Fabrizio Bernardi European University Institute∗ Diederik Boertien Centre d’Estudis Demogràfics Koen Geven World Bank∗∗ Childhood Family Structure and the Accumulation of Wealth Across the Life Course Objective: The aim of this article is to document on the alternative childhood family trajectory how childhood family structure is related to the considered. A subsequent mediation analysis accumulation of wealth. of the “wealth penalty” related to the perma- Background: Childhood family structure is a nent departure of a parent from the household commonly studied determinant of child and adult during childhood points at human capital for- outcomes, but little is known about its effects mation and own family dynamics as the primary on wealth accumulation. Wealth is affected by a channels through which wealth differences are wide variety of factors, including human capi- produced; intergenerational transfers matter tal formation, family dynamics, and intergenera- rather less. tional transfers. Studying wealth therewith sheds Conclusion: Childhood family structure has light on how childhood family structure relates moderate effects on multiple different life to the accumulation of advantages across life. domains that accrue during the life course Method: Data from the 1979 National Longitu- and collectively add up to a more considerable dinal Survey of Youth (N = 7,066) are employed penalty in wealth. to document wealth differences at ages 47 to 55. Growth curve models are estimated to under- stand at what ages these differences emerge. Childhood family structure is related to a variety Results: A median wealth penalty of at least of outcomes in later life including cognitive $61,600 at ages 47 to 55 is observed for ability, educational attainment, income, part- individuals who did not live continuously with nering behavior, and psychological well-being both parents from birth to age 18, depending (Bernardi & Boertien, 2017; Cavanagh & Hus- ton, 2006; Erola, Härkönen, & Dronkers, 2012; Gähler & Palmtag, 2015; Härkönen, Bernardi, Centre d’Estudis Demogràfics, Carrer de Ca n’Altayó, & Boertien, 2017; Kiernan & Mensah, 2009). Edifici E2, Universitat Autònoma de Barcelona, 08193 In this article, we focus on an underresearched Bellaterra, Spain (dboertien@ced.uab.es). outcome that, to some extent, quantifies the ∗ European University Institute, Department of Political and accumulation of various advantages and disad- Social Sciences, Via Roccettini 9, I-50014, San Domenico, vantages across life: wealth. Wealth is affected Italy. by events experienced across a variety of ∗∗ WorldBank, Education Global Practice, 1818 H ST NW, domains in life, such as periods of unem- Office MC11-770, 20433 Washington, DC. ployment, receipt of inheritances, and costly Key Words: economic well-being, family dynamics, family accidents. Similarly, wealth at a given point structure, inequalities, intergenerational. in time is influenced by income received and 230 Journal of Marriage and Family 81 (February 2019): 230–247 DOI:10.1111/jomf.12523 Childhood Family Structure and Wealth 231 costs incurred at all different stages of life wealth penalty of at least $61,600 at ages 47 to (Semyonov & Lewin-Epstein, 2013). Studying 55 exists for individuals who did not live with wealth therefore provides a good opportunity to both parents from birth to age 18, depending on increase our understanding of how family back- the alternative childhood family trajectory con- ground (in our case childhood family structure) sidered. relates to the accumulation of advantages and Second, whereas previous studies relied on disadvantages. Similarly, studying the influence cross-sectional information (Amato & Keith, of childhood family structure represents a novel 1991), three waves of data (Lersch & Baxter, contribution to the literature on wealth, which 2015), or followed respondents until their late has received increasing attention as the distribu- 30s (Keister, 2004), we are able to follow respon- tion of wealth has become considerably more dents from their 20s until their early 50s. Using unequal during the past decades (Killewald, growth curve models, we show that the most Pfeffer, & Schachner, 2017; Piketty, 2014). important differences in wealth emerge between A few studies have reported associations ages 40 and 50. between childhood family structure and wealth Third, a recent review article by Killewald for the United States. Amato and Keith (1991) et al. (2017) made a plea for studies document- investigated differences in the net value of ing the mechanisms that connect background individuals’ homes, real estate, and businesses characteristics with wealth accumulation. In this in 1988 (hence, not considering savings, stocks, respect, we identify the following three main and other forms of wealth) depending on four routes through which childhood family structure types of parental absence experienced during could affect wealth accumulation: (a) human childhood. White females, White males, and capital formation, (b) own family behavior as Black females who experienced a parental adults, and (c) intergenerational transfers. The divorce or never lived with both parents during only two available studies that looked at medi- childhood had fewer assets when compared ators of the effects of childhood family structure with their counterparts who did. Evidence was on wealth found that the most important role less clear-cut for Black males and Hispanic is played by own partnering and childbearing respondents and cases of parental death. Keister behavior (Amato & Keith, 1991; Lersch & Bax- (2004), however, found that the greater likeli- ter, 2015). However, these studies were not able hood of experiencing a parental separation or to look at important mediators such as earnings divorce among Black individuals could partly histories and had no or very limited information explain racial differences in net worth during on transfers received. Our study of the mech- early adulthood. Although these studies on the anisms underlying the observed differences in United States suggest that childhood family wealth is more comprehensive due to the avail- structure and wealth are related, the size of ability of information on earnings and transfers. the overall “penalties” in wealth associated In our mediation analysis, we focus on the most with various childhood family structures is not common childhood family structure trajectory, known. Recent research by Lersch and Baxter namely, the permanent departure of one parent (2015) in Australia found that respondents from the household after birth and before age 18. whose parents separated or divorced before age We find that the primary channels that mediate 15 had on average 38% less wealth in adult- its observed association with wealth are respon- hood when compared with others. The extent dents’ human capital and own family dynamics, to which these findings apply to the United whereas intergenerational transfers matter less. States remains unclear, especially given that the consequences of childhood family structure for child outcomes appear to vary across countries Childhood Family Structure and the (Hampden-Thompson, 2013; Pong, Dronkers, Accumulation of wealth & Hampden-Thompson, 2003). Wealth provides many direct benefits to indi- We make three major contributions in this viduals and households such as economic secu- article. First, we provide estimates of differences rity and status (Pfeffer, 2011) and is positively in wealth around age 50 according to childhood related to well-being outcomes including health family structure in the United States. We employ (Thompson & Conley, 2016) and marriage (Eads data from the 1979 National Longitudinal Sur- & Tach, 2016). Parental wealth is also a key vey of Youth (NLSY79) and show that a median determinant of offspring’s cognitive skills and 232 Journal of Marriage and Family educational attainment (Pfeffer, 2011). Wealth stress (Biblarz & Raftery, 1999; McLanahan is therefore an important dimension of social & Percheski, 2008). Family stress related to stratification and a key channel for the trans- economic hardship has been repeatedly shown mission of advantage across generations (Spiler- to affect child development (Conger, Conger, & man, 2000). Existing studies have documented Martin, 2010). Single parents might also lack individual differences in wealth according to a the time to employ intensive parenting strategies handful of socioeconomic characteristics such that might be conducive of children’s outcomes as age, ethnicity, sex, education, and religion (Astone & McLanahan, 1991). If single par- (Killewald et al., 2017). At the same time, dif- enthood is the result of a break-up between ferences in wealth accumulation according to parents, children will need to emotionally adjust childhood family structure have been largely left to such a new situation, even though parental unstudied. union dissolution can also relieve children from