Three-Generation Households in the U.S.: The First Exit After a Child’s Birth

Abstract

Despite the increased prevalence of three-generation households in the U.S., there is still a limited understanding of the dynamics of dissolution of these households. Using event-history analysis and data drawn from the first five waves of the Fragile Families and Child Wellbeing Study and its restricted contextual data, this study describes how long mothers and their newborns spend in three-generation households after the child’s birth and whether this duration differs by race and ethnicity. Additionally, it examines the correlates of the first transition out and whether they differ by race and ethnicity. Results indicate that children and their mothers are more likely to experience the first exit between birth and age 1, and a formal test does not find significant differences by race and ethnicity. Several non-economic factors are positively associated with the exit out of three-generation households including a mother’s marital status and having a new partner; factors with a negative association include the number of children in the household and whether the focal child is a mother’s first kid. Finally, TANF or SNAP receipt and mothers’ employment are both positively associated with the first exit out of three-generation households. There are not significant differences in correlates of first exit by race and ethnicity. Implications and directions for future research are discussed.

1. Introduction

As families in the U.S. have become more diverse, researchers have been more focused on the potential consequences of family structure and transitions on children’s well-being. Disadvantaged families more commonly experience family changes (Cherlin, 2009), and they have been proven to increase the parents’ stress and create challenges for children’s development and well-being (Mitchell et al., 2015). Recently, this literature has been expanded to include an examination of the role that household structure plays on children’s well-being (in addition to the role of the nuclear family structure). Particularly, children’s living arrangements in the U.S. are now more likely to include other related adults in the household (called “extended-family households”), making this living arrangement even more relevant to study. Of extended-family households, the most common arrangement is the three-generation household in which the child, one or both of his or her parents, and at least one grandparent co-reside (Cross, 2018; Kreider & Ellis, 2011), with recent estimates indicating that at least 25% of all children will experience this living arrangement within their life (Amorim et al., 2017).

A growing body of research focuses on the dynamics of the formation and dissolution of three-generation households. Previous findings indicate that these households are more common early on in children’s lives, particularly between their birth and first birthday (Amorim et al., 2017; Cross, 2018; Pilkauskas, 2012), that they are usually short-lived (Pilkauskas, 2012), and that they are likely to occur multiple times during their childhood with transitions in and out being common (e.g., Harvey, 2020a; Mollborn et al., 2012). When looking at the prevalence and dynamics among different populations, findings indicate that three-generation households are more common among low-income populations, young mothers, single mothers, mothers who have young children and lower levels of education and earnings (Pilkauskas & Cross, 2018; Pilkauskas, 2012), and among racial and ethnic minorities (Amorim et al., 2017; Cross, 2018).

Despite the well-documented prevalence of three-generation households, the duration of this living arrangement, the correlates of its formation, and the significant differences between races and ethnicities, little is known about the correlates of continuity and change for these households ¹ and whether these correlates vary by race and ethnicity. Using the first five waves of the Fragile Families and Child Wellbeing Study (FFCWS), I aim to identify the correlates of the first exit from three-generation households after a child’s birth for a sample of custodial mothers and their focal child. In this study, I investigate whether the correlates of first exit differ by race and ethnicity. By using event history models and standard descriptive survival analysis, I also document the time to first exit out of three-generation households, and whether it differs by race and ethnicity.

A better understanding of children’s living arrangements beyond the nuclear family is relevant given increasing evidence of the association between these arrangements and child outcomes (e.g., Amorim, 2019; Dunifon, 2013; Mollborn et al., 2012; Perkins, 2017). Scholars have started to document the potential effects that three-generation households and their transitions might have on children and adolescent’s wellbeing, with previous studies identifying both positive and negative effects (Amorim, 2019; Deleire & Kalil, 2002; Dunifon, 2013; Dunifon & Kowaleski-Jones, 2007; Dunifon & Ziol-Guest, 2014; Harvey, 2020a; Mollborn et al., 2011; Pittman & Boswell, 2008) and differential effects based on racial and ethnic affiliations.

Taking these findings into account, it becomes relevant to better understand the dynamics of three-generation households, including the correlates of their dissolution among fragile families and racial and ethnic minorities, given that these groups are the ones more likely to experience this arrangement and the transitions out of it. The exit of a grandparent (either because the mother and the child move out or because the grandparent(s) leave(s) the household) could have a positive or negative impact on children’s lives. On the one hand, grandparents often provide substantial resources and time to children (i.e., Amorim, 2019; Kalil et al., 2014), and having them removed from co-residence removes a day-to-day resource which is likely to negatively influence child wellbeing. On the contrary, the presence of a grandparent in the household might add strain to already tight economic resources and increase stress and difficulties in parenting (i.e., Harvey, 2020b); thus, having them removed from co-residence, might have a positive influence on child wellbeing. Either way, understanding more about the correlates of exit is likely to be important for a more thorough understanding of child wellbeing and its trajectories.

This study aims to address previous research gaps by examining the correlates of the first exit out of three-generation households and whether they vary by race and ethnicity. Although the use of a birth cohort in the study of three-generation households is not new, this study aims to use a birth-cohort study together with event history analyses to better understand the influence of time constant and time varying variables in the transitions out of three-generation households after a child’s birth. I focus particularly on three-generation households given their higher prevalence among other extended-living arrangements and that households with grandparents versus other related kin might operate under different mechanisms and dynamics (Glick & Van Hook, 2011; Mollborn et al., 2012).

2. Literature review

Three-generation households have been historically portrayed as a response to social and economic need, as a safety net for the most disadvantaged, and as a common arrangement among racial and ethnic minorities in the U.S. (Cross, 2018; Mollborn et al., 2012; Perkins, 2017; Pilkauskas & Cross, 2018). Although this living arrangement is not new, recent studies indicate that the number of children living with a grandparent in their households has increased and will likely continue to do so given demographic changes, such as the increased availability of kin and the racial and ethnic composition of the U.S. (Cross, 2018), as well as the changing formal safety net in place, particularly for single mothers and their children (Moffitt, 2015). Below, I provide a portrait of the prevalence of this arrangement, followed by a review of the factors that have been associated with three-generation co-residence and a review of the existent studies looking at transitions, and, particularly, at exits.

In the past few years, extended-family households have been on the rise in the U.S. In examining their increase, Pilkauskas and Cross (2018) found that three-generation households are accountable for such an increase. Using the 1996–2008 Survey of Income and Program Participation (SIPP) and the 2009–2016 American Community Survey, estimates indicate that in 1996, 5.7% of all children in the U.S. were living in a three-generation household while in 2016 this figure increased to 9.8%. In the context of children’s living arrangements, three-generation households are more common during early childhood, and the likelihood of entering one decreases as a child ages (Cross, 2018), likely reflecting higher childcare needs at the time of the child’s birth and soon after (Amorim et al., 2017; Cross, 2018; Pilkauskas, 2012). Finally, cumulative probabilities of co-residence during childhood indicate that nearly 25% of children will co-reside with their grandparents (Amorim et al., 2017). In a recent study, Harvey (2020a) estimates that children who live in a multigenerational household between ages 1 through 17 spend, on average, 4.25 years in these arrangements.

2.1. Correlates of three-generation co-residence

Literature around the characteristics and correlates of three-generation co-residence suggests at least three groups of factors: non-economic (culture and generational needs), economic, and policy influences (Cross, 2018; Pilkauskas, 2012). Below, I summarize findings related to these groups of factors, followed by a discussion on whether these correlates differ by race and ethnicity; I particularly focus on non-economic factors.

