The Inheritance of Race Revisited: Childhood Wealth and Income and Black–White Disadvantages in Adult Life Chances

Vast racial inequalities continue to prevail across the United States and are closely linked to economic resources. One particularly prominent argument contends that childhood wealth accounts for black–white (BW) disadvantages in life chances. This article analyzes how much childhood wealth and childhood income mediate BW disadvantages in adult life chances with Panel Study of Income Dynamics and Cross-National Equivalent File data on children from the 1980s and 1990s who were 30+ years old in 2015. Compared with previous research, we exploit longer panel data, more comprehensively assess adult life chances with 18 outcomes, and measure income and wealth more rigorously. We find large BW disadvantages in most outcomes. Childhood wealth and income mediate a substantial share of most BW disadvantages, although there are several significant BW disadvantages even after adjusting for childhood wealth and income. The evidence mostly contradicts the prominent claim that childhood wealth is more important than childhood income. Indeed, the analyses mostly show that childhood income explains more of BW disadvantages and has larger standardized coefficients than childhood wealth. We also show how limitations in prior wealth research explain why our conclusions differ. Replication with the National Longitudinal Survey of Youth and a variety of robustness checks support these conclusions.

V AST racial inequalities continue to prevail across the United States. Given that economic resources almost always differ by race (Shapiro 2017), scholars have long debated how much black-white (BW) disadvantages can be accounted for by economic disparities. Fifty years ago, Otis Dudley Duncan (1969) scrutinized how much of BW inequalities in adult socioeconomic attainment could be accounted for by childhood economic resources. According to Duncan, what was commonly perceived as the "inheritance of poverty" was actually the "inheritance of race" because economic resources failed to explain most of BW disadvantages.
More recently, scholarship has progressed beyond viewing race and economic resources as competing, rival explanations of life chances. Rather, racial inequalities are channeled through inequalities in economic resources (e.g., Altonji and Doraszelski 2005;Bobo 2017;Brown et al. 2016;Pais 2014;Rothstein and Wozny 2013;Sewell 2016;Shapiro 2017;South et al. 2016). The emerging view is that racial stratification causes economic inequalities earlier in and at various stages of life, and these economic inequalities then contribute to and exacerbate racial inequalities later in life (e.g., Addo, Houle, and Simon 2016;Bloome 2014;Brown 2018;Killewald and Bryan 2018;Manduca and Sampson 2019;Phelan and Link 2015;Sharkey 2013;Umberson et al. 2014;Williams 2019). This research has crystallized into a perspective on racial inequality such that the "total" BW disadvantage is composed of both a "direct" disadvantage net of, and "indirect" disadvantage mediated by, economic resources.
One particularly influential argument contends that childhood wealth is the paramount economic resource that explains BW disadvantages in life chances (Conley 1999;Shapiro 2004Shapiro , 2017. This argument builds upon pioneering scholarship showing that BW wealth inequalities are larger than BW income inequalities and that historical and institutional racism underlie BW wealth inequality (Denton 2001;Massey 2016;Oliver and Shapiro 1997). According to this prominent view, one will underestimate and even misunderstand how economic resources mediate racial inequalities by analyzing income rather than wealth. 1 Motivated by this perspective and particular argument, we assess how much childhood wealth and childhood income mediate BW disadvantages in adult life chances. 2 We analyze Panel Study of Income Dynamics (PSID) and Cross-National Equivalent File (CNEF) data on children from the 1980s and 1990s who were 30+ years old in 2015. We also replicate the analyses with the 1997 National Longitudinal Survey of Youth (NLSY). Given extensive literatures exist on both BW disadvantages and intergenerational inheritance, we underline three reasons this study is innovative. First, the relatively recent availability of longer-term high-quality panel data enable us to address data limitations of prior studies. The PSID now has a large enough sample of black and white adults whose wealth and income were observed sufficiently in childhood and who have reached mature adulthood. Second, we more comprehensively assess adult life chances with 18 economic, education, family, well-being, and health outcomes. Third, we measure income and wealth more rigorously by incorporating the prevailing international standards on income measurement. Altogether, this study provides novel evidence on the relationships between race, childhood wealth and income, and adult life chances. In the process, we clarify how wealth and income operate as mechanisms of racial inequality.

Explaining Black-White Disadvantages
Researchers commonly estimate BW inequalities quantitatively by regressing various adult outcomes on an indicator variable for black relative to white people. Although the coefficient for being black may not represent the "causal effect" of race (Sen and Wasow 2016), it indicates the magnitude of the "BW disadvantage" conditional on observables. The coefficient captures the costs of being black in a society that systematically disadvantages black people-rather than the effect of being black per se-and is a relational disadvantage for black people and an advantage for white people (Williams 2019).
Large and pervasive BW disadvantages are consistent with systemic discrimination theory (Bobo 2017;Feagin 2000;Reskin 2012;Small and Pager 2020). According to systemic discrimination theory, racial disparities in one social sphere are interdependent with and reciprocally cause racial disparities in other spheres. These disparities are durably reinforced by and adapted to different organizational settings (Tilly 1998;Tomaskovic-Devey 2014), and ultimately an encompassing system of racism emerges (Bonilla-Silva 1997;Tomaskovic-Devey and Roscigno 1996). Even if only some disparities result from discrimination, those spill over into other spheres and accumulate over the lifetime. Hence, even modest discrimination compounds and cumulates to form large BW disadvantages in life chances (Brown 2018;Korver-Glenn 2018).
Indeed, ample evidence suggests discrimination is more than modest (Goosby et al. 2015;Feagin 2000;Small and Pager 2020;Williams, Lawrence, and Davis 2019). Black children grow up in more segregated (Massey 2016) and disadvantaged neighborhoods (South et al. 2016), which undermine life chances (Manduca and Sampson 2019;Sewell 2016). Neighborhood disadvantage mediates BW gaps in income mobility, wealth mobility, obesity, poor health, depression, anxiety, educational aspirations, and cognitive skill (Sharkey 2013). Black people experience discrimination in health care and stress from stigma and are more exposed to environmental hazards (Manduca and Sampson 2019;Monk 2015;Sewell et al. 2020;Williams et al. 2019). In turn, black people experience higher allostatic loads, more traumatic events, and worse health (Goosby et al. 2015;Hardaway and McLoyd 2008;Phelan and Link 2015;Williams et al. 2019). Black people receive disproportionate sanctions in criminal justice (Legewie and Fagan 2019;Sewell, Jefferson, and Lee 2016;Sewell et al. 2020) and welfare programs (Schram et al. 2009), which worsen life chances (Legewie and Fagan 2019;Sewell et al. 2016;Sewell et al 2020). Partly because of residential (Massey 2016) and school segregation (Clotfelter 2006) and concentration in high poverty schools (Saporito and Sohoni 2007), black children also receive lower quality education.
We refer to the overall BW inequalities in life chances as the "total" BW disadvantage, which has two components. First, the "direct" BW disadvantage refers to the residual mean difference between black and white adults after adjusting for childhood wealth and income. Consistent with direct BW disadvantages, for example, Thomas and Hughes (1986;Hughes and Thomas 1998) find large and stable direct BW disadvantages in psychological well-being and subjective quality of life even after controlling for economic resources (see also Bond Huie et al. 2003;Flippen 2004;Killewald and Bryan 2018;Shuey and Wilson 2008). Second, the "indirect" BW disadvantage is the portion of the total disadvantage mediated and expressed through childhood wealth and income. Consistent with theories of racism as a fundamental cause (Bobo 2017;Phelan and Link 2015;Umberson et al. 2014;Williams et al. 2019), white people have advantages in flexible resources that can be employed for multiple, reliably replaceable mechanisms. Inequalities in resources like wealth and income have accumulated from historical racism interacting with ongoing discrimination (Shapiro 2017). These economic resources are used both strategically and unconsciously by white people to maintain and perpetuate advantages over black people (Shapiro 2004(Shapiro , 2017. This acknowledges that black people may occasionally narrow the gap on one mechanism or in one social sphere. However, white people's advantages in wealth and income fill in to reproduce and maintain BW disadvantages (Bloome 2014). Although wealth and income are not the only resources that white people have at their disposal, wealth and income are very likely salient resources (Phelan and Link 2015). The next two sections explicate why childhood wealth and income could mediate BW disadvantages in life chances.

