The Family Safety Net of Black and White Multigenerational Families

Introduction

One of the most striking and persistent features of the American socioeconomic landscape is the disadvantaged position of African Americans relative to Whites. This study explores some of the present reasons for that disadvantage, focusing on the multigenerational family safety net. We describe black and white multigenerational families’ socioeconomic resources while taking into account information about living biological kin up and down the generational ladder. We ask: (1) Do Blacks have more vertical family ties than Whites?; (2) How do the socioeconomic resources in black multigenerational families differ from those in white multigenerational families?; and (3) What are the implications of race differences in kin availability and socioeconomic resources for the family safety nets of Blacks compared to Whites? We examine Black-White differences in who has family members who may be able to provide financial support to others as well as differences in who has a family member who may require financial support to alleviate hardship.

We show that race differences in the number of vertical family ties are small, although black families have somewhat more living generations than white ones. But, we find large race differences in both the transitory and durable socioeconomic resources of multigenerational families. Over twice as many Blacks as Whites are in multigenerational families in which at least one family member is unemployed, and over three times as many Blacks are in multigenerational families in which everyone is simultaneously unemployed. Whites are more likely to have at least one family member who has a college degree or who owns a home by 28 percentage points and 32 percentage points, respectively. Race differences in multigenerational socioeconomic resources are even larger among families in which one generation is disadvantaged―for example, a family member who is unemployed or does not own a home. We conclude that the family safety net is substantially weaker for Blacks in multigenerational families, with greater potential demands to help others but fewer resources available to provide help.

Background on the family safety net

Family members help each other during times of need. The safety net that a family provides can come in the form of direct assistance such as financial or time transfers, human capital investment, or the inheritance of wealth. Even in the absence of direct assistance, a latent support system can provide an individual with a sense of security against unanticipated economic and health crises (Riley 1983). Moreover, individuals’ behavior throughout life may be affected by the knowledge that one’s family members can protect them against losses or catastrophes. For example, individuals with limited familial safeguards might be risk averse in their education, employment, and financial choices while those with stronger family safety nets may assume more risk (Keister 2000).

Help is common between parents and children, but extended family members in other generations are also important sources of support (Altonji, Hayashi, and Kotlikoff 1992; LaFave and Thomas 2017). Increasingly, US families span multiple generations, contributing to individuals in each generation having more “vertical” family ties than ever before (Bengtson 2001). Yet, there are also significant variations in family structure by race. In the United States, Blacks’ higher fertility compared to Whites contributes to increased vertical ties to adults in younger generations while greater mortality and lower marriage rates diminish the likelihood of vertical ties to older generations (Sarkisian and Gerstel 2012; Szinovacz 1998). At the same time, there are persistent Black-White disparities across a range of socioeconomic measures, with rates of college attainment, homeownership, and labor force participation for Blacks continuing to lag behind those of whites. (Charles and Hurst 2002; Couch and Fairlie 2010; McDaniel et al. 2011). Prior research also has established that these disparities extend beyond a single generation, with strong intergenerational correlations across multiple socioeconomic measures (Charles and Hurst 2003; Mazumder 2005; Pfeffer and Killewald 2017; Solon 1992; Zimmerman 1992). Because the demand for and ability to provide help may vary as a result of differences in both the availability of kin and in the socioeconomic characteristics of these kin spanning multiple generations, the family safety net may be substantially weaker for Blacks than Whites in the United States.

Although both kin and non-kin may provide social support, survey respondents are more likely to report that they would rely on family members for help during emergencies or other difficult circumstances (Antonucci 1990; Hogan and Eggebeen 1995). However, there are substantial race differences in attitudes about intrafamilial support. Between adult children and the elderly, Blacks are more likely to anticipate support from each other than Whites, and have greater expectations for financial help from their children (Burr and Mutchler 1999; Lee, Peek, and Coward 1998). Past research on actual help backs this notion of racial variation in familial support. Blacks are less likely to receive monetary assistance from their parents than Whites, and more likely to provide financial help to their parents than Whites, even at younger ages (Jayakody 1998; Park 2017). Furthermore, Black families are more likely to be enmeshed in kin networks where financial help is one-directional, although it may be reciprocated with non-monetary support (Sarkisian and Gerstel 2004). These studies have also established that differences in family structure―as well as the socioeconomic characteristics of family members―are important determinants of who, within their families, individuals view as reliable sources of help, and conversely, whether any family members would look to them for help. In this section, we review the roles of demography and socioeconomic status in the composition of an individual’s latent family safety net.

Who is in the family?

An individual’s family safety net depends on who is in the family (Cherlin and Seltzer 2014). Yet, the structure of the family has undergone significant change since the start of the twentieth century. Declining fertility has resulted in fewer individuals in a given generation, or fewer “horizontal” ties. In contrast, gains in life expectancy have increased the overlap in “shared lives,” contributing to the growth of multigenerational families (Bengtson 2001). The total years of adult life lived with at least one surviving parent has been steadily increasing (Watkins, Menken, and Bongaarts 1987), and the fraction of individuals with both living parents and children has also increased (Wiemers and Bianchi 2015). From 1960 to 2000, the proportion of individuals with at least one grandparent alive at age 30 rose from 51 percent to 76 percent, and is projected to be 82 percent by 2020 (Uhlenberg 1996; Uhlenberg 2005). The restructuring of multigenerational families to have more vertical ties and fewer horizontal ties has altered the range of family members from whom individuals may ask for help and to whom they expect to provide help if needed.

Race differences in rates of fertility, nuptiality, and mortality contribute to racial variation in the availability of kin (Seltzer and Bianchi 2013; Umberson et al. 2017; Verdery and Margolis 2017). Blacks have higher levels of fertility and have their first child at younger ages than Whites (Martin et al. 2015). Both of these factors increase the number of vertical ties between generations. However, Blacks have lower rates of marriage, and higher mortality than their White counterparts, which decrease the number of vertical ties between generations (Szinovacz 1998; Xu et al. 2015). These differences in fertility, nuptiality, and mortality are likely to increase the availability of younger kin but shrink the availability of older family members for Blacks relative to Whites.

The impact of race differences in demographic processes across multiple generations on the availability of kin is difficult to measure because most surveys cover too short a time period or lack questions about whether extended family members are still alive. Simulation approaches make assumptions about demographic processes to project the size and characteristics of multigenerational families and kinlessness into the future (Verdery and Margolis 2017; Wachter 1997). Formal demographic models also can be combined with regression analysis of long-running survey data to understand how demographic processes interact with intergenerational processes, such as the intergenerational transmission of educational attainment (Maralani 2013; Song and Mare 2017). Our analysis complements simulation approaches by examining kin availability in the long-running Panel Study of Income Dynamics (PSID), which includes in the sample multiple generations of the same family as well as reports about family members who are not, by design, part of the PSID sample.

Who can help?

