before 1984 and those with sufficiently poor health that they die before age 65. After imposing these selection criteria we are left with a sample of 1113 women. The panel as-
11
We also explored a younger initial age but found that in going much earlier, most mothers still had children who were either younger than 18 or still in college. Because we are interested in coresidence for women and their grown children we prefer the age 58 cutoff. 12
Prior to 1984 there is also a problem with correctly identifying residence in a nursing home residence. (See Ellwood and Kane (1990) for suggested methodology for dealing with this issue.)
7
pect of the data provides us with several years of observations for each of our widows from ages 58 on, allowing for repeated observations on living arrangements and other time-varying characteristics. The average number of interviews for the women in our sample from age 58 until the last observation is 15 and all told we have 16,303 person–years of data.
3. Patterns of Living Arrangements In this section we begin with a description of the patterns existing in our data and how the choice of living arrangements relates to characteristics of the older women and their adult children. We then turn to a more formal analysis that considers the question in a multivariate regression context. Throughout the discussion we remain cognizant of the likely endogeneity of financial resources and provide a parallel discussion of the relationship between living arrangements late in life and a pre-determined measure of income. 3.1
Living Arrangements by Age
To obtain a sense of how living arrangements evolve as women age, Figure 1 shows the distribution of living arrangements by mother’s age. 13 Here we use all the observations on our respondents and report their situation at each age. 14 We categorize women at each observation into one of five mutually exclusive and exhaustive groups: (1) women living alone or with their husbands, which we refer to as living “alone”; (2) women living with children (with or without a husband) and the children left at some point since 1968 but are now living with their mother, referred to as “children left home”; (3) women living with children (with or without a husband) and at least one child has never left the home, referred to as “child never left”; 15 (4) women living with “others” (with or without a husband present), including non-
13
All figures and tables are weighted using PSID individual weights which account for attrition and for relative size of the SEO and SRC samples in PSID. We do not weight our regression analyses as presented here, but our results are unchanged if we do. 14
Note that because interviews during this time were conducted biennially, individuals typically contribute an observation at every other year of age. 15
In determining whether or not a child has lived away from home we use all available data. The “child never left” designation is, in practice, an indicator of whether the mother in our sample lived with the child for our entire period of observation – from 1968 until the last time we observe her. The child may have left home prior to 1968 or may
8
relatives and relatives – and importantly those living with grandchildren without the child present; and (5) women living in a “nursing home.” We note that because of the small number of observations at older ages we observe few women in this category. The Figure shows the fraction of women living in each of the five living arrangements at each age. In addition to the five categories listed above, we also include the time trend of women living with children – i.e., the combination of categories (2) and (3) and labeled in the Figure as “All Kids” – because this is the category typically reported in studies of living arrangements. Consistent with other crosssectional evidence, those living alone or with a spouse comprise the largest part of the sample in each age group. The pattern appears to follow an inverse u-shape with age; the fraction living independently increase until about age 70 as children leave the nest (and the “child never left” category decreases) and then decreases as women age and begin to need care themselves. The life cycle pattern in the fraction of women living with children is almost the mirror image of that for the fraction living alone. At the same time, we note that those living with children is rarely less than 20 percent. Separating out the two types of coresidence with children – those living with a child who had left home and with those with children who have left and returned (or with whom a parent began residing) – we see, unsurprisingly, that the fraction living with children who never left declines monotonically over time. The greater prevalence of this arrangement at early ages is indicative of the “failure to launch” phenomenon noted in the popular press and popular culture more generally. However, around 13 percent of women at age 58 live with one or more children who never left their parental household and, even at the oldest ages, approximately 7 percent of women are living with children who never left home. The prevalence of this living arrangement has gone unrecorded in other data sources yet it appears to be an important phenomenon. Moeover, this finding suggests that not all coresidential arrangements are the result of parents needing assistance in older ages, but rather that one of the parties is unable to live independently due to health or financial concerns or simply that coresidence is the preferred arrangement for
leave at some point in the future.
