Reducing Intergenerational Poverty (2024)

1998–2015; and (3) the Census 2000 long form and the 2005-2015 American Community Surveys. The target population comprises all children in the 1978–1983 birth cohorts (a) who were born in the United States or are authorized immigrants who came to the United States in childhood and (b) whose parents were also U.S. citizens or authorized immigrants.¹ After excluding children who cannot be linked to parents or have no address information during childhood, the primary analysis sample comprises 20.5 million children, which covers 96% of the target population. When reporting race-specific estimates, an additional 5% of children for whom race data is missing are excluded.

Parental income is defined as the mean of parents’ household income over 5 years: 1994, 1995, and 1998–2000. Parents’ household income is measured as AGI in years in which a parent files a tax return and is defined as zero otherwise. Children’s annual income is defined similarly, except that data from W-2 forms (which are available in more recent years) are used to impute income for non-filers. Children’s income is measured as the mean of children’s annual incomes in 2014 and 2015 (when they are between ages 31 and 37).

Methods

We use children’s and parents’ income ranks to measure intergenerational mobility. A child’s income rank is measured as his or her percentile in the national distribution of incomes (measured between ages 31 and 37) relative to all others in his birth cohort who are included in our primary analysis sample. Similarly, we measure parents’ income rank as their percentile in the national distribution of parental income for their child’s birth cohort. For any given parental income percentile, we can then directly calculate the mean income percentile of their children, as shown in Figures 2-1 through 2-5.² These relationships provide measures of relative mobility, addressing the question “What are the outcomes of children from low-income families relative to those of children from high-income families?”

___________________

¹ We construct this sample by identifying all children born between 1978 and 1983 who were claimed as a dependent child on a 1040 tax form at some point between 1994 and 2015 by an adult who appears in the 2016 Numident file and was between ages 15 and 50 at the time of the child’s birth. We define a child’s parent as the person who first claims the child as a dependent (between 1994 and 2015).

² From children’s and parent’s income ranks, we can also create a summary statistic measure of intergenerational mobility: the correlation between children’s and parents’ income ranks (rank-rank slope). Chetty et al. (2014a) show that rank-rank slopes provide a more robust measure of relative mobility than another commonly used measure, the elasticity of children’s income with respect to parental income.

by Fisher and Johnson using data from their article provided average economic rank in adulthood for children in the 8th to 12th and 48th to 52nd percentiles of the PSID-based distributions of income, consumption, and wealth.

Poverty Transitions

Table C-2-1 shows intergenerational data on low-income rates and transitions in Internal Revenue Service (IRS) data using the bottom decile or quintile of the child or adult distributions to define low-income thresholds. Overall rates of low-income status for the bottom decile and quintile are, by definition, 10% and 20%, respectively. In both cases, rates are three times as high for Black as for White individuals.

Counterpart numbers for the PSID are shown in the first three columns of Table C-2-2. The Parolin et al. (2022) annual measure of poverty—cash plus tax credits and SNAP benefits relative to the official poverty line—yields an average annual poverty rate of 12.3%. In their main analyses, Parolin et al. (2022) consider children to be poor if their household income fell below the poverty threshold in more than half of their birth-to-age-10 years. Despite this unusual definition, their estimate of childhood poverty—11.3%—is very similar to the average of the annual rates, and it is much higher for Black (38.2%) than White (5.4%) children. Their estimate of poverty in adulthood (10.3%) is also similar, although the racial gap for this adult poverty measure is smaller than for its childhood counterpart.

Given the committee’s focus on reducing intergenerational poverty, the most relevant estimates from these two data sources are of the fraction of children living in poor or low-income families who are also observed to be poor or low-income in adulthood. These are shown in the “Conditional Poverty Persistence” columns of the two tables. Rates are similar for the 20th percentile threshold in tax data (33.7%; also shown in Figures 2-1 and 2-2) and in the PSID data (28.6%; also shown in Figure 2-1). Lowering the threshold to the 10th percentile produces a considerably lower (19.6%) estimate of conditional mobility out of poverty/low-income status. In both datasets, substantially fewer Black than White children escape poverty/low-income status.

