In contrast to 30 years ago when approximately 60% of children lived only with married parents and siblings (Casper and Bianchi 2002; Kreider and Ellis 2011), more than 50% of children now grow up with only one parent, or with parents who cohabit, or with adults other than parents (Casper and Bianchi 2002; Kreider and Ellis 2011). The changes in children’s living arrangements are important to study because children’s wellbeing is related to the types of households in which they are raised. Children in female-headed households, for instance, are poorer, more prone to welfare use, more likely to drop out of school, and less successful as adults than children living in two-parent households (McLanahan and Sandefur 1994). Other research suggests that children living with both biological parents who cohabit are less likely to receive welfare than children living with single mothers or mothers cohabiting with unrelated males (Brandon 1999). Also, studies suggest that children in foster homes do worse at school and in later adulthood than children in parental homes (McDonald et al. 1993). Overall, compelling evidence suggests children’s development, school attainments, economic wellbeing, and later adulthood relate to their living arrangements.

Far from understood, however, are the mechanisms leading millions of today’s children to eventually live with grandparents. Yet, in the United States in 2008, for example, an estimated 6.4 million grandparents lived with grandchildren and among them about 2.6 million (nearly 41%) had responsibility for their grandchildren (U.S. Census Bureau 2011). Since the 1990s, a consistently large proportion of grandparents have had responsibility for grandchildren (Simmons and Dye 2003). Although only a fraction of all children live in the households of grandparents with no parents present the numbers are still large and constitute a major nonparental living arrangement (Kreider and Ellis 2011; Lugailia 1998).

Despite data from several countries showing that many children live with grandparents, few studies have identified socioeconomic and demographic factors which may predispose children to eventual coresidence with grandparents. Understandably, this lack of knowledge stems from a paucity of data that can predict family change, including most of the current sources of longitudinal data.

Notwithstanding drawbacks with existing longitudinal data, addressing this key question of contemporary family life is imperative. Thus, this study aims to discover some of the social levers that are associated with the rise of three-generational households nowadays. In this study, I identify predictors of transitions into three-generation households and provide evidence on the relative effects of these predictors. I conjecture that even with less than full information about the constellation of psychological–emotional factors that bring grandparents and grandchildren together under the same roof, there are socioeconomic and demographic predictors that could increase the chances of two- to three-generational household transitions. I explore our conjectures using data from the Household, Income, and Labor Dynamics (HILDA) survey of Australia, which is a household panel study containing precise and comprehensive measures of all household relationships over time.

Background

Until the late-1980s, many industrialized countries could not count the numbers of children living with grandparents because household surveys omitted specific categories allowing classification of household members as either grandparents or grandchildren, even though the phenomenon of “kinship care” has long been a recognized practice in countries like Australia, the focus of this study (Joslin and Brouard 1995). In Australia, the Australian Bureau of Statistics (ABS) (2001) Census of Population and Housing had a category of “Other not classifiable household,” numbering over 112,000 children, which may have included children living with their grandparents (COTA 2003). Since the early-1990s, however, most developed nations’ household surveys, including Australia’s 2003 Family Characteristics Survey, have incorporated categories permitting classification of household members as grandparents or grandchildren (COTA 2003). The statistics emerging from these more recent surveys for several industrialized countries, like Australia, the United States, and the United Kingdom, suggest that the numbers of children living with grandparents are increasing (Glaser et al. 2010).

In Australia, for example, in 2003 there were 22,500 families in which grandparents were the guardians of grandchildren (31,100 children), under 17 years of age (ABS 2005). In 73% of these grandparent families, the youngest child was between 5 and 14 years and of the 31,100 children, 28,700 of them had biological parents living elsewhere (ABS 2005). The ABS study (2005) reported that in 61% of the grandparent families, the younger grandparent or single grandparent was 55 years of age or older. Almost half of these families were lone grandparent families, many received welfare (62%), and about a third were employed (ABS 2005).

Brandon (2004) in a study of Australian children’s living arrangements using the HILDA survey estimated that at least 60,000 Australian children lived with grandparents in three-generation households in 2001. De Vaus and Gray (2003) estimated that about 30% of Australian children lived in alternative family structures to living with both biological parents at some time during their childhood. Other Australian research strongly suggests that the number of grandparent-headed households is grossly underestimated, despite the major role such households play in the child welfare system (Horner et al. 2007).