The Case for Childhood Wealth as Mediator
One of the most prominent arguments in the literature on racial inequality contends that inequalities in childhood wealth account for BW disadvantages in adult life chances (Bond Huie et al. 2003;Oliver and Shapiro 2019;Orr 2003;Pollack et al. 2007;South et al. 2016). This is partly because wealth inequalities are viewed as particularly tightly linked to historic and contemporary racism (Oliver and Shapiro 2019). For instance, black people have long been disproportionately constrained to reside in segregated neighborhoods (Massey 2016). Segregation then disproportionately reduces opportunities for homeownership and home equity among black people (Denton 2001;Flippen 2004;Sewell 2016), which worsens racial wealth inequality (Shapiro 2004). In turn, some prioritize wealth transfers as the most effective strategy to reduce racial inequality and increase the life chances of disadvantaged children (Oliver and Shapiro 2019). Conley (1999) provides one of the most visible and relevant studies. Conley argues that BW disadvantages in life chances are explained specifically by childhood wealth and explicitly not by childhood income or other aspects of social class. Using the PSID, Conley examines a cohort of adolescents, whose parental wealth was observed in 1984/1985, and various outcomes were observed from 1992 to 1995 when the respondents were aged 18 to 30 years. 3 Conley (1999) writes, "Certain tenacious racial differences-such as deficits in education, employment, wages and even wealth itself among African Americans-will turn out to be indirect effects, mediated by class differences. In other words, it is not race per se that matters directly; instead, what matters are the wealth levels and class positions that are associated with race in America" (P. 7). After adjusting for childhood wealth, Conley finds BW disadvantages largely disappear. In turn, Conley (1999) argues, "Socio-economic variables have a much greater impact in predicting outcomes than does skin color or racial identity for this recent cohort" (P. 134). Shapiro (2004Shapiro ( , 2017 also contends that parental wealth is the crucial mediator of racial inequalities in life chances. Shapiro (2004) argues, "Family inheritances, especially financial resources, are the primary means of passing class and race advantages and disadvantages from one generation to another" (P. 61). Shapiro (2004) analyzes "[h]ow the uses of wealth perpetuate inequality" (P. 2) and frames racial wealth inequality as the cause of broader racial inequalities. He stresses that wealth is the essential mechanism by which "one generation passes advantage and disadvantage to the next" and is "the bedrock of racial inequality" (P. 8). Altogether, Shapiro (2004Shapiro ( , 2017 argues that wealth is the paramount mediator of BW disadvantages in life chances. A principal reason scholars argue wealth is the crucial mediator is that wealth is claimed to be a superior measure of economic resources and uniquely consequential for life chances (Conley 1999;Hällsten and Pfeffer 2017;Keister 2005;Orr 2003;Pollack et al. 2007;Shapiro 2004Shapiro , 2017. Because BW gaps are larger in wealth than in income (Oliver and Shapiro 1997), measuring economic resources as income, rather than wealth, will obscure understanding of how racial inequalities are mediated. For example, Pollack and colleagues (2007) write, "Failure to measure wealth may result in under-estimating the contribution of [socioeconomic status] to health, such as when studying the etiology of racial/ethnic disparities" (P. 250). Oliver and Shapiro (1997) write that wealth "[i]s more encompassing than is income or education, and closer in meaning and theoretical significance to our traditional notions of economic well-being and access to life chances" (P. 2). Compared with income, Keister (2000) claims, "wealth comes closer both theoretically and empirically to our general understanding of well-being" (P. 11). Shapiro (2004) writes, "Wealth is critical to a family's class standing, social status, whether they own or rent housing, the kind of community they live in, and the quality of their children's schools" (P. 31). These claims lead directly to the argument that wealth is the primary mediator for BW disadvantages.
Relatedly, wealth scholars often argue that wealth better measures long-term economic resources than income (Bond Huie et al. 2003;Oliver and Shapiro 1997;Shapiro 2017). Keister (2000) claims that wealth is "more enduring across generations" (P. 4) and writes, "Income is an indicator of short-term security, a type of security that may be lost if markets change abruptly, if the income earner becomes ill or dies, or if one relocates with a spouse. Wealth implies a more permanent notion of security and an ability to secure advantages in both the short and long term" (P. 11). By lifting a "family beyond its own achievements," Shapiro (2004) argues wealth is "transformational," facilitating mobility (e.g., through investments in education) and serving as "life support" protecting households during crises (P. 10). As a result, scholars often claim that wealth is more highly intergenerationally inherited than income (Keister 2005;Shapiro 2004). For instance, Conley (1999) writes, "Wealth is much more stable within families and across generations than is income, occupation, or education" (P. 14). Indeed, Killewald (2013) finds childhood wealth, but not childhood income, significantly predicts adult wealth.

