Widowed Mothers’ Coresidence With Adult Children

Abstract

Objectives.

Coresidence is one way that middle-aged offspring assist vulnerable, aging parents. This study investigated which characteristics of widowed mothers and adult children predict coresidence. When coresidence occurred, the analysis explored how individual children’s characteristics were associated with their coresidence with the mother.

Method.

Survey data from adults 53–54 years old in 1993 (N = 2,324) and a random sibling reported about their living situation, other siblings, and their mother, median age 80.

Results.

Logistic regressions revealed that mothers in poor health, who were older, and who had a daughter were more likely to live with a child. Among coresiding families, results from discrete choice conditional logit models showed that widowed mothers were more likely to live with an unmarried son than an unmarried daughter. Married children were less likely to coreside than unmarried children, but married daughters were more likely than married sons to coreside. Past receipt of financial help from parents was not associated with coresidence. Coresidence was more likely for those with a close relationship with the mother.

Discussion.

The discussion considers coresidence as an intergenerational transfer and its importance for the contemporary aging society. Data are needed on characteristics of all offspring to test theories about parent–child relationships.

Home is where the heart is, but family members also share a home for reasons other than emotional attachment. Historically, sharing a household was a common strategy for alleviating economic hardship and for providing care to infirm elderly family members (Hareven, 1990). Contemporary experience demonstrates that this strategy is still in use today. U.S. families responded to the economic strains of the Great Recession by combining households (Mykyta & Macartney, 2011, 2012). Similarly, when elderly family members are disabled, coresidence with younger kin becomes more likely (Crimmins & Ingegneri, 1990; Silverstein, Parrott, & Bengtson, 1995). Paradoxically, the increase in multigenerational households due to the economic downturn has occurred at the same time as a continuation of the long-term rise in the percentage of one-person households (Kobrin, 1976; Kreider & Elliott, 2009; Lofquist, Lugaila, O’Connell, & Feliz, 2012). Between 1900 and 2008, the percentage of those aged 65 and older who lived alone increased from 5.9% to 27.4% (Pew Social and Demographic Trends, 2010). Both older parents and their adult offspring appear to prefer to live apart from each other when they are able to do so (Klinenberg, 2012).

Yet, there are periods in life when parents and adult offspring benefit from coresidence. When young adults are finishing their formal schooling and seeking steady employment or they have lost their jobs, living with parents is an important way that parents provide material support to offspring (Furstenberg, Rumbaut, & Settersten, 2005; Kaplan, 2012; Schoeni & Ross, 2005). In fact, throughout most of life, intergenerational coresidence benefits the younger generation more than the older generation (Choi, 2003; Kahn, Goldscheider, & García-Manglano, 2013; Keene & Batson, 2010; Logan & Spitze, 1996). Late in life when parents are more likely to be infirm or have health problems, the direction of transfers shifts, and the benefits of coresidence are greater for the older generation (Choi, 2003; Cohen & Casper, 2002; Keene & Batson, 2010; Suitor, Sechrist, Gilligan, & Pillemer, 2011).

This paper investigates the determinants of elderly, widowed mothers’ coresidence with adult children. We focus on coresidence of widows in later life to understand the availability of care and well-being of an especially vulnerable subset of the population. Widows are economically disadvantaged, especially when they reach advanced old age. Compared with those who are married, widows are much more likely to be poor in old age (Angel, Jiménez, & Angel, 2007). Women who have been widowed also are more likely to be poor than men who have lost a spouse (Lin & Brown, 2012). In addition to having greater economic needs, widowed mothers are more likely to need practical help and caregiving from adult children than married parents in old age who have a spouse to whom they can turn should caregiving needs arise. Owing to their greater life expectancy, wives are likely to nurse their husbands through their last illness. Later in life, when disabilities are more common, widowed mothers who need help rely on adult children for assistance, including coresidence (McGarry, 1998; Roan & Raley, 1996).

This paper contributes to an understanding of intergenerational coresidence by examining differences between and within families in widowed mother–child coresidence late in the mother’s life. We investigate factors associated with differences among families in whether a widowed mother lives with any of her children to address the question of how different types of families may respond to an aged parent. We also examine within families how the characteristics of available offspring are associated with who among the widow’s children shares a home with her. We use information about all children in the mother’s family to place coresidence in the context of who is available to help. This focuses attention on the division of labor within families where potential helpers vary in their own needs and resources. Our approach contributes to renewed efforts to treat coresidence with and assistance to an older parent as a family decision rather than as the outcome of a single dyadic relationship between the parent and a child (Henretta, Soldo, & Van Voorhis, 2011; Pezzin, Pollak, & Schone, 2007; Pillemer & Suitor, 2006). We focus on coresidence even though most widowed mothers do not live with a child (McGarry & Schoeni, 2000) because coresidence makes resource sharing and some types of time transfers more efficient. Determining who coresides also is important because much research treats intergenerational transfers as conditional on living arrangements (Bianchi, Hotz, McGarry, & Seltzer, 2008). Knowledge of the determinants of living arrangements informs a more general understanding of intergenerational family ties.