Other industrialized countries have experienced large growth in the numbers of grandchildren living with grandparents, as well (Goodman 2003; Hank and Buber 2009). In the United States, 3.2% of children lived in households maintained by grandparents in 1970. By 1997, the percentage had risen to 6% (Casper and Bryson 1998). Large increases occurred among all types of households maintained by grandparents regardless of the presence or absence of children’s parents, but increases were greatest among children with only one parent in the household (Casper and Bryson 1998; Kreider and Ellis 2011; Minkler and Fuller-Thomson 2005). The number of grandchildren living in households maintained by grandparents with just mothers present increased by 118% from 1970 to 1997, while those living with just fathers increased by 217%. In contrast, smaller increases occurred among those living with both parents (53%) and those living with neither parent (37%). However, since 1990, the greatest growth in the United States has occurred in the number of grandchildren residing with grandparents only, with neither parent present. By 1997, a third of grandparent-maintained families did not contain either parent of the child (Bryson and Casper 1999; Mutchler and Baker 2004; Mutchler and Baker 2009; Silverstein 2006; Park 2005). According to Livingston and Parker (2010), American grandparents are increasingly more likely to raise grandchildren. Using 2008 Census data, they estimated that about 7 million children lived in households that included at least one grandparent. Of that number, 2.9 million children were raised primarily by grandparents—up 16% from 2000.

As one final example, in the United Kingdom, information from the British Social Attitudes Survey for 2001, (and earlier in 1988), suggested that there was approximately 100,000 children under the age of 13 living with a grandparent. (Richards and Tapsfield 2003).

The numbers of children living with grandparents has caught the attention of scholars and policymakers. Indeed, an emerging literature documents the rise of grandparent-maintained households, reasons for the rise, and their caregiving practices (see Bowman 2011; Burton 1992; Chalfie 1994; Dowdell 1995; Dressel and Barnhill 1994; Fuller-Thomson et al. 1997; Jendrek 1994; Joslin and Brouard 1995; Minkler 1998; Minkler and Roe 1993; Rutrough and Ofstedal 1997; Shor and Hayslip 1994). And, the trend has led some policymakers to question whether public policies protect grandparent rights and their economic well-being (see Fitzpatrick and Reeve 2003; Gerard et al. 2006; Hayslip and Kaminski 2005; Tasmanian Parliament 2003; COTA 2003; U.S. Senate, Special Committee on Aging (1992); U.S. House of Representatives, Select Committee on Aging (1992); Congressional Record 2000).

As continued growth of three- and skipped-generation households is expected, more research is needed. Especially urgent is to know which factors push or pull two-generation households towards forming three-generational households (Sorensen and McLanahan 1990). This study hypothesizes that adverse health, immigrant status, and the economic organization of two-parent and single-mother households can increase the odds of establishing three-generation households. By pursuing this line of inquiry, this research offers a backdrop to cross-sectional studies documenting the relatively poor economic situations of three-generation households (Baker et al. 2008; Brandon 1999, 2005; Bryson and Casper 1999; Chalfie 1994; Fuller-Thomson et al. 1997; Rutrough and Ofstedal 1997).

Theoretical Perspectives

The process leading to three-generation household formation is undoubtedly complex. Theoretically, a compound set of biological, psychological, familial, and socioeconomic factors could interact over time and space to transform a household from a two-generation to a three-generation one. Possibly, establishing a three-generation household happens over time and requires a particular alignment among a multifaceted set of factors. Although these sorts of theoretical possibilities and complexities are beyond the scope of this study, I nevertheless posit that among the constellation of factors there is a subset of key sociodemographic and economic factors that can affect transitions from two- to three-generation households. Mindful of the limitations of social science theory for explaining three-generation household formation, there are relevant economic and sociological frameworks that can pinpoint economic and sociodemographic factors that could affect the likelihood of this sort of household transition.

The economic approach to understanding family organization suggests that a two-generation household’s economic resources and the costs of raising children should affect transitions to three-generation households. Economists Becker and Lewis (1974) argued that the costs of raising children lead parents to limit the size of their household so that fewer children have more resources. I conjecture that more children decrease the odds of forming a three-generation household rather than increase the odds because more children in a household imply fewer resources for other family members, including potential live-in grandparents. I think that any economic gains from grandparent coresidence, like child care, fail to outweigh the costs of fewer resources for investments in children, or sustaining children’s consumption patterns and levels without major adjustments to the organization of the household, e.g., increased parental work hours or depleting household savings. Also, more children provide parents with added incentive to help aging parents find substitutes in the housing market, such as, assisted living quarters. Obviously, no perfect market substitutes exist to parental care and the parental home, or are necessarily appropriate, for most young children.