The effectiveness of the family safety net depends on one’s family members’ ability to help. The resources that individuals draw on to provide material help to family members can come from more permanent sources of socioeconomic advantage such as education and homeownership. But, the ability to help family members also requires stability in one’s own current economic situation, such as not experiencing unemployment or not having highly unpredictable work hours. We explain how both permanent resources and economic instability contribute to the efficacy of the family safety net below.

Permanent resources

Permanent sources of socioeconomic advantage such as a college degree, homeownership, and wealth, are important predictors of providing material support to kin. Higher educational attainment is positively correlated with earnings. Additionally, the prevalence and amount of financial transfers given to family members increase with the income and the educational attainment of the donor. Similarly, homeownership among previous generations of the same family lineage is highly correlated with homeownership among subsequent generations through the direct transmission of accessible resources (Prix and Pfeffer 2017). Likewise, parents’ accumulation of wealth through homeownership also increases the likelihood of children attending college (Lovenheim 2011).

Clearly, one’s socioeconomic assets may help family members through the direct transmission of capital or material resources. However, individuals’ socioeconomic resources influence their kin through indirect social mechanisms as well. For example, educated family members provide intangible benefits to others in the extended kin network. Not only can older family members pay for schooling, but kin may communicate critical information about navigating the academic system to prepare the younger generation for college (Crosnoe 2004; Lareau 1987). Family members also assume important roles in socializing younger kin, establishing norms about educational aspirations (Crosnoe 2004) that contribute to higher rates of postsecondary enrollment (Sandefur, Meier, and Campbell 2006). This intergenerational transmission of social and cultural capital surrounding education heavily disadvantages individuals from families who do not possess a legacy of college-educated kin, such as African Americans (Gosa and Alexander 2007; Waite 2009). Family members’ homeownership also may shape expectations about standards of living for future generations, spurring individuals who were raised by home-owning parents or exposed to family who owned homes to behave similarly (Henretta 1984; Prix and Pfeffer 2017).

Prior research provides strong evidence of Blacks’ lower levels of permanent resources compared to Whites. Blacks face persistent achievement gaps in education (McDaniel et al. 2011), and while the proportion with a bachelor’s degree or higher has been increasing for all race/ethnic groups, in 2015, only 22 percent of Blacks had a bachelor’s degree or higher, compared to 36 percent of non-Hispanic Whites (Ryan and Bauman 2016). A majority of Blacks are also the first in their families to be homeowners (Charles and Hurst 2002; Gosa and Alexander 2007; Shapiro 2006). In fact, Wolff (2014) finds that Whites’ rates of homeownership ranged from 68 percent to 75 percent since the 1980s, while only 44 percent to 49 percent of Blacks owned homes. Even among homeowners, non-Whites are more disadvantaged than Whites because the former are more likely to lose their homes. Shapiro, Meschede, and Osoro (2013) find that minorities are twice as likely to receive risky mortgage products and have their homes go into foreclosure.

Economic instability

Stability in one’s current economic resources― such as continuous employment or steady hours of employment―increases the ability to help family members and decreases reliance on kin for help. The converse is also true: economic instability increases an individual’s need for family support and reduces the ability to provide support (Schoeni 1997). In fact, financial transfers often occur in response to short-term income fluctuations, such as when individuals experience economic shocks due to job loss or reductions in full-time employment (McGarry 2016; Park 2017). Unemployment reduces earnings and increases uncertainty and strain in meeting basic financial needs. The irregularity of work schedules, especially among non-salaried employees, also contributes to earnings volatility. Thus, financial uncertainty plays a pivotal role in the activation of the latent family safety net.

Economic instability is much higher among Blacks than Whites. Compared to Whites, Blacks’ unemployment rates are approximately double that of Whites, and among those who work, Blacks have lower annual earnings (Couch and Fairlie 2010; Western and Pettit 2005). Black men also have much more unstable work schedules than White men. For example, McCrate (2018) finds that Blacks were more likely to be vulnerable to variable scheduling practices, to be apprised of their hours with little advance warning, or to be sent home early.

The degree of economic stability in one’s family may also transmit financial advantages through indirect mechanisms. A stably employed family member can serve as a point of contact in securing a job in the same industry or even the same employer through referrals (O’Regan and Quigley 1993; Tassier and Menczer 2008). Because Blacks tend to have less random, more overlapping social networks than Whites, it is more likely that Blacks’ networks exclude individuals who would be able to provide helpful employment information and include those with more precarious work histories (Granovetter 1983; Patterson 1998).

Studying economic instability provides insight into another dimension of (dis)advantage beyond education or homeownership that may affect inequality in family support. Although numerous articles examine the effect of work histories of older generations on the eventual joblessness of younger generations (Couch and Dunn 1997; Ekhaugen 2009), to our knowledge, there is no research on contemporaneous unemployment across generations. Our paper exploits newly-available data to examine race differences in the extent to which economic disruption due to unemployment characterizes family networks, including both coresident and non-coresident kin.

Looking beyond two generations

Most previous research on the degree of presumed familial support focuses on the parent-child dyad. This work underscores the strong intergenerational correlations in income, education, and wealth (Blau and Duncan 1967; Charles and Hurst 2003; Mazumder 2005; Pfeffer and Killewald 2017; Solon 1992; Zimmerman 1992). Taken together, Blacks’ historical socioeconomic disadvantage relative to Whites from an intergenerational perspective suggests that Blacks’ multigenerational family safety net also may be more tenuous than Whites’.

In fact, recent research substantiates the view that extended family members can be a source of support, providing financial resources, time, housing assistance, and emotional help to benefit individuals’ well-being throughout life (Bianchi et al. 2008; Seltzer and Bianchi 2013). In particular, grandparents are often involved in the care and rearing, and sometimes the financial support, of their grandchildren (Ellis and Simmons 2014; Harrington Meyer 2012; Luo et al. 2012). Additionally, emerging scholarship has investigated the influence of extended kin on the socioeconomic well-being of later generations. Both Song (2016) and Chan and Boliver (2013) find a positive association between grandparent and grandchild socioeconomic statuses in the United States and the United Kingdom, respectively, even after controlling for parents’ characteristics. Likewise, Pfeffer and Killewald (2017) find that grandparents’ wealth is associated with grandchildren’s wealth, net of parents’ wealth. In the other direction, grandchildren may aid grandparents as well by providing financial or time-based help (Sheehan and Petrovic 2008). Although empirical studies beyond three generations remain limited, ethnographic research on black families finds that great-grandparents play a role similar to that of grandparents in providing financial and instrumental help to younger generations, and receiving help from younger generations late in life (Burton 1996).

Our study also adopts a broader perspective in examining potential familial support among Blacks and Whites by extending the family to include multiple living generations. We consider both kin availability and a range of socioeconomic characteristics of adult family members across generations to describe race differences in the efficacy of the family safety net. Furthermore, in addition to traditionally-accepted measures of socioeconomic status such as college attainment and homeownership, we introduce a measure of economic instability―unemployment experienced concurrently among family members―to assess a new dimension of race differences in the strength of the family safety net.

