“Excess” Doubling Up During COVID: Changes in Children's Shared Living Arrangements

Abstract

The proportion of U.S. children living in doubled-up households, in which a child lives with a parent plus adult kin or nonkin, has increased in the last 40 years. Although shared living arrangements are often understood as a strategy to cope with crises, no research to date has examined changes in children's living arrangements during the first year of the COVID-19 pandemic. We use the American Community Survey and the Current Population Survey to examine children's doubled-up living arrangements during 2020 and the extent to which children may have experienced “excess” doubling up relative to earlier years. We consider trends by household type (multigenerational, extended with other relatives, and nonrelative households) and differences by demographic characteristics (marital status, race and ethnicity, work status, education, age, and number of coresident children). We find evidence that more than half a million (509,600) children experienced “excess” doubling up in 2020. Greater than expected increases in doubled-up arrangements were driven by increases in multigenerational households, in particular among Black and Hispanic children, young children (under age six), those whose mothers never married, and those whose mothers were not working. Correlates of coresidence remained largely unchanged over time, although having a mother who had never married became a stronger correlate in 2020. Our findings suggest that both economic and instrumental needs likely explained the rise in multigenerational coresidence in 2020.

Introduction

In 2018, more than 15% of U.S. children lived in a “doubled-up” household, in which a child lives with additional adults beyond their parents, parent's partners, or siblings, such as grandparents, aunts or uncles, or nonrelatives (Harvey et al. 2021). The share of children living in doubled-up households has increased dramatically over the last 40 years, driven mostly by increases in a specific type of doubled-up household—“multigenerational” households, in which a child lives with at least one parent and grandparent (Pilkauskas et al. 2020; Pilkauskas and Cross 2018). Shared living arrangements shape the resources available to children, the instability experienced by children, and children's outcomes (Amorim 2019; Cross 2020; DeLeire and Kalil 2002; Dunifon and Kowaleski-Jones 2007; Harvey 2020; Mollborn et al. 2011; Perkins 2019; Pilkauskas 2014).

In early 2020, a global pandemic began, resulting in lockdowns where businesses, schools, and other nonessential operations were shut down. The share of Americans out of work reached levels not seen since the Great Depression (Wheelock 2020). School closings were extensive: almost 90% of public schools closed in April 2020, and about 40% of schools remained closed at the start of the following school year in August/September 2020 (Parolin and Lee 2021). As a result, many families, and especially lower income families, had to scramble to make ends meet and deal with school and childcare closures, while also trying to protect family members from COVID-19. Prior research shows that many young adults in the United States returned to their parents' homes during the lockdown (Fry et al. 2020), but little research has examined children's living arrangements during the first year of the COVID-19 pandemic. On the one hand, parents may have been more likely to double up to take advantage of economies of scale to make ends meet, benefit from shared resources like childcare, avoid more extreme housing hardships, or provide aid to older family members. On the other hand, parents may have been less likely to double up during the early days of the pandemic (when vaccines were not available) owing to the increased risk of transmission of COVID-19, or the increased household crowding when children and adults may have needed to work and do school from home.

In this study, we use the American Community Survey and the Current Population Survey to examine children's doubled-up living arrangements during 2020 and the extent to which children may have experienced “excess” doubling up relative to earlier years. We consider trends by doubled-up household type (multigenerational, extended with relatives other than grandparents, and nonrelative households) and differences by demographic characteristics (marital status, race and ethnicity, work status, education, age, and number of coresident children). Overall, we find evidence of greater than expected increases in multigenerational households in 2020, in particular among young children (under the age of six), Black and Hispanic children, those whose mothers never married, those whose mothers had lower levels of education, and those whose mothers were not working (either unemployed or out of labor force). By investigating the consequences of the COVID-19 pandemic on children's living arrangements, we shed light on the wide-ranging potential impacts of COVID-19 and the ensuing economic downturn, housing instability, and work precarity on children's well-being and outcomes.

Background

Trends in Doubled-Up Households

“Doubled-up” households are those in which a child lives with additional adults beyond their parents, parent's partners, or siblings (Harvey et al. 2021). A growing literature has documented trends in coresidence over time, correlates of coresidence, and the duration of doubled-up households (Harvey et al. 2021; Mykyta and Macartney 2012; Pilkauskas et al. 2014).¹ Despite widespread preference for residential independence in the United States (Harvey 2022; Pilkauskas and Michelmore 2019), the rates of doubling up among households with children have increased over the last couple decades and, today, nearly one in six children live in doubled-up households (Harvey et al. 2021). These arrangements often address crises such as eviction, unemployment, or other economic shocks (Desmond 2016; Wiemers 2014), offering families affordable housing, economies of scale, and child-rearing or health support during moments of need (Edin and Shaefer 2015; Seefeldt and Sandstrom 2015; Skobba and Goetz 2015).