9
this family. We also note that the large fraction of women at relatively young ages who are living with children who have left provides suggestive evidence of the phenomenon of what has been termed “boomerang children” by the popular press (Ramachandran, 2005; Business Week, 2003) wherein grown children move back home for economic reasons. In contrast, the upturn in living with children who were previously independent at approximately age 80 and beyond is more consistent with a care-giving role for children. The fraction of women living with others remains relatively constant across ages. For many of these women the other persons are grandchildren with the child’s parent living elsewhere. The prevalence of living with grandchildren varies significantly by race. Forty-seven percent of our white women in our sample who are living with others are living with a grandchild (without the child present) compared to 76 percent of blacks (not shown). Finally, living in a nursing home only begins to comprise a meaningful portion of the sample at much greater ages. 16 In the late 70s and early 80s the fraction of women living independently falls sharply. It is unfortunate that we do not have more observations at older ages to investigate this choice in great detail. Capturing nursing home use is also difficult in that many stays will be less than the time elapsed between interviews and many patients in nursing homes may not participate in the survey. Other surveys such as the Health and Retirement Study may be more useful in this regard. The life cycle trends displayed in Figure 1 accord with our intuition but also demonstrate that coresidence with children is a much more complex phenomenon than one might have expected and one clearly misses an important element of the interaction between older parents and their children if one views their coresidence as primarily to benefit a parent who can no longer care for herself. Children may return home after a period of independence for their own benefit and many remain at home far into their adult lives. Thus, one must consider the costs and benefits accruing to both parents and their adult children and how the distribution of benefits varies over the life course. 16
Note that the PSID measures only long term nursing home stays and we are therefore not mistakenly counting someone as residing in a nursing home if there are only there for a recuperative stay.
10
3.2
Correlates of Living Arrangements
To learn more about what factors are correlated with living arrangements, in Table 1 we present the means of the characteristics of our sample of mothers and their families by type of living arrangement. The data in this exercise are organized by the living arrangement of women as of the last observation we observed for her in our sample. We use the five living arrangement categories defined above and used in Figure 1, i.e., living alone, with children who left, with children who never left, with others and in a nursing home. Several of our measures of characteristics warrant discussion. For time varying variables such as income, we include values of variables at both age 58 and at our final observation. For women who are married we have a measure of the combined income of husband and wife.17 We also report total family income which includes the incomes of any additional people living in the household. With respect to the characteristics of children for this parent based sample, we sought a parsimonious way to report the resources of children when the number of children varies across families and reaches up to 15 in our data. In Table 1 we therefore report values for the total number of children, the number of daughters (because the literature suggests that daughters figure prominently in the provision of care) and the average income and average schooling of children as summary measures of these variables. We also report the number of grandchildren as they may present competing demands on the middle generation or the grandmother may need/want to provide assistance by caring for a grandchild. Past work has repeatedly shown that living arrangements are strongly correlated with income (McGarry and Schoeni, 2000). With that in mind, we look first at a comparison of the incomes of the mothers in our sample. Consistent with this past work, the mean income of those living alone (or with a spouse) is $50,000 per year, at least 50 percent greater than those living in any of the other types of ar-
17
We later examine differences in financial well-being controlling for household size. In contrast to income, wealth is not measured on an individual or even a couple-level in the PSID, but rather at the household level. Women who are residing with grown children or others will therefore appear to be wealthier than they themselves actually are, provided that there is at least some wealth belonging to others in the household. Furthermore, wealth in the PSID is obtained only intermittently. We therefore limit our analysis to differences by income, but note that differences by wealth explored in earlier versions of the paper show similar patterns.
11
rangement, and nearly three times as large as those living with others. However, it is not clear whether incomes differences are the cause or consequence of the living arrangement as both income and living arrangements are likely jointly determined. For instance, it may well be that income is low among nursing home residents because those with low income can qualify for Medicaid, making nursing home residence much more affordable. Alternatively, individuals may have intentionally divested themselves of assets (and thus asset income) in order to qualify for Medicaid, or they may have spent down their savings coping with a serious health condition prior to nursing home admission. 18 Similarly, income may be low among those living with children because they retired early in anticipation of living with children, assigned assets to children in exchange for coresidence or out of altruism, or because programs such as Supplemental Security Income (SSI) reduce benefits for those living in the home of another. As a way of exploring the likely endogeneity of income, we report the mean values of income measured at age 58, which is, on average, 16 years before the age at all versions of current income were measured. It seems unlikely that adjustments to income have been made in anticipation of future living arrangements as early as age 58. Yet this measure leads to the same conclusions: Individuals who are living alone at the oldest age for which we have data have significantly higher income not just at that point, but at age 58 as well, on average 16 years prior to our measure of living arrangements. The average income for those eventually living alone is $85,500 compared to $59,000 for those with children and $35,000 for those living with others. This finding that income during prime working years is correlated with living independently later in life is consistent with the other evidence (see for example, Smith, 2007) of a strong health/income gradient to the extent that being healthier and remaining independent at older ages and, themselves, positively related. Finally, our finding that fact that those who eventually live in a nursing home have the lowest incomes, even at age 58, also is consistent with higher levels of economic attainment early in life and better health outcomes later in life. 18
Individuals who are institutionalized and who have little in the way of assets can become eligible for Medicaid even if their income is above the Medicaid eligibility limits if the cost of the nursing home exceeds their ability to pay. These “Medically Needy” programs existed in 36 states in 2006. In multivariate regression analyses we controlled for this and other state characteristics but they were typically not important determinants of our outcomes.