Conceptual Approach: Why Some Children and Widowed Mothers Coreside and Others Do Not

Coresidence or space is one among several “currencies” that parents and adult offspring transfer to each other. Time, money, emotional support, and advice are other currencies of exchange in intergenerational relationships (Soldo & Hill, 1993). Coresidence may differ from these other transfers because shared housing is both economically valuable and an opportunity for time help, as when adult offspring assist parents with household tasks and personal care.

Health problems that limit older persons’ ability to care for themselves independently increase an unpartnered parent’s likelihood of coresidence with an adult child (Isengard & Szydlik, 2012). Although living in close proximity may substitute for coresidence under some circumstances, evidence is mixed on whether proximity substitutes for coresidence when an older parent is in poor health. U.S. data show that mothers’ functional limitations are associated with greater proximity to adult offspring (Seltzer, Yahirun, & Bianchi, 2013; Supplementary Tables S2 and S4), whereas European data show that parents’ health problems are not associated with close proximity (Isengard & Szydlik, 2012). Thus, it is unclear if coresidence with adult children is a unique form of intergenerational assistance when elderly parents are in poor health.

Theories about why adult offspring and parents help each other with coresidence are similar to theories about other types of intergenerational transfers (Bianchi et al., 2008). Five general factors affect whether children help parents in old age: parents’ need, children’s ability to provide help, gender socialization and norms, exchange or reciprocity for earlier help from parents, and closeness or affective solidarity between the parent and child. We consider how each factor contributes to an explanation for when an aged, widowed mother lives with an adult child.

Parents’ Needs

Widows vary in their economic and health care needs. Parents with higher incomes and educational attainment are more likely to live alone (Crimmins & Ingegneri, 1990; Spitze & Logan, 1990; Wolf & Soldo, 1988) due to their greater resources. Data from the American Community Survey show that fewer than half of widows without a high school education live alone compared with about two thirds of widows with at least some college education (Seltzer & Yahirun, forthcoming). Poor health and advanced age increase parents’ need for help, and adult offspring respond by helping more (Isengard & Szydlik, 2012; Schmertmann, Boyd, Serow, & White, 2000; Silverstein et al., 1995). Perhaps because they experience a rapid reduction in their economic resources and contraction in their system of social support, widows are more likely to transition to coresidence in the period soon after losing their spouse than after more time has passed (Silverstein, 1995; Strohschein, 2011).

Children’s Ability to Help

Adult children’s ability to help their widowed mother depends on their own needs and family responsibilities. Wealthier offspring are less likely to live with an unmarried, aged mother (Smits, Van Gaalen, & Mulder, 2010; Soldo, Wolf, & Henretta, 1999), but they may substitute financial support for coresidence (Couch, Daly, & Wolf, 1999). However, highly educated offspring may have more space in their homes that would allow room for a widowed mother but still maintain the privacy that both generations value.

Adult offspring who have other family responsibilities, such as those who are married or caring for young children, may not have the space to incorporate their aged mother into their household, or the offspring may be reluctant to give up their privacy (Smits et al., 2010). Single mothers of young children, on the other hand, are more likely to live with parents for at least a short time than are married mothers (Mutchler & Baker, 2009). Their coresidence may be an exchange relationship in which both generations provide each other with time help. From the family perspective, parents who have a larger number of offspring are more likely to have at least one child who is able to help than parents who have fewer offspring (Spitze & Logan, 1990). When an older mother has more adult offspring, she is more likely to live with an adult child than when a mother has fewer adult offspring (Seltzer et al., 2013).

Gender Socialization and Norms

Women are “kin keepers” (Hagestad, 1986). Compared with men, women are more likely to care for older family members and to share a home with them (Henretta, Hill, Li, Soldo, & Wolf, 1997; Hogan, Eggebeen, & Clogg, 1993; McGarry, 1998; Pillemer & Suitor, 2006; Wolf & Soldo, 1988). Parents who have reached advanced old age, a time when they are likely to need assistance, are more likely to live with a daughter than with a son (Schmertmann et al., 2000). The gender composition of adult children in a family also is associated with whether children provide care to older parents. Parents with at least one daughter are more likely to receive care from their children than parents who have only sons (Spitze & Logan, 1990; Wolf, Freedman, & Soldo, 1997). Having a daughter may increase care that parents receive if daughters orchestrate siblings’ assistance through their greater engagement with siblings compared with sons’ (White & Riedmann, 1992). A daughter’s role as kin keeper may compete with responsibilities in her family of procreation, such as the role of wife and mother. Thus, research on gender differences must take into account daughters’ and sons’ marital status and other characteristics (e.g., being a parent, employment) that might be associated with differences in coresidence.

Exchange or Reciprocity

Economic and sociological exchange theories of intergenerational relationships posit that children help their older parents to repay them for assistance that parents gave children earlier in life (Bianchi et al., 2008). Parents may pay for children’s schooling or give them a gift or loan to help them set up their own households to ensure the younger generation’s future economic well-being. Differences among siblings in parents’ transfers earlier in life help explain later variation within the family in who provides help (Henretta et al., 1997).