Doubled-Up Arrangements During the COVID-19 Pandemic

Although there are many studies of the effects of the COVID-19 pandemic on an array of family outcomes, we know little about the living arrangements of children during the pandemic. To our knowledge, no research on the living arrangements of children in the United States during the pandemic has been published (although one research brief considered the implications of the pandemic on children's living arrangements but did not estimate the prevalence of such arrangements) (Vandenberg 2021). The key role that household structure plays in child development (Demo and Cox 2000), child well-being (Augustine and Raley 2013; DeLeire and Kalil 2002; Dunifon 2013), and parental behavior (Amorim 2019), however, underscores the need to document the consequences of COVID-19 for coresidential patterns.

Research has found that households with children are more likely to double up in response to economic recessions (Bitler and Hoynes 2013), suggesting that COVID-19 would increase instances of coresidence. As parents left the labor force (Wheelock 2020), lost access to school or daycare (Parolin and Lee 2021), or died as a result of COVID-19 (Kidman et al. 2021), shared arrangements may have become more common. Since grandparents are often the first line of support for parents in need (Reyes 2018; Swartz 2009), the pandemic may have had the greatest impact on one particular type of doubled-up household: multigenerational households, in which children live with at least one parent and one grandparent. Although the share of U.S. children living in multigenerational households has increased substantially since the 1990s, growing from about 5.5% in 1996 (Pilkauskas and Cross 2018) to 10% in 2018 (Pilkauskas et al. 2020), the pandemic may have led to a disproportionate rate of increase in the share of these households in 2020. Prior studies found that coresidence with grandparents often results from events that either increase parents' need for support (e.g., unexpected pregnancy, divorce, or economic hardships) (Harvey et al. 2021; Swartz 2009) or enhance grandparents' ability to support children and grandchildren (e.g., expansions of Social Security benefits) (Bitler and Hoynes 2013; Pilkauskas and Cross 2018). Multigenerational arrangements can also be motivated by grandparents' instrumental or health needs (Aquilino 2005; Choi 2003; Cohen and Casper 2002; Dunifon et al. 2014; Pilkauskas 2012). Older adults were also particularly vulnerable to COVID-19, which may have reduced their mobility and increased their need for assistance, potentially increasing multigenerational coresidence.

It is also possible, however, that the pandemic had a limited impact on rates of coresidence. Although the unemployment rate rose from 4.4% in March 2020 to 14.7% in April 2020 (Bureau of Labor Statistics 2022b), the COVID recession lasted only two quarters (Center on Budget and Policy Priorities (CBPP) 2022b). In fact, employment levels increased during the third and fourth quarters of 2020 (the unemployment rate fell to 6.7% by November) (Dalton et al. 2021). Additionally, economic recovery was hastened by unprecedented policies at the federal level (i.e., stimulus checks, eviction moratoriums, rental assistance, expanded unemployment insurance, the pandemic electronic benefits program, larger Supplemental Nutrition Assistance Program benefits). Thus, although there are many reasons to expect that doubling up may have increased during the first year of the COVID-19 pandemic, the short-lived nature of the COVID recession and the success of economic relief policies (CBPP 2022b) may have allowed families to retain residential independence.

Differences by Key Socioeconomic and Demographic Characteristics

Doubling up is driven by many factors, in particular, socioeconomic need, generational care needs, or preferences. Those with less education or lower socioeconomic status are more likely to double up (Cross 2018; Dunifon et al. 2014; Pilkauskas et al. 2014; Swartz 2009), although this was not the case historically (Pilkauskas et al. 2020). Doubling up is more common when children (and mothers) are young (Amorim et al. 2017; Pilkauskas 2012; Pilkauskas et al. 2014) and when caregiving needs may be especially high (Harvey and Dunifon 2023; Harvey et al. 2021). Unmarried mothers, who may need more assistance, are also more likely to double up (Dunifon et al. 2014; Harvey and Dunifon 2023; Harvey et al. 2021; Pilkauskas 2012; Pilkauskas and Cross 2018). Lastly, compared with non-Hispanic White families, doubling up is common among all other racial and ethnic groups (Amorim et al. 2017; Cross 2018; Kreider and Ellis 2011; Pilkauskas 2014; Pilkauskas et al. 2020; Pilkauskas et al. 2014). Racial and socioeconomic disparities in doubled-up arrangements stem, partly, from widening racial and socioeconomic differentials in mortality, fertility, and family formation patterns (Furstenberg 2014; McLanahan 2004).

Doubling up during COVID-19 may have been more common among socioeconomic and demographic groups who already had higher rates of shared residence prior to the pandemic. After all, these were also the groups who were most negatively impacted by COVID-19 in terms of employment and wages (Gambau et al. 2022). Women, mothers with the youngest children (under six), racial minorities, lower wage workers, and workers without a college degree were those most likely to experience job loss during the pandemic (Albanesi and Kim 2021; Alon et al. 2020; Barroso and Kochhar 2020; Bureau of Labor Statistics 2022a; Dalton et al. 2021). Balancing work and family life obligations may have also been particularly challenging for single mothers during the pandemic, leading to greater doubling up: 25% of single mothers left the labor force in 2020, compared with 18% of married mothers (Barroso and Kochhar 2020; Bauer 2021). Notably, whereas the pandemic exacerbated hardships most among families with children and Black and Latino communities (CBPP 2022a; Hardy and Logan 2020), federal policies were effective at stemming the economic impacts of the pandemic, and they seem to have had the largest positive effects on these groups (CBPP 2022b).