12
When looking at the income of the entire household (total family income), however, those living with children are much better off than those living alone as befits a family with more prime age workers. The average household income of those living with children is approximately $70,000, significantly larger than the income of those living independently and approximately twice the amount of the woman (and spouse) herself. Those living with children who never left the family home have average income of $57,000, an amount greater than those living alone, but a somewhat smaller increase than for those with children who left. This result provides some suggestive evidence that those children who never leave are either in lower income families or are themselves are less financially well-off then children who are coresiding after a period of independence. On a per capita basis, the advantage of those living with children disappears, and per capita income is nearly one-third smaller than that of those living alone. Additionally, the financial hardship of those living with others becomes apparent. As we noted earlier, many of these women are grandmothers who are living with their grandchildren. The grandparent thus has the extra mouth to feed with little additional income. Differences by education follow the same pattern with those eventually living alone having the most schooling and those living with others or with children who never left, the lowest. Finally, there are very large racial differences across the groups. Just five percent of those living alone and just 7 percent of those in a nursing home are black while over 20 percent of those with children or with others are black. 19 In exploring further the differences in this latter category we find that the large fraction of blacks living with others is due to the prevalence of black grandmothers living with their grandchildren in households in which the grandchild’s mother/father is not present. Perhaps unsurprisingly, women living alone have the fewest children—consistent with both their higher apparent socioeconomic status and the lower potential of a coresident arrangement. Those living in a nursing home have the next fewest children, providing at least some suggestive evidence that children may “protect” one against institutionalization. Despite the 19 The PSID sample contains few Asians or individuals of Hispanic ethnicity. We therefore do not separately tabulate these groups. Asians are included with whites as “non-black” and Hispanics are classified with their selfreported race.
13
overwhelming evidence that daughters provide more care than sons, there appears to be little difference across groups in the fraction of children who are daughters. And finally, following our note that blacks living with others are often living with grandchildren, we find the number of grandchildren to be the largest for the “others” category. One would expect to see differences in health status by living arrangements. The PSID has only limited health information and even that is not available until 1984.20 We rely on self-reported health in the categories excellent, very good, good, fair and poor which we recategorize by combining excellent, very good, and good and combining fair and poor. Again, we are looking at women whose average age is 76 so many are in poor health in each category, but there is a dramatic difference across groups: just 37 percent of those living alone are in poor health compared to approximately 50 percent of those living with children and 64 percent of those living with others and 72 percent of those in a nursing home. (Although in this case one might have expected a much larger number.) If parents were choosing to live with the child who had the largest home or the greatest financial resources, one would expect mothers who live with their children to have children with higher incomes, instead we find the reverse. The average income of children is significantly higher for those women who live alone than for those who live with children or with others. This result is consistent with the overall higher socioeconomic status of the independent living women. Their children have average household income of nearly $106,000 compared to $50,000 - $60,000 for those living with children or others. The second highest average income is for the children of those women living in a nursing home. These differences could be partially explained by an older average age among the children of women in nursing homes but, perhaps these children are also pursuing careers that make caregiving impossible, or have the wherewithal to afford formal care for their mother. Overall, we find that those who live alone appear to be the best off in virtually every dimension. 20
From 1984-1992 and from 1996 onward self-reported health status is available only for household heads and their spouses. Self-reported health status is available for the entire PSID sample for just the years 1992-1996. Similarly, ADL limitations, likely a better predictor of the ability to live independently are available for just the years 19921996 and for 2003-2005 and is therefore of little help in our study which covers an extended period of time.