Despite the attention paid to reciprocity as a motivation for offspring to help parents in old age, qualitative evidence suggests that adults who provide significant help to older parents are reluctant to interpret their assistance as something they do to repay their parents. Instead, adult children describe their motivation as part of their filial responsibility (Funk, 2012). This is consistent with U.S. public opinion data in which adults who hold favorable attitudes toward coresidence with an older mother who needs economic help explain their views as part of their family responsibility (Seltzer, Lau, & Bianchi, 2012).

Relationship Quality

Children who are emotionally closer to their parents may be able to anticipate parents’ needs and more effectively provide for these needs. Having a good relationship, or affective solidarity, also increases children’s motivation to provide care (Lin, 2010; Silverstein, Conroy, Wang, Giarrusso, & Bengtson, 2002). Within families, those with better parent–child relationships receive more support than those with worse relationships (Fingerman et al., 2011; Gans & Silverstein, 2007). Living together entails a loss of privacy compared with maintaining separate households. When a widowed mother and child have a good relationship, both may be more willing to forego privacy to meet the mother’s needs.

Hypotheses

Informed by these conceptual perspectives and past research, we address two questions: What characteristics of mothers and children predict whether an older widowed mother lives with a child? Among families in which widowed mothers live with a child, what characteristics of children predict with which child the mother lives? We designate hypotheses about between-family differences by Between and hypotheses about within-family differences by Within.

We hypothesize that between families:

Between 1. Widowed mothers with greater needs, those who are older, in worse health, economically disadvantaged, and recently widowed, are more likely to live with an adult child.

Between 2. Mothers with more offspring and those with a daughter are more likely to coreside.

Turning to variation within families, we hypothesize that among coresiding families:

Within 1. Widowed mothers and married children are less likely to live together than unmarried children, but widowed mothers are more likely to live with a daughter than a son when both the daughter and son are married (or both are unmarried).

Within 2. Coresidence with an adult child who is a parent is less likely than when the child is not a parent.

Within 3. Mothers are more likely to live with a child to whom she and the child’s father provided financial help at the transition to adulthood.

Within 4. Coresidence is more likely when the child and widowed mother have a close relationship.

We do not specify either between-family or within-family hypotheses about the association between a mother’s coresidence and offspring’s educational attainment because the direction of association is ambiguous. Highly educated offspring have greater economic security but they may use their greater resources to insure residential privacy.

Method

Data

The Wisconsin Longitudinal Study (WLS) is a 50-year longitudinal study of a random sample of 10,317 men and women who graduated from Wisconsin high schools in 1957. We use survey data for 1957, 1964, 1975, 1992–1993, and 2003–2005. In 1994 and 2006, a randomly selected sibling of the graduate was also interviewed. The latter two waves were conducted as telephone and mail surveys. The WLS sample is still generally representative of the original 1957 sample (Hauser, 2005). We use data on widowed mothers’ living arrangements in the 1992–1994 surveys (hereafter the 1993 wave), when most mothers were in their late 70s or older, ages at which the benefits of coresidence are likely to accrue to this older generation. Of the original sample of graduates, 87% responded in 1992–1993. The response rate for the randomly selected sibling was 80% in 1994 (Wisconsin Longitudinal Study Handbook, 2006).

The absence of minorities and immigrants and the restriction to high school graduates limit the generalizability of our findings, although siblings with less than a high school education are included. But the WLS sample does represent a significant portion of the U.S. population of that time and even today. In 2010, approximately three quarters of the U.S. population aged 50–54 years were white and had at least a high school education (U.S. Bureau of the Census, 2011). At the same time, the restricted, state-based sample has the advantage that it limits variation on unobserved factors that influence decisions about coresidence, such as cultural attitudes about filial obligations and, to some extent, housing market variation. The WLS, like other valuable studies of aging families, such as the Family Exchanges Study conducted in the Philadelphia metropolitan area (Fingerman et al., 2011), the study of within-family differences conducted in the greater Boston area (Pillemer & Suitor, 2006), and the Longitudinal Study of Generations drawn from a Los Angeles area sample (Silverstein et al., 1995), complements studies using national samples that include statistical controls to take account of sample heterogeneity.

We refer to original respondents as graduates and to randomly selected siblings as siblings. We restrict analyses to graduates and random siblings who are biologically related (or adopted). More than 90% of the graduates and siblings reported that they were raised in intact families. We restrict our attention to biological siblings because there are too few cases of step and half siblings to support multivariate analyses. We study living arrangements in 1993 because few graduates still had mothers alive in 2005. In 1993, graduates were approximately 53–54 years old, and the median age of their widowed mother was 80. These are the ages when coresidence that helps older parents (vs the adult offspring) is more likely (Isengard & Szydlik, 2012; Schmertmann et al., 2000).

Analysis Samples

We use two analysis samples. The first uses data from 2,324 families to predict between-family differences in the likelihood that widowed mothers live with any of their children. We lack information on who is the householder, but it is likely that mothers at these advanced ages benefit more from coresidence than their children do (Choi, 2003; Speare & Avery, 1993). We use graduates’ reports about whether their mother lives in their household or with one of her other children. Just over 9% of widowed mothers live with a child (n = 215). Mothers who do not live with a child are living on their own.

