Abstract
Using data from the 1996–2008 panels of the Survey of Income and Program Participation and the 2009–2016 American Community Survey, we examine trends in U.S. children living in shared households (living with adults beyond their nuclear (parent/parent’s partner/sibling) family). We find that although the share of children who lived in a shared household increased over this period, the rise was nearly entirely driven by an increase in three-generation/multigenerational households (coresident grandparent(s), parent(s), and child). In 1996, 5.7 % of children lived in a three-generation household; by 2016, 9.8 % did likewise—more than a 4 percentage point increase. More economically advantaged groups (older, more educated mothers, married households) experienced the largest percentage increase in three-generation coresidence, although correlates of coresidence remained largely stable. Decomposition analyses suggest that the rise in Social Security receipt and changes in parental relationship status (less marriage, more single parenthood) most strongly explained the increase in three-generation households. Given the dramatic rise in three-generation households, more research is needed to understand the consequences of these living arrangements for children, their parents, and their grandparents.
Introduction
Many studies have shown the importance of economic, demographic, and policy factors in determining family living arrangements (e.g., Ruggles 2007). Family change (such as high divorce and nonmarital fertility rates) and other demographic factors (such as increased longevity) influence family dynamics and intergenerational relationships, including coresidence (Bengtson 2001). Understanding family change is important because a robust body of research has documented strong links between children’s living arrangements, economic well-being, and long-term child outcomes (e.g., Cherlin 2010). Despite increasing recognition of the diversity in children’s living arrangements (e.g., Carlson and Meyer 2014), to date, no research has studied trends in household sharing (at least one resident adult who is not the child’s parent/parent’s partner/sibling) among families with children and differences by type of household sharing.1 This is an oversight given that nearly 20 % of U.S. children (or 15 million) live in a shared household (Mykyta and Macartney 2012), and adults in shared households play an important role in the lives of children (Amorim et al. 2017; Cherlin and Seltzer 2014; Edin and Lein 1997; Pilkauskas 2012; Stack 1974) that can both positively and negatively affect children’s cognitive and socioemotional well-being (e.g., Dunifon and Kowaleski-Jones 2007; Mollborn et al. 2012; Pilkauskas 2014).
Using data from the 1996, 2001, 2004, and 2008 panels of the Survey of Income and Program Participation (SIPP), as well as the American Community Survey (ACS) (2009–2016), we add to the literature by examining two questions. First, what are the trends in household sharing among families with children over time, and are there differences by type of shared household (e.g., living with an aunt/uncle vs. a grandparent vs. a nonrelative)? Second, what explains the changes in shared household living arrangements? To answer the first question, we plot trends in shared living arrangements by household type and find that the increase in shared living arrangements over time is being driven by an increase in three-generation/multigenerational (coresident grandparent, parent, child) households. We then focus our analyses on three-generation households by examining which demographic groups have increased coresidence, whether the correlates of coresidence have changed over time, and the demographic/socioeconomic factors that help explain the increase. By understanding trends in children’s living arrangements, we can help inform policies and programs that support children.
Data, Measures, and Method
We use the Survey of Income and Program Participation (SIPP), a nationally representative survey of the noninstitutionalized population collected by the U.S. Census Bureau. Data come from the 1996, 2001, 2004, and 2008 SIPP panels.2 We use the Household Relationships Topical Module (HRTM), which was collected as part of the second wave of each panel and asked detailed information on all relationships of individuals in a household; we can identify the relationship of every person in the household to each child in the household. This household detail is uncommon in large nationally representative studies, which typically rely on household rosters that provide data only on the relationship to the reference person and thus miss out on many relationships in the household (Kreider 2008). Data from the HRTM are merged with the core data files and are restricted to the fourth reference month, the SIPP reporting month (Moore 2008). The sample in each panel wave used ranges from 30,000 to 42,000 households, including 20,000–28,000 children.
We also use the 2009–2016 ACS, a nationally representative survey of the U.S. population that samples approximately 3 million households annually. The ACS, collected by the U.S. Census Bureau, uses monthly rolling samples of households to produce annual estimates of the population for people in housing units (and group quarters for most years).3 For our analyses, we exclude individuals living in group quarters. The ACS data for this study were drawn from extracts made by the Integrated Public Use Microdata Sample (IMPUS-USA; Ruggles et al. 2017). The ACS collects household information through a roster with reference to a single person in the household and with less detailed information than the SIPP. We focus our analyses using the ACS on children living with grandparents (three-generation and skipped-generation households).