Other economic research suggests that welfare affects family composition. Posner (1986) stresses that welfare income reduces the cost of raising children and corrects the social problem of parental underinvestments in children when parents lack sufficient income to care for and invest in their children’s development. Honig (1974), Moffitt (1992), and Lundberg and Plotnick (1995) suggest that the generosity of welfare benefits is associated with family structure and composition. Also, Paxon and Waldfogel (2000) argue that decreased welfare benefits raise rates of child maltreatment and thereby increase risks of children entering the foster care system. Drawing upon this research, I hypothesize that government cash transfers should increase odds of three-generation household formation since government cash transfers offset the costs of grandparent coresidence and help maintain income levels for investments in children.

Family adaptation theory also helps explain three-generation household formation. This theory suggests that families try an array of adaptation strategies when confronted with stresses, including doubling up to form three-generation households (Koh and MacDonald 2006). Thus, in response to distress, families adapt by forming three-generation households. Perhaps, this major adaption occurs after other adaptations have failed, e.g., expenditure cutbacks (Conger and Elder 1994), moving to cheaper housing (Aaronson 1995; Fuller-Thomson and Minkler 2003), accepting welfare (Swanson et al. 2008; Yeung and Hofferth 1998), or splitting apart (Brandon and Fisher 2001). Indeed, many types of adversities could make it efficient for generations of a family to band together under one roof.

I identify two stressors confronting a two-generation household that could trigger a transformation into a three-generation household. I hypothesize that parents having to maintain a high level of labor market participation while raising children (Hunts and Avery 1998) and poor health among family members would both increase the likelihood that the family adaptation strategy is forming a three-generation household. If both parents work while raising children, increases in their hours of labor market work will create more work-family challenges, possibly even disrupting the work-family balance. Increased parental work effort I expect will make grandparent coresidence a more attractive option for preserving parental work hours while sustaining work-family balance. Hence, I conjecture that increased work hours among parents raise the chances of grandparents moving into existing two-generation households to form three-generation ones. And, the demands of caring for family members in poor health or with long-term conditions or disabilities might be a sufficient stress to increase odds of forming a three-generation household.

Naturally, I posit that the economic approach and family adaption theory operate within broader systems, including family customs, traditions and history, cultural contexts, life-course events, neighborhood, sociohistorical circumstances, etcetera. I cannot incorporate all these layered systems, but I can hypothesize that indicators of family distinctiveness and geographic context affect the formation of three-generation households (Bronfenbrenner 1979). Among “ecological” elements of the family, I predict immigrant status, one measure of family’s cultural distinctiveness, and geographic location of the household will affect the odds of forming a three-generation household.

Although, the literature highlights that immigrants frequently live in three-generation households (Wilmoth 2001), I expect that strictly foreign-born immigrant families will in fact be less likely, not more likely, to form three-generation households. I expect this outcome because formation of three-generation immigrant households requires first-, second-, and third-generation immigrants to reside in the same host country so that the multi-generational immigrant household can arise. But, I cannot assume that all generations of an immigrant family live in the host country. Hence, I make the reasonable assumption that foreign-born immigrant parents in two-generation households—having left their country of origin—are less likely to form three-generation households in the country of destination because their own parents are most probably back in the country of origin.

Geographic location and proximity to relatives, other family ecological factors, should determine the formation of three-generation households, as well (Nichols and Junk 1997). Two-generation households in rural areas live in a different context compared with two-generation households in urban centers. Proximity to neighbors and relatives, e.g., grandparents, is often easier in rural towns. I conjecture that localized resources, shorter work commutes, and closer-knit community networks as well as informal networks for monitoring aging grandparents who live independently decrease the need to form three-generation households.

Our goal is contributing knowledge to the scant literature on transitions from two- to three-generation households. The two theories highlighted offer valuable frameworks for explaining aspects of this complex phenomenon. I hypothesize that the formation of three-generation households is associated with two-generation households’ economic well-being, health status of its members, labor market demands, and ecological factors. To test our conjectures, I use longitudinal data from Australia and estimate a statistical model measuring effects of the socioeconomic variables on the hazard of forming a three-generation household. The variables capture key aspects of our theoretical considerations, such as, economic well-being, (e.g., income levels), or family stress, (e.g., parent health status or work demands).