We use the second analysis sample to investigate how individual children’s characteristics are associated with whether their widowed mother lives with them, among all coresiding families. The within-family analysis sample excludes 30 single-child families from the 215 coresiding families, resulting in a sample of offspring from 185 families with two or more children.

Measurement of Independent Variables

We investigate betweenfamily differences in coresidence as a function of widowed mothers’ needs and resources, children’s needs and resources, and other family characteristics. We treat a mother’s needs and resources as her age, health, education, and her deceased husband’s occupational status when the children were growing up (a proxy for family socioeconomic status). Health is a dichotomous variable distinguishing mothers in poor or very poor health from those in better health based on the graduate’s response to the question: How would you describe your mother’s health? Excellent, good, fair, poor, very poor. Education is whether the mother completed college. Family economic status is measured by the Duncan Socioeconomic Index (SEI) score (range 0–100). We use occupational status instead of family income because the former fluctuates less from year to year than the latter does (Hauser & Warren, 1997). We take account of whether the mother was widowed within the past 2 years as an indication that the mother is likely to be in the period of adjusting to her loss. Children’s needs and resources are whether at least one child in the family has a college education, and whether all of the children have the same level of schooling. We also include other family characteristics: number of children in the family, age of youngest child, and the presence of daughters. We control for youngest child’s age to minimize the likelihood that coresidence is part of the launching process in which young adults have not left the parental home.

The withinfamily analysis treats children (n = 687) as the units instead of families. We consider each child’s gender, birth order (eldest, youngest), and whether the child completed college. For the graduate and one randomly selected sibling, we extend the measures of children’s needs and resources to include marital status (married or not), the presence of children younger than 18 years in the household, and whether the individual was employed at the time of the interview. All models include the interaction of child’s gender by marital status because wives and husbands have different responsibilities in marriage.

Two variables address hypotheses about motivations for coresidence: one for whether the child’s parents contributed $1,000 or more to the child’s schooling, a house down payment, or to help start a business (exchange motivation) and the other for the quality of the child’s relationship with the mother (affective solidarity). Relationship quality is measured by responses to the question: How close are you and your mother? Very close, somewhat close, not very close, not at all close. An error in the WLS questionnaire logic incorrectly skipped some respondents around questions about transfers, but this error was corrected during the field period. It affects 10 cases in our sample. We distinguish those who received a transfer from all others (i.e., those who did not receive a transfer as well as those who were incorrectly omitted from the question sequence).

Both the graduate and randomly selected sibling reported about relationship quality in 1993. Because these reports are contemporaneous with living arrangements, we cannot discern if close relationships are the reason for coresidence or if coresidence fosters close relationships. We present the results with and without this variable. Questions about relationship quality were asked of a random 50% sample of graduates and siblings. The relationship quality variable distinguishes those who report a very close relationship from those who report a lower quality relationship. We also include a dichotomous variable to identify cases in the 50% random sample that were not asked the question or were nonrespondents for other reasons.

We combine graduates’ and siblings’ reports about their own characteristics with graduates’ reports about their mother and other family characteristics using information from all WLS waves. Father’s occupational status when the child was growing up combines survey reports with information from 1957 tax records to provide more complete and accurate information.

Outline of Analysis

Between-family analysis.—

We estimate a logistic regression in which coresidence with any adult child (coded 1 if coreside; 0 otherwise) is a function of mother’s and offspring’s characteristics. Variables are mother’s poor health, her age, whether the mother has a college education, occupational status of the children’s father, whether the mother was recently widowed, whether at least one of her children is college educated, whether all of the children acquired the same level of schooling, number of children in the family, age of youngest child, and whether the mother has at least one daughter.

Within-family analysis.—

We next investigate how the characteristics of each child affect coresidence, among families in which the widowed mother lives with a child. This requires a model that uses information on all offspring. We estimate a discrete choice conditional logit model that takes account of each offspring’s characteristics to predict with which child the mother lives. Family characteristics, such as mother’s health, are excluded because the characteristics are the same for all children in a given family. In preliminary analyses, we examined interactions of family characteristics with individual child characteristics (e.g., mother in poor health × daughter), but there was no consistent evidence of interactions.

Each child is represented by his or her own report in two-child families (i.e., families with only the graduate and a sibling). For families with more than two children, the characteristics of children who were not interviewed come from proxy reports. Some characteristics, such as marital status and previous transfers from parents, are only available for graduates and randomly selected siblings. Because siblings who were not sampled for the WLS study are missing at random, we are able to include their incomplete data in the analysis without introducing bias.

Our model predicts the relative probability that a particular child lives with his or her mother. In families with two children, the data record which child lives with the mother. In such families, the child who lives with the mother is assigned the value 1 on the outcome variable, and the child who does not live with the mother is assigned the value 0. In families with three children, we know the residential status of the graduate and the randomly chosen sibling. For them, the outcome variable is coded as in two-child families. Knowing the residential status of two children lets us infer the status of the third. Thus, if either the graduate or the random sibling lives with the mother, then the third child does not and thus has a 0 on the outcome variable. If neither the graduate nor the random sibling lives with the mother, then the third child does live with the mother and receives a 1 on the outcome variable.