The Case for Childhood Income as Mediator
A vast literature shows childhood income enhances life chances (e.g., Duncan, Ziol-Guest, and Kalil 2010). Following others (e.g., Fox, Torche, and Waldfogel 2016;Mazumder 2005), Chetty and colleagues (2014) use a massive data set of tax records to demonstrate the positive relationship between childhood and adult income. Using the PSID, Johnson and Schoeni (2011) demonstrate that income at ages 13 to 16 years in 1968 to 1975 influences self-rated health, asthma, hypertension, diabetes, stroke, heart attack, and heart disease at ages 39 to 56 years. This literature draws on enduring arguments about how income is a crucial liquid economic resource that can be used to purchase well-being and invest in child development. Several studies find childhood income predicts life chances, including adult wealth, even net of childhood wealth (Addo et al. 2016;Killewald and Bryan 2018;Pfeffer 2018).
Though not as large as BW wealth inequalities, there are certainly large BW income inequalities as well. Just like wealth research, childhood income accounts for substantial shares of BW disadvantages (Altonji and Doraszelski 2005;Rothstein and Wozny 2013). For example, Chetty and colleagues (2018) find that childhood income accounts for 38.3 percent of the BW disadvantage in adult income.
One argument for childhood income as a stronger mediator than wealth is that greater progress has occurred in measurement on income versus wealth (Brady and Parolin 2020). In the 1990s, the United Nations convened the "Canberra Group," which led to an international consensus on best practices in income measurement (Duncan and Petersen 2001;Rainwater and Smeeding 2004;Smeeding and Weinberg 2001). Among the prevailing standards, one should (1) include all income sources from the entire household, (2) incorporate taxes and transfers, and (3) equivalize for household size. Living in households and accessing transfers are principal ways to share expenses, smooth incomes, manage volatility, and maximize wellbeing (Brady et al. 2018;Brady and Parolin 2020;Mazumder 2016;Rainwater and Smeeding 2004). For example, the Supplemental Nutritional Assistance Program (SNAP; Hoynes, Schanzenbach, and Almond 2016) and Earned Income Tax Credit (EITC; Hoynes and Patel 2018) improve child well-being and have lasting benefits into adulthood. In turn, childhood income should be measured as "post-fisc" (including taxes/transfers) household equivalized income (Brady 2009).
Unfortunately, the wealth literature typically fails to measure either income or wealth according to these criteria (Bond Huie et al. 2003;Conley 1999;Oliver and Shapiro 1997;Orr 2003;Shapiro 2004). 4 When neglecting prevailing international standards on income measurement, it may be premature to conclude wealth is superior. Although family structure and the number of children and workers influence wealth (Keister 2000;Oliver and Shapiro 1997), wealth is often not equivalized by household size (Conley 1999;Keister 2000Keister , 2005Oliver and Shapiro 1997;Pollack et al. 2007). Most data on wealth omit taxes even though wealth is taxed when holding (e.g., property tax) and transferring (e.g., inheritance tax). Moreover, much wealth comes as public transfers. As Feldstein (1974) explained, "For the great majority of Americans, the most important form of wealth is the anticipated social security retirement benefits" (P. 905). Wealth research typically omits public transfers and defined benefit pensions, plausibly because of difficulties estimating at what level and how long benefits will be collected. Analysts often omit both defined benefit pensions and pension savings accounts (e.g., 401(k)s) (Altonji and Doraszelski 2005) or inconsistently include pension savings accounts but omit defined benefit pensions and Old Age Survivor's Insurance (Keister 2000;Oliver and Shapiro 1997). Stemming from all these ambiguities about what counts as "wealth," even the best available wealth data may miss much of what is purported to be measured.
A second major argument for income is that income might actually better measure long-term resources than wealth. One classic definition of long-term economic resources is "permanent income," which can be measured as average income over 20+ years (Brady et al. 2018). To the best of our knowledge, the only study that tests whether wealth or income better predicts permanent income demonstrates that income outperforms wealth. Brady and colleagues (2018) find that a randomly chosen year of income explains about 46 percent of the variation in permanent income in the United States, and a random year of income during childhood explains 55 percent of the variation. By contrast, a randomly chosen year of wealth explains only 27 percent of the variation. 5 Whereas wealth is highly intergenerationally inherited, income is even more so. In the United States, Pfeffer and Killewald (2017) report intergenerational elasticities in wealth of 0.37 to 0.41, which they acknowledge are quite similar to estimates for income (Mazumder 2016). Although Pfeffer and Killewald (2017) do not provide a side-by-side comparison with intergenerational elasticities in income in the PSID, we do so in Appendix I in the online supplement. With three different transformations of income and wealth (log, relative rank, and inverse hyperbolic sine), we find greater intergenerational associations for income than wealth. 6 If indeed income better measures long-term resources and is more intergenerationally inherited than wealth, income could be the more crucial source of childhood wellbeing and means of investment in child development.

Limitations of Prior Research
The present study is uniquely able to address three key limitations of prior research. First, sufficient data on childhood wealth have been scarce. 7 Despite claiming wealth is highly stable intergenerationally and over time, Oliver and Shapiro (1997) use the Survey of Income and Program Participation, and Keister (2005) uses the NLSY 1979. Neither survey measures childhood wealth. Conley (1999) only has one observation of childhood wealth when respondents are already adolescents. Because early childhood might be even more salient than adolescence (Duncan et al. 2010), it would be better to measure wealth and income cumulatively across early childhood and adolescence, including as many time points as possible (Altonji and Doraszelski 2005;Duncan et al. 2010;Mazumder 2005Mazumder , 2016Rothstein and Wozny 2013).
Second, prior analyses have not sufficiently adjudicated between wealth and income (Pollack et al. 2007). Wealth and income are obviously correlated. 8 Income may even be the paramount predictor of wealth (Keister 2005;Killewald 2013). 9 For instance, Altonji and Doraszelski (2005) find that income plus demographics can explain the entire BW wealth gap. Yet, many argue wealth is more salient than income without a model including both income and wealth. Many focus solely on the statistical significance of the wealth coefficient without comparing its substantive magnitude against the income coefficient (which is usually significant as well). Despite his claims otherwise in the text, a careful read of Conley's (1999) appendices reveals that childhood income mediates BW disadvantages even before controlling for childhood wealth. 10 Although he shows that childhood wealth has a unique effect on educational attainment, Pfeffer (2018) finds the effects of income are larger than wealth. Conversely, many studies linking childhood income to adult outcomes do not adjust for childhood wealth (e.g., Bloome 2004;Duncan et al. 2010;Johnson and Schoeni 2011). For example, Chetty and colleagues' (2014) administrative data set on income lacks information on wealth. 11 To properly assess how much childhood wealth and income mediate BW disadvantages, it is essential to model them together.
Third, even when modeling wealth and income together, past studies have been constrained by post-treatment bias. Post-treatment bias occurs when a model includes a mediator, which is affected by the independent variable of interest, and operates between the independent variable and the dependent variable. The mediator conceals ("blocks") and could attenuate (bias) the effect of the independent variable (Elwert and Winship 2014). Post-treatment bias occurs when measures of adult attainment are included in models that intend to assess the effects of childhood wealth/income. Even though adult attainment (e.g., education) is endogenous to childhood wealth/income, studies often control for adult attainment while assessing the effects of childhood wealth/income. 12 This is particularly problematic when studies control for adult income, which could disproportionately obscure the effect of childhood income, and then claim that childhood wealth has a larger impact than childhood income.