14
They have higher income, more education, higher income children, and better health than older women who are in some form of shared living arrangement. 3.3
Living Arrangements Over Time Although we categorize the mothers by their living arrangements at the last observation, we do
not mean to imply that they transit from their status at age 58 directly into this arrangement. Rather, they may have spent some time in several different living arrangements as they moved through retirement and into their older years and as their children too, moved through young adulthood. We report evidence of the prevalence of alternative living arrangements over this part of the life course in Table 2a. We exclude from the tabulations for this Table, and the next two tables, the roughly 10 percent of women who have lived with children who never left as, by definition, they do not change living arrangements. Among those living alone at the final observation, 39 percent of these women had, at some point since age 58, lived with a child who had previously left home. We imagine that coresidence with a child following a period of independence and ending with a return to independence is relatively unlikely to be due to the mother’s failing health and more likely a temporary arrangement for the benefit of the child, akin to the phenomenon of boomerang children and consistent with the above evidence that the incomes of children in coresident families is lower than that of children in other groups. In the second column we see that of those living with children at the last observation 80 percent were previously living alone, but a sizable fraction, 20 percent, had some previous experience living with others. Importantly, among those in a nursing home in the final period, 32 percent lived with a child at some point. In these instances it may well be that the coresident arrangement benefited the mother and suggests a potential progression of care from informal family-based care to formal care. Yet it is worth noting that a previous period of coresidency is not the norm. In Table 2b we examine not just the incidence of various types of living arrangements but their importance in terms of the time spent in that state. For those living alone at the end of the sample period, 93 percent of their lives since age 58 were spent in that state, 10 percent with children following their
15
children leaving home, 3 percent of the time with others. For those living with children, the time is fairly evenly split between time alone and time with children. Interestingly, those who end up in a nursing home spent relatively little time there, just 4 percent of the observation period, with far more time being spent alone. Finally, we look to see whether these spells are associated with a single transition from one state to the other or whether there is a good deal of movement in and out of various arrangements. In Table 3 we show that multiple transitions are fairly common. In our sample, 43 percent remain in the same living arrangement throughout the period, and this figure increases to 50 percent if also include the women who are living with children who never left that were excluded from this Table. Among the women covered by Table 3, 24 percent has just one transition, and 18 percent have three or more. The average number of survey years for which our women are observed is 15 so even these relatively large numbers are not surprising. If we look separately by final living arrangement, those living alone at the last observation were definitely the most stable group – 50% were alone throughout the entire period of observation. And yet even among this group, more than one-quarter had 2 or more transitions, corresponding to at least three different living arrangements in our period of observation. In contrast, for those who are living with children at the end of the period, 56 percent had 2 or more transitions, making them the most mobile. Perhaps unsurprisingly, those who end in a nursing home have a relatively high number of transitions with 61 percent transitting 2 or more times. This number is consistent with a pattern wherein a frail elderly women move from independence, to coresidence with a child, to a nursing home. 3.4
Living Arrangements and Financial Resources
Because much of the previous literature has centered on the relationship between financial resources and the choice of living arrangements, we look further at differences in income across living arrangements. In Tables 4a and 4b, we divide our observations into quartiles based on the woman’s (and spouse’s) own income and examine how the distribution of living arrangements varies across quartiles with living ar-
16
rangements; we do not include income of other household members. Table 4a displays the fraction of those in an income quartile who are living in each of our five living arrangements. The most common living arrangement is living alone, but there is a dramatic rise in the probability of living alone as income increases. Just 49 percent of those in the lowest quartile are living independently compared to 81 percent in the highest. The fraction living with children who left and the fraction living with others mirror this result, falling as the income quartile increases. There are almost no cases of well-off women living with others, with just 1 percent of those in the highest quartile living in such an arrangement. However, we note that the fraction living with children who never left, is relatively constant across quartiles, falling slightly before rising again. And nursing home use is again far more likely for those in the lowest quartile. As we noted earlier, current income is likely to be endogenously determined with living arrangements. Therefore, in Table 4b, we present the tabulations for current living arrangements by income quartiles based on income measured when the woman was age 58. Comparing the results in the two tables, the patterns of living arrangements by the distribution of these two different measures of income are strikingly similar. The one noticeable exception is for living in nursing homes. Whereas nursing home use is highest in the lowest income quartile when based on current income, it is much more evenly distributed with lagged income. The sharp decline in income for these women is again consistent with having spent down assets (either strategically or through medical expenses) and qualified for Medicaid and SSI. However, even with income measured earlier in life, nursing home use remains extremely rare among those in the highest quartile supporting evidence elsewhere that the wealthy can use their resources to remain in the community throughout their lives (Marshall, McGarry and Skinner, forthcoming).
4. Multivariate Analyses of Living Arrangements The descriptive evidence points to differences in the choice of living arrangements by socio-economic status. Yet because factors such as race, income and education are correlated, it is impossible to discern their relative effects from a study of the means. To assess more directly how each of these factors influences the choice of living arrangements in this Section we present results from multivariate analyses of
17
the determinants of widows’ living arrangements. We employ a multinomial logit specification, using our five-category living arrangement classification scheme. In Section 4.1, we present results for specifications in which the determinants might be thought to be more or less exogenous. Then, in Section 4.2, we consider specification that control for factors that, themselves, may be affected by or determined simultaneously with the choice of living arrangements. 4.1
Econometric Specification of the Determinants of Widows’ Living Arrangements
Our baseline models characterize the living arrangement of elderly women as a function of her demographic characteristics (age, race, schooling, number of children), economic resources (income), health status (poor health), and the characteristics of her children (mean age, income, disability status and unemployment). 21 Because of the inverse u-shape in living independently that we see in Figure 1, we employ a spline in age with a knot at age 70. Table 5 reports the marginal effects based on estimates from a multinomial logit model with the omitted category being living alone. The findings accord broadly with our intuition. The number of children is negatively related to the probability of living alone as is having a disabled or unemployed child. Before age 70, an additional year of age increases the probability of living with children less likely but after age 70, age is positively related to living with kids. After age 70, older mothers are less likely to live alone. Older children imply a greater probability of living alone. Surprisingly, the choice of living alone is more strongly related to the income of children than to the income of the parent. As with Table 1, we again see strong differences by race, with black mothers being substantially less likely to live alone. In fact, being black has a larger effect than any other 0/1 covariate. Perhaps surprisingly, the mother’s income itself has little effect on the probability of living alone, but the higher the average income of children, the more likely is the woman to live independently.