Doubling up can result in household crowding (i.e., settings that reduce personal space and privacy, often measured as having more than one occupant per room) (Ghosh et al. 2021). Previous studies found that both multigenerational coresidence and household crowding are associated with higher risks of COVID-19 disease transmission and deaths in the United States (Chen and Krieger 2021; Ghosh et al. 2021) and United Kingdom (Aldridge et al. 2021; Nafilyan et al. 2021). Importantly, not only the prevalence of doubled-up arrangements but also the consequences of crowding during COVID-19 may be stratified by socioeconomic and demographic characteristics, such as age, race, and ethnicity. Among kin living in shared arrangements, older family members such as grandparents are those most at risk of infection and mortality from COVID-19 (e.g., Dowd et al. 2020). For example, in a New York City study, researchers found that household crowding and exposure to COVID-19 were especially common among lower income, Black, and Hispanic households (Almagro et al. 2020). Examination of changes in children's living arrangements during COVID-19 sheds light on how families coped with a multifaceted crisis that increased instability, reduced resources, and exacerbated health risks.

Methods

Data and Sample

Our study uses the American Community Survey (ACS) and the Current Population Survey (CPS) for 2015–2020, both collected by the U.S. Census Bureau. We use data from the ACS in our main analyses (IPUMS USA; Ruggles et al. 2022). The ACS is a nationally representative survey of the U.S. population that samples approximately 3 million households annually. Our main analyses of prevalence of household types use the 2020 ACS, which is compared with ACS 1-year data collected from 2015 to 2019.² By comparing the 2020 data to five years of prior data (individually or averaged), we can better understand how trends diverged during the pandemic. To describe within-year trends of different household types, we use the monthly CPS, a nationally representative survey of the U.S. population, with a sample of about 60,000 households (IPUMS CPS; Flood et al. 2021). In both datasets, our analytic sample is composed of all children younger than 18 living outside of group quarters in the United States. For some analyses by demographic characteristics, we further restrict the sample to children younger than 18 with at least one parent present in the household (96%) because demographic information on nonresident parents is not available in either dataset.

The COVID-19 pandemic affected the Census Bureau's ability to collect data. For example, the CPS is normally conducted face-to-face, but interviews were done exclusively by phone between March and June 2020, when in-person interviewing partially resumed. CPS response rates hovered around 82% prior to the pandemic but ranged from 65% to 70% between March and August of 2020.³ The ACS similarly faced many data collection challenges that resulted in lower response rates, which fell from 86% in 2019 to 71% in 2020. Lower response rates were most pronounced among those living in group quarters (excluded from our study) and more disadvantaged groups.⁴ To correct for the lower response rates among certain populations, the Census Bureau created experimental survey weights using administrative, third-party, and other data sources. The experimental weights were found to correct for nonresponse bias in the 2020 sample when tested against standard benchmarks (e.g., income); the weights were also applied to the 2019 1-year ACS for evaluation on a standard sample and, finally, they reduced abnormal sample differences between 2019 and 2020 to normal levels.

Although experimental weights performed well,⁵ the Census Bureau suggests that the 2020 1-year ACS data be used with caution, as there were known data quality issues. Still, the lack of alternative, high-quality, population-level data means we must make use of these data if we are to understand the impacts of the pandemic on children's living arrangements. Similar to the ACS, the CPS response rates were much lower in 2020 compared with previous years, and attrition from the CPS was more pronounced among younger, lower educated, minority populations (Montenovo et al. 2021). Unlike in the ACS, weights to address the nonrandom attrition and low response rates in the CPS are not available; thus, it is unclear whether the CPS survey provides more or less accurate population estimates than the ACS with experimental weights. The CPS March Annual Social and Economic Supplement (ASEC) survey (for which the Census Bureau also developed experimental weights akin to the ACS) is not suitable for the purposes of this study because it collected 2020 data in March, before the pandemic was in full effect. We also considered alternate data sources, such as the Census Household Pulse survey and the Survey of Income and Program Participation, but the former did not collect detailed household roster information, did not begin until April 2020, and had a far lower response rate (less than 10%), and the latter has a somewhat different data structure and faced similar data quality issues as the ACS and CPS.⁶ Thus, the results from these analyses should be interpreted with caution; however, by using two different datasets (one that relies on experimental weights and another that does not) to demonstrate our results, we increase our confidence in the findings.

Measures

Household Types

In both the ACS and CPS, we follow previous work (e.g., Harvey et al. 2021; Pilkauskas et al. 2020) and identify household relationships by first using parent pointers that identify each resident child's mother or father (or mother or father figures)⁷ regardless of parents' marital status or relationship to the reference person. Parent pointers allow us to accurately identify different sets of parents for coresident children who do not share parents (i.e., cousins). We then rely on household rosters that identify other household members in relationship to the reference person (i.e., the person whose name is on the lease or mortgage). Children who live in a household with their parent(s) and any adult other than (1) their parent's married or cohabiting romantic partner or (2) their parent's adult children (the child's siblings) are considered doubled up.⁸ We do not consider stepparents or cohabiting partners to be doubled up, following many other studies (e.g., Mykyta and Macartney 2012). Adults in a doubled-up household may include children's aunts, uncles, cousins, grandparents, or nonkin (parents' friends or roommates). Other minors do not contribute to the doubled-up classification: for instance, a child who lives with their own mother and a minor cousin is not considered to be living in a doubled-up household.