A second gap in the wealth literature is the exposure to conflict in the household (Dronkers, lack of understanding of the mechanisms that 1999; Fomby & Cherlin, 2007). Many of the connect background characteristics to wealth challenges posed by single parenthood can pos- (Killewald et al., 2017). Previous research has sibly be minimized through, for instance, joint shown that individuals save and accumulate custody, public child care, and policies aimed wealth until at least age 60, after which wealth at equalizing living conditions across families accumulation declines (Killewald et al., 2017). (Bjarnason & Arnarsson, 2011; Christopher, To understand the mechanisms underlying England, McLanahan, Ross, & Smeeding, 2001; wealth differences, one therefore has to consider Pong et al., 2003; Hampden-Thompson, 2013). processes up to advanced adulthood. In the sub- Stepparents could also to some extent relieve sequent sections we discuss the following three pressures on single parents’ tasks, but their pres- main channels that reflect the different ways ence in the household might also require new through which individuals have been shown to emotional adjustments to be made by children accumulate wealth in previous research (Semy- (Hetherington & Kelly, 2002; Sweeney, 2010; onov & Lewin-Epstein, 2013; Vespa & Painter, Thomson, Mosley, Hanson, & McLanahan, 2011, Zagorsky, 2005): (a) human capital forma- 2001). tion and labor income, (b) own family behavior A large body of research has documented as adults, and (c) intergenerational transfers. that children who continuously lived with both biological parents throughout childhood tend to have more cognitive and noncognitive Human Capital Formation and Own Income skills when compared with children who did Generation not (Amato, 2010; Härkönen et al., 2017), Individuals can generate wealth through saving even though it has been questioned whether income gained in the labor market or through this association reflects the causal effect of investing (Semyonov & Lewin-Epstein, 2013; parental absence (McLanahan, Tach, & Schnei- Spilerman, 2000). If human capital is the main der, 2013). Childhood family structure is also factor determining labor market income (Farkas, associated with final educational attainment. 2003), and the basis for human capital devel- Individuals who did not continuously live with opment is built in childhood (Heckman, 2000), both biological parents are less likely to com- childhood experiences are likely to matter for plete high school and to enroll in and complete income generation and wealth accumulation. college (holding grades constant), partly due The lives of children not coresiding with both to differences in parents’ financial resources biological parents differ in some salient aspects (Bernardi & Boertien, 2016; Thomson, Hanson, from those living in other types of family & McLanahan, 1994). structures. Children with only one coresident Cognitive ability, noncognitive skills, and biological parent might have less contact with educational credentials can thus affect wealth the nonresident biological parent, and single par- accumulation through their effects on labor ents often have to take on the task of day-to-day market earnings. They can also affect the accu- parenting by themselves. In many cases, this mulation of wealth more directly through the means that single parents, usually mothers, have management of economic resources because reduced levels of employment and earnings, consumption patterns, saving rates, and invest- which may lead to poverty or other economic ments are likely to differ according to skill levels Childhood Family Structure and Wealth 233 (Ameriks, Caplin, & Leahy, 2002; Lusardi & its effect on individuals’ own relationship stabil- Mitchell, 2007). ity. The break-up of a marriage can affect wealth through direct divorce costs, loss of access to a partner’s wealth, and an elimination of scale Family Dynamics advantages associated with living in a couple (McManus & DiPrete, 2001, Zagorsky, 2005). Wealth can be accessed through partnering a (future) wealthy person, too. Family dynamics such as cohabitation, marriage, and separation Intergenerational Transfers are therefore events of interest for our topic. Pre- A last major channel through which individu- vious studies have shown that family dynamics als obtain wealth is transfers of third parties, of individuals are related to the family dynam- primarily individuals’ own parents (Avery & ics that they experienced during childhood. First, Rendall, 2002; Rauscher, 2016; Semyonov & individuals who experienced a parental divorce Lewin-Epstein, 2013). Such transfers can be (Cherlin, Kiernan, & Chase-Lansdale, 1995) or inheritances, which normally take place in later did not live with both biological parents dur- life once parents pass away or in the form of ing childhood (McLanahan & Bumpass, 1988) financial help while parents are still alive. A are more likely to leave the parental home ear- common way for parents to assist their offspring lier due to friction, to form a union, and to in such a way is to pay for college tuition (Con- have children at younger ages. Second, individ- ley, 2001; Rauscher, 2016), which could lead to uals who experienced a parental divorce are less debt differences among college attendees. likely to marry and therefore more often miss The transfer of wealth from parents to their out on gaining wealth through partnering (Erola offspring is determined by the total amount of et al., 2012; Wolfinger, 2005). Besides the direct wealth parents have, the number of persons access partnering creates to wealth of the partner, wealth has to be shared with, and relational several studies have documented the existence of factors that affect the likelihood of receiving an individual “wealth premium” related specifi- (a given amount of) parental wealth. Just as cally to marriage. The accumulation of wealth own family dynamics, parental family dynam- happens at higher rates among married individ- ics affect the wealth of parents, possibly reduc- uals when compared with cohabiting individu- ing the total amount of wealth available to be als, possibly due to tax benefits, higher wages, transferred to offspring. Parental separation and or increased incentives to save (Lersch, 2017; divorce can have direct effects on parents’ stock Vespa-Painter, 2011). On top of the lower prob- of wealth (Avellar & Smock, 2005; Zagorsky, ability to marry, individuals with divorced par- 2005). Moreover, if parents remarry, new part- ents who do marry are more likely to form a ners (and their children) could be entitled to part union with lower educated individuals and hence of the nonresident parents’ resources (Manning gain access to less wealth upon partnering (Erola & Smock, 1999), reducing the wealth available et al., 2012). for individual offspring (Keister, 2004). Finally, Besides its influence on family formation, individuals might not have (close) contact to a childhood family structure can also affect union parent with whom they did not live continuously dissolution. In this regard, the intergenerational in the same household (Kalmijn, 2015; Manning transmission of divorce has particularly received & Smock, 1999; Peters & Ehrenberg, 2008) and attention. Individuals whose parents divorced therefore become less likely to receive financial or who lived with a single parent have been help from that parent (Pezzin & Schone, 1999). found to have higher chances of divorcing them- selves (Kiernan & Cherlin, 1999; McLanahan & Bumpass, 1988; Wolfinger, 2005). The intergen- The Accumulation of Wealth Across Time erational transmission of divorce could reflect Considerable wealth differences might also shared background characteristics including atti- emerge without a large direct impact of the tudinal factors, but could also be due to par- aforementioned groups of mechanisms. Wealth ents giving an example to children that divorce has the property to accumulate by itself, as is a workable solution to relationship problems returns to wealth increase with its amount (Wolfinger, 2005). Childhood family structure (DiPrete & Eirich, 2006). Higher interest on trajectories can therefore affect wealth through large amounts of savings is the clearest example 234 Journal of Marriage and Family of such different returns, but returns on invest- attrition. In the second stage of the analysis, we ments also appear to depend on the size of excluded respondents who were born to a single the investments made (Piketty, 2014). It could parent, never lived with a parent, experienced therefore be that the aforementioned processes a temporary separation of parents, or of whom create relatively small differences in wealth a parent died before age 18 (excluding 1,244 during early adulthood, but that due to returns cases). In the third dynamic stage of the analysis, to wealth these evolve into large differences as we included all cases that provided information time passes. on net worth at least once between ages 30 and 50 (excluding 2,511 of the original 12,686 cases; 1988 was the first year in which net worth was Method measured) and again concentrated on permanent We used data from the NLSY79, which follow parent–child separations that occurred after a representative set of adolescents since they birth (excluding 1,776 cases, leading to 8,399 were aged 14 to 22 years. By 2012 (the last individuals providing 59,890 person-waves of wave included), the participants were between information). 