2.1.1. Non-economic factors

This cluster includes generational needs (Pilkauskas, 2012), cultural norms and preferences, and other demographic characteristics. First, generational needs as Pilkauskas (2012) suggests, include the needs of one or several of the family members involved. Using the Fragile Families and Child Wellbeing Study, Pilkauskas (2012) finds that the needs of the parent instead of the grandparent are more strongly associated with the formation of three-generation households. Help in three-generation households has a downward flow, moving from grandparents to parents (Bianchi et al., 2006). In this context, mothers who experience teen pregnancy, young mothers, new mothers, or mothers who have a child with particular needs are all more likely to share their household with their parents (Baker & Mutchler, 2010; Trent & Harlan, 1994). Other generational needs include the experience of other life events, such as changes in one’s marital or relationship status, immigration, incarceration, and other circumstances in which individuals might activate their familial network (Baker & Mutchler, 2010; Casper & Bryson, 1998; Cohen & Casper, 2002; Glick & Van Hook, 2011; Kamo, 2000; Keene & Batson, 2010; Pilkauskas, 2012; Turney, 2014).

In regard to family transitions, a few studies have raised awareness about the potential effect of the increasing instability of families in the U.S. These changes have likely influenced the number of three-generation families (Pilkauskas & Cross, 2018) as family members who experience, for example, divorce and separation, seek out help to cope with their family transition (Perkins, 2017). In this context, Mollborn and colleagues (2012) discussed the potential of a “push-pull” effect between marital and relationship status changes with changes in household structure. As a new partner enters the life of the individual living in an extended-family household, the likelihood of the new couple leaving the extended-family household increases. In contrast, experiences of dissolution and the loss of a partner increases the likelihood of extended-family household formation.

Cultural norms and preferences have also played a role in the prevalence of three-generation households. Individuals with the cultural value of familism might be more likely to live in multigenerational households. These values are manifested through multiple behaviors, including participation in large kin networks with frequent visitation and contact, and provide different types of support which are common among some racial and ethnic groups in the U.S., including Hispanics and African Americans (e.g. Landale et al., 2006; Martin & Martin, 1980; Vega, 1995). Despite racial and ethnic differences in preferences for shared-households (including three-generation households), previous research suggests a preference for privacy and residential independence when resources allow it (Harvey, 2020b).

2.1.2. Economic factors

Historically, economic forces have played a role in the formation of different family and household structures. Moving in with family is likely a mechanism to alleviate financial crises and works as a safety net or anti-poverty strategy (Angel & Tienda, 1982; Keene & Batson, 2010; Mykyta & Macartney, 2011; Seltzer et al., 2012; Taylor et al., 2011; Wiemers, 2014). Additionally, rising housing costs in the U.S. have made it hard for families to achieve or keep their residential independence, leading to the formation of shared households, including three-generation households. High local housing costs increase the likelihood of moving into these complex households (Cohen & Casper, 2002; Harvey, 2020b). Relatedly, changes in women’s labor force participation have also had an effect on the increasing reliance on extended kin for childcare (Bianchi et al., 2006).

2.1.3. Policy influences

Prior research has examined the association between public assistance and the formation of three-generation households and other shared arrangements. On the one hand, there is evidence for a negative association between public transfers and the use of familial support while other studies find no support for this association. One of the primary mechanisms through which income support policies might affect the formation and dissolution of these households is by increasing the financial support being provided while “crowding out” the need to activate the familial safety net (Bianchi et al., 2006). An older piece of literature explored the effect of the Aid to Families with Dependent Children (AFDC) program on family structure and living arrangements, including three-generation households, and determined that greater AFDC benefits were associated with a greater likelihood of young mothers moving out of their parental home (e.g., Moffitt, 2003; Ruggles, 1997). Similarly, Pilkauskas and Michelmore (2019) find that increases in Earned income Tax Credit (EITC) reduce the formation of three-generational households among single mothers. In contrast, a more recent study by Pilkauskas and Cross (2018) finds that rises in Social Security receipt between 1996 and 2009 are associated with increases in three-generation households, rather than a rise in individuals living independently. Perhaps this is because the Social Security benefits provide grandparents with a relatively stable source of income, leading the child’s parents to move in with them.

2.2. Differences by race and ethnicity

Three-generation households are usually more common among members of racial and ethnic minorities (Fry & Passel, 2014; Kamo, 2000). In 2009, only 9% of white children were living with a grandparent in their households, compared to 17% of black and 14% of Hispanic children of any race (Kreider & Ellis, 2011). More recently, Cross (2018), using the PSID 1988–2013, estimated that more black children (39%) had ever lived with a grandparent than children of any Hispanic (20%) children, white children (14%), or children of any other race (23%).

The higher prevalence of three-generation households among racial and ethnic minorities has led to the discussion of whether motivations for co-residence differ by race and ethnicity. There are competing hypotheses behind this association: one related to cultural values, one to structural factors or to economic resources, and one related to the occurrence of life events (life-course perspective) (Blank and Torrecilha, 1998). In their study of Latino immigrants, Blank and Torrecilha (1998) find support for the life-course explanation according to which extended household arrangements (not only three-generation households) are determined by life course events, mainly as a strategy to care for young children and older adults. In contrast, Cross (2018), finds no statistically significant differences in the determinants of entry into extended family households (including grandparents and other relatives). In some groups, it might be customary to look out to family members rather than to the formal safety net when they need help (e.g., Cross, 2018; Woodward et al., 2010). Moreover, there might also be barriers (e.g., systemic barriers, fear, mistrust, etc.) that actually limit the access of racial and ethnic minorities to formal programs (e.g., Bitler & Hoynes, 2011; Hacker et al., 2015; Menjivar, 1997). Finally, and in contrast to the cultural preferences and values approach, previous research has also argued that the participation of racial and ethnic minorities in extended kin networks is a response to economic need rather than merely a response to cultural values and norms (Roschelle, 1997). In summary, a wide range of motivations should be considered, ranging from cultural preferences to social and economic need (Mollborn et al., 2012).

In addition to the differential prevalence of three-generational households among racial and ethnic minorities, previous research has also documented differences in the wellbeing of children who live with grandparents (and other relatives) and the (in)stability of these arrangements based on racial and ethnic affiliation. In relation to the consequences for child well-being, for example, Mollborn and colleagues (2012) find that any transition in grandparents’ co-residence was detrimental for non-Hispanic white children’s cognitive development but neutral or positive for nonwhite children of other races. In regard to the (in)stability of this arrangement, Perkins (2017) finds that nonwhite children are disproportionately exposed to changes in household composition, which include the entry or exit of grandparent(s) or other relatives. Similarly, Raley and colleagues (2019) estimated the cumulative number of household changes children experience by the age of 18, by race and ethnicity, and found that white children experienced the lowest (0.5) number of household changes due to grandparent(s) exit or entry, followed by Hispanic and Asian (0.7), and black (1) children.