Data Description and Statistical Approach

The data for this study come from the seven waves of the HILDA survey which began in 2001. The HILDA survey selected a nationally representative sample of 7,682 Australian households, thereby yielding a total household response rate of 66%.Footnote 1 Within the 7,682 sampled households, 19,917 persons were enumerated. Interviews were sought with every member of these households who was over the age of 15 years (Watson and Wooden 2002). Of the 19,917 persons, 4,790 were under 15 years of age and ineligible for an interview in Wave 1. This left 15,127 persons eligible for a personal interview 13,969 of which completed the Person Questionnaire and then sought interviews with every member of those households who were over the age of 15 years (Watson and Wooden 2002).

From these seven waves of HILDA data on the same individuals and households, the sample of children younger than 15 years of age with at least one parent 45 years or younger numbered 6,338. Of them, 4,402 were present in Wave 1; the remainder uniformly entered the panel between waves 2 and 7. Among the 6,388 children, 289 already lived with a grandparent when first observed leaving 6,099 children potentially “at risk” of living with a grandparent. However, before proceeding I also excluded the small number of children living with single fathers (N = 115), living with neither parent (N = 27), or those whose parents refused to answer the survey (N = 83). Thus of the 6,099 children, 5,874 children were included. Of the 5,874 children, about 1.31% (N = 77) eventually lived with a grandparent by wave 7, thereby confirming that this type of event is rare. The rarity of this transitional event from a two- to a three-generation household was expected since the “take-up” rate, resembles the small prevalence of children, (1.4%), who live in three-generation households at any given point in time (Brandon 2004). Though a small proportion of all households raising children, these particular households regularly face enormous difficulties (Bryson and Casper 1999).

Of the 5,874 children, 815 and 5,059 children lived with single mothers and two parents, respectively. By exploiting the panel structure of HILDA, I could create child-year files with time-varying and invariant data for each of the 7 years of data. The analysis file contained 2,829 child-year observations for single mothers and 17,721 child-year observations for those with two parents.Footnote 2

Only limited data were collected on children as HILDA is a not a panel study of children. But, data were collected on children’s ages, health, and sex. This key child information was merged with the rich information on parents and households, including parental place of birth, labor force participation, sources of income, family composition and relationships, and household poverty. Overall, HILDA contained much time-varying socioeconomic data to identify factors associated with children’s transitions from two- to three-generation households.

Statistical Model

The formation of a three-generation household is undisputedly a rare event, but one with huge implications for grandchildren, parents, and grandparents alike. Under such conditions where an outcome, like a three-generation household forming, is a rare occurrence, then random-effects complementary log–log regression is preferred to logistic regression (Agresti 1990; Long and Freese 2001). Also, since the data show that the probability of grandchild-grandparent coresidence approaches zero and one at different rates, assuming a symmetric link using the logit model is unreasonable as its link is symmetric and might bias coefficient estimates. Thus, the random effects cloglog model is again preferable to a random effects logit model because the former has an asymmetric link approaching one faster than zero.

Another advantage of using the random effects complementary log–log regression is that it is suitable for survival analyses—certainly a reasonable approach for studying entries into a three-generation household. The survival function describes the probability that a child fails to have a coresident grandparent in year t. The hazard function describes the probability that a child living with a lone parent or with two parents will coreside with a grandparent in year t given that the child has not coresided before year t. As time in HILDA data is measured discretely, I exploit one of the most common discrete-time hazard functions: the complementary log–log hazard function (Singer and Willett 2003). The complementary log–log assumes that the hazard takes the form,

$$ \lambda (t|m_{\text{it}} ) = 1 - [1 - \lambda_{0} (t)]^{{\exp ({\text{bmit}})}} $$
(1)

where m it is a vector of characteristics for child i in year t and β is a vector of parameters. The baseline hazard is the probability that a child will coreside in year t given that mit is zero and that the child does not coreside before year t.

The cloglog transformation of (1) yields,

$$ \log ( - (\log (1 - \lambda (t|m_{\text{it}} )))) = \alpha_{t} + \beta m_{\text{it}} $$
(2)

where αt is the cloglog transformation of the baseline hazard λo(t).