47 and 56 years old (https://www.nlsinfo.org/ content/cohorts/nlsy79). Our analysis proceeded Measures in three stages. First, we estimated the over- all “wealth penalty” in 2012 related to vari- The dependent variable of the study was total ous childhood family structures as compared household net worth of respondents in dollars to having lived continuously with both parents. (inflation adjusted to 2012 levels). “Net worth” This initial comparison of nine different child- represents the sum of all assets (i.e., cars, hous- hood family structures (which we describe in ing, pension, stocks, etc.) minus debts (i.e., detail in the next section) gave insight into credit card loans, mortgages, etc.). In 2012, whether wealth penalties were related to spe- the mean total net worth of households was cific family structures (or transitions) or whether $354,999, whereas the median was $128,000. they held for alternative family structures more Natural logarithm transformations are often used generally. Second, we performed a mediation to account for the skewed distribution of income analysis for these wealth differences to inves- and wealth variables. However, log transforma- tigate the potential mechanisms underlying the tions require excluding negative and zero values. relationship between childhood family structure Studies on net worth have therefore often used and wealth accumulation. It would have been IHS transformations (Killewald et al., 2017; Ler- too ambitious, however, to thoroughly study sch & Baxter, 2015). We used Inverse Hyber- how wealth differences are produced for each bolic Sine (IHS) transformed wealth with a scale type of childhood family structure considered parameter of 1, which is the most similar to in the first part of the analysis. In our media- a logged transformation (Friedline, Masa, & tion analysis, we therefore focused on the most Chowa, 2015; see also online Appendix A; the common childhood family structure trajectory: results were robust to using a scale parameter the permanent departure of a parent from the of 0.0001, which smoothens the distribution for household during childhood (also referred to as values around 0). parent–child separations). Third, we switched The main independent variable was respon- to a dynamic analysis to investigate at what dents’ family structure trajectory experienced ages wealth differences emerge and how they from birth until age 18 based on nine categories. accrue across the life course (data are used for We distinguished among those who had (a) lived the years 1988–1990, 1992–1994, 1996, 1998, with both biological parents from birth to age 18; 2000, 2004, 2008, 2012). (b) lived with both biological parents at birth, The original NLSY79 sample included experienced a permanent moving out of a par- 12,686 cases. In 2012, 7,301 respondents par- ent before age 18, but never lived with a step- ticipated in the survey. The cross-sectional parent; (c) lived with both biological parents at analysis was restricted to these individuals, who birth, experienced a permanent moving out of were also required to have provided information a parent, and lived with a stepparent; (d) lived on net worth in 2012 (excluding a further 235 with one biological parent at birth, but never cases), leading to a sample of 7,066 individuals. with both biological parents or a stepparent; (e) We used sample weights to correct for panel lived with one biological parent at birth, never Childhood Family Structure and Wealth 235 with both biological parents, but lived with a partner; (k) second or higher order marriage, stepparent; (f) did not live with either parent at higher educated partner. A second variable birth; (g) lived with both biological parents at combined parenthood and age at first childbirth birth, but experienced a temporary parental sep- (no children in 2012, had a child before age 25, aration; (h) lived with one biological parent at had a child after age 25), as children might be birth but with both biological parents at some a major cost influencing (the timing of) wealth point; (i) experienced a parental death by age accumulation (Keister, 2004). In the dynamic 18 (all cases that fulfilled this condition were part of the analysis, the variable indicated included in this category). Note that the NLSY79 whether the respondent had a child at the time only provides information on coresidence. We of measurement. could therefore not distinguish between types Finally, our measure of transfers was the nat- of parental union (married or cohabiting) and ural log of the sum of any money transfers whether parents moved out due to separation, received by respondents from other individu- divorce, or other motives. als up to the year of measurement. We also The first channel of mediating variables cap- controlled for whether a parent had died by tured human capital and own income genera- the time of measurement to capture differences tion. Main components of human capital are in expected inheritances. In additional analysis, cognitive ability, noncognitive skills, and educa- we looked at the influence of whether someone tion (Heckman, 2000). We used Armed Forces else paid (partly) for the respondents’ college Qualification Test (AFQT) test scores to cap- attendance, but this variable had no explana- ture cognitive ability (Measured in 1980) and tory power beyond the ones already included in included the following two measures of noncog- the analysis. In all of the models, we controlled nitive skills: the Rotter Internal-External Locus for ethnicity, number of siblings, maternal and of Control Scale based on four items, such as paternal years of education, own age, mother’s “What happens to me is my own doing” (mea- age at birth, and religion. We also looked at the sured in 1979) and a scale indicating various role of physical and mental health (using the 12 risky behaviors of respondents from 1979 to Item Short Form Survey health scales), which 1984 (based on 15 questions regarding drug use, has been shown to be an important determinant illegal and violent behavior; Heckman & Kautz, of wealth (Thomson & Conley, 2016). Table 1 2012). We employed a categorical measure of displays descriptive statistics and the percentage the highest education completed by the respon- of the 7,066 cases that had missing information dent (no qualifications, GED, high school, some on each variable. We multiply imputed data for college, degree). To monitor the overall impor- missing information on all independent variables tance of the channel of human capital we also of the analysis. (A total of 20 imputations were included a measure of the logged average wage created; all variables of Table 1 were included in income of the respondent across measurement the multiple imputation procedure, and this pro- points (average across 1982–2012). In the lon- cedure included cases with missing information gitudinal analysis, the latter variable took on the on net worth, too. These cases were subsequently average income recorded between 1982 and the dropped during the analysis [Johnson & Young, year of measurement. 