2.3. Transition patterns and (in)stability of three-generation households

Several studies have looked at trends in the duration and prevalence of three-generational households and other shared households during childhood, while fewer studies have looked at correlates of exit. Below, I summarize findings of both bodies of research. First, studies related to the transitions and duration of three-generation households in chronological order include that of Pilkauskas (2012), who used the FFCWS to examine transitions, duration and the correlates of formation of three-generation households. Findings indicate that a large percentage (60%) of single mothers rely on the support of grandparents at least once over the first 9 years of the focal child’s life and that its prevalence is higher around birth (26%), decreasing over time (11% at age 9). Starting out in a three-generation household and then leaving that household was the most common pattern observed, likely reflecting a need for childcare at the time of the child’s birth. Findings of this study also indicate that transitions into three-generation households after a child’s birth may be driven by different mechanisms than those motivating co-residence at birth.

Next, Mollborn and colleagues (2012) studied the impact of extended household transitions (including grandparents and other adults) on age 2 cognitive development. Through the use of the Early Childhood Longitudinal Study Birth Cohort, their findings show that most children do not experience any transition between 9 and 24 months old (with significant differences by race and ethnicity as mentioned above). Through an examination of factors related to transitions, their research indicates that among Latino children, three-generation households dissolve when parents achieve some degree of financial stability (Mollborn et al., 2012). Similar to these findings, Perkins (2017), using the SIPP, identifies that co-residence with grandparents is more stable over time than co-residence with other relatives and non-relatives, and that changes in household composition (beyond the nuclear family) are more common among children with one parent or no parents (during the beginning of the observation period). Finally, and in line with Perkin’s findings, Raley and colleagues (2019), also using the SIPP, delineate that changes in household composition (with relatives and non-relatives) are common for children with low-educated mothers.

In addition to the studies mentioned above which focus on the prevalence of three-generation households and other shared arrangements over time, there are at least two studies that focus on the correlates of dissolution. Oberlander and colleagues (2009) look at urban, low-income African American adolescent mothers in their 24 months postpartum to examine transitions out of multigenerational households. Findings show that continued co-residence in multigenerational households is common, and that few mothers make the transition out. Finally, Glick and Van Hook’s (2011) study examines the potential triggers for the dissolution of extended-family households, both multigenerational households (parents and adult children) and other arrangements, including kin and non-kin. They pay particular attention to intrahousehold inequality, which indicates the presence of a potential benefactor and a potentially needy individual or couple. Specifically, for the case of multigenerational households (adult children and their parents), the authors identify that 45% of households remain together after 1 year and that they are less likely to dissolve when there is intrahousehold income inequality (one person or couple holding fewer financial resources than other people in the household). In contrast, intrahousehold income inequality predicts the dissolution of other extended household arrangements. Additional factors associated with changes in extended household composition in Glick and Van Hook’s study include life course events, such as changes in marital or relationship status. Finally, they present the differences in the likelihood of exiting extended-family households by race and ethnicity: African American householders and householders of Mexican origin were less likely to experience a change in composition compared to non-Hispanic whites.

In summary, there is a growing body of research about the dynamics and prevalence of three-generation households in children’s lives. There are three main findings. First, three-generation households are more common between a child’s birth and first birthday, and its likelihood of formation decreases over time. Second, they are short-lived and likely experienced multiple times during childhood. Third, this arrangement is more common among disadvantaged populations, particularly single mothers and ethnic and racial minorities. Taken together, these findings demonstrate that the presence of grandparents in a children’s household represents a source of change in household structure, particularly for the groups mentioned above. In this context, less is known about the potential correlates of these changes, and there is still a limited understanding of the mechanisms by which these transitions have a differential impact on the trajectories of children of racial and ethnic minorities.

3. Current study

Based on prior research and with the intent of filling the gap that exists in the understanding of three-generation households, this study seeks to document the potential correlates of their dissolution and whether they vary by race and ethnicity. In doing so, I also document their duration from the time of the child’s birth, and differences by race and ethnicity. From this perspective, I am able to document the (in)stability of three-generation households early in life, taking a longitudinal approach.

My first research aim is to describe how long mothers and their newborns spend in three-generation households after the child’s birth and whether this duration differs by race and ethnicity. There are two main hypotheses related to this aim. First, I hypothesize that non-Hispanic white mothers will have the shortest time in three-generation households, followed by non-Hispanic black and Hispanic mothers, as there may be cultural preferences that lead to longer co-residential spells. Additionally, as initial poverty levels and access to the formal safety net might differ across racial and ethnic groups, I expect the duration of the transition out of a three-generation household to take longer for members of racial and ethnic minorities which are more likely to be economically disadvantaged. Second, and in contrast to the previous hypothesis, non-Hispanic white mothers and their children will have the longest time in three-generation households, followed by the other two groups, if there is a different selection process among non-Hispanic white mothers in that those who enter three-generation households might be the most disadvantaged and, therefore, might take the longest to transition out. Relatedly, previous research indicates that nonwhite children are disproportionately exposed to changes in household composition, leading non-Hispanic black and Hispanic children and their mothers to stay the shortest time in three-generation households compared to non-Hispanic white mothers.

My second research aim is to describe the correlates of the first transition out of three-generation households after the birth of a child and how they differ by race and ethnicity, paying particular attention to how non-economic and economic factors operate. I first hypothesize that non-economic factors are relevant for the transition out of three-generation households and that their significance remains even after controlling for economic factors. This hypothesis follows previous research highlighting the impact of both family transitions and life course events on these households, shifting the focus from economic factors to recent trends in family instability and the increasing number of fragile families. Second, based on previous research (Cross, 2018), I hypothesize that correlates of dissolution are similar across racial and ethnic groups, particularly for mothers who start out by living in a three-generation household at the time of their child’s birth.

4. Data, sample, method, and measures

4.1. Data and sample

The data source for this study is the Fragile Families and Child Wellbeing Study (FFCWS), a nationally representative birth cohort study of 4,898 children from birth to age 15, born in large U.S. cities between 1998 and 2000 (https://fragilefamilies.princeton.edu). It uses a stratified random sample strategy, with an oversampling of nonmarital births. This population-based survey includes information from both mothers and fathers (Reichman et al., 2001), collected through interviews at the hospital, followed by phone calls to unavailable participants. In addition to the information collected at birth, information from interviews conducted following the child’s birth and when the child was around 1, 3, 5, 9, and 15 years old is also available. Additional modules include home visits, teacher surveys, and saliva samples. City weights are used to adjust the descriptive statistics for the oversample of nonmarital births, and this makes results representative of births in the 20 sample cities that occurred between 1998 and 2000. Weights are used only in the descriptive context, as in the regression analysis, the main variables the weights account for are included (mother’s marital status at birth, age, race, and education).