I assume that \( \beta m_{\text{it}} \) is a linear function of child, parent, and household characteristics indexed by time (t = 1, …, 7) and a child-specific random component that is normally distributed with mean 0 and variance \( \sigma_{\mu }^{2} \) Measures for children include age, sex, and health status. I include measures of parental age, sex, health status, nativity status, educational attainment, work hours, and employment status. Household measures include number of children under 14 years of age, household income, receipt of government transfers, homeownership status, and the number of bedrooms in the house.

Findings

Table 1 displays informative descriptive statistics on three groups of children in HILDA who are relevant to this study: (1) 289 already living in three-generation households; (2) 77 who entered three-generation households; and, (3) 5,797 who by the end of the survey stayed in two-generation households. Table 1 shows substantial differences across the three groups of children when they were first surveyed. The 289 children already in three-generation households were younger than the other two groups of children. These same children also live in three generation households that have more total household income with a larger proportion of that income coming from government cash transfers. Table 1 indicates children already with grandparents are more likely group among the groups of children to receive government aid. Their households also contain more people, and grandparents are much more likely to provide child care services. Interestingly, these children already in three-generation households and the children who eventually enter into three-generation households are the most disadvantaged children among the three groups according to the index of relative disadvantage, which fits with existing research.

Table 1 Descriptive statistics on child and household characteristics when child first observed, by living arrangements
Full size table

Parental characteristics for the three groups of children differ too. Table 2 shows that parents of those children who enter into three-generation households or already living in three-generation households are younger than parents in two-generation households. Though they are relatively younger parents, parents who enter into three-generation households are more likely to report a long-term health condition, disability, or impairment. Given the high rates of compromised health status, finding that these specific mothers and fathers are the least likely to have full-time employment compared with parents who do not form three-generation households is unsurprising. Parents who eventually enter three-generation households are also the least likely to have a university degree or completed high school compared with the parents who do not form three-generation households; also, they are more likely to cohabit rather than marry among any of the three groupings of parents. On the other hand, these fathers tend to work as many hours as the fathers of the other two groups of children. Parents not entering into three-generation households by the end of the panel were also the ones earning more in total and, predictably, the less likely to receive government cash transfers. Lastly, reflecting the literature on immigrants’ living arrangements (Wilmoth 2001), children already in three-generation households were more likely to live in non-English speaking households and have at least one foreign-born parent.

Table 2 Descriptive statistics on parental characteristics when child first observed, by living arrangements and gender
Full size table

Tables 1 and 2 suggest differences between two- and three-generation households and possible socioeconomic and demographic variables associated with three-generation household formation. Notwithstanding data limitations, Table 3 presents results from regression models showing estimated effects of selected socioeconomic and demographic measures on rates of three-generation household formation among two-parent and single-mother households. Coefficient estimates in the model show the proportional effect on the hazard ratio, i.e., the changes in the logarithm of hazard ratio resulting from a unit increase in an explanatory variable. Thus, the exponentiated coefficients are interpretable as hazard ratios (Allison 1995).

Table 3 Random effects complimentary log–log regressions exterminating entries into three-generation households, by family structure (standard errors in square brackets)
Full size table

Column 2 of Table 3 presents findings for the hazard of entry into three-generation households for children living with two parents. The model failed to confirm our conjecture that a child in poor health, with a chronic health condition, or disabled compared to a child without such health concerns would increase the hazard of forming a three-generation household as a family adaption. I had no prior hypothesis about a child’s gender, but this demographic trait was found unassociated with the hazard of a three-generation household forming. Yet, a child’s age was associated with the hazard of three-generation household formation: a 1 year increase in child age yielded about a 20% increase in the hazard of three-generation household formation.

Findings for effects of parental characteristics on the hazard of three-generation household formation were also mixed. While a year increase in maternal age is associated with a nearly 15% decrease in the hazard of three-generation household formation, her husband’s age, though a negative effect as well, had no significant effect on the hazard.

Two of the three estimated coefficients for husbands’ educational attainments suggest that husbands with high school diplomas or associates degrees compared with husbands with less than a high school diploma decrease the hazard of forming three-generation households. Though statistically nonsignificant, university-educated husbands share the same negative effect as other less-educated husbands. Overall, husbands who have attained higher education credentials appear disinclined to form three-generation households. For wives, the education attainment findings are inconclusive. While some educational attainment measures are nonsignificant, the estimated coefficient for wives with high school diplomas compared with wives without less than high school diplomas increases the hazard of forming three-generation households by two and a half fold. Like husbands, wives with associate diplomas or university-educated seem less inclined to form three-generation households.