We then categorize doubled-up children into three mutually exclusive household types: (1) multigenerational households, in which the child lives with at least one parent and one grandparent; (2) other extended households, in which the child lives with at least one parent and one non-grandparent adult extended family member (but no grandparent); and (3) nonkin households, in which the child lives with at least one parent and one nonrelative (but no other extended family members). Our measures are at the child level, not the household level. Thus, within the same household, some children may be categorized as being in a doubled-up household and others might not, or children within the same household may be identified as belonging to different types of doubled-up households (e.g., one child may be in a “multigenerational household” whereas another may be in an “other extended household”).⁹ We also privilege multigenerational households followed by extended households and, finally, nonkin households; thus, if a child lives with both a grandparent and an aunt, the child would be categorized as living in a multigenerational household (following previous research) (Harvey et al. 2021).

There are a few differences between the ACS and the CPS that lead to some differences in the estimates of multigenerational coresidence (and of all doubled-up arrangements). In the ACS it is possible to identify in-laws, whereas in the CPS it is not; thus, we are able to identify more multigenerational households in the ACS. Additionally, some differences in estimates may arise because the CPS and ACS definitions of who resides in the household differ slightly. The CPS asks respondents to include people who “usually” live in the household. In comparison, the ACS asks respondents to include anyone living in the household for two or more months, or anyone who has been in the household for less than two months, but has nowhere else to stay. These differences suggest that the share of children coresiding in doubled-up arrangements will be higher in the ACS than in the CPS (e.g., Michelmore and Pilkauskas 2022).

Demographic Characteristics

We measure both parents' and children's demographic characteristics. For parents' characteristics, we use information primarily from the mother, relying on data from fathers only when mothers are not present in the household. We refer to these as “maternal characteristics” throughout, because more than 95% of children in both the ACS and CPS are living with their mothers. We use demographic characteristics of both mothers and children to describe stratification in the observed trends in coresidence. For children, we consider: (1) children's race and ethnicity (non-Hispanic White, non-Hispanic Black, Asian, non-Hispanic other race or ethnicity, Hispanic), (2) children's age (0–5, 6–12, 13–17), and (3) the number of children younger than 18 in the household (one, two, or three or more).¹⁰ For maternal characteristics, we examine: (1) maternal education (less than high school, high school, some college, or bachelor's degree or higher), (2) maternal marital status (married, cohabiting, previously married [divorced, separated, or widowed], or never married), and (3) maternal employment status (working, not working).

Table 1 describes the characteristics of our ACS and CPS samples (2015–2020 pooled). The sample characteristics across the two datasets are quite similar. On average, children are 8.7 years old, and about half identified as White, one quarter as Hispanic, 13% as Black, and about 5% as Asian. Approximately 70% of children's mothers are married, two thirds are employed, and more than one third have a college education. In online appendix Table A1, we show the same sample characteristics contrasting the two samples and comparing pre-COVID (2015–2019 pooled) to the 2020 data. We find that changes in the sample composition between 2015–2019 and 2020 are small in both the ACS and the CPS, and that the characteristics of the 2020 samples are similar, despite challenges faced in the 2020 data collections and different approaches to weighting the data.

Analytic Strategy

First, we conduct descriptive analyses using the ACS akin to those used to identify excess mortality (Beaney et al. 2020; Rossen et al. 2022; Stang et al. 2020). To calculate “excess” doubling up, we compare estimates of the prevalence of doubled-up household types in 2020 to a counterfactual no-COVID scenario created by forecasting the 2015–2019 pre-pandemic annual trends linearly.¹¹ The “excess” doubling up during the COVID-19 pandemic is calculated as the difference between the share of doubled-up households during 2020 and the expected share in the absence of COVID.¹² Essentially we identify the share of doubled-up arrangements above and beyond what would be expected in the absence of the COVID pandemic. To provide additional evidence that the “excess” doubling up is not simply a result of noise, we then look at trends by quarter (January–March, April–June, July–September, October–December) in the 2020 CPS compared with average within-year trends for 2015–2019 (we pool across these five years to reduce noise; in online appendix Figure A1, we show this trend for each year separately). In these analyses, we expect that within-year trends in doubled-up arrangements will differ only after the first quarter, since lockdowns started in March 2020.

Second, we consider whether changes in children's living arrangements during the pandemic were distributed equally across key demographic groups defined by children's race or ethnicity, children's age, number of children in the household, maternal education, marital status, and employment. We take the same approach to examine “excess” doubling up as described above but within each demographic group. In these analyses, we focus on multigenerational households not only because they are the most common type of doubled-up arrangement, but also because we find them to be the key drivers of “excess” doubling up during the pandemic (as shown later).