This study uses mothers’ surveys since they have lower attrition rates, and usually the mother is the custodial parent. In addition, I have access to the restricted contextual data of this study which includes a set of characteristics of the census tract at each wave (e.g., median house value and rent as percent of income), and it is rarely available in other data sources. I do not examine the age-15 interview due to the gap between the age-9 and age-15 surveys being relatively long and there being no available information about changes in household structure between these time points². To construct my analysis sample, I start with 4,898 mothers who had valid interviews and responses to at least one survey wave, and I keep the information from the waves in which they were at risk of a first transition out of a three-generation household (excluding the years of data after they have experienced the transition out and are therefore no longer at risk of experiencing the event). I do not include in my analysis mothers who may have skipped a survey wave but have been placed back into the sample. I exclude 53 mothers who are not the child’s primary caregiver. Of the 4,845 mothers left, I drop 37 with information missing from the household roster to construct the three-generation household indicator, 203 with missing information in key correlates, and 106 with missing information on city weights for a final sample of 4,499 mothers and their focal child, including those living in three-generation households at the time of the child’s birth. I then divide the remaining mothers into those not living in a three-generation household when the focal child was born (n=3,358) and my analysis sample, those who were (n=1,141)³ and are thus at risk of making a transition out of a three-generation household. I arrange these 1,141 mothers into 2,964 person-period observations, in which a mother is at risk of first moving out of a three-generation household after the focal child is born. I use this analysis sample for the descriptive section, in order to make full use of the data available. Next, for my regression analysis, I further exclude cases of individuals who were living in a three-generation household at the time of the child’s birth but who dropped out of my sample at year 1 (next survey wave after the child’s birth). Therefore, my regression models include a sample of 1,031 mothers and their focal child, arranged in 1,823 person-period observations, all of whom are at risk of first moving out of a three-generation household. Some contextual variables have more missing values than the variables in my main analyses; thus, when I include median house value as a sensitivity test in my analysis, the sample drops from 1,823 to 1,546.

4.2. Method

I begin the analysis by performing a standard descriptive survival analysis. Life tables show the proportion of mother-child households that exit a three-generation household after the birth of the child. They indicate the probability at each interval that a mother and her child will first transition out of a three-generation household and the proportion that will remain (survive). It also includes the number of censored cases between intervals, accompanied by the hazards of exiting. I perform a series of tests (log-rank and Wilcoxon) to compare the survival distributions by race and ethnicity. Next, in order to study the influence of a series of independent variables on the duration and transition out of three-generation households after the birth of a child, I use event-history modeling. Specifically, I create an event-history dataset from the Fragile Families and Child Wellbeing study to estimate complementary log-log models for discrete-time event-history analysis. This model is preferred over a traditional Logit to model binary outcomes because the complementary log-log model is ideal when the phenomenon studied is generated by a proportional hazard model in continuous time, as it is the case for my study ⁴.

I prefer event-history methods over standard regression models because the latter do not account for important issues with these data. Specifically, I am using data up to when the child is nine years old, but I do not observe mothers and their children beyond this point, when some of them might experience the exit out of the three-generation arrangement. Consequently, for some families, their transition out of extended households is unobserved due to right censoring. Event history models can properly account for right censoring (Allison, 1984; Box-Steffensmeier & Jones, 2004). Additionally, event history models account for timing, in the measurement of the outcome variable and on the time-varying covariates (Allison, 1984). Therefore, I can model the length of time spent in three-generation households after the birth of a child and the transition out of them, while including time constant and time varying covariates. However, in contrast to fixed effects, these models do not account for unmeasured time-constant individual characteristics.

With this approach, I model the risk or probability of the event—transition out or exit—occurring as a function of a series of covariates. In this context, since more research has focused on the role that economic factors play in the dynamics of three-generation households and taking into consideration the influence of family trends (e.g., reduced marriage and increased single parenthood) on the prevalence of shared households, I first examine the role of non-economic factors (Model 1), followed by an examination of how the inclusion of economic factors (Model 2) changes the marginal effect of non-economic factors in the sample overall.

These models are followed by a subgroup analysis by race and ethnicity for which I also follow a step-by-step model construction, including the non-economic factors followed by the economic cluster of variables. Finally, and following the second part of my research aim, I perform a Chow test using the contrast command in Stata to assess whether the coefficients across the three racial and ethnic groups of interest are equal.

Besides my main analysis, I include a series of sensitivity tests to check for the robustness of my results. First, I replicate all my main analysis using only data from the child’s birth and interviews from when the child was around 1, 3, and 5, ignoring data from the last wave (age 9), in order to test whether keeping only data with closer gaps influences my results. Second, I investigate the role of contextual factors as a proxy for housing costs on the transition out of three-generation households (i.e., lagged median house value and rent as percent of income).

4.3. Measures

My main outcome is whether there is a transition out of a three-generation household. Therefore, I construct an indicator of whether the custodial mother and the child live in a three-generation household (with one or more grandparents) at each wave. This measure is constructed based on the household roster reported by the custodial mother at each wave. Therefore, I consider a transition out when the mother and child no longer report living with grandparent(s) in the household, either because the mom and child moved out of the household, or because the grandparent(s) did. In addition, I have a duration variable indicating the approximate length of time the mother and her child live in a three-generation household after the child’s birth. Based on previous research, my models account for non-economic and economic factors, giving particular relevance to the measures related to the needs and characteristics of the parent. Additional variables include two contextual factors measured at the census tract where the mother and the child lived at each wave of the survey. These variables will allow me to look at the association between housing costs and the living arrangements of mothers and their children.

4.3.1. Duration

I construct the duration variable (time in a three-generation household) using the child’s age at the wave in which the mother and child stop co-residence with grandparent(s) for the first time after the child’s birth, given that at birth grandparents are in the household. Since the first interview occurs when the children are newborns, the proceeding intervals are determined based on the age ranges of the children at each wave (i.e., at wave 2, children range from about 1 year of age to 2.5 years of age; at wave 3, children range from about 3 years of age to 4 years of age). Therefore, I use four categories to match the four waves of the survey after baseline (>0 & <2.5 years old, >=2.6 & <4 years old, >=4 & <=6 years old, and >=7 & <=11 years old). I perform a series of tests to identify whether the process of transition is time dependent and to identify the ideal specification for the time variable. Although the tests reveal that the process is not time dependent, I include year dummies (defined according to the age ranges of the children at each wave as mentioned above) to control for the differential length of intervals between survey waves; the inclusion of the duration variable does not change the magnitude or direction of any of my estimates.

4.3.2. Non-economic factors

I include several variables that have been identified as correlates of three-generation households at the mother, father, and child level. This set of variables includes time-constant variables: the mother’s race and ethnicity (Non-Hispanic white, Non-Hispanic black, Hispanic⁵, and Other); and mother’s nativity (U.S. born). It also includes time varying variables, such as the mother’s age (under 20, 20–24, 25–35, and over 35 years old⁶), indicators of whether the mother’s health was excellent or very good, and whether the father was in jail at the time of the survey (yes, no, missing), which are all potential proxies for the need of the mother to co-reside with the child’s grandparents if additional support is needed. Next, I include variables related to the mother’s current and past family experiences: a time varying measure of the mother’s relationship status with the child’s father (single, married or cohabiting); an indicator of whether the mother has a new partner; a continuous measure of the number of children in the household; and an indicator of whether the child is in good health at each wave. Finally, I include additional time constant variables, such as an indicator of whether the mother is living with two parents at age 15 and an indicator of whether the baby is the mother’s first child.

4.3.3. Economic factors

This cluster captures the mother’s socioeconomic characteristics by measuring her changing economic resources and needs over time. It includes time-varying measures of the mother’s education (less than high school, high school or equivalent, some college or technical school, college or graduate studies); an indicator of whether the mother received Temporary Assistance for Needy Families (TANF) or the Supplemental Nutrition Assistance Program (SNAP) in the year prior to the survey, and whether the mother was employed the week prior to the survey, starting in wave two (when the child was about 1 year old).