21
We have also experimented with including measures of the economic environment--the parameters of the Medicaid program in her state of residence to proxy the price and availability of institutional care and measures of the economic conditions the state (state/year specific variables on employment rates and wages) to proxy potential labor market income. However, these factors were not significantly different from zero and we exclude them here for to simplify the discussion.
18
Living with children who left is more likely if there is a (perhaps newly) disabled child or unemployed child, pointing strongly to the living arrangement benefiting the child rather than the parent. Having a disabled child also increases the probability of living with a child who never left.22 Similarly, the higher the average income of the children, the less likely is coresidence. Even after controlling for income, family size, and health, older black women continue to be significantly more likely to live with children. Living with others is also negatively related to income of both the parent and the child. Finally, as we have seen repeatedly, nursing home use is negatively related to income, and unsurprisingly positively related to poor health. Because of the differences in the effect of aging on living arrangements before and after age 70, and in order to provide more information on who is likely to be helping whom with respect to shared living arrangements, we estimate the regression separately for women under age 70 and for those ages seventy or older to note how the effects vary. For the under age 70 category we omit nursing home as an option in the MNL model. Tables 6 panels a and b show the marginal effects for the estimation of a multinomial logit model for mothers under 70 and for mothers age 70 and over, respectively. Three interesting patterns emerge from the results. The first is that when the sample is stratified by age, the effects of mothers’ income become larger and statistically significant. For women under 70, having more income makes living alone less likely and living with kids more likely. For women 70 and over, having more income makes living alone more likely and living with kids less likely. This points to the phenomenon of boomerang kids when at young ages, children return home if their parents are able to support them. Conversely, at older ages when the parent is more likely to be frail and the children established in their careers, a mother’s income may buy independence. The second notable result is that the effect of having a disabled child differs by whether a child has previously left the home, particularly for women over 70. In this age group, having a disabled child nearly doubles the probability of living with a
22
In results not shown here we verified that it was the disabled child who was typically the coresident child.
19
child who never left home. Though the effect is also strong for living with a child who has returned to the parental home, it is only ¼ as large as for those who have remained coresident for their entire lives. Finally, having an unemployed child predicts coresidence with a child who has previously left the parental home only for younger mothers (with younger children). These results suggest that at younger ages, coresident children are more likely to be receiving support from their parent rather than giving support to their parent and that before age 70, coresident living arrangements between adult children and their mothers largely benefits the child. 4.2
The Influence of Lagged Income and Wealth
We have shown that many of the results presented in the simple cross-tabulations are robust to the time at which income is measured, either at the time at which living arrangements are measured or at age 58. Based on these results, one might argue that living arrangements are a product of differences in SES across the life-course rather than a product of shocks to parental or child well-being. However the large effects of children’s income and unemployment on maternal living arrangements found in Tables 5 and 6 are consistent with the possibility that living arrangements may have a direct influence on the income of their adult children. For example, children may remain in school longer or may remain unemployed longer while searching for a good job match if they can live with a parent. To further assess these two differing assessments of the possible causal links between income and living arrangements, we examine two variants on Table 5. First, in Table 7, we exclude the income and unemployment status of the child as covariates in order to examine whether the effects of parental resources change when we do not account for the resources of the child. Second, we repeat the exercise, again excluding child’s income and replacing current parental income with parental income when the mother was age 58 long before her income may have been altered due to living arrangements. Table 6 shows that the parental income effects differ dramatically before and after age 70 so, as before, we stratify by age. Table 7 shows that the results do not change substantively when we exclude children’s income and unemployment status, although the coefficients on child disability and race become larger. Child dis-
20
ability is correlated with the income of children so we would expect to see larger coefficients when child income and unemployment are excluded. Table 8 shows the results using the mothers’ income at age 58. The differences between table 7 and 8 are very small suggesting that the relationship between economic well-being and living arrangements in our original results is largely reflective of factors such as SES that are consistent across the life-cycle rather than a result of shocks that occur when parents are older. The one exception is the important of child unemployment in predicting coresidence between parents and children. However, even these spells of unemployment could reflect chronic instability in the labor market outcomes of children rather than unexpected labor market shocks.