Data

The PSID is commonly used for multigenerational analyses because it allows for the direct observation of family ties across several generations. The PSID began in 1968 with a national sample of roughly 18,000 people in over 5,000 households. Its genealogical design follows the original 1968 sample members and their descendants as they form their own households. The original sample members have the “PSID gene” and the biological and adopted children of individuals with the “PSID gene” are followed along with their offspring.

Our analysis combines data from multiple PSID sources: the 2013 Parent Identification File (PID), the 2013 Rosters and Transfers (R&T) Module, and the 2013 core interview. The PID links parents and children in the PSID. It summarizes the information on parent-child relationships gathered from the PSID from several sources: fertility histories for most respondents of childbearing age (summarized in the Childbirth and Adoption History File), the 1988 Time and Money Transfer File which obtained information about parents and parents-in-law without the “PSID gene” who had never lived in the household, interviewer notes about parentage, and information on parents in the Child Development Study and child support sequences. The PID also includes information from the R&T Module, which asked all 2013 respondents and their spouses/partners the names and characteristics of parents, stepparents, parents-in-law, children, and stepchildren age 18+, irrespective of whether the individual lived in the same household as the respondent. Previously, information about stepparents, nonresident biological and stepchildren, nonresident parents, and parents-in-law was limited, because the PSID does not include information about the characteristics of these kin unless they live with a sample member or are themselves a sample member.

We use the R&T Module to obtain demographic and socioeconomic characteristics of family members. The R&T Module is extremely useful for augmenting vertical generational ties with their associated characteristics. However, it does not augment family ties within a generation such as siblings or cousins with their demographic and socioeconomic traits. For this reason, in what follows, we focus exclusively on family members linked through vertical ties and refer to these as multigenerational families. All references to generations refer to vertical generational ties. The 2013 core interview is the source of information on race and is also used to augment generational links using codes on relationships within the household that are not included in the PID. We focus on relationships between biological/adoptive kin who are alive and responding or reported about in the R&T Module in the 2013 PSID.

Strengths and limitations of the PSID

The PSID is unique in its inclusion of information on multiple generations across the life course. It is the premier US survey on households’ and individuals’ socioeconomic well-being and on intergenerational ties among adults of all ages. Its wave-to-wave response rates are the highest of any panel survey in the world (Schoeni et al. 2013). Weights are provided that correct for attrition. Using these weights, the PSID has been shown to produce similar estimates to contemporary samples of US households including the American Time Use Survey for time use behaviors, the National Health Interview Survey for health status and health behaviors, and the Current Population Survey for income (McGonagle et al. 2012) even with its genealogical design.¹

Despite these great strengths, multigenerational ties observed in the PSID are an undercount of total generational ties of PSID families for three reasons. First, for most families, only one side of the family is in the genealogical structure of the PSID. The R&T Module ameliorates this problem by enumerating parents and parents-in-law (both sides of a family) but only does so for a single generation. Second, although the R&T Module in the PSID includes step-kin relationships, the Module only identified current stepchildren and current stepparents.² Third, even with the extensive genealogical information in the PSID, it is not possible to comprehensively estimate the total number of living relatives of PSID respondents because of attrition. For example, if a family member attrits from the survey and then subsequently dies, his or her death is never recorded in the PSID. Similarly, if a child exits from the survey prior to having children of his/her own, these children would not be recorded in the PSID. Although wave-to-wave response rates are very high, cumulative attrition over the entire 45-year history of the PSID was about 38 percent as of 2009 (Schoeni and Wiemers 2015). As with many other long-running studies, there are higher attrition rates among younger respondents from more disadvantaged backgrounds in the PSID. The total impact of attrition on estimated kin counts is unknown.

We significantly reduce the impact of attrition on our estimates with data from the R&T Module. It captures the information about the mortality and fertility of family members who have attrited for one older generation and one younger generation (18+). When successive generations are themselves respondents, the R&T Module allows us to fill in missing information due to attrition, but grandparent and great-grandparent mortality and child and grandchild fertility information remain incomplete.

Although the R&T data ameliorate the primary weaknesses of the PSID design, we cannot address two problems. The PSID data do not allow us to fully account for horizontal family members such as siblings, nor do the data allow us to document the existence and characteristics of former stepkin. Both horizontal and former step-family members are potential sources or recipients of help. In principle, ignoring these relationships will underestimate the family safety net in terms of the number of kin and the availability of resources, particularly among Blacks who tend to have more siblings and stepkin than Whites (Seltzer and Bianchi 2013). However, based on prior research, it is unclear whether these kin would in fact be considered a member of an individual’s safety net. Step-relationships tend to be weaker than those of biological relationships (Ganong and Coleman 2017), and they are less likely to be involved in intergenerational support (Pezzin, Pollak, and Schone 2008). Although full siblings typically are counted as part of the family, siblings are less likely to provide help than family members who are connected through vertical ties, such as parents and children (Kahn, McGill, and Bianchi 2011; Schoeni 1997), perhaps because socioeconomic resources among siblings are similar (Mazumder and Levine 2003). For these reasons, we believe vertical ties to be of greater importance in studying the family safety net.

Linking generations to identify multigenerational families

We anchor our analyses with the 2013 responding households whom we refer to as the zero generation, or “G0.” For households headed by a married/cohabiting couple, we examine the relatives of both the husband and wife/partner. We count consecutive generations as well as skipped generations. We count the total generations going down the younger generations consecutively, considering the number of children, grandchildren, great-grandchildren, and great-great-grandchildren. We refer to these as G1, G2, G3, and G4, respectively. We refer to these four younger generations collectively as “lower generations.” When there are no remaining responding lower generations, we count the number of parents and parents-in-law, grandparents, and great-grandparents, whom we refer to as G-1, G-2, and G-3. We collectively refer to these three older generations as “upper generations.” These linkages are shown by the white arrows in Figure 1.

FIGURE 1.

Linking consecutive and skipped generations

We count nine possible skipped generations, that is, generations missing a middle tie, such as a grandparent and grandchild without the middle parent generation. Skipped generations occur in the PSID if there are responding children of non-responding parents (or grandparents or great-grandparents). The skipped generations we consider are shown by the black arrows in Figure 1. Because we anchor our analyses with head/spouse/ partner respondents from the 2013 interview, there are only nine feasible combinations of skipped generations: G0*G2, G1*G3, G0*G3, G0*G4, G1*G4, G2*G4, G-2*G0, G-3*G0, G-1*G-3. We update the total number of generations initially calculated from the consecutive counts to include those generations added with the consideration of skipped generations.