Also of interest is the general prevalence of intergenerational poverty—the fraction of all children who live in low-income families in both childhood and adulthood. This is simply the product of childhood poverty rates and intergenerational persistence. So, if, for example, 12% of children grow up in poor families and one-third of poor children are also poor in adulthood, then the fraction of all children poor in both generations is one-third of 12%—or 4% in all. Estimates of intergenerational poverty/low-income prevalence are highest (6.7%) for the IRS-based 20th-percentile cutoff for

low-income status, lowest (2.0%) for the IRS-based 10th-percentile threshold, and in between (3.2%) for the PSID-based poverty measure. In both cases where race-specific estimates are available, the prevalence of persistent poverty/low-income status is higher for Black than for White children.

Income Mobility

Text Figure 2.6 shows adult outcomes for children with household incomes that placed them on the 10th or 50th rung of the childhood economic ladder based on data from tax records. Table C-2-3 and Figure C-2-1 provide data on income, consumption, and wealth from the PSID as described above and in Fisher and Johnson (2023). The two data sources and various measures of economic well-being tell a broadly similar story about intergenerational economic mobility:

• Children starting out on the 10th rung of the economic ladder, on average, climb to considerably higher rungs in adulthood. This is especially true for White children, for whom rungs average between the 36th and 45th, depending on the measure.
• The average adult destinations for Black children are between 11 and 18 rungs lower than for White children.

Translating Percentiles of the Adjusted Gross Income Distribution to Incomes Relative to the Poverty Line

To support the report’s data efforts, commissioned consultants were asked to match the AGI of children (based on their families’ incomes) to a measure of their families’ incomes relative to the poverty line (measured under the SPM). The consultants produced estimates of this correspondence by determining children’s position in the AGI percentile distribution and then calculating the average income-to-needs of children within each percentile, with income-to-needs ratios assessed using the SPM. They first describe the steps they took to match the percentiles of the AGI distribution to income-to-needs ratios in greater detail and the challenges encountered in this process. They use the Historical Supplemental Poverty Measure Data Series (Wimer et al., 2022) and the Annual and Social Economic Supplement to the Current Population Survey, retrieved from IPUMS³-CPS (Flood et al., 2022) to produce these results.

___________________

³ “IPUMS” stands for Integrated Public Use Microdata Series.

3. If their parents are non-filers and the children do not live with any relatives who are filers, a non-filing tax unit is created that links the children to their parents.
4. If they do not live with their parents, and none of their relatives files taxes, a non-filing tax unit is created that links the children to their eldest relative.

Once we identified these tax units, we calculated the total AGI across all members of the tax unit using the ASEC-CPS AGI variable. The AGI variable was constructed by the Census Tax Model and is described in the IPUMS-CPS codebook as including “an individual’s total gross (pretax) income from taxable sources minus certain items, such as individual retirement plan contributions (payments to a Keogh plan or a deductible Individual Retirement Account), alimony paid, medical savings accounts, and non-reimbursed employee business expenses.”⁴ In the case of joint-filers, we also ensured that income sources were not double-counted when totaling AGI across tax unit members. The total AGI of the tax unit represented the AGI of each child’s family and is the primary variable used in the subsequent analyses described below. We produced these estimates for data representative of 1994, 1995, 1998, 1999, and 2000.

Determining Children’s Income-to-Needs Ratios

Next, we determine the income-to-needs ratio of all children in the data using variables available in the Historical SPM Data Series (Wimer et al., 2022). As described by Fox et al. (2015) and Wimer et al. (2016), the Historical SPM Data Series includes the necessary inputs for measuring poverty under the SPM from 1967 to 2009 (before the data were available in the ASEC-CPS). We define children’s income-to-needs ratio as the total resources for the SPM poverty unit divided by the poverty threshold for their unit, all measured using the SPM and available in the Historical SPM Data Series. See Fox et al. (2015) for a more extensive discussion of the construction of the SPM resources variables, and Nolan et al. (2017) for a discussion of the SPM poverty thresholds construction. Again, we produced these results for data representative of 1994, 1995, 1998, 1999, and 2000.