I conjectured that increasing hours of work for both spouses and spouses’ health status would have impacts on the hazard of three-generation household formation. Regression results do not support our conjectures. Neither spouse’s work hours nor poor health status was associated with the hazard of three-generation household formation. Perhaps, our measures for work hours and health status are inadequate to detect effects, but in any event I cannot assert that a transition to three-generation households is a family adaptation response to poor health status or maintaining higher levels of labor force participation in a two-generation household.

Measures of economic well-being were also not statistically significant. In two-parent households, the log of household income, receipt of government transfers, and renting a home rather than owning or paying off a mortgage was not associated with the hazard of three-generation household formation. Our theory predicted that these were important predictors, but it was unsupported by the regression analyses.

Nevertheless, the number of children in a two-parent household confirmed the prediction of economic theory about the effect of increasing numbers of children on three-generation household formation. The estimated coefficient for the number of children in the household showed that an additional child decreased the hazard of three-generation household formation by about 35%. Further, other results for two-parent households confirmed the importance of measures of the family ecology. Geographic location of the two-parent household mattered as the coefficient for “Regional area” suggests that two-parent households in rural areas compared to two-parent households in cities lowered the hazard of forming a three-generation household by 71%. And, findings for wives suggest that if they are foreign-born, they are less likely to form three-generation households presumably because when they left their country of origin they left behind their own relatives, including parents. (The coefficient for foreign-born husbands is statistically nonsignificant and positive.) Analyses did not detect that either parent having been married before was associated with the hazard of three-generation formation.

Column 3 of Table 3 displays findings for single-mother households. In single-mother households, a child’s age, gender, and poor health status were not found associated with the hazard of entry into three-generation households. By contrast, however, estimated coefficients for several characteristics of single mothers indicate strong influences on the hazard of three-generation household formation.

Though only the coefficient for vocational training or an associate degree is significant, coefficients for the education attainment variables collectively suggested that the hazard of entry into three-generation households was decreased as single mothers attained progressively higher levels of education. The statistically significant effect of an associate degree or vocational training indicates that a single mother with this sort of an education beyond high school compared with a mother with less than a high school diploma is about 98% less likely to transition into a three-generation household.

Whereas maintaining hours in the labor force was not statistically significant for married mothers, for single mothers an hour increase in the number of hours worked was associated with nearly a 13% increase in the hazard of forming a three-generation household. Thus, from the perspective of adaptation theory, a single mother sharing housing with at least one of her children's grandparents, and thereby forming a three-generation household, helps preserve her labor market attachment; indeed, the transformation from single-mother household to three-generation household might help avoid welfare dependency and poverty.

Also consistent with our proposed theoretical hypotheses, for a percentage increase in the log of income, the hazard of forming a three-generation household fell by 51%. Similarly, fitting with our prediction and shown earlier for two-parent households, the number of children under 14 years of age in the household mattered for single mothers, as well. For an additional child in a single-mother household, the hazard of forming a three-generation household decreased by 80%, which is a larger effect than that found for married mothers.

Finally, family ecology matters for single mothers. Geographic location of her household was associated with the hazard of forming a three-generation household. Single mothers heading households in rural areas were 97% less likely to form three-generation households compared with single mothers living in metropolitan or suburban areas. And, like the estimated coefficient for foreign-born married mothers, the coefficient for foreign-born single mothers indicates that they are only half as likely to form a three-generation household compared with single-mothers who are not foreign-born. I argue the consistency of the foreign-born effect across mothers is because mothers’ parents are probably in the country of origin. Lastly, single mothers heading households were more likely to form three-generation households if they had never been married. In other words, unmarried childbearing plays a pivotal role in determining the formation of three-generation households, which clearly aligns with the literature.

To complete the portrait of newly-formed three-generation households, Table 4 compared characteristics of the 77 grandparents who became members of three-generational households with characteristics of grandparents already identified as members of three-generational households. Grandparents who formed three-generation households compared with grandparents who were already members of three-generation households were older and were less likely to have completed high school, have attachments to the labor market, or receive welfare. When working, however, those who formed three-generation households earned similar amounts as grandparents already in such households. By contrast, grandparents who enter three-generation households were more likely to have been foreign-born, and married or widowed.