Finally, contextual variables measured within the census tract where the mother resides include time varying measures of the median house value (divided in terciles from the unweighted data in the following categories $0– $64,800; >$64,800– $115,400; and >$115,401) and rent as percent of income (divided in terciles from the unweighted data in the following categories 0%– 24.8%; >24.8% <= 28.3%; and >28.3%). These two variables are lagged one year to better account for timing. Because these variables have more missing data, I do not include them in the base model. However, due to the potential importance of the housing market in decisions about living arrangements, I include them in the sensitivity tests.

5. Results

5.1. Descriptive results

Table 1 provides mean and percentage distributions for all the variables at the time of the child’s birth for the entire sample (n=4,499) (first column). The next two columns show my entire sample divided into two groups: those not born into a three-generation household (n=3,358) and those born into a three-generation household (n=1,141) with the corresponding test of whether the differences in characteristics between these two groups are statistically significant. The last sample is then further divided into two groups: those born into a three-generation household but who exit (n=907) and those born into this household type who do not transition out within the observation period (n=234), with the corresponding tests of whether differences between these two groups are statistically significant. All columns in Table 1 use city sampling weights to adjust for the oversample of nonmarital births (and the corresponding differences by age, race, and education). First, looking at the three-generation households at the time of the child’s birth and similar to previous research, I find that these households are more common among more disadvantaged mothers. Mothers in three-generation households are more likely to be younger, members of racial and ethnic minorities, and single, as well as have a lower level of education, a lower lagged median house value, and their child’s father in jail. Additionally, they were less likely to live with two parents at age 15 and have other children prior to the focal child.

Table 1.

Sample Characteristics at Baseline by Subgroups: All, Not in Three-Generation Household at Birth, in Three-Generation Household at Birth, Ever Exited, Never Exited¹

Variables at baseline	All T1-T5 n=4,499	Not in three gen at birth T1 n=3,358	T-test/Chi-square	In three gen at birth T1 n=1,141	Ever out T2-T5 n=907	T-test/Chi-square	Never out T2-T5 n=234 (Censored)
	Mean or % SE	Mean or % (SE)		Mean or % SE	Mean or % SE		Mean or % SE
Demographic and non-economic factors
Mother’s age	27.00 (0.11)	27.90 (0.16)	***	22.80 (0.43)	22.03 (0.73)		24.78 (1.67)
Mother’s age			***
Under 20	12.98	9.01		31.42	34.95		22.31
20–24	25.42	21.36		44.27	46.59		38.29
25–35	51.36	58.36		18.86	14.03		31.33
>35	10.24	11.27		5.45	4.43		8.07
Mother’s race and ethnicity			**
Non-Hispanic white	29.95	33.47		13.62	12.61		16.21
Non-Hispanic black	34.69	31.43		49.78	54.71		37.07
Hispanic	28.59	27.87		31.96	29.06		39.42
Other	6.766	7.22		4.64	3.62		7.29
Mother is U.S. Born	72.36	71.58		75.98	80.62		63.98
Mother’s relationship with baby’s father			***
Single	24.06	15.97		61.59	64.54		53.95
Married	52.82	59.9		19.94	17.44		26.37
Cohabiting	23.13	24.13		18.48	18.01		19.67
Number of kids in the household	1.15 (0.05)	1.13 (0.05)		1.23 (0.09)	1.23 (0.08)		1.23 (0.23)
Mother’s health is excellent or very good	69.41	69.82		67.53	68.09		66.07
Mother was living with two parents at age 15	54.38	57.29	*	40.87	40.27		42.42
Baby is mother’s first baby	39.47	35.1	**	59.74	60.23		58.47
Baby was low weight at birth (less 3% missing, multiple births)	7.45	6.16		13.43	12.13		16.78
Child is in good health	89.83	91.03		84.23	84.89		82.55
Father in jail at time of survey			*
Yes	2.14	1.62		4.51	4.28		5.12
Missing (<1%)	0.60	0.63		0.44	0.45		0.4
Economic Factors
Mother’s education			***
Less than high school	28.41	25.46		42.08	40.17		47.02
High school or equivalent	31.98	31.69		33.34	36.35		25.58
Some college/Technical school	19.52	19.14		21.28	20.62		23
College or graduate studies	20.09	23.71		3.3	2.87		4.4
Mother received TANF or SNAP past year	22.98	22.73		24.13	27.02		16.66
Contextual Factors 2
Lagged median house value
$0 – $64,800	27.45	25.4		36.8	38.37		30.07
>$64,800 – $115,400	27.1	26.61		29.33	29.83		27.19
>$115,401	45.45	47.98		33.87	31.8		42.73
Lagged median house value	149,728.70 (5427.58)	157,213.10 (5952.12)	**	115,497.50 (9106.15)	113,012.00 (11578.95)		126,159.90 (13120.31)
Lagged rent as percent of income
0% – 24.8%	39.24	40.81		32.06	31.67		33.7
>24.8% <= 28.3%	29.33	29.24		29.74	30.03		28.5
>28.3%	31.43	29.95		38.2	38.3		37.8
Lagged rent as percent of income	27.07 (0.25)	26.95 (0.42)		27.61 (0.72)	27. 55 (0.72)		27.87 (0.83)

¹City sampling weights used to adjust for the oversample of nonmarital births (and the corresponding differences by age, race, and education). Sample numbers are not weighted.

²Contextual factors n=3,909; 2,893; 1,016; 881; 135

^*p<0.05

^**p<0.01

^***p<0.001

Next, I compare the characteristics of mothers who were in a three-generation household at the time of the child’s birth and exited to those who did not exit during my observation period. My observations show no clear pattern of socioeconomic advantages or disadvantages at baseline, nor statistically significant differences. The lack of differences might be due to the fact that those who do not exit during my observation period might just take longer to experience the transition rather than never experience an exit out of three-generation households.

5.2. Descriptive results aim 1: to describe how long mothers and their newborns spend in three-generation households after the child’s birth and whether this duration differs by race and ethnicity.

Table 2 presents the life table analysis for the overall sample of the proportion of mother-child households that exit a three-generation household after the birth of the child. They indicate the probability at each interval that a mother and her child will first transition out of a three-generation household and the proportion that will remain (survive). It also includes the number of censored cases between intervals, accompanied by the hazards of exiting. Overall, children are more likely to experience the first exit at very young ages, which is consistent with prior research. The life table shows that time in a three-generation family is likely to be short. The hazard rate of first exiting a three-generation household is very high between birth and age 1, and it declines between ages 1 and 3 before stabilizing. The majority of households experience a transition out in the first year, and very few remain beyond age 5. The survival column highlights the proportion of mother-child households that continue co-residing with grandparents after the child’s birth, showing that by year 9, only 5% of mother-child households remain in a three-generation household.

Table 2.

Life Table: First Exit out of Three-Generation Households for the Sample Overall¹

Interval		Beg.	Exit	Censored	Survival	SE	95% Conf. Int.		Hazard	SE	95% Conf. Int.
0	1	1,141	522	147	0.51	0.02	0.48	0.54	0.65	0.03	0.59	0.70
1	3	472	234	16	0.25	0.01	0.23	0.28	0.34	0.02	0.30	0.38
3	5	222	104	20	0.13	0.01	0.11	0.15	0.33	0.03	0.27	0.38
5	9	98	47	51	0.05	0.01	0.03	0.06	0.24	0.03	0.18	0.30

¹Numbers and percentages not weighted.