Analytic samples and measures

We restrict our analyses to single-race non-Hispanic White and single-race non-Hispanic Black families, where the family-level racial classification is based on the race of the adult household head and spouse/partner (if present). In the case of single-headed households, we take the head’s race. In the case of two-headed households, we keep couples where both identify as either single-race non-Hispanic White or single-race non-Hispanic Black and are in a mono-racial union.³ We exclude the immigrant refresher sample added to the PSID in 1997 because this sample does not include the long generational histories that are available for respondents in the original 1968 PSID sample. This results in 7,261 households (4,278 White; 2,983 Black), which covers over 80 percent of all 9,063 families interviewed in 2013. We use this sample of 7,261 households to first describe the distribution of black and white families in terms of the number of living generations as shown in Table 1. Then, we restrict our analytic sample to multigenerational families (families where there is more than one generation alive in 2013, who may live in or outside G0’s household) to conduct analyses of families with at least one living vertical tie. Our sample is representative of US non-Hispanic black and white households in 2013 that descend from US households in 1968. We show in Appendix Tables A-1 and A-2 that our estimates of kin counts are broadly consistent with fertility and mortality patterns in more contemporary data.

TABLE 1.

Distribution of the number of generations in White and Black PSID families (2013)

	All Ages		G0 Age: 55 years old and younge		G0 Age: Older than 55 years old
Number of generations	Whites %	Blacks %	Whites %	Blacks %	Whites %	Blacks %
1	9.0	9.3	4.4	5.8	14.4	17.0
2	22.3	25.2	26.2	28.5	17.8	17.8
3	48.8	44.7	53.5	46.7	43.4	40.2
4	19.6	19.4	15.7	17.9	24.1	22.8
5	0.3	1.4	0.2	1.1	0.5	2.3
G0 mean age	52.7	46.5	39.2	38.5	66.5	62.4
N	4,278	2,983	2,824	2,344	1,454	639
Black-White Test of difference	N.S.		*		N.S.

NOTES: Data are from the 2013 PSID. Weighted using family weights adjusted using Freedman and Schoeni (2016). N.S.: Not statistically significant at the p < 0.10 level.

^†p < .10

^*p < .05

^**p <.01

^***p < .001

For multivariate analyses, we further condition by G0’s age, as well as their kin, as described in more detail below. All analyses were conducted using svy commands in STATA 14 to adjust for the complex sample design of the PSID. We use the family weights augmented by the adjustments to PSID weights for Blacks (Freedman and Schoeni 2016) to provide robust estimates of Black-White differences in kin availability and socioeconomic resources.

We examine three indicators of socioeconomic assets/liabilities: having a college degree, owning a home, and unemployment. We conceptualize having a college degree and owning a home as measures of permanent socioeconomic resources and unemployment as a measure of transitory economic instability. Unemployment is defined as in the labor force and looking for work. Individuals who are not working but searching for work are treated as unemployed. However, individuals who are not in the labor force (retired, homemaker, or student status) are not considered unemployed. Each characteristic for individuals is taken from the 2013 core interview where possible and from the R&T Module for individuals who are not 2013 respondents.

We aggregate the socioeconomic resources in G0’s multigenerational network in two ways: (1) whether any family members in any of the generations have (do not have) each asset (liability); and (2) whether all individuals in the entire generational structure have (do not have) each asset (liability). For each of these measures, we place different age restrictions on the individuals who are counted as “in the multigenerational network.” Because educational attainment is traditionally measured among individuals who are at least age 25, our analytic sample when examining educational differences consists of family members ages 25 and older. For homeownership, our analytic sample is based on family members 18 and older. Finally, our analytic sample consists of 25–60 year-olds for unemployment.

Methods

To address the question of whether Blacks have more vertical kin ties than Whites, we report the distribution of black and white families in terms of the number of living generations, describe the prevalence of having upper and lower generations alive, and show the age structure of 2, 3, and 4+ generation families. We test for race differences using chi-square and t-tests.

To examine the differences in socioeconomic resources by race among multigenerational families, we pool the black and white samples and estimate the linear probability model:

$$I_{ic}={\beta }_{0c}+{\beta }_{1c}BlackG{O}_i+u_{ic}$$ (1)

where I_ic = 1 if kin in family i has characteristic c where c = {any family member has a college degree, any family member does not have a college degree, all family members have a college degree, no family members have a college degree, any family member owns a home, any family member does not own a home, all family members own a home, no family members own a home, any family member is unemployed, all family members are unemployed} with the age restrictions outlined above and I_ic = 0 otherwise. Black GO_i = 1 if the G0 generation in family i is Black and Black GO_i = 0 otherwise. We report predicted probabilities from Equation (1). We also estimate similar models in which we control for characteristics of the family X_i that may be correlated with I_ic and with Black GO_i as follows:

$$I_{ic}={\alpha }_{0c}+{\alpha }_{1c}BlackG{O}_i+{\alpha }_{2c}{X}_i+e_{ic}$$ (2)

where X_i includes G0’s mean age, whether G0 is partnered (married or cohabiting), G0’s total number of siblings, the total number of individuals in the lower generation, the total number of individuals in the upper generation, the mean age of lower generation, the mean age of upper generation, and indicator variables for the presence of any lower generation and the presence of any upper generation. We report predicted probabilities from these models, which we refer to as adjusted probabilities. Statistical differences between Blacks and Whites are denoted with asterisks in all tables.

Finally, we consider whether the differences in socioeconomic resources of multigenerational black and white families vary with the socioeconomic resources of G0. For example, we test whether race differences in the likelihood of having at least one family member who has a college degree are greater when G0 does not having a college degree. For this analysis, we estimate Equations (1) and (2) and include an interaction term between race and the socioeconomic characteristic in question as follows:

$$I_{ic}={\gamma }_{0c}+{\gamma }_{1c}BlackG{O}_i+{\gamma }_{2c}G{0}_{ci}+{\gamma }_{3c}BlackG{O}_i*G{0}_{ic*}+{\gamma }_{4c}{X}_i+e_{ic}$$ (3)

where I_ic = 1 if kin in family i has characteristic c where c = {any family member has a college degree, any family member does not have a college degree, any family member owns a home, any family member does not own a home, any family member is unemployed} and I_ic = 0 otherwise, and G0_ic* = 1 if G0 family i has characteristic c* where c* = {has a college degree, does not have a college degree, owns a home, does not own a home, unemployed} with the age restrictions outlined above, and G0_ic* = 0 otherwise. In these regressions c and c* need not be the same. For example, we consider whether there are race differences in the likelihood of having a least one family member with a college degree conditional on G0 not having a college degree, and the likelihood of having at least one family member without a college degree conditional on G0 having a college degree. We also consider a set of cross-characteristic regressions where I_ic = 1 if individual i is unemployed and I_ic = 0 otherwise and G0_ic* = 1 if G0 has at least one family member who either has a college degree or owns a home and G0_ic* = 0 otherwise.

Results

Race differences in the structure of multigenerational families

Table 1 shows the distribution of the number of vertical generations for black and white households in the PSID for the overall sample as well as for those ages 55 and under and those over 55. For the overall sample, the distributions look similar. The modal number of generations is 3 for both Blacks and Whites but there are slightly higher percentages of two- and five-generation families among Blacks. Differences between the distribution of number of vertical generations between Blacks and Whites are not statistically significant. However, when we condition by age, we find that for G0 who are ages 55 and younger, there are statistically significant race differences. There are more four- and five-generation families and fewer three-generation families among Blacks. The differences between Blacks and Whites in the number of vertical generations in the kin network are only statistically significant within the younger age group, which suggests the importance of controls for age in the adjusted multivariate results in Tables 4 and 5.