2011]; a robustness check retaining these cases The second channel of mediating variables is displayed in Model 5 of online Appendix B.) captured the family dynamics of respondents in adulthood. These included a measure of respondents’ partnership history: (a) never Analysis married; (b) separated; (c) widowed; (d) never The analysis proceeded in several steps. We married, cohabiting with a lower educated first established overall differences in house- partner; (e) never married, cohabiting with a hold net worth in 2012 using Ordinary Least college educated partner; (f) first marriage, Squares (OLS) regressions and unconditional lower educated (less than college) partner; (g) quantile regressions at the median. We subse- first marriage, college educated partner; (h) pre- quently focused on having experienced a per- viously married, now cohabiting with a lower manent departure of a parent from the house- educated partner; (i) previously married, now hold during childhood and performed a media- cohabiting with a college educated partner; (j) tion analysis following the approach of Kohler, second or higher order marriage, lower educated Karlson, and Holm (2011; khb command in Stata 236 Journal of Marriage and Family Table 1. Descriptive Statistics (Weighted) of the Sample Used in 2012 N = 7,066 Variable Mean SD Minimum Maximum % missing Net worth in 2012 (× 1000) 354.9 690.2 −1950 3691 0a Lived with both parents throughout childhood 0.66 0 1 5.4 Permanent moving-out of parent, no stepparent ever 0.11 0 1 5.4 Permanent moving-out of parent, stepparent ever 0.08 0 1 5.4 Born to single parent, no stepparent ever 0.02 0 1 5.4 Born to single parent, stepparent ever 0.02 0 1 5.4 Never lived with biological parents 0.01 0 1 5.4 Parent died before age 18 0.07 0 1 5.4 Biological parent moved out and back in 0.01 0 1 5.4 Biological parent moved in after birth 0.004 0 1 5.4 Male (ref. female) 0.51 0 1 0 Black (ref. White) 0.14 0 1 0 Hispanic 0.07 0 1 0 Education of the father (years) 11.8 3.6 0 20 14.9 Education of the mother (years) 11.6 2.8 0 20 6.3 Maternal age at birth 27.0 6.3 15 55 4.8 Own age at interview 51.6 2.3 47 56 0 Raised other Christian (ref. no religion) 0.28 0 1 0.3 Raised Baptist 0.24 0 1 0.3 Raised Catholic 0.32 0 1 0.3 Raised Jewish 0.01 0 1 0.3 Raised other religion 0.11 0 1 0.3 Number of siblings (top coded at 10) 3.3 2.2 0 10 0.2 Cognitive ability (AFQT test) 0.0 1.0 −1.5 2.0 3.8 Risky Behavior scale 0.0 1.0 −0.7 10.3 10.6 External Locus of Control Scale 0.0 1.0 −1.9 3.0 0.9 GED (ref. no qualification) 0.11 0 1 0 High school 0.34 0 1 0 Some college 0.36 0 1 0 College degree 0.12 0 1 0 Ln(average wage income 1982–2012) 10.2 1.4 0 12.7 0 Does not have children in 2012 0.20 0 1 0.2 Had a child before age 25 0.43 0 1 0.2 Had a child after age 25 0.37 0 1 0.2 Not partnered, never married in 2012 0.10 0 1 0.3 Cohabiting, lower educated partner, never married 0.01 0 1 0.3 Cohabiting, higher educated partner, never married 0.01 0 1 0.3 No partner, widowed in 2012 0.02 0 1 0.3 No partner, separated/divorced in 2012 0.21 0 1 0.3 First time married, lower educated partner 2012 0.18 0 1 0.3 First time married, higher educated partner 2012 0.25 0 1 0.3 Cohabiting, lower educated partner, previously married 0.03 0 1 0.3 Cohabiting, higher educated partner, previously married 0.01 0 1 0.3 Second time married, lower educated partner 2012 0.09 0 1 0.3 Second time married, higher educated partner 2012 0.10 0 1 0.3 Ln(transfers received up to 2012) 5.0 4.9 0 15.7 0 A parent died by 2012 0.12 0 1 0 Mental health at age 40 0.1 1.0 −5.0 1.9 1.9 Physical health at age 40 0.0 1.0 −5.4 2.3 1.9 ref. = reference; AFQT = Armed Forces Qualification Test; Ln = natural logarithm. a Refers to percent of final sample, of the 7,301 cases interviewed in 2012, 2.0% provided no info on net worth. Childhood Family Structure and Wealth 237 [StataCorp., College Station, TX). This media- analysis, parental income). At the same time, tion analysis established what part of the total we lacked measures of other possible sources of effect on wealth goes through each of the three endogeneity traditionally discussed in the litera- general channels discussed previously. In the ture on parental separation. Notably, we did not final stage of the analysis, we identified the measure family conflict, which might affect both life stages where differences in wealth emerged childhood family structure and individuals’ out- using growth curve models. Growth curve mod- comes, including wealth. Therefore, we could els are multilevel models with random intercepts not separate possible effects of childhood family and random slopes, where observations at differ- conflict on wealth from effects of childhood fam- ent points in time are nested within individuals. ily structure. Similarly, our mediation analysis We constructed time-varying versions of all our relied on associations between mediators, child- variables to fit this part of the analysis (unless hood family structure, and wealth. The role of mentioned that they were measured at a certain mediators should therefore not be directly inter- age or time only). Our baseline model explain- preted as causal but, rather, as the channels that ing net worth N for individual i in survey year j connect family structure to wealth accumulation can be expressed as the following: (independently of the actual cause). ∑ 4 Nij = 𝛽0 + 𝛽1k Ageijk + 𝛽2 Xi + uj + eij Findings k=1 We first documented the association between The variable Age was split into four equally childhood family structure and the wealth of spaced splines that allowed the growth of net respondents in 2012 (at around age 50). Model worth with age to vary across the life course 1 of Table 2 presents the results from a regular (31–35, 36–40, 41–45, 46–50, age 30 being the OLS regression, Model 2 displays results from year at which initial wealth differences are mea- an unconditional quantile regression ran at the sured). X represents a vector of time-constant median, and Model 3 displays results from an control variables. Unobserved stable differences OLS model using the IHS transformation of between individuals across years were captured wealth. Individuals who continuously lived with by the person-specific term uj. (random effect), both biological parents during childhood had and eij is a time-varying error term. Our base- more wealth as adults than those who did not. line growth curve model was estimated as a This observation held for all of the different non- multilevel, mixed-effects model (using Stata’s intact childhood family structures considered. xtmixed command). Mediating variables were For children who experienced the permanent added one by one to this baseline model. Finally, departure of a parent from the household, we we again employed the khb procedure to perform observed a median wealth penalty of $61,600 a formal mediation analysis of growth curves if a stepparent ever moved in, and of $66,600 estimated through OLS regression (see online if they never lived with a stepparent. This Appendix D). corresponds to 17% and 19%, respectively, of average wealth observed in 2012 (see Table 1), which is lower than the 38% wealth penalty Endogeneity observed for parental separations in Australia A main difficulty in research on family dynamics in earlier research (Lersch & Baxter, 2015). is endogeneity due to nonrandom selection into For other childhood family structures, wealth family structures. Disadvantaged families are penalties were in general higher, but additional more likely to be overrepresented within given tests revealed that differences in wealth among family structure types (McLanahan & Percheski, the different nonintact family structures were 2008). It could therefore be that pre-existing not statistically significant. The results using the disadvantages affect both wealth accumulation IHS transformation were similar to those from and childhood family structure. We addressed the quantile regression. Additional analysis this possible source of endogeneity by including (online Appendix B) showed that these results measures of pre-existing socioeconomic disad- were also robust to using listwise case deletion, vantage (parental years of education, race, reli- were similar for different ages at parental sep- gion, mother’s age at birth, and, in additional aration, and did not change once considering 238 Journal of Marriage and Family Table 2. Models Explaining Net Worth in 2012 by Childhood Family Structure Net worth Net worth in dollars in dollars IHS of net IHS of net (×1000) (OLS) (×1000) (quantile 0.5) worth (OLS) worth (OLS) Model 1 Model 2 Model 3 Model 4 Variable Coefficient SE Coefficient SE Coefficient SE Coefficient SE Family history (ref. always with both parents)a Biological parent moved −69.8* 31.9 −66.8** 14.1 −1.56** 0.34 out, no stepparent ever Biological parent moved −118.8** 29.1 −61.6** 17.3 −0.91* 0.40 out, stepparent ever Born to single parent, no −103.8* 42.9 −81.3** 24.4 −2.46** 0.67 stepparent ever Born to single parent, −149.8** 42.6 −101** 28.6 −2.55** 0.86 stepparent ever Never lived with −10.6 144.8 −123.2** 39.0 −0.94 0.80 biological parents Parent died before age 18 −70.5* 35.1 −61.9** 16.6 −1.97** 0.44 Biological parent moved −158.2** 55.0 −77.3* 38.6 −1.64† 0.97 out and back in Biological parent moved −207.3** 64.3 -132* 60.9 −4.08† 2.29 in after birth Biological parent moved −1.32** 0.27 out permanently (ref. always living with both parents) Male (ref. female) 13.4 20.1 −12.4 8.5 0.23 0.19 0.27 0.20 Black (ref. White) −146.6** 16.3 −135.8** 10.5 −3.22** 0.28 −3.45** 0.31 Hispanic 4.5 28.7 −62.3** 14.1 −1.45** 0.34 −1.79** 0.36 Education of father 23.5** 3.8 8.2** 1.7 0.10** 0.04 0.10* 0.04 (years) Education of mother 27.9** 4.8 10.6** 2.1 0.17** 0.05 0.18** 0.05 (years) Maternal age at birth 6.8** 1.8 4.4** 0.7 0.06** 0.02 0.05** 0.02 Own age at interview 17.7** 4.6 8.4** 1.9 0.15** 0.04 0.14** 0.04 Raised other Christian 34.2 50.9 32.3 22.9 0.50 0.52 0.64 0.55 (ref. no religion) Raised Baptist −39.6 46.6 −15.2 22.8 0.34 0.53 0.32 0.57 Raised Catholic 41.0 50.6 52.2** 22.8 0.67 0.51 0.78 0.55 Raised Jewish 408.6** 173 73.7 42.2 0.85 1.03 0.71 1.08 Raised other religion 12.7 57.1 14.4 25.0 0.04 0.58 −0.13 0.62 Number of siblings −15** 4.5 −9.5** 2.1 −0.12* 0.05 −0.10** 0.05 Constant −1261.3** 252.2 −579.8** 104.0 −1.72 2.38 −1.36 2.50 N 7, 066 7, 066 7, 066 5, 822 Note. Robust standard errors, missing data on independent variables multiply imputed. Sample weights are included. Model 2 unconditional quantile regression at the median, using Stata’s rifreg command. IHS = Inverse Hyperbolic Sine; OLS = Ordinary Least Squares Regression. a Additional analysis changing the reference category revealed no statistically significant differences in effects among the various nonintact family structures in Models 1 to 3 (available upon request). ref. = reference. † p < .1; * p < .05; ** p < .01. Childhood Family Structure and Wealth 239 wealth measured in 2008 (i.e., measured before Growth Curve Analysis of Wealth Accumulation the financial crisis). We subsequently turned to a dynamic study Model 4 of Table 2 shows the of wealth accumulation. Figure 1 displays pre- IHS-transformed wealth penalty for individ- dicted values based on a growth curve model uals who experienced the permanent departure explaining wealth, as displayed in Table 4 of a parent after childbirth and before age 18 (Model 1). IHS-transformed wealth grew (regardless of whether a stepparent subsequently steadily with age, but stalled and even declined moved in). In the remainder of the analysis, we in the late 40s. This is probably due to the aimed to explain this penalty by looking at the time period studied. Once excluding postcrisis role of the three general channels that could con- cases, increases in wealth were observed dur- nect childhood family structure to wealth: own ing respondents’ late 40s for individuals who human capital formation, own family dynamics, lived with both parents throughout childhood and intergenerational transfers. Column 1 of (Figure C2 in the online appendix). Figure 1 Table 3 presents the effect of each mediator also displays how wealth differences devel- considered in our analysis on wealth (i.e., each oped with age. Respondents who experienced a model only included one mediator). We found parent–child separation had less wealth already that cognitive ability, education, wages, being at age 30, and this gap grew with age. The key in a first marriage, transfers, and having good periods where differences emerged were during health were strongly associated with wealth. respondents’ early 30s and particularly in their Column 1 of Table 3 also allowed us to bench- late 40s; during the early 40s, the gap seemed to mark the parent–child separation penalty of have tightened temporarily. approximately 1.32 of IHS transformed wealth To what extent could mediating variables documented in Model 4 of Table 2 to effect explain the development of this wealth gap? sizes of other covariates. The effect of having Models 2 to 5 of Table 4 summarize growth experienced a permanent departure of a parent curve models that included the following four was about one third of that of having achieved a mediating variables that appeared to be most college degree. important for our analysis: education, wages, The second column of Table 3 documents having children, and partnership history. Addi- the results of the mediation analysis, indicating tional analysis displayed in online Appendix D for each mediator and block of mediators the tested whether these four factors mediated each percentage of the “wealth penalty” they could of the terms of the basic growth curve model explain. Human capital variables explained 39% (Model 1) to a statistically significant extent. of the penalty, with a major role played by edu- Education and wages had a positive influence cation (which on its own explains 37%), fol- on initial wealth, and given their negative rela- lowed by wages and cognitive ability. Family tionship to having experienced a parent–child dynamics practically played an equally impor- separation (Table D3 in the online appendix), tant role and explained 34% altogether. Being they mediated part of the initial wealth penalty. in a first marriage with a college-educated part- Additional analysis found these mediating ner (individuals who experienced a parent–child effects to be statistically significant with p val- separation were less often so) played an impor- ues below .01 (Table D2 in the online appendix). tant mediating role (30%). The other partner- Having children positively predicted initial ship histories mattered less, indicating that it is wealth and was also positively related to having not the lower probability to marry that matters experienced a parent–child separation. The pres- but, rather, who one marries and whether one ence of children therefore initially dampened the stays in that marriage. Lower wealth accumula- wealth gap (statistically significant at the 99% tion due to having a child before age 25 played a level). However, as respondents aged, having nonnegligible role, too (11%). The third channel children was related to a lower accumulation of intergenerational transfers explained wealth of wealth. Finally, being in a first marriage accumulation in general but did not mediate the with a higher educated partner mediated initial parent–child separation penalty. Altogether, the wealth differences and childhood family struc- mediators explained 56% of the penalty, sug- ture accumulation differences during ages 30 gesting considerable overlap between the effects to 35 (significant at the 95% and 99% levels, of human capital formation and own family respectively). It was also the only variable that dynamics. could to some extent explain childhood family 240 Journal of Marriage and Family Table 3. Percentage of Parental Departure Effect on Wealth (IHS) Explained by Mediators % of parental Effect of mediator departure penalty on wealth explained by mediator Column 1 Column 2 Variable Coefficient SE Block: human capital 39.3** Cognitive test scores 1.00** 0.14 7.4** External locus of control −0.48** 0.11 2.4† Risky behavior −0.63** 0.13 10.1** GED (ref. no education) 0.23 0.55 −2.7 High school 2.58** 0.39 9.8** Some college 3.23** 0.40 18.8** College degree 4.15** 0.40 11.4** Wage income 1982–2012 1.09** 0.10 8.2* Block: family dynamics 34.1** Union history (ref. single, never married) Cohabiting partner is lower educated 1.57 1.07 −0.3 Cohabiting partner is higher educated 2.45** 0.34 0.9 No partner, widowed 1.11** 0.34 −0.1 No partner, separated/divorced −0.35 0.35 1.3 Married to first partner, lower educated 3.77** 0.35 1.0 Married to first partner, higher educated 4.39** 0.34 30.2** Cohabiting partner lower educated, previously married 0.77* 0.35 −0.6 Cohabiting partner higher educated, previously married 3.00** 0.34 −0.6 Married partner lower educated, second marriage 1.93** 0.35 −4.0* Married partner higher educated, second marriage 3.76** 0.34 −1.2 No children (ref. child after age 25) −0.79** 0.27 −2.8* Child before age 25 −1.35** 0.16 10.6** Block: transfers −1.0 Ln(transfers) 0.15** 0.02 −1.0 Parent died by 2012 −1.21* 0.52 0.0 Other Physical health 1.01** 0.13 5.4† Mental health 0.98** 0.12 7.3* All mediators combined 55.7 ** Note. N = 5,822. Percentages taken from decompositions of total effects into direct and indirect effects using the method of Kohler and others (2011): Indirect Effect/Total Effect × 100%. Separate models ran for each mediator, subsequently for each block, and finally for all mediators together. Based on 20 multiply imputed data sets for missing data on independent variables, robust standard errors. All models control for maternal years of education, paternal years of education, own age, maternal age at birth, number of siblings, religion, gender, and ethnicity. ref. = reference; IHS = Inverse Hyberbolic Sine. Italics indicate estimates for blocks of variables. † p < .1; * p < .05; ** p < .01. structure accumulation differences at ages 46 to initially more likely to be in several partnered 50, but this mediating effect was not statistically situations. The influence of family dynamics significant. The results for other partnering therefore seems to be partly a story of timing: dynamics are displayed in online Appendix D. Individuals who did not live continuously with Several of these had a small dampening effect both parents during childhood started their fami- on initial wealth differences, as individuals who lies earlier and therefore gain access to partners’ experienced a parent–child separation were wealth at younger ages. This initial dampening Childhood Family Structure and Wealth 241 Figure 1. Predicted Wealth Trajectories based on Growth-Curve Models. Note. Predicted wealth based on Table 4, Model 1. Dashed lines are 95% confidence intervals. N = 59,890. IHS = Inverse Hyperbolic Sine. influence of family dynamics on wealth differ- compared to respondents who lived with both ences fades as also those from “intact” families biological parents during their entire childhood, start to partner (and often marry higher educated depending on the family structure trajectory con- partners). sidered. We subsequently focused on the penalty