Household Sharing
Household sharing is coded to examine many types of shared living arrangements. Additional details on how the measures were constructed are available in the online appendix.
Shared household: a child lives with an additional adult beyond the parent/stepparent, sibling, or the cohabiting partner of the parent (also known as doubled-up households; Pilkauskas et al. 2014).
Extended family: a child lives with any adult relative beyond the parents/stepparents, sibling, or parent’s cohabiting partner. This includes grandparents, aunts/uncles, nieces/nephews, and other relatives (e.g., cousins).
Grandparents: a child lives with at least one grandparent.
Aunt/uncle: a child lives with at least one adult aunt/uncle.
Other relative: a child lives with at least one adult relative who was not the grandparent, aunt/uncle, sibling, or parent/parent’s partner.
Nonrelative: a child lives with at least one adult nonrelative.
Categories are not mutually exclusive; a child who lives with both a grandparent and nonrelative is included in both categories. In each case, a parent may or may not be present.
Prior research has noted large differences between three-generation/multigenerational households (children living with at least one grandparent and at least one parent) and skipped-generation households (at least one coresident grandparent and grandchild, no parent present; Dunifon et al. 2014). Thus, we also examine these two households separately.
Correlates
Following previous research (e.g., Kamo 2000), we examine the following indicators in the correlates analysis and the Oaxaca-Blinder decomposition: child’s age (<5, 6–11, 12–17), sex, and race/ethnicity (non-Hispanic black, non-Hispanic white, Hispanic, Asian, other race/ethnicity); mother’s relationship status (married, unmarried, divorced/widowed/separated), age (<18, 18–29, 30–39, 40–49, 50+), education (less than high school, high school, some college, college or higher), and labor force participation (employed, unemployed, not in labor force); whether the mother is an immigrant; and family income-to-needs ratio (using Census Bureau income-to-needs thresholds <101% of poverty, 101 to 200 %, 201 % to 300%, 301 % to 400 %, and 401 % or greater.4 We also include measures of whether the child lives in an urban area; the region of the country (Northeast, Midwest, South, West); whether the home is owned; and whether anyone in the household received Temporary Assistance for Needy Families (TANF), food stamps/Supplemental Nutrition Assistance Program (SNAP), Social Security (also including survivors benefits), Supplemental Security Income (SSI; including both children and adults), unemployment insurance, veteran’s payments (VA), child support, Special Supplemental Nutrition Program for Women, Infants, and Children (WIC), or public housing/housing assistance. Further, we include measures of the child’s health insurance (Medicaid, private health insurance, none) and whether the child received free or reduced-price school lunch/breakfast. In the online appendix (Table A1), we provide descriptive statistics on all the covariates for 1996 and 2009. Compared with 1996, in 2009, fewer children were white, and more mothers had college degrees, were over the age of 40, and were immigrants. Children in 2009 were more likely to live in a household receiving public assistance (especially Medicaid and school breakfast/lunch, although far fewer received TANF), but a higher share also lived in households above 401 % of poverty.
Method
To answer our first research question, regarding trends in household sharing among families with children over time, we provide weighted descriptive statistics. To answer our second research question, aimed at explaining the observed changes in household sharing—specifically, the increase in three-generation coresidence, which is the only shared living arrangement that increased over this period—we conduct three additional analyses. First, we examine which demographic groups have increased three-generation coresidence, using weighted descriptive statistics and calculating the percentage change by group. Second, we compare the correlates of three-generation coresidence in 1996 with those in 2009 to see whether different factors predict coresidence over time. To do this, we run logistic regression models (reporting odds ratios) in which we regress three-generation coresidence on the covariates detailed earlier separately for 1996 and for 2009.5 We run Chow tests to test for significant differences in beta coefficients across the two periods. Last, to examine which factors might explain the change in coresidence over time, we conduct Oaxaca-Blinder decomposition (the Fairlie extension for binary outcomes; Fairlie 2005) using the same covariates detailed earlier (those used in the correlates model). This statistical method decomposes how changes in various factors—say, the share of children living with single parents—might explain mean differences in household composition over time. (See Van Hook et al. (2004) for a detailed application of the Oaxaca-Blinder decomposition.) We report results using a pooled sample for 1996 and 2009 so that the results are not affected by the selection of comparison year (Neumark 1988; Oaxaca and Ransom 1994); however, extensions using a shift-share approach (for either year) were nearly identical to those presented here. We compute the decomposition using normalized effects (deviations from a grand mean), which allows us to calculate effects for every category (including the excluded category; for more details, see Jann 2008; Yun 2005a, b), but substantive findings were the same when we ran models excluding a reference category.