Table 3 presents the life table analysis by race and ethnicity of the proportion of mother-child households that exit a three-generation household after the birth of the child. The table includes complete data only for non-Hispanic and Hispanics because an inadequate number of observations prevented a full analysis for the other groups (I am showing rows only where the beginning total is 30 or more observations). Non-Hispanic mothers and their children have the highest hazard rate of exiting three-generation households between birth and year 1, followed by non-Hispanic mothers, Hispanics, and those of the “Other” category. For both subgroups with complete data, the hazard of exiting a three-generation household is higher between birth and year 1 (about 0.6), and it decreases until reaching a low of 0.2 at year 9.

Table 3.

Life Table: First Exit out of Three-Generation Households by Race and Ethnicity¹

Interval		Beg Total	Exit	Censored	Survival	SE	95% Conf. Int.		Hazard	SE	95% Conf. Int.
NH-black
0	1	647	298	84	0.51	0.02	0.47	0.55	0.65	0.04	0.58	0.72
1	3	265	142	5	0.23	0.02	0.2	0.27	0.37	0.03	0.31	0.43
3	5	118	55	13	0.12	0.01	0.09	0.15	0.33	0.04	0.25	0.41
5	9	50	26	24	0.04	0.01	0.02	0.06	0.26	0.04	0.17	0.35
Hispanic
0	1	295	129	37	0.54	0.03	0.47	0.59	0.61	0.05	0.51	0.71
1	3	129	63	3	0.27	0.03	0.22	0.32	0.33	0.04	0.25	0.41
3	5	63	26	3	0.16	0.02	0.11	0.20	0.27	0.05	0.17	0.37
5	9	34	15	19	0.06	0.02	0.03	0.10	0.22	0.05	0.12	0.32
NH-white
0	1	153	79	18	0.45	0.04	0.37	0.53	0.76	0.08	0.60	0.91
1	3	56	23	4	0.26	0.04	0.19	0.34	0.27	0.05	0.16	0.38
Other
0	1	46	16	8	0.62	0.07	0.46	0.75	0.47	0.11	0.25	0.69

¹Numbers and percentages not weighted.

Next, the survival column shows that by year 9, very few mothers remain in a three-generation household, across both subgroups with complete data. Finally, I compare the survival distributions between the subgroups with a log-rank and Wilcoxon (Breslow) test for the equality of survivor functions and found no statistically significant differences (p=0.12 and p=0.10). These tests compare estimates of the hazard functions for my groups of interest at each time. Also, it is important to consider the fact that all three groups have overlapping confidence intervals, which indicates that these estimates are not statistically different from each other.

5.3. Regression results aim 2: to describe the correlates of the first transition out of three-generation households after the birth of a child and how they differ by race and ethnicity.

As mentioned above, I estimate complementary log-log models to identify the correlates of the transition out of a three-generation household for those mothers and children who were co-residing with grandparents in the household at the time of the child’s birth. The first two columns in Table 4 show the marginal effects of Models 1 and 2 for the sample overall, which are relevant for the first part of my research aim. Model 1 predicts the transition out of a three-generation household including demographic and non-economic factors, while Model 2 adds to Model 1 the cluster of economic variables. First, Model 1 suggests that U.S. born mothers are more likely to exit these households. Married and cohabiting mothers are more likely to exit a three-generation household compared to single mothers. Similarly, mothers with a new partner are more likely to exit a previously formed three-generation household. Moreover, other indicators of potential need are significant predictors of the continuity of this living arrangement. Mothers with more children are less likely to exit as are those for whom the focal child is their first child. Finally, regarding the mothers’ age, a marginally significant category is between 20 to 24 years of age, with these mothers being more likely to transition out than mothers who are over 35 years of age. Even though this finding is somewhat surprising since previous research has indicated that younger mothers are more likely to enter (and potentially remain) in three-generation households (e.g., Trent & Harlan, 1994), my finding might be related to the selection process of older mothers in these households. Therefore, older mothers who are living with the child’s grandparent(s) at the time of the child’s birth might be experiencing additional needs or providing care to the grandparents in the household, among other circumstances, that could delay their transition out.

Table 4.

Marginal Effects: Complementary Log-Log Models Predicting First Exit Out of Three-Generation households

Variables	Model 1 Non-Economic	Model 2 Non & Economic	Model 3 NH-white	Model 4 NH-black	Model 5 Hispanic
	MEa (SE)	ME (SE)	ME (SE)	ME (SE)	ME (SE)
Duration Child’s age (>0 &<2.5 years old as reference)
>=2.6 & <4 years old	−0.00 (0.03)	−0.00 (0.03)	−0.09 (0.07)	0.03 (0.04)	0.02 (0.05)
>=4 & <=6	−0.02 (0.04)	−0.03 (0.04)	0.13 (0.09)	−0.04 (0.05)	−0.05 (0.07)
>=7 & <=11	−0.01 (0.06)	−0.02 (0.05)	0.05 (0.16)	0.00 (0.07)	−0.07 (0.09)
Non-economic factors Mother’s age (Over 35)
Under 20	0.06 (0.07)	0.07 (0.07)	0.19 (0.15)	0.03 (0.11)	0.07 (0.12)
20–24	0.13* (0.06)	0.13* (0.06)	0.34** (0.12)	0.07 (0.09)	0.09 (0.10)
25–35	0.11 (0.06)	0.10 (0.06)	0.16 (0.11)	0.05 (0.09)	0.11 (0.10)
Mother’s race and ethnicity (NH White as reference)
Non-Hispanic Black	0.03 (0.03)	−0.00 (0.04)	- -	- -	- -
Hispanic	−0.02 (0.04)	−0.04 (0.04)	- -	- -	- -
Other	−0.07 (0.06)	−0.09 (0.06)	- -	- -	- -
Mother is U.S. Born	0.09* (0.04)	0.07 (0.04)	0.32 (0.17)	0.09 (0.08)	0.05 (0.05)
Mother’s relationship with baby’s father (Single as reference)
Married	0.24*** (0.04)	0.26*** (0.04)	0.12 (0.08)	0.31*** (0.06)	0.34*** (0.06)
Cohabiting	0.26*** (0.03)	0.27*** (0.03)	0.34*** (0.07)	0.28*** (0.04)	0.25*** (0.05)
Mother has a new partner	0.34*** (0.03)	0.35*** (0.03)	0.36*** (0.09)	0.31*** (0.04)	0.48*** (0.06)
Number of kids in the household	−0.06*** (0.01)	−0.06*** (0.01)	−0.07* (0.03)	−0.06*** (0.01)	−0.06** (0.02)
Mother’s health is excellent or very good	0.01	0.01	0.10	0.00	0.02
	(0.02)	(0.02)	(0.07)	(0.03)	(0.04)
Mother was living with 2 parents at age 15	−0.05	−0.04	−0.03	−0.05	−0.04
	(0.02)	(0.02)	(0.06)	(0.03)	(0.04)
Baby is mother’s first baby	−0.11***	−0.10***	−0.36***	−0.09*	−0.10
	(0.03)	(0.03)	(0.08)	(0.03)	(0.05)
Child is in good health	−0.06	−0.06	−0.16	−0.05	−0.05
	(0.03)	(0.03)	(0.12)	(0.04)	(0.06)
Father in jail at time of survey (No as reference)
Yes	0.02	0.01	0.24	0.02	−0.02
	(0.04)	(0.04)	(0.17)	(0.05)	(0.10)
Missing	0.04	0.04	0.07	0.05	0.02
	(0.05)	(0.05)	(0.10)	(0.06)	(0.09)
Economic Factors Mother’s education (Less than high school as reference)
High school or equivalent	-	−0.02	−0.03	0.02	−0.09
	-	(0.03)	(0.08)	(0.04)	(0.05)
Some College/Technical school	-	0.01	−0.00	0.03	−0.01
	-	(0.03)	(0.09)	(0.04)	(0.05)
College or graduate studies	-	0.03	0.16	0.09	−0.05
	-	(0.06)	(0.13)	(0.08)	(0.13)
Mother received TANF or SNAP past year	-	011***	0.16*	0.10**	0.11*
	-	(0.02)	(0.06)	(0.03)	(0.05)
Mother worked last week	-	0.06**	0.06	0.07*	0.05
	-	(0.02)	(0.07)	(0.03)	(0.05)
Constant	−0.65	−0.81	−1.52	−0.76	−0.88
	0.28	0.294	1.03	0.44	0.50
Log Likelihood	−1147.67	−1135.307	−128.78	−632.88	−299.214
Period-Person Observations	1,823	1,823	235	1,013	498