Figure 2 shows how much of the wealth dif- related to the experience of a parent–child sepa- ferences remained once controlling for both ration during childhood. A mediation analysis of human capital variables and family dynamics this penalty revealed a role for a wide variety of simultaneously. Although differences in wealth processes and outcomes ranging from cognitive trajectories between respondents with separated and noncognitive skills to education, wages, fer- parents and those from intact families were con- tility, and own partnering behavior. Educational siderably reduced, part of the gap remained attainment and respondents’ own family dynam- unexplained. In particular, the mediators of the ics played a key role in mediating the observed analysis could not account for how the wealth difference between childhood family structure penalty grew during respondents’ late 40s. We and wealth. Wealth differences were not equal return to this issue in the discussion. at different stages of life. Respondents who experienced a parent–child separation during childhood accumulated less wealth, especially Discussion during ages 46 to 50. Wealth plays a central role in the social stratifica- A first main contribution of our article has tion of societies, but only selected determinants been to show that childhood family structure is of wealth accumulation have been studied so an important predictor of wealth. Interestingly, far (Killewald et al., 2017). In this article, we statistically significant differences in wealth used data for a cohort of U.S. respondents born were observed for all types of alternative child- around the early 1960s to show that childhood hood family structures (as compared to intact family structure is an important factor related families) including parental death, temporary to wealth. We documented median “wealth separations, and the experience of stepparent- penalties” of at least $61,600 at ages 47 to 55 as hood. Importantly, a wealth penalty was also Table 4. Summary of Growth Curve Models Explaining Net Worth (N = 59,890) 242 Model 1 Model 2 Model 3 Model 4 Model 5 Variable Coefficient SE Coefficient SE Coefficient SE Coefficient SE Coefficient SE Ages 31–35 0.16** 0.02 0.15** 0.02 0.02 0.16 0.18** 0.02 0.10** 0.04 Ages 36–40 0.16** 0.02 0.16** 0.02 0.33 0.17 0.15** 0.03 0.13** 0.05 Ages 41–45 0.00 0.02 −0.01 0.03 0.48** 0.17 −0.02 0.04 −0.08 0.06 Ages 46–50 −0.05 † 0.03 −0.04 0.03 −0.49* 0.20 −0.01 0.06 −0.09 0.08 Parent/child separation (ref. always with both parents) −0.68** 0.16 −0.59** 0.16 −0.62** 0.16 −0.72** 0.16 −0.66** 0.16 Parent/Child Separation × Ages 31–35 −0.06† 0.04 −0.05 0.04 −0.06 0.04 −0.06 0.04 −0.04 0.04 Parent/Child Separation × Ages 36–40 0.01 0.04 0.02 0.04 0.01 0.04 0.01 0.04 0.00 0.04 Parent/Child Separation × Ages 41–45 0.07 0.05 0.07 0.05 0.06 0.05 0.07 0.05 0.08 0.05 Parent/Child Separation × Ages 46–50 −0.17** 0.07 −0.18** 0.07 −0.17** 0.07 −0.17* 0.07 −0.15* 0.07 Education in years 0.15** 0.02 Education in Years × Ages 31–35 0.02** 0.01 Education in Years × Ages 36–40 0.01 0.01 Education in Years × Ages 41–45 0.01 0.01 Education in Years × Ages 46–50 −0.02 0.01 Average wages up to measurement 0.69** 0.06 Wages × Ages 31–35 0.01 0.02 Wages × Ages 36–40 −0.02 0.02 Wages × Ages 41–45 −0.05** 0.02 Wages × Ages 46–50 0.04* 0.02 Has children in year of measurement 0.62** 0.13 Children × Ages 31–35 −0.06† 0.03 Children × Ages 36–40 0.02 0.04 Children × Ages 41–45 0.03 0.05 Children × Ages 46–50 −0.07 0.07 Married, higher educated partner (coefficients for other partnership trajectories not shown; ref. single, never married) 2.21** 0.20 Married, Higher Educated × Ages 31–35 0.06 0.05 Married, Higher Educated × Ages 36–40 0.09 0.09 Married, Higher Educated × Ages 41–45 0.13† 0.13 Married, Higher Educated × Ages 46–50 0.12 0.10 Note. Full results including full partnering dynamics results shown in Table D1 in the online appendix. Growth curve models estimated using xtmixed in STATA based on 20 multiple imputations. Controls included in all models but not shown are maternal years of education, paternal years of education, own age, maternal age at birth, number of siblings, religion, Journal of Marriage and Family gender, and ethnicity. ref. = reference. † p < .1; * p < .05; ** p < .01. Childhood Family Structure and Wealth 243 Figure 2. Predicted Wealth Trajectories Controlling for Education, Wages, and Own Family Dynamics. Note. Dashed lines indicate 95% confidence intervals. Controls were maternal age at birth, paternal education, maternal education, ethnicity, religion, gender, age in 2012. Mediators were own education in years; average own ln (yearly wage in 2012 dollars) up to that age; partnership status and history; parent of child in reference year. Mediators interacted with age splines. N = 59,890. IHS = Inverse Hyperbolic Sine. observed for individuals who grew up in a (when compared with earlier research), but sur- stable, single-parent family. It therefore appears prisingly transfers did not mediate the wealth unlikely that what matters for offspring’s wealth penalty. It could be that respondents were still is the stability of childhood family structures too young (approximately age 50) to capture the (Fomby & Cherlin, 2007). The adjustment to full lifetime impact of transfers. changes in family composition during child- We also included measures of cognitive and hood is therefore unlikely to be the main factor noncognitive traits in our analysis. Although responsible for the wealth differences observed these appeared to mediate the wealth penalty, during adulthood. their role was relatively minor compared to edu- A second main contribution of our article has cational attainment and own partnering behavior. been to look at a more extensive set of mediators Possible ways in which childhood family struc- ture might affect educational attainment include when compared with previous research (Amato financially driven decisions to not continue edu- & Keith, 1991; Lersch & Baxter, 2015). Pre- cation (Bernardi & Boertien, 2016) and the vious studies primarily found a role for part- desire to start a family earlier in life (McLanahan nering and childbearing decisions in explaining & Bumpass, 1988). wealth gaps according to childhood family struc- An alternative explanation for the patterns ture. We also found that own family dynamics, in observed is endogeneity. Although causality was particular who one marries, partly explained the not among the major concerns of this article, we lower wealth of individuals who experienced a think our analysis provided some indications in parent–child separation. However, we also found that regard. The two usual suspects that produce a considerable role played by own education endogeneity in the study of parental separation and wage histories (the latter factor was not are socioeconomic disadvantage and family measured in earlier studies). We included an conflict during childhood. The socioeconomic improved measure of intergenerational transfers control variables used within this article were 244 Journal of Marriage and Family relatively rich. Furthermore, in additional The initial wealth gap observed at age 30 grew analysis we estimated Model 4 of Table 2 with over time, and particularly so during respon- a control for (logged) family income in the dents’ early 30s and late 40s. The differences first wave (1979) for a selected subsample of in wealth started to dramatically diverge when 4,630 individuals who lived with their parents the respondents were ages 46 to 50; a pattern at that time. Although the wealth differences we could not explain in our main analysis. We were smaller (1.05 IHS transformed wealth, performed some additional tests to understand but including family income might control the growing wealth gap during ages 46 to 50. away both endogeneity and the causal effect of First, the financial crisis hit part of the cohort parent–child separations mediated by changes in under study at these later ages and could have family income), the role of explanatory factors enlarged the differences in wealth observed up was similar. We were not able to control for to age 45. Figure C2 in the online appendix parental wealth, and therefore cannot entirely is a replication of our growth curve analysis rule out the influence of unobserved socioeco- but restricted to observations up to 2008. This nomic disadvantage (but direct wealth transfers “precrisis” model also showed an unexplained were accounted for in our study). At the same increase in the parent–child separation penalty time, previous research concluded that the neg- at later ages. Second, we investigated whether ative association between parental separation returns to accumulated wealth started playing and educational attainment often persists once an increasingly important role at later ages, but accounting for endogeneity (McLanahan et al., we did not find evidence in support of that 2013). If the effects of parental