In families with more than three children, if either the graduate or the random sibling lives with the mother, then the remaining children do not live with the mother and are assigned a 0 on the outcome variable. However, if neither the graduate nor the random sibling lives with the mother, then our information is incomplete. We know that one of the remaining children lives with the mother, but we do not know which one. This problem is analogous to the problem of assigning ranks to survival times in the Cox proportional hazards model when survival times are tied (Allison, 2010; Box-Steffensmeier & Jones, 2004). In such cases, one knows the survival times of the tied cases relative to all other cases in the data, but one cannot rank within a set of ties. The incomplete information in our design also is analogous to incomplete ranked preference data in which one knows the most and least preferred of a set of options, but not the details about intermediately ranked options (Allison & Christakis, 1994).

For 49 of the 185 families, we do not know with which child the widowed mother lives. These 49 families are families with more than three children in which neither the graduate nor the random sibling lives with the mother. To address the problem posed by incomplete information on who coresides, we employ the same method as is used in analogous studies, that is, a marginal likelihood approach to “tied” data, where we regard the children other than the graduate and the random sibling as “tied” observations. In this approach, our model assumes that each of these children has a nonzero probability of living with the mother and that, within this group of children, their probabilities of living with the mother sum to 1.0, conditional on knowing that one of them lives with the mother. This method uses all of the information we have—namely that one of these children lives with the mother—but we make no further assumptions about whether any particular child lives with the mother. Additionally, our model reduces the probability that any of these children lives with the mother in accordance with the size of the sibship; that is, the larger their sibship, the smaller their probabilities of coresidence. Because these probabilities are constrained by the model to sum to 1.0 among these children, there is no bias in the approach toward or away from inferring that such children live with their mother.

Our models include each child’s birth order (eldest, youngest, vs other), gender, and whether the child is a college graduate. For the graduate and randomly selected sibling, we also include whether the respondent is married, presence of a child younger than 18 years, employment status, whether the respondent had received a financial transfer from parents, and how close the respondent feels to the mother. We identify siblings who were not interviewed with a dichotomous variable in all models. The availability of information on a subset of characteristics for all offspring provides a more complete picture of the potential family network with whom the mother might live than would be possible with only two children (the graduate and randomly selected sibling).

Results

Do Widowed Mothers Live With a Child?

Table 1 shows the characteristics of families of widowed mothers. As noted earlier, 9.3% of these mothers live with one of their children. Family sizes were large for mothers who had a child in the 1957 high school cohort. Mothers had on average almost four living children. The vast majority, nearly 91%, had more than one child, and almost 45% had four or more children (not shown). Most had at least one daughter. The increase in educational attainment across generations is evident in the fact that fewer than 6% of mothers had a college education, but nearly half of mothers had at least one child with a college education. In 23% of the families, all of the children had the same amount of schooling. The family’s socioeconomic status also is measured by the occupational status of the deceased husband (the graduate’s father). The means on the Duncan SEI are consistent with occupations like plumber, machinist, and some lower level managers. Despite their advanced age, only 10% of mothers were described by their children as being in poor or very poor health. The percentage in poor health is higher among widowed mothers who live with a child (22%) than among those who do not (9%).

Table 1.

Characteristics of Widowed Mothers and Their Families by Coresidence With a Child, 1993

	All mothers	Mother lives with a child
		No	Yes
Lives with child (%)	9.3
Mother’s needs
Mean age, years (SD)	80.7 (5.0)	80.5 (4.9)	82.7 (5.1)
Poor or very poor health (%)	10.2	9.0	22.3
Mother college educated (%)	5.6	5.6	5.6
Deceased husband’s occupational status (SD)	33.0 (21.2)	32.9 (21.2)	34.2 (21.2)
Widowed in past 2 years (%)	12.3	12.8	7.9
Family characteristics
Mean number of children (SD)	3.8 (2.2)	3.8 (2.2)	3.3 (2.0)
One or more daughters (%)	86.4	86.2	88.4
Mean age of youngest child, years (SD)	29.1 (5.9)	28.9 (6.0)	31.1 (5.1)
Children’s resources
At least one child college educated (%)	46.0	45.5	50.2
All children have same education (%)	22.8	22.1	29.8
N	2,324	2,109	215

Notes. Excludes institutionalized parents and those living with a relative other than their child. Variables are defined in the text.

Source: Wisconsin Longitudinal Study.

Table 2 shows the odds ratios (ORs) from the logistic regression models predicting widowed mothers’ coresi dence with a child for families of all sizes and with at least two children, the starting point for the within-family analysis. The results for the two samples are very similar, as expected given the small number of mothers with only one child. Compared with other widowed mothers, those who are older and those in poor health are more likely to live with a child. Being in poor health increases a mother’s odds of living with a child by 180% (184% for all families, 179% for families with two or more children). Neither mother’s education nor the occupational status of her deceased husband affects a mother’s coresidence with a child, but having at least one child with a college education increases the chance of coresidence among mothers with two or more children. Having at least one daughter increases a mother’s chance of living with a child (but this is only significant at the p ≤ .10 level). Number of children is negatively associated with coresidence, but this association becomes statistically insignificant in the sample with two or more offspring. Age of youngest child is not associated with coresidence. In analyses not shown, we also examined whether mothers were more likely to live with a child if they had a child younger than 18 years or a child younger than 21 years, ages when some young adults might not yet have left their parents’ homes. Age of youngest child is not associated with coresidence, regardless of the age cutoff we use.