Results
What Are the Trends in Household Extension?
Figure 1 displays trends in children’s household extension between 1996 and 2009. Overall, the proportion of children living in shared households during this period increased by 3 percentage points (from 17.6 % to 20.8 %). This increase did not occur evenly across all types of shared households. The largest growth in household sharing took place among children living with grandparents (from 7.5 % to 10.5 %), whereas the share of children living with aunts/uncles (approximately 5 %), other relatives (7 %), and nonrelatives (5 %) remained relatively stable during these years.
To further explore trends in grandparent coresidence, Fig. 2 plots skipped-generation and three-generation coresidence over time. Although skipped-generation households increased somewhat over this period, we find that nearly the entire rise in household extension was attributable to a growth in three-generation households (5.7 % to 8.0 %).
To study whether the upward trend in grandparent coresidence continued beyond 2009, in Fig. 3, we use the ACS to examine trends in three-generation and skipped-generation coresidence through 2016.6 The share of children living in three-generation households continued to rise, from 8.1 % in 2009 (8.0 % in the SIPP) to 9.8 % in 2016. To better understand this trend, the remainder of our analyses focuses on three-generation households.
Who Is More Likely to Live in a Three-Generation Household?
To further unpack the rise in three-generation households, we examine whether particular demographic groups were more likely to live in such a household. In Table 1, we examine which demographic groups experienced the largest increase in three-generation coresidence (by row).7 In keeping with the fact that three-generation households increased over this period, we see that most demographic groups experienced an increase in coresidence, but a few trends stand out. For example, black and Asian children experienced the smallest percentage increase in coresidence, despite more generally having the highest rates of three-generation coresidence (Pilkauskas 2014). Hispanic and children of “other” race/ethnicity experienced the largest increase in coresidence. Mothers with high school or greater education experienced much larger percentage increases than those with less than a high school education. Interestingly, the share of young mothers (under 18) living in a three-generation household declined (as did the share of mothers older than 50), whereas the share of coresident mothers aged 40–49 in three-generation households nearly doubled. Children in married-parent households increased three-generation coresidence by 45 % (2 percentage points), whereas children in unmarried parent households increased by 15 % (3 percentage points). Households with income above poverty experienced greater percentage increases in three-generation coresidence. In terms of government assistance, households receiving school lunch, unemployment income, and housing assistance experienced the largest percentage increases, although change in percentage point terms was relatively small (2–3 percentage points).
Have Correlates of Three-Generation Coresidence Changed Over Time?
The rise in three-generation households may be explained in part by a change in the characteristics correlated with coresidence. In Table 2, we examine whether the correlates of three-generation coresidence changed over time. Concordant with prior research, we find that children living in families with greater economic need (income or social safety net participation), younger children, racial/ethnic minorities, and those with unmarried and younger mothers (especially teenage mothers) had higher odds of living in a three-generation household.8 We find that only three predictors changed over time: Chow tests suggest that being Asian and receiving SNAP/food stamps was less predictive of coresidence, whereas having an immigrant mother became more predictive.
What Factors Explain the Observed Changes in Three-Generation Households?
To examine factors that explain the increase in three-generation households, Table 3 displays the results from Oaxaca-Blinder decomposition detailing the percentage of the difference in three-generation households between 1996 and 2009 (8.38 % – 5.72 % = 2.66 %; because this analysis is restricted to children living with at least one parent, the rate in 2009 is slightly higher than that for all children) that can be attributed to observable differences in characteristics of households with children between these two time points.
Two factors stand out as major contributors to the increase in three-generation households: the change in mother’s relationship status, and the share of households receiving Social Security. (Note that a negative coefficient suggests that if compositional changes had not occurred over this period, the share of children in three-generation households would have been lower; in other words, this compositional shift contributes to the increase in coresidence). The change in mother’s relationship status accounts for approximately one-half a percentage point of the increase (sum of the individual beta coefficients for each union status) in three-generation family households (approximately 20 % of the total 2.66 percentage point change over time). Likewise, Social Security receipt accounts for a one-half percentage point increase in three-generation family households (19 % of the total percentage point change). Changes in the racial/ethnic composition of children also increased the share of three-generation households, as did the decline in the share of households with private health insurance.