separation on hypothesis (see online Appendix E for more education are indeed causal, we can expect the details). Third, we looked at changing sources of part of the parental separation penalty that was wealth accumulation across the life course. Sav- mediated through education to also be causal. ings, consumption patterns, or pension wealth could start playing a bigger role as individu- The associations presented in this article may als come closer to retirement. In general, dif- thus overestimate the causal effect of parental ferences in various types of assets arose with separation on wealth, but they are unlikely to be age, but especially total housing worth (the main entirely explained by endogeneity (especially component of individual wealth) diverged at given that education explained a large part of later ages between individuals who did and did wealth differences). not experience the permanent moving out of a A third main contribution of this article has parent (see online Appendix F). Additional anal- been to document how wealth differences devel- ysis indicated that the growing gap in housing oped with age. Initial differences in wealth could wealth at later ages was driven by changes in the already be observed when respondents were value of housing assets rather than by changes in 30 years old. Part of these initial wealth differ- mortgage values (see online Appendix F; addi- ences remained unexplained in the analysis. This tional analysis did not indicate a role for home variation in “starting capital” could be due to ownership per se, i.e., as a binary variable). differences in early saving behavior by young Given that housing wealth constitutes the great adults. Alternatively, it could be that persons majority of overall wealth, explanations based whose parents separated transfer money to their on changes in the value of housing are likely parents once they start working, a stream of candidates to explain the growing “wealth penal- money not captured in this study. Another possi- ty” at later ages. For instance, individuals whose bility would be unmeasured forms of economic parents stayed together might have owned homes support by parents to their children. Although we in areas where housing prices rose faster than found that money transfers did not contribute to other areas (or declined less during the crisis). the observed parental separation gap in wealth Finally, a last explanation that we have not been during ages 30 to 50, such economic support able to test for is that during mid-adulthood could also be provided in other ways not cap- transfers to parents might start playing a role. tured by the survey. For instance, parents can Separated parents might require more financial provide guarantees once children get a mortgage help from their children as they age when com- and can function as a safety net allowing them pared with others. Future studies could look fur- to take more risk and invest (Charles & Hurst, ther into the issue of growing wealth differences 2002). during individuals in their late 40s, which would Childhood Family Structure and Wealth 245 also be a relevant avenue for studies on wealth Note in general. We thank Katy Morris for her excellent comments and sug- To conclude, the results of this article have gestions as well as Michael Grätz, Léa Pessin, and those documented that childhood family structure present at presentations given at the Stockholm University Demography Unit, and the European Consortium for Soci- matters for wealth accumulation. Given the ological Research conference in Tallinn. This work was central role of wealth in social stratification supported by funding from the European Union’s Seventh (Spilerman, 2000) and the effects of wealth on Framework Programme for the research project Families a number of life outcomes including marriage, and Societies (FP7/2007-2013 grant agreement no. 320116). Diederik Boertien acknowledges research funding from the age at retirement, and longevity (Killewald Beatriu de Pinos program of the Generalitat de Catalunya et al., 2017), the question arises how wealth (2016-BP-00121) as well as the EQUALIZE project led penalties related to childhood family structure by Iñaki Permanyer (ERC-2014-STG grant agreement no. could be reduced. We have discussed different 637768). channels that connect childhood family struc- ture to net worth, and, among others, education Supporting Information appeared to play a key role. To minimize the Additional supporting information may be found online wealth penalty experienced by individuals, a in the Supporting Information section at the end of the article. fruitful approach would be to minimize child- Online Appendix. Additional Models and Information hood family structure’s effects on educational attainment. Previous cross-national research has shown that the association of childhood References family structure with educational attainment is Amato, P. R. (2010). Research on divorce: Continuing relatively large in the United States (Bernardi trends and new developments. Journal of Marriage & Boertien, 2017; Pong et al., 2003) and that and Family, 72, 650–666. https://doi.org/10.1111/ the lower levels of household income among j.1741-3737.2010.00723.x single parents are a major mechanism explain- Amato, P. R., & Keith, B. (1991). Separation from a ing the lower educational attainment of their parent during childhood and adult socioeconomic children (Amato, 2010). Reducing income attainment. Social Forces, 70(1), 187–206. https:// doi.org/10.1093/sf/70.1.187 differences across families or reducing the Ameriks, J., Caplin, A., & Leahy, J. (2002). role of family income in access to education Retirement consumption: Insights from a sur- could therefore be an effective step to reduce vey (National Bureau of Economic Research later wealth disparities between groups of indi- Working Paper No. 8735). Cambridge, MA: viduals. At the same time, given that many National Bureau of Economic Research. different mechanisms appear to jointly pro- Astone, N. M., & McLanahan, S. S. (1991). Fam- duce wealth differences, no single strategy is ily structure, parental practices and high school completion. American Sociological Review, 56, likely to be able to deal with wealth disparities 309–320. https://doi.org/10.2307/2096106 entirely. Avellar, S., & Smock, P. J. (2005). The economic In sum, the results of this article have shown consequences of the dissolution of cohabiting that childhood family structure is connected unions. Journal of Marriage and Family, 67, to a considerable accumulation of advantages 315–327. https://doi.org/10.1111/j.0022-2445 (or disadvantages) across different spheres .2005.00118.x of life and across time. Experiencing the Avery, R. B., & Rendall, M. S. (2002). Lifetime inher- itances of three generations of Whites and Blacks. moving out of a parent during childhood is American Journal of Sociology, 107, 1300–1346. negatively related to skills, education, wages, https://doi.org/10.1086/344840 and own family behavior as an adult. Indi- Bernardi, F., & Boertien, D. (2016). Understanding vidually each of these channels produces heterogeneity in the effects of parental separation relatively modest penalties in terms of wealth, on educational achievement in Britain: Do chil- but together they sum up to a more consid- dren from lower educational backgrounds have erable one. The heavily intertwined nature of less to lose? European Sociological Review, 32(6), 807–819. https://doi.org/10.1093/esr/jcw036 advantages and disadvantages accumulated Bernardi, F., & Boertien, D. (2017). Non-intact fami- across spheres of life therefore makes child- lies and diverging educational destinies: A decom- hood circumstances highly influential for adult position analysis for Germany, Italy, the United outcomes. Kingdom and the United States. Social Science 246 Journal of Marriage and Family Research, 63, 181–191. https://doi.org/10.1016/j Forces, 90, 1323–1345. https://doi.org/10.1093/sf/ .ssresearch.2016.09.004 sos073 Biblarz, T. J., & Raftery, A. E. (1999). Family struc- Farkas, G. (2003). Cognitive skills and noncognitive ture, educational attainment, and socioeconomic traits and behaviors in stratification processes. success: Rethinking the pathology of matriarchy. Annual Review of Sociology, 29(1), 541–562. American Journal of Sociology, 105(2), 321–365. https://doi.org/10.1146/annurev.soc.29.010202 https://doi.org/10.1086/210314 .100023 Bjarnason, T., & Arnarsson, A. M. (2011). Joint Fomby, P., & Cherlin, A. J. (2007). Family instabil- physical custody and communication with parents: ity and child well-being. American Sociological A cross-national study of children in 36 western Review, 72(2), 181–204. countries. Journal of Comparative Family Studies, Friedline, T., Masa, R. D., & Chowa, G. A. (2015). 