Table 2.

Parameters From Logistic Regression of Widowed Mother’s Coresidence With a Child, Between-Family Analysis

	Widowed mothers
	All		Two or more children
	Odds ratio exp()	|z()|	Odds ratio exp()	|z()|
Mother’s needs
Age, years	1.074**	4.26	1.078**	4.13
Poor or very poor health	2.840**	5.64	2.790**	5.08
Mother college educated	0.712	1.02	0.617	1.28
Deceased husband’s occupational status	0.998	0.54	0.998	0.56
Widowed in past 2 years	0.644+	1.65	0.672	1.41
Family characteristics
Number of children	0.918+	1.67	0.922	1.56
One or more daughters	1.557+	1.86	1.772+	1.86
Age of youngest child, years	1.029	1.50	1.025	1.29
Children’s resources
At least one child college educated	1.271	1.52	1.401*	2.00
All children have same education	1.159	0.73	1.039	0.17
Log likelihood	−672.8		−591.0
N	2,324		2,126

Notes. Excludes institutionalized parents and those living with a relative other than their child. Variables are defined in the text.

Source: Wisconsin Longitudinal Study.

⁺ p ≤ .10. *p ≤ .05. **p ≤ .01.

With Whom Do Widowed Mothers Coreside, If They Live With a Child?

Table 3 shows the characteristics of children in families where one of the children and the widowed mother live together. We report these characteristics for the cases with full information from the survey. The table also shows the number of cases for which the characteristic is observed. The family composition variables have levels generally consistent with those available for the between-family analysis. Just over half of the offspring are daughters. Over a quarter are college graduates. Two thirds of offspring are married and 17% have children younger than 18 years in their households. Sixty-six percent of the offspring are employed. Rates are slightly higher for sons, 70%, than for daughters, 64% (not shown). Only 8% of children reported that they received $1,000 or more from their parents for their education, a house down payment, or to help start a business. Most children, 62%, say that they are very close to their mother, but a substantial minority have less salutary relationships.

Table 3.

Characteristics of Children in Families in Which a Widowed Mother Lives With a Child

Variable	Percenta	Number of observationsb
Daughter	51.7	687
Eldest child	26.9	687
Youngest child	26.9	687
College graduate	26.2	687
Marriedc	65.8	363
Has child <18 in householdc	17.4	294
Employedc	66.5	367
Received financial transfer from parentsc	7.9	357
Has close relationship with motherc	61.9	176

Notes. N = 687 children from 185 families with two or more children. Variables are defined in text.

Source: Wisconsin Longitudinal Study.

^aPercentages are for children with complete data.

^bObservations missing owing to siblings sampled at random, questions asked of a random subsample, and nonresponse.

^cAvailable only for graduate and randomly selected sibling.

We report results from three conditional logit models of coresidence on children’s characteristics. Model 1 takes account of demographic characteristics: gender, birth order, education, marital status, the interaction of gender by marital status, whether the adult child has minor children, and employment status. Model 2 adds whether the child received financial support from parents, and Model 3 adds relationship quality. All models control for whether the child was neither the graduate nor the randomly selected sibling. The models also include an indicator for whether information on the randomly selected sibling was missing in 1993.

Table 4 shows the ORs for the associations between children’s characteristics and whether they live with their widowed mother. Being an eldest child does not increase the odds of coresidence. Youngest children are more likely to live with their mother than are middle children. Few WLS respondents have siblings who are still in the early stages of adulthood, but we investigated whether the birth order difference occurs because the youngest child was still being launched from the parent’s home. We reestimated the model with a variable to indicate whether the child was in the young adult years (e.g., younger than 21 years, younger than 25 years), but including this variable did not explain the association between being the youngest child and coresidence (not shown).

Table 4.

Parameters From Conditional Logit Model of Which Child Lives With Widowed Mother, Within Family Analysis

Variable	Model 1		Model 2		Model 3
	Odds ratio exp()	|z()|	Odds ratio exp()	|z()|	Odds ratio exp()	|z()|
Daughter	0.376*	2.41	0.374*	2.42	0.331**	2.65
Eldest child	1.212	0.91	1.211	0.90	1.321	1.29
Youngest child	1.495+	1.93	1.508*	1.97	1.638*	2.25
College graduate	0.748	1.16	0.771	1.02	0.811	0.80
Married	0.024**	7.54	0.023**	7.50	0.019**	7.81
Married daughter	13.67**	4.64	13.91**	4.63	17.63**	4.99
Has child <18 in household	0.655	1.02	0.640	1.07	0.593	1.29
Employed	1.242	0.74	1.212	0.66	1.131	0.41
Received financial transfer from parents			1.648	1.03	1.654	1.03
Has close relationship with mother					3.213**	2.92
Log likelihood	−176.3		−175.8		−167.3
Number of parameters	11		12		14

Notes. All models include a dummy variable identifying children who are not graduates or randomly selected siblings. Models also identify randomly selected siblings who were nonrespondents. The models control for the interactions between each missing indicator variable and daughter. Model 3 also includes a dummy variable controlling for missing data on the closeness indicator. Variables are defined in the text.