Data and Methods
We use data from the PSID and the CNEF (Frick et al. 2007). This data set was used in many related studies (e.g., Conley 1999;Killewald 2013). The CNEF, which is a supplement to the PSID, provides higher quality standardized measures of income incorporating taxes, tax credits, and transfers (Frick et al. 2007). The first year of wealth data in the PSID is 1985, and 2015 is the last available year of CNEF data. Although the wealth data began in 1985, we use all available waves from respondents' childhoods for other variables. All outcomes are measured in 2015.
We analyze approximately 2,300 black and white children (i.e., aged <18 years) observed in the 1980s and 1990s who were 30+ years old in 2015. 13 Our sample was aged 0 to 13 years in 1985 and 30 to 43 years in 2015 (i.e., birth cohorts 1972 to 1985). Appendix II in the online supplement displays descriptive statistics for key variables. For replication and transparency, our code for data set construction and analyses is publicly available. 14 Our data set has at least two unique advantages. First, past research was constrained by small samples of black respondents, which makes the black coefficient vulnerable to type II errors. 15 This could explain why, for example, Conley (1999) does not find initial BW differences for several outcomes (e.g., hourly wages, unemployment, being held back a grade in school, and high school graduation) and why initially significant black coefficients become insignificant when he controls for childhood wealth or income. Second, past studies often measure outcomes too early in adulthood. For example, Conley (1999) observes employment and wages in respondents' early 20s, only seven to nine years after parental wealth was observed, and only among those who have formed independent households (see endnote 3). Because education may be ongoing, attainment remains unreliable and there is less heterogeneity in outcomes like health, the intergenerational mobility literature recommends that respondents be at least 30 years old (Duncan et al. 2010;Fox et al. 2016;Mazumder 2016;Pfeffer and Killewald 2017).

Dependent Variables
We examine a broader variety of adult life chances than prior studies. We include six economic, three education, two family, and seven well-being/health outcomes.
Beginning with economic outcomes, wealth is measured as household (HH) net worth, which includes home equity and is calculated as assets minus debts. 16 Income is measured with the CNEF HH post-fisc (e.g., including SNAP and the EITC) variable. 17 Both outcomes are in real 2015 dollars. Following prevailing international standards on income (Brady 2009;Brady et al. 2018;Brady and Parolin 2020;Duncan and Petersen 2001;Rainwater and Smeeding 2004;Smeeding and Weinberg 2001) and some wealth studies (Altonji and Doraszelski 2005), we equivalize both wealth and income for household size by dividing by the square root of household members. 18 Wealth and income are both highly skewed. Whereas income researchers almost universally employ a log transformation, wealth analysts often do not apply any transformation (Keister 2000(Keister , 2005Oliver and Shapiro 1997:130, 220). 19 Because highly skewed dependent variables often lead to heteroscedasticity and high-influence points, violations of ordinary least squares assumptions are likely. Unfortunately, however, wealth has a large share of cases with zero or negative values. The log transformation bottom codes those cases near zero. Killewald (2013) shows that logging disproportionately inflates black households' wealth because black households are more likely to be net debtors. Thus, logging wealth probably biases the black coefficient. As a result, we use two transformations that preserve zero and negative values (Fox et al. 2016): the inverse hyperbolic sine (IHS) and relative rank percentiles.
Home ownership is coded 1 if the respondent owns or is buying a home and paying a mortgage (renter or neither owns nor rents = 0). Employment is coded 1 if the respondent is currently employed (unemployed, disabled, housewife, student, retired, or other = 0).
We analyze three education outcomes. High school graduate is a measure of whether the respondent completed 12+ years of schooling. College graduate is a binary measure of whether the respondent completed 16+ years of schooling. We also analyze years of schooling.
We examine two family outcomes. Single parenthood is coded 1 for those with children and not married/cohabiting (reference = married and/or no children). Partnered is coded 1 for those currently married/cohabiting (reference = never married/divorced/widowed). We also found similar results when analyzing single motherhood solely on females.
We analyze seven well-being/health outcomes. Life satisfaction is measured as 1 = not at all satisfied, 2 = not very satisfied, 3 = somewhat satisfied, 4 = very satisfied, and 5 = completely satisfied. Self-rated health is measured as 1 = excellent, 2 = very good, 3 = good, 4 = fair, and 5 = poor. Poor/fair self-rated health codes excellent, very good, and good as zero, and fair and poor as 1. Psychological distress is scored 0 to 24 using Kessler and colleagues' (2002) K6 nonspecific scale. 20 Chronic condition is coded 1 if the respondent reports asthma, high blood pressure, cancer, diabetes, arthritis, or lung disease (reference = none). We also include if the respondent has ever experienced a stroke or heart attack (reference = neither). Finally, mortality is measured as death by 2015. As the sample is aged 30 to 43 years in 2015, this is "early" mortality (only 0.8 percent have died).

Race, Childhood Wealth, and Childhood Income
We include a binary measure for being black (reference = white). The 2015 PSID sample includes a reasonably large sample-18 percent (i.e., roughly 420)-of black adults.
Childhood wealth and childhood income use the same definitions as above. We average wealth and income over multiple years. We require that wealth and income be observed at least twice but use all available childhood years. This follows Killewald (2013) and differs from Conley (1999), who only observes childhood wealth once in adolescence. On average, respondents had 2.9 observations of wealth and 15.6 observations of income during childhood. Because income is observed more often than wealth, this could make income more reliable than wealth. Appendix IX in the online supplement shows consistent results with only one observation of income and wealth required. Appendix X in the online supplement shows consistent results if income is observed only in childhood waves when wealth is also observed. Like the dependent variables, we transform childhood wealth and income into IHS and relative rank. Appendix V in the online supplement reports even stronger results when we use the raw (untransformed) values of childhood wealth and income.
The case for measuring childhood wealth and income according to prevailing international standards is not merely theoretical (Brady et al. 2018;Brady and Parolin 2020). Appendix IV in the online supplement compares simple regression models with alternative measures of wealth (nonequivalized) and income (HH earnings before taxes/transfers and nonequivalized) versus models with our preferred measures. For 14 of 15 outcomes, preferred measures attenuate the wealth coefficient. For 13 of 15 outcomes, preferred measures magnify the income coefficient. Furthermore, for eight of 15 outcomes, preferred measures attenuate the black coefficient. These measurement decisions are empirically consequential. Although we use the same PSID data set as studies claiming larger effects of childhood wealth versus income (e.g., Conley 1999), our improvements in measurement of income and wealth are a clear advantage. Nevertheless, even with these improvements in measurement, we acknowledge that measurement is still likely to be more challenging for wealth than income (see, e.g., endnote 7).