Finally, we conduct a series of multivariate regressions to examine whether the correlates of coresidence changed between 2015–2019 and 2020. This analysis also allows us to consider whether differences in the trends by demographic characteristics remain once we control for all other covariates. In these analyses, we use linear probability models, which allow us to directly interpret the point estimates as percentage-point changes. However, in supplemental analyses (available upon request), we ran logit models and found substantively similar results. To test whether the differences in correlates were significantly different between 2015–2019 and 2020, we ran tests of significance (using seemingly unrelated estimation). Standard errors are clustered at the household level (results remain unchanged if standard errors are not clustered or if we run models that randomly select one child per household). All descriptive and multivariate analyses are weighted using ACS and CPS individual weights.

Results

“Excess” Doubling Up During COVID-19

We begin by showing the change in the share of children living in different household types, comparing 2015–2019 to 2020 in Table 2 for both the ACS and CPS. First, we find that the share of children who are doubled-up is greater in the ACS than in the CPS. This is consistent with findings from other studies (Michelmore and Pilkauskas 2022) and likely arises from small differences in data collection approaches, household rosters, and question wording (described earlier). We find that there was a small increase in doubling up in 2020 in both datasets, although the increase is more pronounced in the ACS (5.8%) than in the CPS (2.3%) and driven largely by an increase in multigenerational households (10.2% vs. 3.9%). The ACS and CPS differ in the trends documented for other extended and nonkin households: whereas the ACS suggests that the share of other extended households decreased slightly and that the share of nonkin households remained constant in 2020 compared with the previous years, the CPS suggests that the share of other extended households increased slightly and that the share of nonkin households decreased in 2020 compared with previous years. Notably, the overall share of other extended households and nonkin households in both datasets is small.

Although Table 2 reports average changes, it tells us little about whether changes diverge from a prior trend or simply reflect a continuation of annual trends in doubled-up households, which have been documented elsewhere (e.g., Harvey et al. 2021). To shed light on whether the increase in the share of children living in any type of doubled-up arrangement was disproportionate to what would be expected in the absence of the COVID-19 pandemic, in Figure 1 we plot annual estimates over time and include an estimate from a linear extrapolation of the 2015–2019 trends. In doing so, we find that doubling-up appears to have increased at a higher rate during 2020 than would have been expected based on prior trends. Figure 1 shows that the share of children in doubled-up households increased by about 0.4 percentage points (hereafter, “pp”) between 2015 and 2019 (an average annual increase of 0.1 pp). Between 2019 and 2020, however, the share of children in doubled-up households increased by 0.6 pp (from 15.7% in 2019 to 16.3% in 2020), an increase six times greater than the average annual increase in the previous years. A comparison between the expected share of children to be living in doubled-up households in 2020 (15.6%, denoted by the dashed line) and the actual share of children who lived in doubled-up households in 2020 (16.3%) suggests an “excess” share of 0.7 pp of children entered this arrangement during the COVID pandemic. This represents more than half a million children, based on a 2020 population of 72.8 million children younger than 18.

Looking at the other three panels of Figure 1, the disproportionate increase in the share of children in doubled-up arrangements is driven by a large increase in the likelihood of being in a multigenerational household. Whereas the share of children in multigenerational arrangements increased by 0.5 pp between 2015 and 2019 (0.12 pp per year), it increased by 0.7 pp from 2019 to 2020 (a nearly sixfold increase compared with previous years). Our linear prediction (dashed line) suggests that 10.2% of children would be living in multigenerational households in the absence of COVID-19, but 10.8% of children actually did so during the 2020 year of the pandemic. This represents an “excess” of about 460,000 children experiencing multigenerational arrangements during the pandemic.

The share of children in nonkin households fluctuates between 2.1% and 2.3% in the observed period, without following a linear trend. The share of children in other extended households declined slightly from 3.5% to 3.2% between 2015 and 2019 (− 0.075 pp per year). In 2020, the share of children living with other extended kin increased slightly to 3.3%. A comparison of the share of children living in other extended kin living arrangements in 2020 with our projected expected share (dashed line) suggests a small “excess” share of children in these arrangements during the COVID pandemic (0.15 pp). Notably, the projected downward linear trend is driven mostly by small declines in the share of other extended kin households between 2015 and 2017. In fact, between 2017 and 2020, the share of children in these arrangements remains virtually constant, at 3.3%, similar to the rate in 2020. Overall, trends of “other extended households” and “nonkin households” impact a smaller share of children, are noisier, and are less linear. Altogether, this hinders our ability to interpret deviations from projected linear trends as a consequence of the COVID-19 pandemic.¹³

Figure 2 shows the share of children living in each doubled-up arrangement by quarter using data from the CPS. Trends in 2015–2019 (dashed lines) suggest that the prevalence of doubled-up households is greater in the first and fourth quarters of the year compared with the second and third quarters. We are not aware of any prior research that has investigated seasonal variation in doubled-up arrangements; thus, this finding was a surprise, but in online appendix Figure A1 we show that this trend generally holds when we look at each of the 2015–2019 years separately. Although explaining the seasonality of living arrangements is beyond the scope of this study, we conjecture that this pattern may result from seasonal variation in births (e.g., Darrow et al. 2009), divorces (e.g., Brines and Serafini 2016), homeless shelter usage (Colburn 2017), and public assistance receipt (e.g., Pilkauskas and Michelmore 2019).