Source: Wisconsin Longitudinal Study. N = 687 children from 185 families with two or more children.

⁺ p ≤ .10. *p ≤ .05. **p ≤ .01.

Table 4 also shows that college-educated children do not differ from their siblings who did not complete college in the likelihood of living with their mothers. Daughters and sons, however, do differ in the odds of coresidence, depending on their marital status. Unmarried sons are more likely than unmarried daughters to live with their widowed mother. The odds of an unmarried daughter living with her widowed mother are about one third the odds for an unmarried son (0.376). Marriage reduces coresidence for both sons and daughters. The odds that a married son lives with his mother are only 2.4% of the odds of an unmarried son (0.024), and for a married daughter, the odds of coresidence are 12.3% of the odds of an unmarried son (0.376×0.024×13.67). Another way to interpret these results is to compare the predicted probabilities of coresidence for the different gender by marital status combinations, with other variables evaluated at their sample means. The probability that a married daughter coresides is 0.529, about 5 times higher than for a married son, 0.103. The difference between married sons and daughters is statistically significant (not shown). Neither having children in the household nor employment status is associated with coresidence.

We investigated gender differences in the association between employment and coresidence because daughters’ employment may make them less available to provide care to older mothers. The interaction of gender by employment status is small and statistically insignificant (not shown). Because very few unmarried sons lived with minor children in this sample, we could not examine gender differences in coresidence among single parents.

Model 2 shows that having received a large financial transfer from parents is not associated with coresidence. The pattern of associations between other child characteristics and coresidence remains the same when the transfer variable is included (Model 1 vs Model 2). In contrast, emotional closeness has a large, statistically significant association with coresidence, as shown in Model 3. The odds of coresidence for children who feel very close to their mothers are 3.2 times the odds for children who do not have as close a relationship. We interpret this association cautiously because relationship quality is measured at the same time as coresidence, and therefore closeness may be the result of sharing a home as well as a reason to coreside. This association also is sensitive to the treatment of missing data, unlike our other results.

Summary and Conclusions

Our between-family analysis shows that characteristics associated with widowed mothers’ needs, advanced age and being in poor health, are associated with living with a child, consistent with our expectations. We find mixed evidence on the association between socioeconomic status and coresidence. Mothers who are economically disadvantaged are no more likely to live with a child than mothers who are economically advantaged. Children’s education, a rough proxy for the collective ability of offspring to provide for mothers, is associated with coresidence in families with two or more children. Having a college-educated child increases the likelihood that a mother will live with one of her children. Recent widowhood and family size are not consistently associated with coresidence.

As in previous research, we find support for our hypothesis that having a daughter is associated with mother–child coresidence. The within-family results, however, demonstrate that the child with whom the mother lives is not necessarily the daughter herself. This is in contrast to the finding from census data that elderly, unpartnered mothers and daughters are increasingly likely to coreside after the mother reaches advanced old age (Schmertmann et al., 2000). Reconciling these findings requires data on the availability of all adult daughters and sons and their marital statuses, whether they live in a multigenerational household or some other type of household.

Within families, children’s gender and other family responsibilities affect coresidence with mothers. Being married reduces coresidence, but more for sons than for daughters. Unmarried sons are more likely than other children to live with their widowed mother, raising the question of who benefits from coresidence. Widowed mothers and their unmarried sons may be most likely to coreside because the traditional gendered division of labor facilitates exchanges between a mother and unmarried son. Mothers are able to exchange help with cooking and laundry for sons’ financial assistance (Ward & Spitze, 1996). That the widowed mothers in our sample are at advanced ages suggests that even when they are living with an unmarried son, he may be doing more than simply sharing expenses.

Our finding about the importance of having at least one daughter suggests that the processes that explain why having a daughter increases coresidence occur at the family level. Although we cannot investigate the mechanisms involved, perhaps all children in families with at least one daughter have had more caregiving socialization during childhood. Alternatively, daughters may directly encourage their siblings to care for older mothers. Quantitative evidence suggests that, compared with brothers, sisters are closer to and more involved with their siblings (Connidis, 2009; White & Riedmann, 1992). Qualitative evidence also suggests that when parents are older, sisters coordinate parental care with their brothers (Matthews, 2002). What factors account for gender differences in intergenerational coresidence and how these vary at the individual and family levels are important topics for new research (Smits et al., 2010). A recent study of older adults in the Netherlands suggests that new research should consider how similarity in siblings’ characteristics affects who help parents and the degree to which siblings share responsibilities for parent care (Tolkacheva, van Groenou, & van Tilburg, 2013).