Table 4 Comparisons between grandparents based on if observed to live in a three-generation household
Full size table

Discussion and Conclusions

Dramatic changes that have reshaped families since the 1970s make this study useful and timely (Casper and Bianchi 2002). The study produced new findings, especially for single-mother households, that support the literature, generate new insights, serve the research community, and have policy implications. Most of the significant findings fit with our theoretical predictions and confirm that sociodemographic and economic factors are indeed associated with the rise of three-generation households. Those findings include indications that odds of three-generation household formation decreased with mothers’ ages, fathers’ higher educational attainments, and more children, but increased as children grew older for two-parent households. Meantime, for a single-mother household, the transition from a two-generation to three-generation household also decreased with her higher educational attainment, increasing income, and more children, but increased if she was never married and worked more hours in the labor market. Furthermore, regardless of family structure, (i.e., mother-only versus two-parent household), living in a rural community decreased odds of three-generation household formation.

Oftentimes, estimated effects of coefficients on the hazard for three-generation household formation for single-mother households were more sensitive to changes in the various regressors compared with two-parent households and no doubt the findings have implications for welfare use and spells of poverty among mother-only households. Equally interesting are the comparisons possible between mother-only and married-mother households of directions of estimated effects on the hazard of three-generation household formation. Overall, the estimated effects for educational attainments, number of children, foreign-born status, and physical location are similar. Clearly, grandparents who share housing with their adult unmarried daughters raising children are crucial to the household’s economic well-being.

Besides producing findings that expand the literature, the study calls for future research to pursue issues left unaddressed in this study. For example, a central concern of ours was having no measures of the proximity of grandparents to grandchildren. Grandparents, most likely, who lived nearby grandchildren would have less incentive to form a three-generation household. I also lacked information on whether grandparents were deceased; and, for grandparents still living I possessed no data on their demographic and socioeconomic circumstances until they moved in with grandchildren because HILDA is a household-based panel survey, not a panel study of kinship networks. Thus, I am censored on the supply of grandparents available or alive to form three-generation households; the grandparents in this study forming three-generation households, (or already in such households), were most likely those that might possess the highest propensity for coresidence; in other words, our HILDA data probably yields a somewhat selective sample. Also, today’s families include a mix of blended and cohabiting families. I suspect that a study designed to examine generational networks, enumerate the stock of potential grandparents to form three-generation households, and track intergenerational resource flows might show that our coefficients are upwardly biased. Yet, until such a study is conducted our supposition is simply speculative.

There are also other analytical concerns that could affect our hazard estimates of transitions into three-generation households. Within the broader kinship network of a child, there are aunts, uncles, and cousins that grandparents could potentially choose to share housing with rather than the focal grandchildren in this sample. Our child in the HILDA panel is possibly only one of several grandchildren among whom grandparents could live. The grandparent might even have other residential options in the housing market that they might wish to consider. Alternatively, grandparents might feel that cash, in-kind transfers, and future bequests (Cao 2006; Sheng 2009) rather than joint living arrangements are more effective means for securing their grandchildren’s well-being and helping their adult children adapt to distress. I have no measures of economic transfers that would assist in reconciling the issue. Although certainly not the last concern, the families that are not observed living with grandparents might be different inter-generationally from those I observe living with grandparents. Theoretically, if there are intergenerational transmissions of wealth, income, and health, the families failing to live with grandparents may be predisposed that live close-by or far away. So, overall there are push and pull factors that I simply cannot address; I suspect that the overwhelming majority of existing longitudinal data sources on households could not address such factors either. I speculate that these issues are no doubt related to several of the nonsignificant results I report.

Notwithstanding these issues that should stimulate future research and encourage new data collection efforts to remedy problems faced by this study, because of the profound changes families have undergone in the past three decades and the implications for child and family well-being, I chose to proceed now rather than wait for the ideal data source.

In summary, the overarching messages from this study are that: (a) new theories on the formation of three-generation households are needed; (b) a new round of innovative data collection efforts, which transcend traditional household survey designs are required; and (c) greater attention to the lives of children who have higher odds of living in three-generation households is essential. In the long run, new longitudinal data sources need to carefully measure the formation of these households as well as identify which grandparents and which grandchildren are at the highest risk within these types of households. The future of families requires that researchers better understand the factors propelling some children into three-generation households that can either promote well-being or worsen it. This study attempts to make a contribution towards this understanding of how such three-generation households arise while highlighting the challenges associate with using existing sources of household panel data.