TABLE 4.

Socioeconomic resources of White and Black multigenerational families

	Probability			Adjusted Probability
	Whites	Blacks	White-Black difference	Whites	Blacks	White-Black difference
College degree (all family members 25+)
At least one with a college degree	0.50	0.22	0.28***	0.49	0.31	0.18***
At least one without a college degree	0.76	0.91	0 15***	0.77	0.87	−0.10***
All with a college degree	0.24	0.09	0 15***	0.23	0.12	0 11***
None with a college degree	0.49	0.77	−0.28***	0.51	0.67	−0.16***
Home ownership (all family members 18+)
At least one who owns a home	0.81	0.49	0.32***	0.79	0.56	0.23***
At least one who does not own a home	0.56	0.76	−0.20***	0.56	0.73	0 1y***
All own a home	0.44	0.24	0.20***	0.44	0.27	0 17***
None own a home	0.19	0.51	−0.32***	0.21	0.44	−0.23***
Unemployment (all family members 25–60)
At least one unemployed	0.09	0.24	0 15***	0.09	0.25	−0.16***
All unemployed	0.03	0.10	0.07***	0.03	0.09	−0.06**

NOTES: Data are from the 2013 PSID. Weighted using family weights adjusted using Freedman and Schoeni (2016). Probabilities are predicted values from bivariate regressions. Adjusted probabilities are predicted probabilities from a regression including race, G0’s mean age, G0’s partnership status, G0’s total number of siblings, the total number of individuals in the lower generation in the given age category, the total number of individuals in the upper generation in the given age category, the mean age of individuals in the lower generation in the given age category, the mean age of individuals in the upper generation in the given age category, and indicator variables for the presence of a lower and upper generation. N.S.: Not statistically significant at the p < 0.10 level.

^†p < .10

^*p < .05

^**p < .01

^***p < .001

TABLE 5.

Socioeconomic resources of White and Black multigenerational families conditional on G0’s resources

	Probability			Adjusted Probability
	Whites	Blacks	White-Black difference	Whites	Blacks	White-Black difference
G0 is economically disadvantaged
At least one F.M. with a college degree conditional on G0 not having a college degree	0.38	0.18	0.20***	0.34	0.23	0 11***
At least one F.M. who owns a home conditional on G0 not owning a home	0.82	0.47	0 35***	0.78	0.51	0.27***
At least one F.M. who is unemployed conditional on G0 being unemployed	0.09	0.29	−0.20†	0.11	0.31	−0.20*
At least one F.M. who has a college degree or owns a home conditional on G0 being unemployed	0.84	0.60	0.24***	0.83	0.64	0.19**
G0 is economically advantaged
At least one F.M. without a college degree conditional on G0 having a college degree	0.63	0.80	0 17***	0.63	0.81	−0.18***
At least one F.M. who does not own a home conditional on G0 owning a home	0.60	0.80	−0.20***	0.56	0.69	−0.13**
At least one F.M. who is unemployed conditional on G0 being employed	0.05	0.17	−0.12**	0.05	0.16	−0.11**
At least one F.M. who is unemployed conditional on G0 having a college degree or owning a home	0.09	0.25	−0.16***	0.08	0.24	−0.16***

NOTES: F.M.: Family Member. Data are from the 2013 PSID. Weighted using family weights adjusted using Freedman and Schoeni (2016). Probabilities are predicted values from bivariate regressions. Adjusted probabilities are predicted probabilities from a regression including race, G0’s mean age, G0’s partnership status, G0’s total number of siblings, the total number of individuals in the lower generation in the given age category, the total number of individuals in the upper generation in the given age category, the mean age of individuals in the lower generation in the given age category, the mean age of individuals in the upper generation in the given age category, and indicator variables for the presence of a lower and upper generation. N.S.: Not statistically significant at the p < 0.10 level.

^†p < .10

^*p < .05

^**p < .01

^***p < .001

We also consider other characteristics of the extended family. Table 2 shows several descriptive characteristics of the extended family and descriptive characteristics of the G0 generation. Table 2 includes the percent of families with only lower generations, with only upper generations, and with both upper and lower generations among Blacks and Whites in families with at least two generations. For both Blacks and Whites, the most common arrangements are to have at least one upper and at least one lower generation or to have only lower generation family members. Black and white families are not statistically different from one another in the number of generations when there are no controls for race differences in age. Table 2 provides more information about race differences in the age structure of multigenerational families by the number of generational ties along with the mean age of the G0 generation. The G0 generation and the mean age of the upper and lower generational family members across generation length are all younger for Blacks than Whites. The span in mean age is also slightly smaller for Black families (for example, 26 to 62 for Black two-generation families compared to 31 to 68 for White two-generation families), and race differences in the spans are slightly larger among three- and four+-generation families. Although Black and White multigenerational families are of similar generational length and have a similar number of members, only 21 percent of Blacks in the G0 generation are married or cohabiting compared to 50 percent of Whites. As a result, white families are more likely to have an in-law relationship in the upper generations. Finally, Table 2 highlights the large race differences in socioeconomic characteristics. Black G0s are nearly three times as likely to be unemployed, and about half as likely to own a home or to have a college degree or higher.

TABLE 2.

Characteristics of G0 and vertically extended families

	Whites	Blacks	Test of difference
Characteristics of the vertically extended family
% Only lower generations	37.6	35.6	N.S.
% Only upper generations	23.0	22.4
% Lower and upper generations	39.4	42.0
# Individuals in lower generations (all)	3.4	3.5	N.S.
# Individuals in upper generations (all)	1.8	1.4	***
Average age of lower generations
2 generations	30.6	26.2	*
3 generations	18.9	15.8	***
4+ generations	18.4	16.0	*
Average age of upper generations
2 generations	68.4	61.7	**
3 generations	70.2	62.8	***
4+ generations	75.6	69.3	***
Characteristics of G0
Age of G0 (18+)	52.7	46.5	***
Whether G0 married or cohabiting (18+)	50.1	20.5	***
G0 employed (25–60)	87.1	69.5	***
G0 unemployed (25–60)	5.2	14.7	***
G0 owns home (18+)	66.1	35.8	***
G0 college or higher (25+)	44.6	17.6	***

NOTES: Data are from the 2013 PSID. Weighted using family weights adjusted using Freedman and Schoeni (2016). The percents in “only lower,” “only upper,” and “lower and upper” are among families with at least two generations. N.S.: Not statistically significant at the p < 0.10 level.