Our finding that parents’ help with educational expenses or a house down payment is not associated with coresidence suggests that sharing a household with an older parent is not simply a matter of reciprocity for earlier assistance. That children with close relationships with their mothers are more likely to share a home with her also is consistent with a view of family relationships as more altruistic or motivated by a common culture rather than a direct system of exchange. A limitation of our study is that relationship quality is only measured contemporaneously and for a random half of the graduates and siblings, which contributes to the instability of this finding across models.

Affective solidarity is a critical concept in theories of intergenerational relationships (Merz, Schuengel, & Schulze, 2007; Silverstein, Bengtson, & Lawton, 1997), but adult parent–child relationship quality is not measured for multiple children in two of the most commonly used large national data sets used to study contemporary intergenerational relationships, the Panel Study of Income Dynamics and the Health and Retirement Study (HRS). The HRS does ask respondents how many of their children they feel close to, but the survey does not ask the quality of relationship with individual children. The National Survey of Adolescent Health (Add Health), a cohort study much like the WLS, includes questions about parent–child relationship quality, but respondents are still in early adulthood and their parents are only now approaching old age. Studies that focus on families from one region of the country, such as the WLS and the Longitudinal Study of Generations, help address this gap in national data. We see the relationship quality, which may be positive, negative, or ambivalent, as a critical construct to be included in new, national data collections. The benefit of obtaining information about the quality of parent–child relationships can best be realized in prospective, longitudinal designs to address such questions as whether close relationships increase intergenerational coresidence and how the loss of privacy associated with coresidence affects parent–child relationships.

Our findings point to the importance of looking beyond a parent–child dyad to take account of all of a parent’s offspring. The characteristics of families that predict intergenerational coresidence—having a daughter, having a highly educated child—differ from the characteristics that predict coresidence within families at the child level. Within families, unmarried sons are more likely to live with a widowed mother than either unmarried or married daughters. There also is no association between a child’s education and coresidence. A way to move beyond the dyad is to broaden the conceptualization of childhood socialization to take account of socialization processes that affect all children in a family, either because parents interact with all of their children at the same time or because how parents treat one child affects the parents’ relationships with other children. These family dynamics have been investigated in childhood (McHale & Crouter, 2008) but are too infrequently examined in adult family relationships.

The WLS data we use represent a cohort born when fertility was much higher than it is today. Just over 2 out of 10 women born in 1910, around the time many of the mothers of WLS respondents were born, had four or more children, compared with more than 1 out of 10 women born in 1960 (Kirmeyer & Hamilton, 2011). Mothers with fewer children are less likely to live with an adult child in later life (Seltzer et al., 2013). In addition, increases in remarriage and nonmarital childbearing mean that higher percentages of families have siblings who do not have the same parents (Seltzer & Bianchi, 2013). This may make coresidence a less common form of intergenerational transfer, even when parents are in need. Public opinion data on perceived obligations to share a home with an older parent suggest that obligations to stepparents are much weaker than obligations to biological parents (Coleman & Ganong, 2008). Consistent with these attitudes, mothers are less likely to live with adult stepchildren than biological children in late life (Pezzin, Pollak, & Schone, 2008). Stepfamily life may include more strain and conflict, which limit both generations’ willingness to give up privacy for shared residence. Stepfamilies also do not use close geographic proximity to substitute for coresidence. Stepmothers and adult offspring are less likely to live near each other than biological mothers and offspring (Seltzer et al., 2013).

Another limitation of the WLS data is the lack of race–ethnic diversity in the sample. Racial and ethnic diversity in aged parents’ relationships to adult offspring is a vital topic for research in light of growth in the minority composition of the old age population. By 2050, Hispanics will make up one fifth of those aged 65 and older (Vincent & Velkhoff, 2010, Figure 5). African Americans and Hispanics are more likely than non-Hispanic whites to live in intergenerational households (Cohen & Casper, 2002; Mykyta & Macartney, 2011). Yet, they also are more likely to have stepkin and quasi-kin whose ties to each other are less durable than those between biological kin and may be more difficult to activate in late life when a vulnerable, widowed parent needs help. It is important to consider both race–ethnic diversity and the diversity of marital and partnership experiences in early and middle adulthood to anticipate how adult offspring are likely to provide for their aged mothers when they are most vulnerable due to the loss of a spouse. Notwithstanding these limitations of the WLS data, the study provides valuable information about widowed mothers’ coresidence with adult offspring. The WLS cohort, who came of age during the late 1950s, are now in their 70s, approaching the ages at which they too are more likely to need assistance from adult offspring. The model they set by caring for their own aged parents may be a significant role model for how their adult children will help them in old age (Cox & Stark, 2005).

Funding

This work was supported, in part, by the California Center for Population Research , which receives core support from the National Institute of Child Health and Human Development at the National Institutes of Health (R24-HD41022). The research was also supported by Friedman’s traineeship in the University of California, Los Angeles Interdisciplinary Relationship Science Program sponsored by National Science Foundation and by the Robert Wood Johnson Foundation Health & Society Scholars program. The Wisconsin Longitudinal Study has been supported principally by the National Institute on Aging (AG-9775, AG-21079), with additional support from the Vilas Estate Trust, the National Science Foundation, the Spencer Foundation, and the Graduate School of the University of Wisconsin-Madison.