Control Variables
Some models adjust for demographics and family background. Because family background has cumulative effects, we average those variables over all available years during childhood. We carefully guard against post-treatment bias by never controlling for adult attainment variables that could mediate the relationship between childhood wealth/income or race and life chances.
Being female is coded 1, and age and age-squared are in years. We include indicators for South, East, and West (reference = Midwest). Sibship size is the average number of other children in the household during childhood. We define the household head as the highest earner, with ties broken by age (and chosen randomly if age is tied) (Brady et al. 2018;Oliver and Shapiro 1997:58). During the respondent's childhood, we calculate the average age of head in years and the average education of head in years of schooling. Finally, we include the proportion of years in a single mother HH during childhood.

Analytical Strategy
In the mediation technique we use, the total, direct, and indirect BW disadvantage maps onto "total, direct, and indirect effects." Although "effect" may imply causality, and causal effects of nonmanipulable characteristics like race are controversial (Sen and Wasow 2016), we only mean "effect" here in the terminology of the mediation literature.
The mediation technique is based on three equations. First, we model the outcomes as a function of being black compared with white. Second, we model outcomes as a function of being black while adjusting for the two key mediators, childhood wealth and income. Third, we model the mediators as a function of being black.
The "direct effect" of being black is β 2 , and the indirect effect of being black is From these estimates, we calculate the "mediated share" as ((β 1 − β 2 )/β 1 ) × 100. The standard errors of the indirect effects use the Sobel test.
For binary outcomes, we employ the Karlson-Holm-Breen (KHB) correction (Breen et al. 2013), which ensures the estimates are unaffected by rescaling or attenuation bias. For both continuous (ordinary least squares) and binary (logit) outcomes, we use the KHB command in Stata (Kohler et al. 2011).
Of course, Kitagawa-Blinder-Oaxaca or Fairlie decomposition techniques would be a reasonable alternative approach. We prefer KHB for several reasons. First, Kitagawa-Blinder-Oaxaca estimates both group-specific levels (i.e., "endowments") and coefficients of the independent variables, whereas Fairlie only estimates levels, which makes continuous and binary outcomes difficult to compare. Second, Appendix XIV in the online supplement shows largely similar coefficients for black and white adults, which supports our focus solely on levels. Third, Kitagawa-Blinder-Oaxaca tends to underestimate the residual, direct black effect and overstate the mediating contribution of childhood wealth/income (Elder, Goddeeris, and Haider 2010). Therefore, KHB is more conservative for our conclusions. Finally, Breen, Karlson, and Holm (2013) show that KHB performs as well or better than alternative mediation techniques in Monte Carlo simulations. In analyses available upon request, we nevertheless replicated all analyses with Kitagawa-Blinder-Oaxaca or Fairlie decomposition techniques and reached consistent conclusions.
We report the following for the main models: (A) the total effect of being black, (B) the direct effect, (C) the percentage of the total that is mediated, (D) the percentage of the mediated share accounted for by childhood wealth, (E) the effect for childhood wealth, (F) the percentage of mediated share accounted for by childhood income, and (G) the effect for childhood income. In the robustness checks, we report A, C, D, and F. If A is significant, this indicates a total BW disadvantage. Comparing D and F clarifies whether childhood wealth or income explains more of the BW disadvantage. Comparing E and G shows whether childhood wealth or income has a larger effect on the outcome. For continuous outcomes, we display y-standardized coefficients for the black coefficient and fully standardized coefficients for childhood wealth and income. For binary outcomes, we display average marginal effects (AMEs) for being black and x-standardized AMEs for childhood wealth and income. All analyses use weights.
The main tables display two set of analyses for every outcome. The first only controls for sex, and the second includes the full set of controls. The full set of controls could introduce post-treatment bias to the key coefficients of interest. For example, parent's education is endogenous to being black and could mediate some of the BW disadvantages. Indeed, including the controls results in smaller coefficients and higher p values for the direct effects of being black. Nevertheless, we show both sets, and the conclusions are generally consistent. Table 1 displays the economic outcomes: wealth, income, home ownership, and employment. The first row, containing the black total effects, shows significant BW disadvantages for all six outcomes. Across outcomes, the largest coefficients and AMEs for being black are for the two income outcomes, whereas the smallest are for employment. These total BW disadvantages are substantively large. For instance, being black is associated with 0.36 standard deviations lower IHS wealth.