The prevalence of all doubled-up households in the first quarter of 2020 (January–March) is substantively similar to and statistically indistinguishable from that of prior years. We find, however, significantly higher rates of doubled-up arrangements during the second and third quarters (April–September) of 2020 compared with prior years. First, this finding provides suggestive evidence that the beginning of the pandemic lockdowns may have contributed to a unique seasonal pattern in children's living arrangements in 2020.¹⁴ Second, results from the ACS (Figure 1) and from the CPS (Figure 2) corroborate each other, suggesting “excess” doubling up during COVID was driven primarily by increases in multigenerational coresidence. Notably, Figure 2 shows the same upward trend during the second and third quarters of the year for both multigenerational households and other extended households. Levels of nonkin arrangements, however, follow a trend similar to that of previous years (i.e., declines in prevalence during the second and third quarters), but they are substantively and significantly lower in 2020 compared with prior years. Corroborating Figure 1, these latter results indicate that the pandemic did not promote “excess” nonkin arrangements.

Finally, Figure 2 also shows that by the fourth quarter (October–December), rates of shared coresidence are similar to those in earlier years. This finding suggests that families with children were more likely to double up during the first six months of the pandemic. To further interrogate this finding, we examined the share of children in each type of doubled-up household using the 2021 ACS data (see online appendix Figure A2). Once again, we find that the impacts of the pandemic on shared arrangements were short-lived: by 2021, shares of doubling up had returned to pre-pandemic levels (15.8%).

The Stratification of Multigenerational Coresidence During COVID-19

Children were not equally likely to enter doubled-up arrangements before the pandemic. The pandemic also did not impact all children equally. In this section, we examine which children were most likely to enter doubled-up arrangements in 2020. We focus on multigenerational households since this arrangement seems to be the main driver in overall trends in doubled-up arrangements both before and during the pandemic.

Figure 3 depicts trends by child characteristics: race and ethnicity, age, and number of children in the household. In the first panel, the solid colored lines represent the observed share of children in multigenerational households in each year and the dashed lines near each colored line represent the predicted share of children in each racial or ethnic group expected to live in multigenerational arrangements in 2020 in the absence of COVID. The dark blue line indicates that the share of White children living in multigenerational households grew slightly, from 6.6% in 2015 to 7.1% in 2020, and is lower than the share of Black (green), Hispanic (orange), and Asian (light blue) children in these arrangements in all years. The dashed line predicts the share of White children expected to live in multigenerational arrangements in 2020 (7.2%) in the absence of COVID. A comparison between the actual (7.1%) and expected (7.2%) shares of White children in multigenerational arrangements suggests that the pandemic had no substantive or significant impact on the formation of multigenerational households among White children. Asian children were the most likely to live in multigenerational arrangement in all years (16.1% in 2015 and 17.5% in 2020). Yet, like White children, comparing the actual (light blue line) and expected (dashed line) trends in multigenerational arrangements among Asian children suggests that the pandemic also had no substantive or significant impact on the formation of multigenerational households among this group.

Black and Hispanic children, on the other hand, experienced high levels of multigenerational coresidence pre-pandemic and “excess” multigenerational coresidence. From 2015 to 2019, the share of Black and Hispanic children in multigenerational households grew slightly: from about 12.6% to about 12.9%. Our overlapping linear prediction (dashed lines) suggests that similar percentages (12.9%) of Black and Hispanic children were expected to live in multigenerational arrangements in 2020. However, the share of Black and Hispanic children in these arrangements increased by more than 10% during 2020, to 14.9% and 14.4%, respectively. Overall, this represents a significant “excess” of about 275,000 Hispanic children (from a baseline of 18.9 million) and more than 200,000 Black children (from a baseline of 10.2 million) living in multigenerational arrangements as a result of the pandemic.

In the second panel of Figure 3, we show trends by child's age. Here we find that multigenerational households significantly increased across all three age groups, but that “excess” coresidence was largest for children younger than 6 (an additional 1.1 pp compared with 0.3 pp among children 6–12 and 0.5 pp among children 13–17). The disproportionate increase in multigenerational households among families with very young children (red line) is especially pronounced because the predicted trend shows that multigenerational coresidence among children ages 0–5 had been downward trending before the pandemic. The proportion of such children living in a multigenerational household was expected to decrease by 0.2 pp on the basis of previous linear trends, but instead it increased by 0.9 pp (from 12.9% to 13.8%).

In the third panel of Figure 3, we show trends by the number of children in the household: although multigenerational households are far more common when there is only one child in the household, their shares increased significantly across households with different numbers of children (and relatively more among children in households with three or more children). The shares of children living in families with one, two, and three or more children in multigenerational households were projected to be, respectively, 12.2%, 9.6%, and 9.6% in 2020. Instead, we find that 12.7%, 10.1%, and 10.5% of children in such families lived in multigenerational arrangements during the pandemic. This represents an “excess” coresidence of about 0.5 pp among children in families with two or fewer children and of about 1 pp among children in families with three or more children.