5. Implications for Well-Being We have shown strong differences in income across living arrangements and examined the factors that are associated with the choice of such arrangements, demonstrating repeatedly that there are strong differences in living arrangements by socio-economic status, and importantly, that these differences hold for measures of SES taken well before the observed outcome. The prevalence of shared living arrangements among the low income population suggests that it may be used as a way to assist family members without the need for cash transfers while at the same time exploiting the returns to scale existing for larger household size. We end here with a formal analysis of how the choice of living arrangements affects well-being and the role played by household size. In Tables 9a and 9b we examine the distribution of income relative to the poverty line across living arrangement types. We first use just the income of the mother (and spouse) and the poverty line for a single person (or couple) asking, in effect, how would she fare on her own? Of course, for those living with others, income might well be higher if she chose to live independently either because of work effort, programmatic changes, or more careful conservation of wealth. In the second panel we base our calculations of income relative to the poverty line on the total income of the household and the poverty line appropriate for a household of that size. This procedure takes advantage of the equivalence scales implicit in the calculation of the poverty line.23 Based solely on own income and 23
There is disagreement about whether the poverty lines are adjusted properly for family size. We ignore any such
21
needs (and the income and needs of a spouse), we again see that those living alone are faring the best financially with 10.6 percent of the sample having income below the poverty line. In 2009 the poverty rate for women 65 or older 10.7 percent (U.S. Bureau of the Census) so our numbers are on target. The next best off group is those living with children who were previously independent, with a poverty rate of 18 percent, similar to the 19 percent for those living with children who never left. Approximately one-third of those living with others or in a nursing home are poor based on own income and needs. Similar patterns exist for the probability of having income below two or three times the poverty line. In living with a child or others a woman may be able to avoid poverty. In Table 9b we add to income any income from other household members and adjust the poverty threshold accordingly. When doing so we find that coresidence improves the condition of our elderly women substantially. The poverty rate for those living with a child falls from 18 to 10 percent, a decline of 44 percent. For those living with children who never left, the decline is from 19 to 13 percent and for those with others, the decline is similar in percentage terms but the probability remains high at 25 percent. As we mentioned earlier, many of those living with others are grandparents living with their grandchildren so the additional income from such individuals is likely to be low. We cannot, however that those women in coresident arrangements experiencing income above the poverty line would have been impoverished had they not had the opportunity to coreside. They may have left work to help their daughter care for their grandchild, foregoing income from earnings and / or pension benefits or they may have transferred assets (and asset income) to a child in exchange for coresidence. Finally, in an attempt to see how these women were faring in their prime, we instead calculate poverty status using own (and spouse’s) income at age 58 (Table 9c). We see roughly the same results as reported in Tables 9a and 9b. Those who live alone were the best off, even very early on with poverty rates of just 6 percent, while those who are living with others or with a child who never left have poverty rates of 16 and 14 percent. The largest difference is in the well-being of eventual nursing home residents issues here and use the Census’s poverty lines because they are a widely cited metric. We do not use poverty lines specific to the age distribution of household
22
who were faring particularly well at age 58 but who may have lost financial resources due to health / nursing home expenses by the time we see them in an institution.
6. Conclusion and Directions for Future Work In this paper we have begun to shed some new light on the living arrangements of women throughout the latter portion of their lives. There is a vast literature examining living arrangements at a point in time as well as the changes in living arrangements across cohorts. What we provide here is a focus on the evolution of living arrangements over time, and how financial factors in particular affect these choices. Our work points to important differences in behavior by socio-economic status. We find strong racial differences in the propensity to live with others and more specifically in the propensity to live with children, and with grandchildren with an absent mother or father. We also find evidence suggesting that a large fraction of the mother-child coresident arrangements are primarily for the benefit of the child. We find strong differences associated with the income of children, their employment and their health status. In a result that speaks to the role of health care costs in precautionary savings models (e.g., De Nardi, French and Jones, 2010) we find evidence that high income individuals are able to avoid nursing homes as well as coresidence and perhaps use their financial resources to retain their independence. This result suggests that there may be a strong motivation for individuals to retain wealth as they age. Although we were concerned about the simultaneous determination of income and living arrangements, our examination of living arrangements as a function of either current or lagged income suggests that the relationship between living arrangements and financial resources at older ages is present far earlier in life, long before resources are likely to be altered because of living arrangements.