Main Models
For five of six outcomes in both sets of models, the direct effect of being black is statistically insignificant. A few near-significant direct effects of being black are substantively nontrivial, and the direct effect of being black is significant for home ownership. For home ownership, for example, the y-standardized coefficient for the direct effect of being black is much larger than the standardized coefficients for childhood wealth or income.
Childhood wealth and income account for 75 to 87 percent of the BW disadvantage for five outcomes and 43 percent of the BW disadvantage in home ownership. With the full controls, the pattern is similar, although the percentage mediated is slightly higher. For the wealth-related outcomes, childhood wealth explains more of the BW disadvantage than childhood income. Childhood wealth explains 63 to 64 percent of the mediated share of the BW disadvantage in IHS net worth, 74 to 79 percent of the mediated share in relative net worth, and 58 to 76 percent of the mediated share in home ownership. Compared with childhood income, childhood wealth also has larger standardized coefficients for IHS and relative net worth and a larger standardized AME for home ownership. For example, a standard deviation higher childhood wealth is associated with 0.20 standard deviations higher adult IHS wealth. By contrast, childhood income explains 36 to 50 percent of the mediated share of the BW disadvantage in IHS wealth, and childhood income has standardized coefficients of 0.09 to 0.13. The models with full controls show similar results, although childhood income matters slightly more. For IHS and relative income, and employment, childhood income explains more of the BW disadvantage than childhood wealth. Childhood income explains 76 to 95 percent of the mediated share of the BW disadvantage in IHS income, 59 to 71 percent of the mediated share in relative income, and 92 to 98 percent of the mediated share in employment. Compared with childhood wealth, childhood income also has larger standardized coefficients for IHS and relative income and a larger standardized AME for employment. For instance, a standard deviation higher childhood income is associated with a 0.30 to 0.35 standard deviations higher adult IHS income. By contrast, childhood wealth only explains five to seven percent of the mediated share of the BW disadvantage in IHS income. Also, childhood wealth has small, statistically insignificant coefficients/AMEs for IHS income and employment. The results are similar in the models with minimal and full controls. Table 2 shows the analyses of education, family, and life satisfaction outcomes. The total effects reveal significant BW disadvantages in all six outcomes. The black direct effect is statistically significant for three outcomes with minimal controls (high school graduate, single parenthood, and partnered), and two of four outcomes with the full controls (single parenthood and partnered). For single parenthood and partnered, the y-standardized AME for the direct effect of being black is much larger than the standardized AMEs for childhood wealth and income. The direct effect of being black is nearly significant for life satisfaction, but its y-standardized coefficient is larger than the standardized coefficients for childhood wealth and income, and those variables only explain 40.4 percent of the BW disadvantage.
The mediators explain most of the BW disadvantages for three outcomes with minimal controls and four outcomes with full controls. For high school graduation and years of schooling, childhood wealth and income explain more than 100 percent of the BW disadvantage-suggesting there is a black advantage net of these mediators.
With minimal controls, for all six outcomes, childhood income explains more than childhood wealth of the mediated share of the BW disadvantage. For five of six outcomes, the percentage of explained accounted for by childhood income is more than three times larger than is accounted for by childhood wealth. Relatedly, childhood income has much larger standardized coefficients/AMEs than childhood wealth for all outcomes except partnered (for which they have similar magnitudes). With full controls, childhood income explains more than childhood wealth of the mediated share of high school graduation, college graduation, years of schooling, and life satisfaction. The standardized coefficients/AMEs are larger for childhood income for those outcomes as well. However, in the full controls models, childhood income explains less than childhood wealth of the mediated share for single parenthood and being partnered outcomes. 21 Table 3 shows the results for well-being/health outcomes. The first row shows significant BW disadvantages only for self-rated health and poor/fair self-rated health. For instance, being black is associated with 0.23 standard deviations lower self-rated health. The signs of the total effects suggest BW disadvantages in psychological distress, having chronic conditions, and having experienced a stroke or heart attack. However, these coefficients are insignificant. Therefore, one needs to be cautious about interpreting the percentage mediated as the total effects are small. For example, the 793 to 988 percent of the BW disadvantage in psychological distress explained is a byproduct of the small, insignificant total effect of being black. For these outcomes, it is more informative to simply compare the coefficients/AMEs for childhood wealth and income.
For all six outcomes, childhood income accounts for far more of the mediated share of BW disadvantages. Also, the coefficient for childhood wealth is not statistically significant in any of the 12 models. By contrast, childhood income has significant coefficients/AMEs for self-rated health, poor/fair health, and psychological distress. As well, childhood income's coefficients/AMEs are nearly significant for the other three outcomes (at least with minimal controls). Even slightly larger samples reveal statistically significant effects for childhood income. 22 For all outcomes, the size of the standardized coefficients/AMEs for childhood income are much larger than those for childhood wealth.
Across Tables 1 to 3, we can summarize three major conclusions. First, there are large BW disadvantages for most outcomes. The total effect of being black is statistically and substantively significant for 14 of 18 outcomes. Second, childhood wealth and income explain a substantial share of most BW disadvantages, although there are several significant and large direct BW disadvantages even after adjusting for childhood wealth and income. Third, the evidence mostly contradicts the prominent claim that childhood wealth is more important than childhood income. Indeed, the analyses mostly show that childhood income explains more of BW Notes: The numbers in parentheses are standard errors. Models with minimal controls adjust for sex. Models with full controls also adjust for age, age squared, South, Northeast, and Midwest regions (the region dummies are omitted from the mortality models), and average childhood HH characteristics: sibship size, age of head, education of head, and single motherhood. † p < 0.01; * p < 0.05. disadvantages and has larger standardized coefficients/AMEs than childhood wealth.

Robustness Checks
A variety of other analyses confirm that the three major conclusions are robust. For brevity, we now only report the black total effect, the percentage mediated, and the percentage of the mediated share accounted for by childhood wealth and income. For all robustness checks, we replicate the models with minimal and full controls but only display those with minimal controls.
Our first replication uses the NLSY 1997 (Bureau of Labor Statistics 2015). The NLSY is a nationally representative sample of 12-to 18-year-olds in 1997, including an oversample of low-income, black, and Latino respondents. Respondents were interviewed annually from 1997 to 2011, then every two years after. We analyze the data from 2015 when respondents were aged 30 to 36 years for most outcomes. The sample includes respondents with multiple observations of childhood income and full information on the controls. We mimic the PSID as much as possible. Childhood income is averaged from 1997 to 2001, but childhood wealth is only measured in 1997. Adult wealth is only available at age 30. Also, income is not measured as comprehensively, only includes some transfers, and omits taxes and tax credits. Life satisfaction and psychological distress are not available, and there is insufficient variation in mortality (only 23 respondents had died). However, we include a scale based on the five-item Mental Health Inventory (ranging from 0 to 15 [worst]). The measure of chronic conditions includes asthma, heart disease, cancer, and diabetes in 2013. Altogether, the NLSY analyses include 15 outcomes.
In Table 4, the NLSY results confirm our major conclusions, with some qualifications. First, the total effects for being black reveal significant and large BW disadvantages for 13 of 15 outcomes. Unlike the PSID, there is a significant BW disadvantage in mental health. However, the black total effect is not significant for high school graduation, and we observe a significant BW advantage in being less likely to have a chronic condition.
Second, childhood wealth and income explain a substantial share of the BW disadvantages. However, childhood wealth/income explain slightly less of the BW disadvantages in several outcomes in the NLSY compared with the PSID. In the NLSY, childhood wealth/income explain more than 50 percent of the BW disadvantage for eight outcomes, 20 to 50 percent of the variation in four outcomes and less than 20 percent of the variation for three outcomes.
Third, the NLSY analyses also contradict the claim that childhood wealth is more important than childhood income. Childhood income explains a larger share of the percentage mediated than childhood wealth for 14 of 15 outcomes. Unlike the PSID, childhood income even explains a larger share of adult IHS wealth and home ownership in the NLSY. The lone exception is that childhood wealth explains more of the percentage mediated for relative wealth. 23 Returning to the PSID, Appendices V to XV in the online supplement show several robustness checks. Appendix V shows the results without transforming childhood wealth and childhood income (i.e., the raw real 2015 dollar values; also see endnote 23). Appendix VI includes Latino and other race respondents (including black and white Latinos). Appendix VII measures wealth excluding home equity (i.e., net financial assets). Appendix VIII measures wealth solely as home equity. 24 Appendix IX only requires respondents have reported childhood wealth and income at least once (as opposed to a minimum of twice). Appendix X observes childhood income only when childhood wealth is also observed. Appendix XI adjusts for Erikson-Goldthorpe (EGP) class. 25 Appendix XII restricts the sample to the PSID's Survey Research Center (SRC) subsample (omitting the Survey of Economic Opportunity). Appendix XIII does not equivalize wealth/income for Notes: The numbers in parentheses are standard errors. All models adjust for sex. Models with full controls available upon request. The mortality models are not estimable because of insufficient variation in the sample. † p < 0.01; * p < 0.05. household size. Appendix XIV includes interactions between being black with childhood wealth/income. The robustness checks buttress the first conclusion. Nearly every appendix in the online supplement shows a significant black total effect for 14 of 18 outcomes. The lone exception is that the SRC sample analyses reveal statistically insignificant total black effects for four other outcomes. The SRC sample is much smaller and less racially diverse, so we treat these estimates cautiously.
The robustness checks support the second conclusion as most robustness checks show that childhood wealth/income explain a substantial share of BW disadvantages. As in the main analyses, childhood wealth and income explain less than 50 percent of the BW disadvantage in home ownership, being partnered, and life satisfaction. Also, childhood wealth/income explain less of adult wealth when wealth is measured only as home equity (Appendix VIII in the online supplement).
The robustness checks confirm and even strengthen our third conclusion that childhood wealth is not more important than childhood income. For most out-comes, childhood income still has larger standardized coefficients/AMEs than childhood wealth. Unlike Table 1, childhood income even has larger standardized coefficients/AMEs than childhood wealth for wealth-related outcomes when (a) not transforming childhood wealth/income (Appendix V in the online supplement) and when (b) excluding home equity (i.e., measuring wealth as net financial assets; Appendix VII in the online supplement). Recall that childhood income had larger standardized coefficients than childhood wealth for two wealth outcomes in the NLSY analyses. Thus, the precise finding that childhood wealth is more important than childhood income to adult wealth outcomes is not as robust.
Finally, Appendix XIV in the online supplement tests for significant interactions between being black and childhood wealth/income. If there are significant interactions, interventions on childhood wealth/income would be less beneficial for black children (Brown 2018;Brown et al. 2016;Finnigan 2014;Flippen 2004). However, of the 36 interaction effects in Appendix XIV, only three (relative wealth, single parenthood, and poor/fair health) imply that black children receive significantly weaker returns from childhood wealth/income. Because 33 of 36 coefficients are statistically insignificant, we conclude it is reasonable to omit the interactions in Tables 1 to 3. 26