^†p < .10

^*p < .05

^**p < .01

^***p < .001

Our analysis of multigenerational ties shows that these relationships are very common. Nearly 90 percent of both black and white families in the PSID are multigenerational with at least two generations. Three generations is the most common structure of multigenerational families. Moreover, approximately 60 percent of families have at least one generation below them and 50 percent have at least one generation above them (not shown). Race differences in the multigenerational structure of families are quite small. These results stand in contrast to recent work using the PSID that suggests that a relatively large proportion of individuals have no living parents, that rates of childlessness are high, and that race differences are large (Daw, Verdery, and Margolis 2016). Because our results differ so dramatically from this recent article on kin counts in the PSID, we show in Table 3 the simple counts of parent and child relationships for individuals in the PSID by race and age, using a format similar to that used by Daw, Verdery, and Margolis (2016). Table 3 includes the distribution of the number of biological or adopted parents and children for non-Hispanic black and white individuals who are heads of household (or spouses/partners) and PSID sample members in 2013. Appendix Tables A-1 and A-2 compare our estimates from the PSID to those using other population representative data sources (the National Study of Adolescent to Adult Health, the Health and Retirement Study, and the Current Population Survey June Fertility Supplement), while additional detail describing these data and samples can be found in the Appendix. As we note below, our PSID estimates match well with those from external sources.

TABLE 3.

Observed number of kin by race and age among sample individuals in 2013

		Own parents			Bio/adopted children
	Age	0	1	2	0	1	2	3	4+
Whites
	25–34	1.0	9.7	89.3	60.4	15.3	16.0	5.3	3.1
	35–44	3.1	22.4	74.5	24.7	18.2	35.0	15.1	6.9
	45–54	15.2	42.7	42.2	21.6	18.4	38.1	16.4	5.5
	55 +	75.0	19.5	5.5	6.7	31.7	35.0	18.2	8.5
Blacks
	25–34	0.7	25.3	74.0	39.1	22.6	22.1	9.7	6.5
	35–44	8.2	40.7	51.0	24.5	21.6	22.1	20.3	11.5
	45–54	30.0	47.0	23.0	14.6	23.9	29.5	20.8	11.3
	55+	79.6	16.8	3.6	8.4	29.3	34.7	15.4	12.3
Black-White Test of difference	25–34		***				***
	35–44		***				†
	45–54		**				*
	55+		N.S.				N.S.

NOTES: Data are from the 2013 PSID. Weighted using individual cross-sectional weights adjusted using Freedman and Schoeni (2016). N.S.: Not statistically significant at the p < 0.10 level.

^†p < .10

^*p < .05

^**p < .01

^***p < .001

In Table 3 of Daw, Verdery, and Margolis (2016),⁴ they estimate that 9 percent of Whites and 23 percent of Blacks between ages 25 and 34 do not have any living biological parents. Our estimates suggest that very few Blacks or Whites age 25–34 have no living parents (0.7 percent and 1 percent, respectively) and that race differences are quite small. However, more Blacks have one living parent than Whites, so race differences at younger ages in the likelihood of having two living parents are large (89 percent for Whites versus 74 percent for Blacks). We find much lower rates of having no living parents in middle age than Daw, Verdery, and Margolis (2016).⁵ We also find that 15 percent of Whites and 30 percent of Blacks age 45–54 have no living parents compared with 25 percent and 52 percent respectively in Daw, Verdery, and Margolis (2016), but higher rates than Daw, Verdery, and Margolis (2016) of having no living parents among older individuals, which is consistent with our lower estimates of having no parents at young ages. We find that race differences in the number of living parents tend to be greater for older age groups (as differential mortality becomes more salient).

Daw, Verdery, and Margolis (2016) also estimate that childlessness is very common among young adults (72 percent of Whites and 59 percent of Blacks ages 25–34 have no children) and that childlessness remains common at older ages as well (22 percent for Whites and 29 percent for Blacks age 55 + ). We find, by contrast, that only 60 percent of Whites and 39 percent of Blacks are childless between the ages of 25 and 34, and that by older ages, childlessness is relatively uncommon (7 percent for Whites and 8 percent for Blacks).⁶ By using more recent data that better account for mortality and attrition than Daw, Verdery, and Margolis (2016) used,⁷ our kin counts show much smaller race differences in the existence of generational ties and that few individuals are isolated in the sense that they have few ties to kin in adjacent generations. Appendix Tables A-1 and A-2 show that our PSID estimates are very close to estimates from external data sources. This further supports our finding in this article that differences in the number of living generations between Blacks and Whites are small.

Race differences in multigenerational socioeconomic resources

We now consider the differences between Blacks and Whites in the socioeconomic resources of multigenerational families. Table 4 shows the predicted probability and adjusted probability from Equations (1) and (2). For easier interpretation, we also show the Black-White differences. Unadjusted for differences in family structure, 50 percent of Whites have at least one family member aged 25 or older who has a college degree compared to only 22 percent of Blacks, and these differences are statistically significant. The race differences in the adjusted probabilities are smaller but remain very large―an 18 percentage point gap in the likelihood of having at least one family member with a college degree. Whites are also 15 percentage points more likely to be in families in which all family members have a college degree and are 28 percentage points less likely to be in an extended family in which no family members have a college degree. These differences also are statistically significant.

In a similar analysis of homeownership, Table 4 shows that 81 percent of Whites compared with only 49 percent of Blacks have at least one family member who owns a home, a 32 percentage point gap. Adjusted for some demographic characteristics of the generational structure, this gap shrinks somewhat to 23 percentage points. Race differences in the likelihood of all family members owning a home are also very large (20 percentage points) and Blacks are nearly three times more likely than Whites to have families in which no kin own a home.

Finally, Table 4 shows a parallel analysis examining unemployment. Only 9 percent of white families have at least one family member who is unemployed compared with 24 percent of black families. These differences are larger than the race differences in the likelihood that G0 is unemployed shown in Table 2. About 10 percent of Blacks are in a family in which all of the family members ages 25–60 are unemployed. Whites are as likely to be in a family in which at least one person is unemployed as Blacks are to be in a family in which all family members are unemployed.

In sum, despite small race differences in the availability of multigenerational kin, there are large socioeconomic differences in the kin network. Blacks are less likely to have family members with socioeconomic assets and are more likely to have family members who have socioeconomic liabilities. As in prior work, the differences in socioeconomic resources in multigenerational families are very large and consistent across measures of socioeconomic resources. For the two measures of permanent socioeconomic resources, the differences in the likelihood of having at least one extended family member lacking the socioeconomic resource is approximately the same size as the difference in socioeconomic resources of the G0 generation. But, for unemployment, the differences among extended family members are larger than the differences in the G0 generation. This suggests that the intergenerational persistence of unemployment is greater for Blacks than Whites. These findings remain largely unchanged with and without controls for G0’s union status.