23
References Business Week, 2003. “Mom And Dad, I’m Home—Again,” Business Week, Personal Business, November 3, 2003, www.businessweek.com/magazine/content/03_44/b3856124.htm. Card, David and Thomas Lemieux. 1997. “Adapting to Circumstances: The Evolution of Work, School, and Living Arrangements Among North American Youth,” NBER Working Paper No. W6142, August. Costa, Dora, 1999. “A House of her Own: Old Age Assistance and the Living Arrangements of Older Nonmarried Women,” Journal of Public Economics, 72(1): 39-59. De Nardi, Mariacristina, Eric French, and John B. Jones, February 2010.Why Do the Elderly Save? The Role of Medical Expenses. Journal of Political Economy 116(1): 39-75. Ellwood, David and Thomas Kane, 1990. “The American Way of Aging: An Event History Analysis,” in Issues in the Economics of Aging, Wise, David A., ed., Chicago: The University of Chicago Press. Ermisch, John. 1999. “Prices, Parents, and Young People's Household Formation,” Journal of Urban Economics, 45(1): 47-71. Garber, Alan M. and Thomas E. MaCurdy, 1993. “Nursing Home Discharges and Exhaustion of Medicare Benefits,” Journal of the American Statistical Association, 88(423): 727-736. Kobrin, Frances E., 1976. “The Fall in Household Size and the Rise of the Primary Individual in the United States,” Demography, 13(1): 127-138. Kaplan, G. 2010. “Moving Back Home: Insurance Against Labor Market Risk,” Unpublished manuscript, University of Pennsylvania, February. Manacorda, Marco and Enrico Moretti. 2006. “Why do Most Italian Youths Live with Their Parents? Intergenerational Transfers and Household Structure,” Journal of the European Economic Association, 4(4): 800-829. Marshall, Samuel, Kathleen McGarry and Jonathan Skinner, forthcoming. “The Risk of Out-of-Pocket Health Care Expenditures at the End of Life” in Explorations in the Economics of Aging, David A. Wise, ed. Chicago: University of Chicago Press. Matsudaira, Jordan. 2010. “Economic Conditions and the Living Arrangements of Young Adults,” unpublished manuscript, Cornell University. McGarry, Kathleen and Robert F. Schoeni, 2000. “Social Security, Economic Growth, and the Rise if Independence of Elderly Widows Unmarried Women in the 20th Century,” Demography, 37 (2): 221-236. McGarry, Kathleen, 1998. “Caring for the Elderly: The Role of Adult Children,” in Inquiries in the Economics of Aging. David A. Wise, ed. Chicago: University of Chicago Press, 463-485. MetLife, 2007. The MetLife Market Survey of Nursing Home and Home Care Cost, September, 2006. Westport, CT: MetLife Mature Market Institute. Michael, Robert T., Victor R. Fuchs and Sharon R. Scott, 1980. “Changes in the Propensity to Live
24
Alone: 1950-1976,” Demography, 17(1): 39-56. Ramachandran, Nisha, 2005. “The Parent Trap: Boomerang Kids” U.S. News and World Report, December 4, 2005. Ruggles, Steven, 2007. “The Decline of Intergenerational Co-residence in the United States, 1850 to 2000.” American Sociological Review, 72(6): 964-989. Ruggles, Steven. 1994, “The Transformation of American Family Structure.” American Historical Review, 99(1): pp. 103-128. Schwartz, Saul, Sheldon Danziger and Eugene Smolensky, 1984. “The Choice of Living Arrangements among the Aged,” in Retirement and Economic Behavior, Henry Aaron and Gary Burtless, eds., Washington: Brookings Institution Press. Smith, James, 2007. “The Impact of Socioeconomic Status on Health over the Life-Course,” Journal of Human Resources, 42 (4) : 739-764. Wolf, Douglas A. and Beth J. Soldo, 1988. “Household Composition Choices of Older Unmarried Women.” Demography, 25(3): 387-403.
25
90 80 70 60 50 40 30 20 10 0 58
60
62 Alone
64
Child who Never Left
66
68 69 70 71 72 74 76 78 80 82 84 85+ Others Child Left Home Nursing Home
All Kids
Notes: Ages after 73 are combined in two year intervals. Weighted using 1968 family weights.