Discussion
This study provides novel evidence on how economic resources mediate racial inequalities. Specifically, we analyze how much childhood wealth and childhood income mediate BW disadvantages in adult life chances with Panel Study of Income Dynamics and Cross-National Equivalent File data on children from the 1980s and 1990s who were 30+ years old in 2015. We advance beyond prior research by exploiting longer panel data, more comprehensively assessing adult life chances with 18 outcomes, and measuring income and wealth more rigorously.
The analyses yield three major conclusions. First, there are large BW disadvantages for most outcomes. Second, childhood wealth and income explain a substantial share of most BW disadvantages, although there are several large direct BW disadvantages. Third, the evidence mostly contradicts the prominent and influential claim that childhood wealth is more important than childhood income. Mostly, childhood income explains more of BW disadvantages and has larger standardized coefficients/AMEs than childhood wealth.
The first conclusion, coupled with the significant and large direct BW disadvantages for some outcomes, buttress theories of systemic discrimination. According to this account, racial disparities are pervasive across social spheres, at least some of these disparities result directly from discrimination, and disparities in each sphere are reciprocally linked to disparities in other spheres. Indeed, we find large BW disadvantages across wealth, income, education, family, well-being, and health outcomes. Unlike childhood wealth or income, which are more associated with certain outcomes, large BW disadvantages prevail across almost all outcomes. Audit studies identifying discriminatory behavior in one concrete setting at a time have made crucial contributions. As a complement to those contributions, there remains a need for big-picture, comprehensive demonstrations of BW disadvantages like this study.
BW disadvantages are as large in magnitude as they are pervasive, including many direct BW disadvantages. For home ownership, single parenthood, and being partnered, the direct BW disadvantages are much larger than the standardized coefficients/AMEs for childhood wealth and income. These respondents were children in the 1980s and 1990s, long after the decay of explicit, legalized discrimination. Thus, we demonstrate persistent, pervasive, and large BW disadvantages among those who grew up in recent decades.
On the second conclusion, childhood wealth and income explain the majority of the BW disadvantages for most outcomes. There are exceptions (e.g., home ownership and partnered), but much of the BW disadvantages in adulthood can be traced back to inequalities in wealth and income during childhood. This conclusion is consistent with a fundamental cause theory of racial inequality (Phelan and Link 2015). Wealth and income are flexible resources that white parents employ both strategically and unconsciously for multiple, reliably replaceable mechanisms to facilitate their children's attainment. Because white people have much greater wealth and income than black people, white parents have myriad opportunities to perpetuate and maintain their children's advantages over black children. Of course, wealth and income are obviously not the only resources white people disproportionately have at their disposal. Also, wealth and income are obviously not the only consequential aspects of social class. Nevertheless, wealth and income are pivotal to the transmission of racial inequalities from generation to generation.
On the third conclusion, the evidence mostly contradicts the claim that childhood wealth is more consequential than childhood income to BW disadvantages and adult life chances. Contrary to the strong claims of some wealth scholars, there is little evidence that income fails to deliver benefits as large as those of wealth. On balance, both childhood income and wealth exert independent influences on several outcomes. Nevertheless, for most outcomes, childhood income explains more of the mediated share of BW disadvantages and has larger standardized coefficients/AMEs than childhood wealth. This conclusion is reinforced by the fact that income is one of the most important predictors of wealth. Although the wealth literature has often made bold claims about the unique salience of wealth, few side-by-side tests exist of well-measured income and wealth for a variety of outcomes. Moreover, this study demonstrates how a number of limitations in prior wealth research bias conclusions (see, e.g., Appendix IV in the online supplement). By measuring wealth and income more rigorously (e.g., equivalizing household size, including taxes and transfers, incorporating more observations, correcting the skew in wealth), modeling them together, and avoiding post-treatment bias, this study reveals different conclusions about the intergenerational inheritance and effects of wealth versus income.
We conclude with two policy implications. First, there have often been arguments that reducing wealth or income inequalities will reduce racial inequalities because black people are disproportionately economically disadvantaged. Our analyses confirm that reducing childhood wealth/income inequalities could improve life chances. However, given the magnitude and pervasiveness of BW disadvan-tages, it is likely that large BW disadvantages will persist even if childhood income and wealth inequalities are reduced. Second, partly because of scholarship on racial wealth inequality, there has been enthusiasm for asset-based social policy. We encourage caution about the potential impact of increasing childhood wealth for equalizing life chances. Raising the incomes of families with children-for example, through cash transfers-could yield larger impacts than raising their wealth. Moreover, childhood wealth transfers are unlikely to improve certain outcomes, and even if they did, the improvement would be modest compared with improvements that could result from childhood income transfers.