Although most compositional changes suggest that three-generation households should have increased over this period, a few changes also worked against the observed increase. In particular, the increase in Medicaid receipt over this period worked to decrease three-generation coresidence by nearly one-half a percentage point (0.46): if Medicaid had not increased, the share of children living in three-generation households likely would have been even higher. Likewise, shifts in maternal age composition (toward an older age) reduce coresidence slightly (by 0.21 percentage points).
Discussion
In this study, we examine trends in shared households among children. We find that although the share of children living in a shared household increased between 1996 and 2009, the increase was driven by a rise in the share of children living in three-generation households. The share of children living in such households increased from 5.7 % in 1996 to 8.0 % in 2009. We may be concerned that the Great Recession drove the increase, but this study shows that the increase started well before the recession (before 2007) and continued beyond its end (2010), increasing to 9.8 % of children in 2016 (7.1 million children).
The correlates analyses show that few correlates changed over time and that three-generation coresidence was more common among lower-income households. Yet more economically advantaged groups experienced the largest percentage increases in three-generation coresidence: children in married households, with mothers who were more educated, and mothers who were older. Future research should examine why this might be the case.
The decomposition suggests that changes in mother’s relationship status represented the strongest explanatory factor in the increase in three-generation coresidence over time; however, an increase in the share of households receiving Social Security also explains a large share of the increase. Like reduced marriage and increased single parenthood, increased receipt of Social Security explained almost one-half a percentage point increase in three-generation coresidence.
The increase in three-generation coresidence associated with increased Social Security receipt may be related to the fact that individuals who receive Social Security are likely to be a source of economic stability (and compared with many other government assistance programs, Social Security payments are generally larger and more stable over time). Children and grandchildren may move in with grandparents receiving Social Security if they are more economically stable.
Although most public assistance programs explained little of the change over time, increased Medicaid receipt reduced three-generation coresidence by almost one-half a percentage point. Similar to Social Security but unlike most public assistance programs, which have smaller economic value and often have shorter recertification periods (e.g., an individual may lose eligibility month-to-month on food stamps), Medicaid provides families with children with a relatively significant in-kind income transfer, which may permit families to live independently and thus reduce coresidence. The difference in direction of the effect of Social Security and Medicaid may in part be driven by differences in who receives the assistance. If mothers or children receive Medicaid, the economic benefit may allow them to live independently. In comparison, Social Security is largely received by the grandparent generation; thus, these grandparents may be able to provide assistance to their children and grandchildren through coresidence.
Although our analysis includes a large set of characteristics likely related to three-generation coresidence, other unmeasured factors (such as preferences, health of household members, or housing markets) may also be important drivers of coresidence. Increased coresidence with grandparents may be explained in part by increased longevity (World Bank 2018), the greater share of individuals who are grandparents (Monte 2017) or stepgrandparents (Yahirun et al. 2018), or an increased length of grandparenthood (Margolis and Wright 2017)—factors we could not observe here. Future research that can examine these factors would be a useful next step. Nonetheless, these findings suggest that public policies likely shape the living arrangements of children.
Given the large increase in the share of children living in three-generation households over the last 20 years, more research should study the implications of these living arrangements. Prior research has linked coresidence with positive outcomes for older children (e.g., Deleire and Kalil 2002), but the evidence is mixed for younger children (e.g., Mollborn et al. 2011, 2012; Pilkauskas 2014). Understanding how policy and demographic changes influence coresidence and the implications for children will be important areas for future research.
Acknowledgments
The authors thank Liana Fox, Katherine Michelmore, and Mariana Amorim for their feedback and assistance.
Notes
Research has documented prevalence of particular types of shared living arrangements, such as extended-family households (children living with adult relatives; Kreider and Ellis 2011) or grandparent coresidence (e.g., Dunifon et al. 2014; Ellis and Simmons 2014), but has not examined trends comparing across types of shared living arrangements for children.
We do not use earlier waves of SIPP (pre-1996) or the 2014 SIPP because changes in sampling and questionnaire design make comparisons difficult and inaccurate.
Additional details on the ACS are available online (https://www.census.gov/programs-surveys/acs/methodology/design-and-methodology.html).
If a child is not living with the mother, we use information on the father.
We checked for high correlations between the variables. Where we found high correlations, we ran extensions excluding those variables. The findings were unchanged.
We do not use the ACS for our main analyses because we cannot distinguish as many household types, and the (nonpilot) data only start in 2005.
Percentage/percentage point inconsistencies in the table are due to rounding.
Mother’s education was not consistently related to coresidence, likely due to collinearity with income. In an extension, we exclude family income from the models and find that greater education is strongly and negatively associated with three-generation coresidence.