___________________

⁴ Additional information on this variable is available at https://cps.ipums.org/cps-action/variables/ADJGINC#description_section

___________________

⁵ The number of teenagers flagged as filers by the Census Tax Model likely does not line up with the tax data, but that is a limitation that we cannot avoid.

⁶ When combining minor filers and children claimed as dependents, the median AGI is $38,004.

these bottom percentiles have higher income-to-needs ratios. This pattern is depicted in Figure C-2-4, which plots the income-to-needs from our primary results (also presented in Figure C-2-2), and when we include minor filers in the analysis. We do not include this group of children in our primary estimates but provide supplemental results inclusive of them in Table C-2-5.

1970.⁷ Income distributions for parents with children in other birth cohorts are estimated analogously.

In their baseline series, which we show in Figure 2-10, Chetty et al. adjust for inflation using the benchmark CPI-U-RS. Choices of the inflation index can affect measures of absolute mobility significantly, as pointed out by Strain (2020) and by others; Chetty et al. demonstrate that the qualitative conclusion of declining absolute mobility is robust to reasonable choices of the inflation index, but the magnitudes of the decline remain debatable.

Methods

Chetty et al. (2017) focus on a measure of absolute mobility to address the question: “What fraction of children born in a given year (e.g., 1940, 1970) grow up to have a household income that exceeds that of their parents?” To do so, they combine children’s and parent’s marginal income distributions constructed as described above with a nonparametric rank-rank copula that measures relative mobility. This copula is a 100 × 100 matrix, where each (i,j) cell in the matrix indicates the probability that a child born to parents with income percentile i will grow up to have income percentile j.

The copula is constructed using modern tax records based on near-population data in Chetty et al. (2014a). The sample used to construct the copula comprises children in the 1980 to 1982 birth cohorts who are linked to parents by being claimed as dependents on federal income tax forms. This comprises more than 10 million parent-child pairs. To construct the copula, children’s and parent’s income are first measured using concepts similar to those described above. For tax filers, income is defined as adjusted gross income plus the nontaxable portion of Supplemental Security Income and Social Security Disability Income. For non-filers, income is the sum of W-2 wage earnings, Supplemental Security Income, Social Security Disability Income, and unemployment insurance income. If individuals’ incomes are missing in these data, they are assigned a value of zero. Children’s incomes are mean income in 2011 and 2012 (when they are between the ages of 30 and 32) and parents’ incomes are mean taxable income between 1996 and 2000, the first 5 years for which population tax records are available.⁸

To produce their baseline series of absolute mobility—shown in Figure 2-10—Chetty et al. apply the copula estimated for the 1980 to 1982 cohorts

___________________

⁷ This approach assumes that the income distribution of those who have children after age 35 is representative of the income distribution of the general population. Chetty et al. (2017) show that results are robust to restricting attention to parents who have children between ages 25 and 35.

⁸ Parents are between ages 30 and 60 when we measure their incomes. Since the distribution of income ranks is fairly stable between ages 30 and 60 (Chetty et al., 2014a), this approach provides a reasonable estimate of the copula that one would obtain using income ranks at age 30 for all parents.

to all cohorts, effectively assuming that relative mobility did not change across cohorts. Using children’s and parent’s marginal income distributions, for each child’s income percentile r^k and parent’s income percentile r^p, they calculate whether children at r^k earn more than parents at r^p. The copula provides the probability that each children-parent rank pair (r^k, r^p) occurs. They then measure absolute mobility as the fraction of cases where children at r^k earn more than parents at r^p, integrating over the copula.