Implications for the family safety net

To further understand the implications of differences by race in the socioeconomic resources of multigenerational families for the family safety net, we consider whether there are differences by race in the prevalence of socioeconomic advantage when an individual in G0 is socioeconomically disadvantaged, as well as in the prevalence of socioeconomic disadvantage when an individual in G0 is socioeconomically advantaged. We consider not having a college degree, not owning a home, and unemployment as measures of socioeconomic disadvantage and having a college degree, owning a home, and being employed as measures of socioeconomic advantage. Table 5 shows the probability of having advantaged family members conditional on G0 being disadvantaged and the probability of having disadvantaged family members conditional on G0 being socioeconomically advantaged. We also show one additional characteristic: the likelihood of having other unemployed family members conditional on being unemployed oneself. This outcome allows us to examine directly the race differences in the multigenerational correlation in unemployment implied by Table 4.

The race differences in the likelihood of having a family member with a permanent resource, be it education or housing, conditional on not having that resource oneself are also very large. Table 5 shows that, unadjusted for differences in family structure, even among white G0s who do not have a college degree, 38 percent have at least one family member with a college degree compared to only 18 percent of black G0s, a difference of 20 percentage points. In terms of homeownership, Table 5 shows that Blacks are 35 percentage points less likely to have at least one family member who owns a home conditional on G0 not owning a home. Only half of Black G0s who do not own a home have at least one family member who does. Table 5 also shows the probability of having at least one family member who is unemployed conditional on G0 being unemployed. Twenty-nine percent of unemployed black G0s have at least one other family member who is unemployed compared with only 9 percent of white G0s. The already very large gap between Blacks and Whites in the likelihood of having an unemployed family member that we saw in Table 4 (15 percentage points) is even larger for individuals who are themselves unemployed.

Table 5 suggests a dire picture for Blacks facing unemployment insofar as they are quite likely to be part of a multigenerational family in which at least one other family member is in similar financial distress. Family members are likely to draw on their permanent socioeconomic resources when another family member is unemployed. For this reason, we show the probability that at least one family member owns a home or has a college degree conditional on G0 being unemployed. Over 80 percent of Whites who are unemployed have at least one family member with some permanent socioeconomic resources. However, only about two thirds of Blacks are in this situation. Unemployed Blacks are 24 percentage points less likely to have a family member with some permanent socioeconomic resources than their White counterparts. These differences combined with the higher likelihood that multigenerational Black families experience several unemployed spells at the same time highlight the race differences in the potential family safety net.

Another way to think about race differences in the family safety net is to consider the likelihood that an individual may have family members in need of resources. Table 5 shows the likelihood of having at least one family member without a college degree among G0 individuals who have a college degree, the likelihood of having at least one family member who does not own a home among G0s who own a home, the likelihood of having at least one unemployed family member among employed G0s, and the likelihood of having at least one unemployed family member among G0s who have a college degree or own a home. Across measures, Blacks are between 10 and 20 percentage points more likely than Whites to have a socioeconomically disadvantaged family member. Again, the differences in the employment outcomes are particularly large. Blacks are three times more likely to have at least one unemployed family member when they are employed and almost three times more likely to have an unemployed family member when they either have a college degree or own a home. Taken together, the results in Table 5 suggest that when Blacks need to draw on their families for help, there is less help available and yet, when they have socioeconomic resources, the potential for family demands on these resources is much higher, compared to Whites.

Discussion

Our analysis shows that multigenerational families are common among both Blacks and Whites, with about 90 percent of families having at least two generations and the most common structure of multigenerational families being three generations. Moreover, we find few race differences in the multigenerational structure of Black and White families. In particular, by taking advantage of the more complete accounting of parents and offspring with the data collected in the 2013 R&T Module of the PSID than is available in the PSID genealogical design alone, we find much smaller Black-White differences in the number of parent and child relationships compared to previous research (Daw, Verdery, and Margolis 2016).

By contrast, we find that socioeconomic resources within multigenerational families vary substantially by race. With respect to permanent socioeconomic resources and transitory economic instability, Blacks are much less likely to have at least one family member in their extended kin network who has access to permanent economic resources such as a college degree or homeownership, and significantly more likely to have any extended kin who do not have such resources. Additionally, Blacks are more likely to have a family member who is experiencing transitory instability, in the form of unemployment, than Whites. Black multigenerational families also are more likely to experience socioeconomic disadvantage at the same time in which they do not have any members in their vertically extended family with a college degree, or who own a home. Blacks who are disadvantaged are also more likely than Whites to have all their kin unemployed. In brief, when experiencing socioeconomic disadvantage, individuals in Black families are less likely to have a family member in their network on whom they can rely. Conversely, when experiencing socioeconomic advantage, Blacks are more likely to have someone in their family who needs help, thus facing higher and more frequent demands to draw upon their own resources to help kin.

Our findings support prior research on the higher likelihood of having financially disadvantaged family members who might need help in black families compared to their white counterparts, particularly among the middle class (Chiteji and Hamilton 2002; Heflin and Patillo 2006). Although our study does not examine actual, reported financial help provided to family members, other research confirms that middle-class Blacks are more likely to provide financial assistance compared to Whites (O’Brien 2012), and that financial help is not restricted to help provide to younger family members; rather, help also flows from younger to older generations (Goldscheider and Goldscheider 1991; Park 2017). Our results also underscore the fragile safety net among black families, consistent with prior studies showing that family members who help others are more likely to experience hardship themselves after giving a transfer (Pilkauskas, Campbell, and Wimer 2017), and that Blacks generally give lower amounts of financial help compared to Whites (Jayakody 1998).

Race differences in marriage are also relevant for understanding race differences in the multigenerational family safety net. Consistent with prior research, we find that Blacks are less likely than Whites to be married (Mare and Winship 1991; Raley, Sweeney, and Wondra 2015), which may be both a cause and consequence of Blacks’ more disadvantaged family network. Because spouses can provide security to one another during periods of difficulty, such as when one becomes unemployed (Oppenheimer 1988), unmarried individuals are particularly vulnerable, and unlikely to have the means to help others. On the other hand, possessing in-laws may increase demands for resources with accompanying economic strains for the couple. These financial tensions may contribute to marital dissolution and further weaken the family safety net (Dew, Britt, and Huston 2012; Poortman 2005). We find that race differences observed in the availability of resources are not directly due to race differences in marriage which we control for in the analysis.

Our work exploits advances in data collection efforts, such as the PSID R&T Module, that have moved us past household-based studies by identifying family members no matter where they live. However, there remains a gap in knowledge of attitudes about responsibilities to help. For example, we find that Blacks are more likely to have family members who need help. With competing obligations to help multiple kin, it is unclear how individuals decide whom to help, especially given scarce resources. New data collection should address individuals’ attitudes about helping family members and their attitudes about asking their family for help. This is especially important for historically economically disadvantaged groups such as Blacks, whose resources are significantly constrained, compared to Whites.