Figure 4 presents trends by mothers' educational attainment, work status, and marital status. Results suggest that children at greatest risk of experiencing multigenerational coresidence before the pandemic were also the ones who experienced disproportionate increases in grandparental coresidence during the pandemic. For example, although “excess” multigenerational coresidence was experienced by children regardless of mothers' educational attainment, children of the least educated mothers were most at risk of experiencing “excess” multigenerational coresidence during COVID. Among children whose mothers had a high school diploma or less, 14.4% were expected to live with grandparents in 2020 on the basis of 2015–2019 trends (dashed line), but a significantly higher share, 15.9%, did so in 2020 (representing a 1.5-pp greater increase than anticipated).¹⁵

When comparing trends among children whose mothers were working or not, we find that, although both groups experienced “excess” coresidence in 2020, children whose mothers were not working (i.e., out of the labor force or unemployed) experienced a disproportionate increase in the likelihood of being in multigenerational arrangements compared with children whose mothers worked.¹⁶ Whereas we expected that 10.9% of children whose mothers were not working would live with grandparents in 2020 in the absence of COVID (dashed line), the share of children with nonworking mothers in these arrangements was actually 12.3% (1.5 pp higher) in 2020. Additional analyses looking at the work status of all resident parents find that increases in the share of multigenerational households are driven by households in which all resident parents are not working (see online appendix Figure A5).

Finally, the COVID-19 pandemic had no impact on the share of children in multigenerational households with previously married or partnered mothers (married or cohabiting). However, we see a significantly disproportionate increase in multigenerational coresidence among children with never-married mothers in 2020. Whereas our linear prediction (dashed line) suggests that 30.3% of such children were expected to live in multigenerational arrangements in 2020 in the absence of COVID, 35% did so (5 pp higher than anticipated).

Changes in the Correlates of Multigenerational Coresidence

So far, our analyses have described shifts in coresidential arrangements over time overall and by demographic characteristics. Table 3 presents multivariate analyses examining whether the key correlates of multigenerational coresidence changed between 2015–2019 and 2020. This analysis allows us to examine (1) whether key correlates are more or less salient predictors of coresidence in 2020 compared with 2015–2019 and (2) whether the descriptive differences observed in the trends in Figure 3 remain after controlling for other characteristics.

Consistent with other research (e.g., Pilkauskas et al. 2020), these findings show that younger children and children whose mothers never married, lived with only one child, or had lower educational attainment are more likely to live in multigenerational households. We also find that race and ethnicity are significant correlates of coresidence, with White children being the least likely to live in a multigenerational household compared with Asian and Hispanic children or children of “other” race or ethnicity. Net of controls, Black children are less likely to live in multigenerational arrangements than White children both before and after COVID—a result that suggests the salience of socioeconomic resources and family structure for Black–White differences delineated above.

With regard to change over time, the results in Table 3 show that net of controls, most correlates of multigenerational coresidence did not change in meaningful ways in 2020 from prior years, with one exception—marital status. We find that marital status is a stronger correlate (in particular, being a child with a never-married mother) of multigenerational coresidence in 2020 than in the prior years.¹⁷

Discussion

The share of children in doubled-up arrangements has increased steadily over the past 40 years. Such shared living arrangements shape both the availability of resources and the outcomes of children (Amorim 2019; Cross 2020; DeLeire and Kalil 2002; Dunifon and Kowaleski-Jones 2007; Harvey 2020; Mollborn et al. 2011; Perkins 2019; Pilkauskas 2014). The expected impacts of the COVID-19 pandemic on doubled-up arrangements were unclear as the pandemic may have increased the need to double up due to greater economic stress, school closures, poorer health, or job losses. At the same time, the pandemic may have decreased shared coresidence in an attempt to decrease the risk of transmitting COVID-19. The responsiveness of the public safety net during the pandemic may have also reduced families' need to coreside. The push and pull factors shaping coresidential decisions during 2020 likely varied depending on the number of coresident children and their age and race or ethnicity, as well as parents' education, marital status, and work status.

In our descriptive analyses, we find that rates of doubling up increased sixfold in 2020 compared with what would have been predicted based on 2015–2019 annual trends. We estimate an “excess” of more than half a million children in doubled-up arrangements in 2020 due to COVID-19. This increase is driven mostly by significant rises in coresidence with grandparents (i.e., multigenerational arrangements) and, to a lesser extent, with other kin. We estimate an “excess” of about 460,000 children in multigenerational arrangements in 2020. These results hold across datasets and projection specifications and are in line with prior research, which suggests that grandparents are a key informal safety net for parents and children in moments of need (e.g., Swartz 2009). We also find, however, that the impacts of the pandemic on shared coresidence were short-lived. Specifically, we find that families were more likely to double up during the first six months of the pandemic (the second and third quarters of the year; see Figure 2) and that by 2021, rates of doubling up had returned to pre-pandemic levels (see online appendix Figure A2).