The strong assumption of constant relative mobility over time is based on evidence from Chetty et al. (2014b) that relative mobility has remained stable overall in recent cohorts. Furthermore, Chetty et al. show that their estimates of absolute mobility are insensitive to assumptions about the degree of relative mobility in early cohorts (in the 1940s and 1950s), since virtually all children out-earned virtually all parents in those cohorts, implying that absolute upward mobility rates were close to 100% during that time period irrespective of the degree of relative mobility.⁹

___________________

⁹ Formally, they compute bounds on absolute mobility using linear programming methods to search over all plausible copulas for the maximum and minimum levels of absolute mobility consistent with the marginal income distribution, and show that these bounds are very tight in early cohorts.

The second statement above suggests a tight focus on policies and programs specifically for Black boys and men, even though the first statement identifies the cycle of intergenerational poverty for females, with Black girls each generation growing up in lower-income households than White girls in each generation. Ignoring investments in Black girls and women would maintain this cycle. The committee also highlights the importance of a relational, historical, intersectional, and (qualified) resilience approach to understanding these gendered patterns of Black-White disparities.

First, as stated in Chapter 2, the fact that Black women exhibit similar individual mobility but not household mobility as White women “is probably due to the fact that Black women are more likely than White women to live in single-parent families, and thus also more likely to be the main breadwinners in their families.” Since Black women are more likely to be heads of households than White women, their average socioeconomic position will be more disadvantaged than that of similar White women. Within coupled households (that are same-race and opposite-gender), Black women’s earnings act as compensatory for the relatively low earnings of Black men, whereas White men’s higher earnings compensate for the lower earnings of White women. Too close of a focus on Black men’s outcomes ignores the need for policies that support low-income Black girls and women in their need to generate even greater socioeconomic resources.

Second, the gendered contours of racial discrimination and racism are historically specific and change over time. Black women disproportionately experienced rape during slavery, whereas Black men disproportionately experienced lynching after slavery. The 15th Amendment, ratified in 1870, granted only Black men the right to vote, whereas Black (and White) women were disfranchised until 1920 with the ratification of the 19th Amendment. Black men and women both experienced racism in the labor market in the early- to mid-20th century, but Black women were disproportionately confined to domestic labor whereas Black men were relegated to unskilled blue-collar jobs (outside of agriculture). The historical violence, exclusion, and exploitation explored in this chapter were experienced differently by Black men and women. Similarly, the forms of discrimination and racism today are different and may yield different disparities across race and gender. For example, while low-income Black boys have lower rates of upward mobility than Black girls, Black men on average have substantially higher wealth levels than Black women (Chang et al., 2021).

While attention to Black men may be relevant to the narrow question of intergenerational income mobility, such a focus reflects a presentist bias regarding the changing nature of gendered racism and does not allow for discussions of policies and interventions that can support Black women’s mobility while also improving Black men’s outcomes.

Third, the comparison by race within genders obscures the fact that both Black and White women trail White men in achieving upward mobility. Roughly 39% of Black and White women who grew up poor rise to the top three quintiles of the individual income distribution as adults, compared with 54% of White men. Gender-based discrimination and what might be called “structural sexism” (Homan, 2019) reduces the socioeconomic mobility and well-being of all women (Ridgeway & Correll, 2004), just as structural racism negatively effects the outcomes of all Black people, with Black men and women showing different outcomes relative to White people in different domains (see, e.g., Kim, 2009; Paul et al., 2022). Only comparing Black women to White women and Black men to White men controls away the workings of racialized gender structures in education, health, housing, families, law, and the labor market. An intersectional approach recognizes that Black women from low-income backgrounds still do not enjoy the same opportunities for upward mobility as similar White men.