^aMarginal effects

^*p<0.05

^**p<0.01

^***p<0.001

Next, adding the economic factors to Model 1 changes only the significance for the nativity variable. All the other non-economic factors in the model are not explained by adding in the economic factors. The economic factors in model 2 suggest that mothers receiving TANF or SNAP in the year prior to the survey are more likely to exit three-generation households, as are mothers who worked the week prior to the survey.

The finding related to TANF or SNAP receipt is consistent with prior research if we consider the receipt of these benefits as a support to the mother’s household that might allow her to transition to living independently. However, given the problematic timing of the survey, there might be an issue of reverse causality, in which those mothers who exit three-generation households request TANF or SNAP to compensate for the loss of the grandparent in the household. To get a closer look at this association, I examine the status of TANF or SNAP receipt in a descriptive context to better understand the timing of both events: receipt and exit out of three-generation households. I compare mothers who remain in a three-generation household for the first two waves (at child’s birth and around the time when the child is 1 to 2.5 years old) with those who are in at the first wave and exit by the second wave. This approach is not perfect, as events might still be mistimed. However, this is the best approximation I have to the right ordering, since these two waves are the ones with the closest time gap (around 1 to 2 years). At the time of the child’s birth, both groups of mothers present similar figures of TANF or SNAP receipt in the 12 months prior to the survey (29.01% vs 22.29%). However, when looking at the second wave, those who remain in a three-generation household have lower levels of receipt (33.50%) compared to those who are no longer living with grandparent(s) (43.98%). If the timing is right and TANF or SNAP receipt predates the exit of the three-generation household, this might suggest the possibility of welfare receipt allowing some degree of financial independence that leads mother to living independently (without grandparents in the household), ignoring all other events that might be simultaneously occurring like changes in marital status. On the other hand, if the timing is not correct and mothers exit three-generation households before receiving TANF or SNAP, they might pursue welfare receipt as a way to compensate for the absence of grandparent(s) in the household.

Finally, the indicator of whether the mother worked in the week prior to the survey indicates that mothers who work are more likely to exit three-generation households.

5.3.1. Racial and ethnic differences in the first exit after a child’s birth

The last three columns of Table 4 show the marginal effects of the final model (Model 2 including economic and non-economic factors) for the subgroup analysis, excluding mothers in the “Other” category due to sample size ⁷.

First, in terms of the non-economic factors that are significant in the overall model, most of them retain their significance across the three groups of mothers (non-Hispanic white, non-Hispanic black and Hispanics), with few exceptions. Married mothers are still less likely to exit a three-generation household compared to single mothers, with the exception of non-Hispanic white mothers for whom this category is not significant. In contrast, cohabiting mothers in all racial and ethnic groups are more likely to exit these households compared to single mothers. Similar to the results for the sample overall, mothers from these three groups are more likely to exit a three-generation household if they have a new partner, and they are less likely to exit if they have more children in the household. In contrast, whether this baby was the mother’s first child is not significant for Hispanic mothers.

Finally, in terms of the economic factors that are significant in the main model, some of them retain significance across groups including TANF or SNAP receipt in the year prior to the survey. In contrast, mother’s employment status remains only significant for non-Hispanic black mothers, although this difference might be due to the small sample size for the other two groups (despite the marginal effect having a similar size and going in the expected direction).

Finally, I move beyond comparing estimates to assessing whether the coefficients are equal across all three racial and ethnic groups. The results of a Chow test indicate that the only significant difference in the predictors of the transition out of a three-generation household across the three groups is whether the mother received TANF or SNAP in the 12 months prior to the survey (p=0.03).

A finding to highlight from the subgroup analyses is the lack of variation and significant differences both in the duration and the predictors of the first transition out of three-generation households. Across all subgroups, even after controlling for socioeconomic factors, the marginal effect of most of the variables are similar, and a formal test of differences only indicates one significant difference. One reason behind this finding might be the limited sample size particularly for non-Hispanic white mothers. Additionally, mothers who start out in three-generation households at the time of the child’s birth might share similar motivations: the need for childcare, while mothers who move in with grandparent(s) later in a child’s life might do it as a way to cope with a variety of shocks or needs. An examination of that group of mothers might yield more variation and, therefore, different results. Future research should explore these associations with a bigger more diverse sample size to confirm this lack of variation or to shed light on additional dynamics.

5.3.2. Sensitivity tests

As a sensitivity test, I run two additional models, both including contextual measures related to housing costs within the census tract. One test adds the lagged median house value, and a second one adds a lagged measure of rent as the percent of income in the tract. Mothers with higher lagged median house value in the previous wave are less likely to exit three-generation households compared to those with the lowest lagged median house value. Similarly, mothers in the highest category of the lagged measure of rent as percent of income (>28.3%) are less likely to exit three-generation households compared to those with the lowest rent as percent of income measure (0–24.8%). The significance of these estimates fades away or becomes only marginally significant once the non-economic factors are included in the full model (results not shown; available upon request). The inclusion of these variables does not change the size, direction or significance of any other variable in the model. Given the large number of missing values for this variable, results are not shown in the main table. After conducting a similar test for racial and ethnic groups, by including contextual variables, rent as percent of income remains significant only for Hispanic mothers, even after controlling for other economic and non-economic factors.

6. Discussion

Using longitudinal data from the FFCWS, the present study examines the duration and factors associated with the first transition out of a three-generation household after the birth of a child given that at the time of the birth, the custodial mother and her child are co-residing with grandparents. Additionally, it looks at how the time to the first transition and the factors associated with the transition out varies by race and ethnicity in a sample of disadvantaged mothers. This paper aims to address limitations of previous research by using a birth cohort, event history analysis, and by focusing specifically on the correlates of the first exit out of three-generation households. Additionally, it pays particular attention to non-economic factors that might be related to transitions out of three-generation households.

As children’s living arrangements are now more likely than before to include grandparents in the household, the study of these complex households take even more relevance. An increasing number of studies have started to look at the dynamics of these arrangements, by examining the triggers of their formation and at some of the effects on child well-being. However, less is known about the continuity of three-generation households and about the potential triggers of their dissolution, as the focus in this area has been on the duration and prevalence of these arrangements over time. The current study aims to fill in this gap. First, by looking at the characteristics of those in and out of three-generation households at the time of the child’s birth, there is a consistent pattern of disadvantage among those mothers living in a multigenerational household at the time of the child’s birth, which is consistent with prior research (Cross, 2018; Pilkauskas, 2012; Zonta, 2016). Next, comparing those who ever exited a three-generation household during the observation period to those who do not, there are no significant differences, likely due to sample size and right censoring for those who do not transition out of three-generation households during my observation period. Below, I summarize and discuss my findings for each research aim.