To understand the heterogeneous impact of the pandemic on children's living arrangements, we investigated descriptive trends in multigenerational coresidence by key demographic characteristics. We find that the children at greatest risk of experiencing multigenerational coresidence before the pandemic were also the ones who experienced disproportionate increases in grandparental coresidence during the pandemic: younger children, non-White children, children with less educated mothers, and those with never-married mothers. These findings suggest that families with fewer socioeconomic resources may have needed more economic or instrumental assistance during the first year of the pandemic.

Did economic or instrumental needs drive this increase in multigenerational coresidence? Our analyses by demographic characteristics suggest the rise in coresidence was likely driven by both economic and caregiving needs. First, although working parents experienced challenges balancing work and childcare responsibilities during 2020, we find that children of nonworking mothers, who were at greater risk of experiencing multigenerational coresidence before the pandemic, were also more likely to move into multigenerational arrangements during 2020. Although children with both working and nonworking mothers were more likely to live in a multigenerational household in 2020, the far larger increase among children with nonworking mothers points to the notion that economic needs were a particularly important driver of the increase in coresidence. In an additional analysis (see online appendix Figure A5), we considered the work status of all parents in the household and similarly found that the coresidence was disproportionally driven by households in which all parent figures were unemployed or out of the labor force, further corroborating the notion of the importance of economic need in motivating coresidence.

Second, we also find larger increases in multigenerational households among children with mothers with the lowest levels of education (high school or less) and among those whose mothers were never married—that is, both representing groups of mothers who typically have fewer economic resources. Never-married mothers may also be less able to rely on the child's other parent for economic resources through formalized support arrangements. Together, these findings also support the notion that economic need may drive the increase in multigenerational households. Yet, it may also be the case that never-married mothers also have a greater need for instrumental assistance that leads them to coreside in greater numbers, especially if nonresident parents are less likely to engage in childcare. That we also observe a greater than expected increase in coresidence among households with the youngest children (under age six) further supports the salience of childcare needs. Thus, descriptive evidence suggests that both caregiving and economic needs drove the increase in multigenerational households during 2020.

Our multivariate analyses indicate that correlates of multigenerational coresidence were similar in the pre-pandemic period as in 2020. The only variable that became a stronger correlate of multigenerational coresidence in 2020, as compared with earlier years, was mother's marital status—in particular, the gap between never-married mothers and other mothers became more salient in 2020 net of controls. The fact that we do not observe other factors (say, employment status, low level of education, or child's age) as more highly correlated with coresidence in 2020 despite observing descriptive differences is likely because these factors are themselves highly correlated. As noted earlier, that being a never-married mother became an even stronger correlate may reflect both the economic and instrumental needs faced by these mothers, who may have been especially unlikely to get assistance from the second parent owing to COVID-19 restrictions or challenges.

Taken together, the descriptive and multivariate results suggest that overall levels of coresidence increased during 2020 among groups that were already at greater risk of grandparental coresidence before COVID. Levels of multigenerational coresidence increased disproportionally among groups who were also most vulnerable to the pandemic in terms of disease and mortality, such as Black and Hispanic families (Almagro et al. 2020). Despite the potential dangers of crowding during a pandemic (Chen and Krieger 2021; Ghosh et al. 2021), this finding suggests that family needs (for childcare, shared resources, housing, or other economic or instrumental support) were a more salient determinant of living arrangements than avoiding potential COVID-19 household exposures.

This study has important limitations, in particular the data quality, and thus results should be interpreted with caution. The pandemic made data collection challenging, and both the ACS and the CPS faced issues in collecting data early in the pandemic that resulted in lower response rates among non-White groups, those with lower levels of education, and those with lower incomes, all groups that have high rates of doubling up in non-pandemic times. Although the ACS estimates were corrected for nonresponse bias with experimental weights, the groups who were least likely to respond are those who are most likely to coreside, suggesting that our estimates may be biased downward. Despite data quality issues, that results hold across two different datasets helps increase our confidence in the findings. Although we considered alternative data sources on children's living arrangements, most data collection efforts were hindered by the pandemic. Another important limitation is that neither the ACS nor the CPS allows us to account for the mortality of parents or other kin. Mortality of family members may have both motivated doubled-up arrangements or reduced the availability of kin to partake in doubled-up arrangements in ways that are hard to predict.

Despite these limitations, our study allows scholars and policymakers to identify the children most likely to enter doubled-up arrangements during periods of economic instability (see also Bitler and Hoynes 2013). It also sheds light on the growing share of families with children in doubled-up arrangements, who may face greater barriers to accessing services or to economically recovering from the pandemic. Future studies should investigate the role of pandemic-related policies (e.g., restrictive shutdowns) on living arrangements. In addition, although we could not test the effects of coresidence on COVID-19 mortality, understanding if family responses to the pandemic affected mortality is an important area for future research (e.g., Ghosh et al. 2021).

Acknowledgment

The authors thank Liana Fox for her thoughtful advice.

Notes

References