Finally, the performance of Black women also reflects their effort and resilience, and individual- and community-level protective factors. This success goes even further in some domains. For example, at every parental income level, Black women have higher college attendance rates than even White men (Chetty, 2020, p. 716). These outcomes are attained in spite of racism and discrimination (and sexism), not because racism and discrimination do not exist. Black girls are more likely than White girls to experience school discipline (Morris, 2016); less likely to be seen by adults as needing nurturing, care, and support (Epstein et al., 2017); and less likely to receive a substance abuse treatment referral in the juvenile justice system (Johnson et al., 2022), to name just a few examples. Yet they show greater achievement in some areas. Upward mobility is also not without its health costs. Research on “skin-deep resilience” has found that the positive mindsets and behaviors of upwardly mobile low-income youth pay off in educational attainment and other outcomes, but also take a toll on physiological health (Chen et al., 2020; Miller et al., 2015). This finding holds across race and gender, but studies have not looked simultaneously at the effortful striving required for upward mobility and experiences of racial (and gender) discrimination. The resilience shown by low-income Black girls who climb the socioeconomic ladder must be qualified by these factors and does not obviate a discussion of the contexts of discrimination and structural racism within which Black girls and boys grow up.

framework on inequality has several advantages. According to National Academies (2022b, p. 29):

In other words, racial inequality is often used to refer to system-level racial gaps whereas racial disparities are used to characterize differences in a single outcome or variable.

Studies of racial discrimination using statistical, audit, and experimental methods (Pager & Shepherd, 2008), as well as several meta-analyses, find clear patterns of racial discrimination against Black Americans in areas such as employment, housing, health and mental health care, criminal prosecution, conviction, and sentencing, consumer markets, and in children’s placement into gifted classes (Anwar et al., 2012; Bertrand & Mullainathan, 2004; Besbris et al., 2015; Card & Giuliano, 2016; Faber & Mercier, 2022; Gaddis, 2015; Ge et al., 2020; Kugelmass, 2016; O. Mitchell, 2005; Pager & Shepherd, 2008; Quillian et al., 2020a; Wu, 2016). There is less research on racial discrimination against Native Americans, but audit, self-report, and correlational studies show frequent experiences of discrimination across a range of settings (Abramson et al., 2015; Findling et al., 2019; Hurst, 1997; Puumala et al., 2016; Robert Wood Johnson, 2018; Stepanikova & Oates, 2017; Turner & Ross, 2004; Wilmot & Delone, 2010; Weber et al., 2018; however, see Button & Walker, 2020, which does not find employment discrimination in a large audit study). Discrimination is relevant for intergenerational poverty because it excludes Black people and Native Americans from access to and participation in contexts that enhance opportunities or exposes them to practices that reduce opportunity.

The term “racism” is sometimes used to describe individual dispositions linked to beliefs in racial stereotypes and negative sentiments against a racialized outgroup. Survey researchers have designed scales to measure racism among individual respondents (National Research Council, 2004), and psychologists have designed experiments to detect racial bias among research subjects (see, for example, Eberhardt et al., 2004; Geller et al., 2021). This evidence demonstrates the ways in which belief, sentiment, and cognition operate at the individual level to drive decision making and other behavior in a direction that is harmful to racialized outgroups. Beyond the

level of the individual, social organizations and institutions (e.g., neighborhoods, families, markets, health care systems) are often run or structured in a way that is harmful to racialized outgroups, even in the absence of individual-level racism. Scholars have described this configuration of social relations as “structural racism.”

Some scholars emphasize the historical character of structural racism, in which “whiteness, a privileged racial category justified by negative racist stereotypes, [is] passed down from generation to generation, so as to become acceptable, normal, and part of the public common sense” (Marable, 2001, p. 13; see also Rucker & Richeson, 2021). Whereas racial inequality describes (perhaps enduring and multidimensional) group-based differences, structural racism attributes such inequality to social organization and institutions. Against this background, structural racism is defined as the operation of race as an organizing social force to enact, codify, or enable the oppression of one or more groups. Once a society becomes racialized, invidious racial distinctions affect “social relations and practices” at “all societal levels” (Bonilla-Silva, 1997). Structural racism is not defined by individual bigotry, prejudice, or discrimination but rather is based on how social, economic, and political institutions of government and civil society are organized by law, policy, practice, and norms. In this way, the argument that posits structural racism as a force contends that inequalities by race occur specifically because of social and institutional factors that perpetuate racial inequality.