First aim: to describe how long mothers and their newborns spend in three-generation households after the child’s birth and whether this duration differs by race and ethnicity.

I find a clear pattern of this arrangement being more common during early childhood. The hazard of transitioning out is quite high between a child’s birth and year 1, and it decreases as the child ages. These findings are consistent with prior research documenting the prevalence of three-generation households during the first year of a child’s life (Amorim et al., 2017; Cross, 2018; Pilkauskas, 2012). Finally, after performing a formal test for equality of survivor functions, I find no statistically significant differences across racial and ethnic groups. These results are contrary to both of my hypotheses of differential time to dissolution by nonwhite and non-Hispanic white mothers and their children. As mentioned in the results section, this lack of variation might be driven by the characteristics of my sample: mothers who start out in three-generation households at the time of the child’s birth and who might share the need for childcare. As found by Pilkauskas (2012) using the FFCWS, mothers who live with grandparents at the time of the child’s birth and then move out are to some extend different from those who do not co-reside with grandparent(s) at the time of the child’s birth but move in afterwards, likely following some kind of transition or shock.

Second aim: to describe the correlates of the first transition out of three-generation households after the birth of a child and how they differ by race and ethnicity.

Findings indicate that the mother’s characteristics and her needs, and life-course events have a strong association with the dissolution of three-generation households. In terms of some particular mother’s needs, those for whom the focal child is their first child and those with more children in the household are all less likely to exit.

Life-course events, such as changes in marital and relationship status, take on a particular relevance in the transition out. This finding is consistent with Perkins’s research (2017), which also highlights the likelihood of children experiencing both parental (i.e., exits or entries of one or both parents from/into the child’s home) and non-parental changes (i.e., exits or entries of other related or nonrelated people from/into the child’s home, different from the nuclear family) early in life. In addition, with this finding I am able to document the “push-pull” effect described by Mollborn and colleagues (2012) in which the presence of a new partner is associated with the exit out of three-generation households and other extended living arrangements.

Finally, in regard to economic factors, mothers who received TANF or SNAP in the year prior to the survey are more likely to exit three-generation households. In terms of welfare receipt, my findings relate to the “crowding out” effect of the formal safety net over the private safety net (Bianchi et al., 2006). In this context, we could interpret the receipt of TANF or SNAP not just as a sign of disadvantage but also as a form of support being provided by the formal safety net rather than by extended family members, similar to what previous research has indicated about the effect of increased AFDC benefits (Moffitt, 2003) or EITC receipt (Pilkauskas & Michelmore, 2019). However, future research should better assess the timing of events, since mothers could be pursuing TANF or SNAP benefits to compensate for the loss of grandparent(s). In addition to the positive association with TANF or SNAP receipt, I also find that mother’s employment in the week prior to the survey is positively associated with the transition out of three-generation households. This association might be also pointing out in the direction of achieving financial independence, although mothers could also have the need to work if they had stopped co-residing with grandparent(s). Finally, my findings support my hypothesis under aim 2 that non-economic factors are significant even after controlling for economic factors, and that family transitions are some of the most salient correlates of exit.

In the comparison of economic and non-economic factors and the study of three-generation households, it is relevant to highlight the fact that none of the marginal effects for the mother’s race and ethnicity are statistically significant. In contrast, economic factors such as TANF or SNAP receipt and employment status are. In that sense, it is important to study the dynamics of three-generation households beyond the effect that race and ethnicity might have, as the results indicate that other factors might be playing an important role in the dissolution of these households.

Next, by looking specifically at the subgroup analysis, a formal test for the differences across groups indicates very few significant differences. The only marginally significant difference in the factors related to the transition out of three-generation households by race and ethnicity is TANF or SNAP receipt in the 12 months prior to the survey. In this area, my results are similar to Cross (2018), who examines predictors of household extension by race and ethnicity and finds no statistically significant differences across groups.

Although my findings are somewhat puzzling, particularly when looking at the economic factors, there are a few relevant highlights. First, there is evidence that life course events, such as changes in marital and relationship status, are important correlates of the transition out of three-households. As such, I am able to add to the debate about the importance of looking at parental and non-parental changes in the home when thinking about the potential effects of family and household composition on children’s early development, particularly among children from racial and ethnic minorities. Future research should continue attending to the interaction between nuclear family households and three-generation households as competing or complementing one another. Second, my measures of welfare receipt and employment are significant, and descriptive results suggest that welfare receipt might predate the transition out of three-generation households. It would be relevant to better attend to the timing of these associations.

7. Limitations and future research

The present study is among the first to look at the correlates of the first transition out of a three-generation household after the birth of a child, adding new information to the understanding of complex extended households. However, it carries several limitations that open up the opportunity for future research. First, the Fragile Families study does not collect data between waves on the exact day or year in which the transition out of a three-generation family household occurs. Therefore, I am missing data on shorter transitions in and out since I rely on the indicator at each wave to know if the mother and the child are living with grandparents in the household. Second, I am only looking at the first transition out of a three-generation household after the birth of a child. In future research, it would be relevant to look at the correlates of multiple transitions in and out of multigenerational households, as these experiences are indicators of household instability. In addition, mothers who start out in a three-generation household at the time of the child’s birth might be different from mothers who do not live with grandparent(s) at the time of the child’s birth and move in afterwards. Therefore, expanding the analysis to this group of mothers might reveal different results. Third, I have issues of reverse causality in my models, since many of the potential covariates I am including are measured at the same time as the household structure and therefore, are endogenous to my outcome of interest. Fourth, omitted variable bias might be an issue, since I am not able to include, for example, measures of mothers’ preferences or the availability of grandparents, among other variables.

Despite these limitations, this study is one of the few to examine the correlates of the first exit out of a three-generation household after a child’s birth in a sample of fragile families, taking a fully longitudinal approach. Findings underscore the need to examine both parental and non-parental changes in children’s homes, as these transitions are increasingly more common and likely to happen early in a child’s life. Moreover, this study examines racial and ethnic differences in the duration and correlates of dissolution for three-generation households, which in turn helps us get a better sense of the differential role that these households play for each group. Taken together, these results add to the knowledge of the dynamics of three-generation households and the complexity of children’s living arrangements early in life, and they have the potential to inform future research in several areas.

Finally, future research should continue exploring whether the factors that explain entries into three-generation households and those that explain exits differ, as those two sets of factors might not be parallel, as this study suggests. More research is needed to help us understand the process of dissolution of three-generation households, their potential implications for child wellbeing, and the role that policies might play in promoting or not the continuity of these households.

Highlights.

• Custodial mothers in three-generation households are more disadvantaged.

• The first exit out of three-generation households is more likely to happen between birth and age 1.

• There are not significant differences in survival distributions by race and ethnicity.

• Non-economic factors are associated with the first exit out of three-generation households.

• TANF or SNAP receipt and mother’s employment are associated with the first exit out of three-generation households.

• There are not significant differences in the correlates of first exit by race and ethnicity.