Figure 1. Fraction of women in each living arrangement by age
26
Means Proportion Income (1000$) Total Family Income (1000$) Per Capita Income (1000$) Income at 58 (1000$) Total Family Income at 58 (1000$) Education (Years) Black Number of Children Number of Grandchildren Number of Daughters Fraction of Daughters Age Poor Health Has a Disabled Child Has an Unemployed Child Mean Kids Income (1000$) Mean Age of Children Mean Kids Education (Years) N=1113
Child Left Home 15.78 35.23* (7.69) 69.38** (8.95) 22.07*** (2.07) 49.66*** (4.87) 58.94*** (5.08) 11.53*** (0.23) 0.24*** (0.03) 3.98*** (0.18) 4.92 (0.36) 1.90* (0.12) 0.49 (0.03) 74.21 (0.62) 0.54*** (0.04) 0.24*** (0.04) 0.17*** (0.03) 52.10*** (3.33) 45.53 (0.66) 13.01*** (0.15) 243
Alone 66.59 50.15 (3.11) 50.15 (3.11) 32.56 (1.97) 78.14 (5.17) 85.51 (5.38) 12.24 (0.12) 0.05 (0.008) 3.40 (0.08) 4.66 (0.17) 1.67 (0.06) 0.49 (0.01) 74.59 (0.29) 0.37 (0.02) 0.05 (0.009) 0.07 (0.01) 105.94 (5.24) 46.42 (0.29) 14.01 (0.08) 637
Child Never Left 6.85 31.75*** (3.19) 56.88 (4.45) 18.86*** (1.75) 54.30*** (6.11) 77.19 (5.88) 10.95*** (0.38) 0.19*** (0.04) 4.67*** (0.33) 4.69 (0.63) 2.13** (0.20) 0.44 (0.04) 74.11 (0.91) 0.43 (0.06) 0.21*** (0.05) 0.14 (0.04) 61.35*** (12.83) 43.53 (1.07) 12.84*** (0.23) 109
Others 4.51 17.75*** (2.40) 26.38*** (3.33) 10.57*** (1.43) 35.22*** (4.97) 43.21*** (5.89) 10.95*** (0.45) 0.23*** (0.06) 4.22 (0.53) 5.82 (0.89) 1.77 (0.30) 0.41 (0.05) 73.42 (1.02) 0.61*** (0.08) 0.14 (0.06) 0.07 (0.04) 55.32*** (6.82) 46.35 (1.34) 12.55*** (0.29) 58
Nursing Home 6.27 21.11*** (2.83) 21.11*** (2.83) 21.11*** (2.83) 54.03*** (6.11) 55.88*** (4.38) 11.32** (0.43) 0.07 (0.03) 3.71 (0.32) 4.13 (0.57) 1.73 (0.18) 0.44 (0.04) 83.28*** (0.93) 0.72*** (0.06) 0.06 (0.03) 0.05 (0.03) 100.25 (16.17) 52.08*** (0.72) 13.77 (0.28) 66
Notes: Weighted using average individual weights. Standard deviations in parentheses underneath means. Tests for differences in means are all conducted relative to living alone. *significant at 10%, **significant at 5%, ***significant at 1% level
Table 1. Selected Means by Women’s Most Recent Living Arrangement 27
Fraction who were ever previously observed as: Alone Child Left Home Child Never Left Others Nursing Home
Alone 100.00 39.47 12.07 1.20 100.00
Status at Final Observation Children Left Home Others 79.60 70.87 100.00 49.21 19.49 100.00 0.89 0.00 79.60 70.87
Nursing Home 86.31 32.67 17.72 100.00 86.31
Notes: Weighted by individual weights.
Table 2a: Propensity for other living arrangements by final status
Fraction of time previously observed as: Alone Child Left Home Child Never Left Others Nursing Home
Alone 92.98 9.68 3.15 0.00 92.98
Status at Final Observation Children Left Home Others 41.59 39.27 58.22 10.03 4.67 56.11 0.00 0.00 41.59 39.27
Nursing Home 66.38 12.04 4.23 11.84 66.38
Notes: Weighted by individual weights
Table 2b. Fraction of Time in Different Living Arrangements
Total of All Transitions 0 1 2 3+ Overall Mean
Alone 50% 22% 13% 15% 1.05
Notes: Weighted by individual weights.
Status at Final Observation Children Left Home Others Nursing Home 13% 10% 0% 32% 45% 39% 23% 15% 30% 33% 30% 31% 2.25 2.15 2.23
Table 3. Number of Transitions in Living Arrangements
28
Total 43% 24% 15% 18% 1.29
Status at Final Observation Alone Child Left Home Child Never Left Others Nursing Home
Quartile of Woman’s/Spouse’s Own Income Quartile 1 Quartile 2 Quartile 3 Quartile 4 49% 60% 77% 81% 21% 22% 12% 8% 8% 7% 5% 7% 9% 6% 3% 1% 13% 5% 3% 4%
Notes: Living arrangements and quartiles are weighted using individual weights.
Total 67% 16% 7% 5% 6%
Table 4a. Fraction of Women in Each Living Arrangement in Each Income Quartile
Status at Final Observation Alone Child Left Home Child Never Left Others Nursing Home
Quartile of Woman’s/Spouse’s Own Income at 58 Quartile 1 Quartile 2 Quartile 3 Quartile 4 52% 68% 68% 80% 23% 15% 15% 10% 10% 4% 6% 7% 8% 7% 2% 1% 8% 6% 9% 2%
Notes: Living arrangements and quartiles are weighted using individual weights.
Total 67% 16% 7% 5% 6%
Table 4b. Fraction of Women in Each Living Arrangement in Each Income Quartile at Age 58
29
Marginal Effects Number of Children Age Spline