Notes
1 Although this perspective includes many contributions, it would be difficult to overstate the influence of this particular argument. Google Scholar shows that Oliver and Shapiro (1997) have about 4,000 citations, Conley (1999) has more than 1,600 citations, and Shapiro (2004) has more than 1,200 citations. Second editions have been published of Oliver and Shapiro and of Conley, and all three have been among the most widely taught books in sociology courses.
2 "Childhood" refers to the household/parents when respondents were aged 0 to 17 years. Wealth is the stock of assets minus debts at one point in time, and income is flows (e.g., earnings and transfers minus taxes) over time.
3 Specifically, he analyzes net worth from 1992 to 1994; high school or college graduation by 1995; self-reports of having been held back a grade, expelled, or suspended from school in 1995; unemployment and employment in 1992; hourly wages among the fulltime employed in 1992; premarital childbearing among women aged 18 to 21 years in 1992; and welfare receipt among those originating from low-income families in 1992.
4 For instance, Keister (2005) argues that wealth is superior to income based on analyses of the NLSY. However, her measure of income is not equivalized for household size, does not consistently include transfers, and omits taxes, tax credits, and some transfers entirely.
5 They also find that 2 to 5 randomly chosen years of income data explain 62 to 88 percent of the variation in permanent income in the United States. By contrast, long-term averages (two to nine observations) of net worth only explain 36 percent. Brady et al. (2018) find the same patterns in parallel German data.
6 The confidence intervals overlap slightly for log, but not for IHS and relative rank. Although our estimates are high, Mazumder (2016: Table 1) similarly reports elasticities of 0.65 or greater when childhood family income is observed for 12+ years and adult income is observed once. Our elasticities are high partly because of our post-fisc equivalized household income definition (see below), which improves on even Mazumder.
7 Wealth also has more missingness and requires greater respondent sophistication than income (Rothstein and Wozny 2013). Oliver and Shapiro (1997) acknowledge, "Home equity determination, for instance, presumes knowledge of local housing markets" (P. 57).
8 Appendix III in the online supplement shows that childhood wealth and income correlate 0.46 to 0.76. Although there are modest differences between black and white respondents (partly owing to differing sample sizes), the correlation between wealth and income is generally similar within each group.
9 Oliver and Shapiro (1997) acknowledge, "Income makes up the largest component of potential wealth" (P. 89) and is second only to age as a predictor of net worth (P. 131). Shapiro (2004) finds that "[i]ncome is the most important factor in net financial assets variation" (P. 51) and "[i]ncome is the most important variable determining net worth" (P. 52) . . . "Change in family income is the most important factor in wealth changes" (P. 54).
10 In Conley's (1999) model C for adult net worth (Table A2.5), the coefficient for being black becomes insignificant (from -1.63 to 0.01) when childhood income is included (before childhood wealth is added in model D). This pattern of an insignificant black coefficient when controlling for income and before controlling for wealth also occurs for high school and college graduation (Tables A3.1 to A3.2), being held back a grade (Table A3.3), school expulsion or suspension (Table A3.4), hourly wages (Table A4.3), and welfare receipt (Table A5.2).
11 Chetty et al. (2018) do not have wealth data either, but "proxy for parents' wealth . . . using information on home ownership, monthly mortgage payments, home value, and the number of vehicles from the 2000 Census long form and the [American Community Surveys]" (P. 14).
12 Conley (1999: 14 See https://bradydave.wordpress.com/code-data/. 15 For example, Conley (1999:Appendix A1.1) reports that 11 percent of the sample is black, although his samples vary across outcomes. Assuming 11 percent of every sample is black, he has 119 to 142 black respondents in the largest samples, 69 to 71 black respondents for some outcomes, and only 26 to 29 black respondents for premarital childbearing, unemployment, and welfare receipt.
16 In the online supplement, Appendix VII excludes home equity from net worth, and Appendix VIII only includes home equity. Both yield similar conclusions.
17 Past research typically omits taxes and many transfers. For example, Chetty and colleagues (2014) cannot measure low-income transfers like food stamps. Studies using the NLSY omit taxes and some transfers (Keister 2005;Rothstein and Wozny 2013). Even studies using the PSID without the CNEF do not typically include taxes and all transfers (Bloome 2014;Conley 1999;Duncan et al. 2010).
18 Because adult wealth/income are HH variables, the results could be affected by the small share of the sample who have not formed independent HHs in 2015. However, the results (including in Appendix I in the online supplement) are consistent if we confine the sample to heads and wives/partners (not shown).
19 Keister (2005) writes, "Logging the variable, double logging it, taking the square root, and otherwise transforming it does not markedly reduce the skew because a large portion of households have zero net assets . . . I use the un-logged value of net assets" (P. 265).
In contrast, we find that untransformed net worth has a skew of 15.75, but the inverse hyperbolic sine and relative rank of net worth have skews of -1.1 and -0.09. Equivalizing for household size also reduces skew. Nonequivalized (untransformed) wealth has a skew of 16.37. 20 Following Duncan et al. (2010), we also dichotomized psychological distress into high (>12) or low (<13). The results were consistent (e.g., childhood income was significantly negative).
21 The bivariate correlations between those outcomes and childhood income are actually slightly stronger than the correlations with childhood wealth. However, childhood income (vs. wealth) is more strongly correlated with childhood head's education and childhood single motherhood. Thus, the inclusion of the full controls complicates the mediation process.
22 Including Latino and other race respondents, the N increases to 2,348 for heart attack/stroke, and childhood income is significant (z = −2.42). Including respondents with only one observation of childhood wealth/income, the N increases to 3,025 for chronic conditions and to 3,068 for heart attack/stroke, and childhood income is significant (z = −2.51 and −2.65).
23 In the NLSY, childhood wealth has greater explanatory power when not IHS transformed, but such results still do not support the argument that childhood wealth is more consequential than childhood income for most outcomes.
24 Homes are often claimed to be the largest asset for most Americans. However, in the 2015 PSID sample of all adults (i.e., aged 18+ years), home equity is 45.5 percent of net worth at the median and 26.4 percent at the mean. Because most black adults are not homeowners (unlike the majority of white adults), the median black adult has zero home equity. Only 25.6 percent of sample black adults are homeowners, whereas 59.5 percent of sample white adults are homeowners. For the mean black adult, home equity is 48.9 percent of net worth. In our sample, childhood home equity is 39.4 percent of childhood net worth at the mean and 49.3 percent at the median. For black adults in our sample, median childhood home equity is zero and 49 percent of childhood net worth at the mean.
25 We use the modal EGP class of the lead earner in the HH during childhood. Because EGP class (and the underlying occupation data) has more missing values, samples are smaller. Other analyses (not shown) adjust for two-digit occupation and show similar results. Although there are obviously other aspects of social class not captured by EGP or occupation, the full controls models also include childhood education (and wealth and income).