42, 871–890. Transforming wealth: Using the inverse hyper- Cavanagh, S., & Huston, A. C. (2006). Family insta- bolic sine (IHS) and splines to predict youth’s bility and children’s early problem behavior, Social math achievement. Social Science Research, 49, Forces, 85, 551–581. https://doi.org/10.1353/sof 264–287. https://doi.org/10.1016/j.ssresearch .2006.0120 .2014.08.018 Charles, K. K., & Hurst, E. (2002). The transi- Gähler, M., & Palmtag, E. L. (2015). Parental tion to home ownership and the Black-White divorce, psychological well-being and educational wealth gap. The Review of Economics and Statis- attainment: changed experience, unchanged effect tics, 84(2), 281–297. https://doi.org/10.1162/ among Swedes born 1892–1991. Social Indicators 003465302317411532 Research, 123(2), 601–623. https://doi.org/10 Cherlin, A. J., Kiernan, K. E., & Chase-Lansdale, .1007/s11205-014-0768-6 P. L. (1995). Parental divorce in childhood and Hampden-Thompson, G. (2013). Family policy, fam- demographic outcomes in young adulthood. ily structure, and children’s educational achieve- Demography, 32(3), 299–318. https://doi.org/10 ment. Social Science Research, 42(3), 804–817. .2307/2061682 https://doi.org/10.1016/j.ssresearch.2013.01.005 Härkönen, J., Bernardi, F., & Boertien, D. (2017). Christopher, K, England, P., McLanahan, S., Ross, Family dynamics and child outcomes: An K., & Smeeding, T. M. (2001). Gender inequality overview of research and open questions. Euro- in poverty in affluent nations: The role of single pean Journal of Population, 33(2), 163–184. motherhood and the state. In K. Vleminckx & T. M. https://doi.org/10.1007/s10680-017-9424-6 Smeeding (Eds.), Child well-being, child poverty, Heckman, J. J. (2000). Policies to foster human cap- and child policy in modern nations (pp. 199–219). ital. Research in Economics, 54(1), 3–56. https:// Bristol, UK: Policy Press. doi.org/10.1006/reec.1999.0225 Conger, R. D., Conger, K. J., & Martin, M. J. (2010). Heckman, J. J., & Kautz, T. (2012). Hard evidence Socioeconomic status, family processes, and on soft skills. Labour Economics, 19(4), 451–464. individual development. Journal of Marriage and https://doi.org/10.1016/j.labeco.2012.05.014 Family, 72, 685–704. https://doi.org/10.1111/j Hetherington, E. M., & Kelly, J. (2002). For better .1741-3737.2010.00725.x or for worse: divorce reconsidered. New York: Conley, D. (2001). Capital for college: Parental assets Norton. and postsecondary schooling. Sociology of Educa- Johnson, D. R., & Young, R. (2011). Toward best tion, 74, 59–72. practices in analyzing datasets with missing data: DiPrete, T. A., & Eirich, G. M. (2006). Cumulative Comparisons and recommendations. Journal of advantage as a mechanism for inequality: A review Marriage and Family, 73, 926–945. https://doi of theoretical and empirical developments. Annual .org/10.1111/j.1741-3737.2011.00861.x Review of Sociology, 32, 271–297. https://doi.org/ Kalmijn, M. (2015). How Childhood circumstances 10.1146/annurev.soc.32.061604.123127 moderate the long-term impact of divorce on Dronkers, J. (1999). The effects of parental con- father–child relationships. Journal of Marriage flicts and divorce on the well-being of pupils and Family, 77, 921–938. https://doi.org/10.1111/ in Dutch secondary education. European Socio- jomf.12202 logical Review, 15(2), 195–212. https://doi.org/10 Keister, L. A. (2004). Race, family structure, and .1093/oxfordjournals.esr.a018260 wealth: The effect of childhood family on adult Eads, A., & Tach, L. (2016). Wealth and inequality in asset ownership. Sociological Perspectives, 47(2), the stability of romantic relationships. The Russell 161–187. https://doi.org/10.1525/sop.2004.47.2 Sage Foundation Journal of the Social Sciences, .161 2(6), 197–224. Kiernan, K. E., & Cherlin, J. A. (1999). Parental Erola, J., Härkönen, J., & Dronkers, J. (2012). More divorce and partnership dissolution in adulthood: careful or less marriageable? Parental divorce, Evidence from a British cohort study. Population spouse selection and entry into marriage. Social Studies, 53, 39–48. Childhood Family Structure and Wealth 247 Kiernan, K. E., & Mensah, F. K. (2009). Poverty, Demography, 36(3), 287–297. https://doi.org/10 maternal depression, family status and children’s .2307/2648053 cognitive and behavioural development in early Pfeffer, F. T. (2011). Status attainment and wealth in childhood: A longitudinal study. Journal of Social the United States and Germany. In T. Smeeding, R. Policy, 38(4), 569–588. https://doi.org/10.1080/ Erikson, & M. Jäntti (Eds.), Persistence, privilege, 00324720308068 and parenting. The comparative study of intergen- Killewald, A., Pfeffer, F. T., & Schachner, J. N. erational mobility (pp. 109–137). New York: Rus- (2017). Wealth inequality and accumulation. sell Sage. Annual Review of Sociology, 43, 35.1–35.26. Piketty, T. (2014). Capital in the 21st century. Cam- https://doi.org/10.1146/annurev-soc-060116- bridge: Harvard University 053331 Pong, S., Dronkers, J., & Hampden-Thompson, Kohler, U., Karlson, K. B., & Holm, A. (2011). G. (2003). Family policies and children’s Comparing coefficients of nested nonlinear prob- school achievement in single- versus two-parent ability models using khb. Stata Journal, 11(3), families. Journal of Marriage and Family, 65, 420–438. 681–699. https://doi.org/10.1111/j.1741-3737 Lersch, P., & Baxter, J. (2015). Parental separation .2003.00681.x during childhood and adult children’s wealth (Life Rauscher, E. (2016). Passing it on: Parent-to-adult Course Centre Working Paper Series 2015-15). child financial transfers for school and socioeco- Queensland, Australia: Institute for Social Sci- nomic attainment. The Russell Sage Foundation ence Research, The University of Queensland. Journal of the Social Sciences, 2(6), 172–196. Retrieved from https://espace.library.uq.edu.au/ Semyonov, M., & Lewin-Epstein, N. (2013). Ways to view/UQ:364673 richness: Determination of household wealth in 16 Lersch, P. M. (2017). The marriage wealth premium countries. European Sociological Review, 29(6), revisited: Gender disparities and within-individual 1134–1148. https://doi.org/10.1093/esr/jct001 changes in personal wealth in Germany. Demog- Spilerman, S. (2000). Wealth and stratification raphy, 54(3), 961–983. https://doi.org/10.1007/ processes. Annual Review of Sociology, 26(1), s13524-017-0572-4 497–524. https://doi.org/10.1146/annurev.soc.26 Lusardi, A., & Mitchell, O.S. (2007). Baby Boomer .1.497 retirement security: The roles of planning, finan- Sweeney, M. M. (2010). Remarriage and stepfami- cial literacy, and housing wealth. Journal of Mon- lies: Strategic sites for family scholarship in the etary Economics, 54, 205–224. 21st century. Journal of Marriage and Family, Manning, W. D., & Smock, P. J. (1999). New fami- 72, 667–684. https://doi.org/10.1111/j.1741-3737 lies and nonresident father-child visitation. Social .2010.00724.x Forces, 78(1), 87–116. https://doi.org/10.1093/sf/ Thompson, J., & Conley, D. (2016). Health shocks 78.1.87 and social drift: examining the relationship McLanahan, S., & Bumpass, L. (1988). Intergenera- between acute illness and family wealth. The tional consequences of family disruption. Ameri- Russell Sage Foundation Journal of the Social can Journal of Sociology, 94(1), 130–152. https:// Sciences, 2(6), 153–171. doi.org/10.1086/228954 Thomson, E., Hanson, T. L., & McLanahan, S. S. McLanahan, S., & Percheski, C. (2008). Family struc- (1994). Family structure and child well-being: ture and the reproduction of inequalities. Annual Economic resources vs. parental behaviors. Social Review of Sociology, 34, 257–276. https://doi.org/ Forces, 73(1), 221–242. https://doi.org/10.1093/sf/ 10.1146/annurev.soc.34.040507.134549 73.1.221 McLanahan, S., Tach, L., & Schneider, D. (2013). The Thomson, E., Mosley, J., Hanson, T. L., & McLana- causal effects of father absence. Annual Review of han, S. S. (2001). Remarriage, cohabitation, and Sociology, 39, 399–427. https://doi.org/10.1146/ changes in mothering behavior. Journal of Mar- annurev-soc-071312-145704 riage and Family, 63, 370–380. https://doi.org/10 McManus, P. A., & DiPrete, T. A. (2001). Losers .1111/j.1741-3737.2001.00370.x and winners: The financial consequences of sepa- Vespa, J., & Painter, M. A. II. (2011). Cohabita- ration and divorce for men. American Sociological tion history, marriage, and wealth accumulation. Review, 66(2), 246–268. Demography, 48, 983–1004. https://doi.org/10 Peters, B., & Ehrenberg, M. F. (2008). The influence .1007/s13524-011-0043-2 of parental separation and divorce on father–child Wolfinger, N. H. (2005). Understanding the divorce relationships. Journal of Divorce & Remar- cycle: The children of divorce in their own mar- riage, 49(1–2), 78–109. https://doi.org/10.1080/ riages. New York: Cambridge University Press. 10502550801973005 Zagorsky, J. L. (2005). Marriage and divorce’s impact Pezzin, L. E., & Schone, B. S. (1999). Parental mar- on wealth. Journal of Sociology, 41(4), 406–424. ital disruption and intergenerational transfers: An https://doi.org/10.1177/1440783305058478 analysis of lone elderly parents and their children.