Our study is part of a growing literature on multigenerational family ties that considers how demographic differences and socioeconomic resources interact to create a reliable or tenuous family safety net. The size and strength of the safety net across extended families may create durable inequalities in the transmission of advantage or disadvantage across multiple generations (Hall and Crowder 2011; Mare 2011), thereby contributing to the persistence of Blacks’ socioeconomic disadvantage relative to Whites in the United States. The weaker safety net for black multigenerational families may impede the socioeconomic progress of more well-to-do Blacks as they face higher and more frequent demands to draw upon their own resources to help kin. Furthermore, the cognizance of having fewer family members on whom to rely may also play a factor in Blacks’ lower levels of risk-taking in occupational choices, lower selectivity during their job search which results in less advantageous skills matches, or more conservative financial investments which limit their socioeconomic advancement. Whether the efficacy of one’s family safety net induces these types of decisions should be examined as a possible contributor to persistent Black-White differences in economic well-being in the United States.

APPENDIX

Comparisons between PSID estimates of surviving parents and children and estimates from external data sources Benchmarking the number of own parents by race and age

To benchmark our estimates of individuals’ number of own living parents, we used the National Study of Adolescent to Adult Health (Add Health) and the Health and Retirement Study (HRS). We also considered several other population-representative surveys: the National Longitudinal Survey of Youth (NLSY) cohorts 1979 and 1997 and the National Survey of Families and Households (NSFH), but excluded them for different reasons.¹

Table A-1 summarizes the comparisons of our estimates of number of own parents by race and age to two other nationally representative surveys for two age groups: 25–34 and 55+. For the 25–34 age group, we compare our distribution of the number of parents to Add Health Wave IV because the study’s respondents were virtually the same age (24–34) at the time of the interview (2008). We find small differences between our results and Add Health for both Blacks and Whites. For the oldest age group 55+, we used the HRS (2010). For Whites, our PSID estimates are very close to the HRS estimates, with slightly higher estimates of no living parents for Blacks (79.6 percent to 72.8 percent).

Benchmarking the number of biological and adopted children by race and age

To check counts of biological or adopted children in our analytic sample, we rely on two surveys: the Current Population Survey (CPS) June Fertility Supplement (2012) and the HRS (2010).

For the 25–34 and 35–44 age groups, we compare our estimates to the CPS, but with two important adjustments. First, we limit our analytic sample from the PSID to females because the questions from the 2012 Fertility Supplement were only administered to female civilian household members aged 15–50 (estimates for the 45–54 age group are not available because of this age restriction for the Fertility Supplement). Second, we further restrict the CPS sample to women who are household heads or spouse/partners of the household head for comparability to our PSID sample.

Table A-2 shows that for our samples, our estimates are very similar to those in the CPS, with one exception. The percentage of childless White women aged 25–34 is almost 13 percentage points higher in the PSID (51.9 percent compared to 39 percent). This can be partially attributed to higher rates of attrition among young adults of lower socioeconomic status in the PSID (Schoeni and Wiemers 2015). Young adults of low socioeconomic status are more likely to be parents at these ages than the young adults with higher socioeconomic status who remain in the sample.

For the 55+ age group, we find small differences between our estimates and estimates from the HRS with respect to the percent of childless individuals. Greater differences exist in the number of own children among those with at least one child. We expect the counts of children from the HRS will vary somewhat from our estimates because the HRS assigns longitudinal Other Person Numbers for respondents’ children, but these identifiers were not consistent across waves. Thus, relationship codes are known to have occasionally changed, introducing measurement error in the number of own children (Campbell et al. 2014). In contrast, the PSID preloads the person numbers of previously identified family members, improving the tracking and subsequent enumeration of offspring. On the other hand, the PSID is more likely to retain parents than nonparents, which may be a function of the genealogical design. For this reason, we use the weight adjustment for analyses of multigenerational families (Freedman and Schoeni 2016).

TABLE A-1.

Benchmarking the number of own parents by race and age in the PSID to estimates from external data sources

		PSID (2013) [1]			External sources (identified in notes)				Difference between PSID and external sources
Whites		0	1	2	0	1	2		0	1	2
	Age
	25–34	1.0	9.7	89.3	0.4	13.1	86.6	[2]	0.6	−3.4	2.7
	35–44	3.1	22.4	74.5	-	-	-	[3]	-	-	-
	45–54	15.2	42.7	42.2	-	-	-	[3]	-	-	-
	55+	75.0	19.5	5.5	75.9	19.0	5.1	[4]	−0.9	0.5	0.4
Blacks
	25–34	0.7	25.3	74.0	2.8	22.3	74.9	[2]	−2.1	3.0	−0.9
	35–44	8.2	40.7	51.0	-	-	-	[3]	-	-	-
	45–54	30.0	47.0	23.0	-	-	-	[3]	-	-	-
	55+	79.6	16.8	3.6	72.8	23.5	3.7	[4]	6.8	−6.7	−0.1

NOTES:

^[1]Authors’ calculations. Weighted using individual cross-sectional weights adjusted using Freedman and Schoeni (2016).

^[2]National Study of Adolescent to Adult Health, Wave IV (2008). Weighted using individual weights.

^[3]Recent, external data unavailable for age interval.

^[4]Health and Retirement Study (2010). Weighted using individual weights.

TABLE A-2.

Benchmarking the number of own children by race and age in the PSID to estimates from external data sources

		PSID (2013) [1]						External sources (identified in notes)						Difference between PSID and external sources
Whites		0	1	2	3	4+		0	1	2	3	4+		0	1	2	3	4+
	Age
	25–34	51.9	17.4	20.5	6.0	4.2	[2]	39.0	21.8	24.9	9.8	4.6	[3]	12.9	−4.4	−4.4	−3.8	−0.4
	35–44	21.2	17.7	36.0	17.1	8.0	[2]	16.1	19.0	37.0	18.7	9.3	[3]	5.1	−1.3	−1.0	−1.6	−1.3
	45–54	21.6	18.4	38.1	16.4	5.5		-	-	-	-	-	[4]	-	-	-	-	-
	55+	6.7	31.7	35.0	18.2	8.5		12.6	13.7	35.1	21.6	17.0	[5]	−5.9	18.0	−0.1	−3.4	−8.5
Blacks	25–34	26.1	26.6	25.8	11.7	9.7	[2]	27.1	24.0	22.8	13.2	13.0	[3]	−1.0	2.6	3.0	−1.5	−3.3
	35–44	14.3	18.3	28.9	24.3	14.2	[2]	16.6	17.9	29.0	19.7	16.9	[3]	−2.3	0.4	−0.1	4.6	−2.7
	45–54	14.6	23.9	29.5	20.8	11.3		-	-	-	-	-	[4]	-	-	-	-	-
	55+	8.4	29.3	34.7	15.4	12.3		13.0	16.6	24.1	17.0	29.3	[5]	−4.6	12.7	10.6	−1.6	−17.0

NOTES:

^[1]Authors’ calculations. Weighted using individual cross-sectional weights adjusted using Freedman and Schoeni (2016).

^[2]Sample restricted to females for comparability to Current Population Survey (2012), June Fertility Supplement.

^[3]Current Population Survey (2012), June Fertility Supplement. Sample restricted to female household heads, spouses of heads, or unmarried partners of heads. Weighted using individual weights.

^[4]Recent, external data unavailable for age interval.

^[5]Health and Retirement Study (2010). Weighted using individual weights.