Understanding the historical roots of structural racism is crucial to recognizing its effects today and how it has evolved over time (Glenn, 2015; Harris, 1993). Structural racism is reflected in the distribution of political power, economic wealth, material conditions, and equal access to, or fair treatment by, social systems over time, from housing to health care to the criminal justice system (Feagin & Elias, 2013). However, these impacts are neither linear nor constant. Since laws change and social forces are dynamic, shifting with politics, demographics, economics, and social movements, structural racism has evolved over time (see. e.g., Alexander, 2010).

A classic example of structural racism is the passage of voter disenfranchisement laws after the passage of the 15th Amendment. Laws establishing poll taxes, literacy tests, and grandfather clauses disproportionately excluded Black men from registering and casting a vote (Baker, 2022; Manza & Uggen, 2006). They were written without any explicit reference to race in order not to violate the 15th Amendment, which disallowed racial discrimination in voting. Yet their intentions to exclude Black voters were clear, and these laws produced large racial disparities in voter participation between White men and Black men, and often complete Black disenfranchisement. These disparities were the evidence of legally sanctioned second-class citizenship for Black Americans. The 24th Amendment to the Constitution

eliminated poll taxes, and the Voting Rights Act of 1965 added enforcement powers to combat the structural racism of ostensibly race-neutral disenfranchisement laws. Yet law and practice are not equivalent, and later policies, such as partisan redistricting, felon disenfranchisement, and voter identification restrictions have continued to disproportionately curtail the franchise for Black citizens. Thus, structural racism encompasses both those neutral policies that are motivated by racist intent and those that reinforce racial hierarchies resulting from past intentional racism, regardless of motivation (Roithmayr, 2014).

Researchers have developed novel measures to study contemporary structural racism, including one that combines indicators of political participation, employment and job status, educational attainment, and judicial treatment (Lukachko et al., 2014). In this study, structural racism is defined by state-level racial disparities across those four domains. Using this measure, the researchers found that Black people living in states with high levels of structural racism were more likely to experience myocardial infarction relative to their counterparts living in states with low levels of structural racism (Lukachko et al., 2014). Another group of researchers developed a measure of state-level structural racism that combines indicators of residential segregation, incarceration rates (though not adjusted for crime), educational attainment, economic indicators, and employment status. This latter study found that higher levels of structural racism were associated with a larger disparity between Black and White victims of fatal police shootings (Mesic et al., 2018; National Academies of Sciences, Engineering, and Medicine [National Academies], 2021).

A final illustration of structural racism is how ostensibly objective artificial intelligence algorithms produce racially disparate outcomes (Benjamin, 2019; Noble, 2018). Obermeyer et al. (2019) document how such algorithms in health care result in doctors providing fewer medical interventions and less care for Black patients. Manifold structures in American society—e.g., the tax code, criminal fines and fees, and the child welfare system (Brown, 2021; Jacobs, 2014; U.S. Commission on Civil Rights, 2017; Williams, 2022)—have built-in, unstated stereotypes, biases, and rules that contribute to the ongoing impoverishment of Black and Native American people.

Contemporary scholars view racial inequality as at least partly structural, cumulatively generated through the mutual and reciprocal interaction of institutions (Powell, 2008; Ray, 2019; Sampson, 2012; Williams & Collins, 2001; Wilson, 1987). The mechanisms of structural racism today can be (although they are not always) found in an array of public policies such as zoning laws and in the pricing of goods and services, as well as in credit risk scoring to limit access to loans or rental housing (using income, zip codes, and arrest records). Still, it is often the case that contemporary forms

of structural racism can be traced back to racially exclusive or racially targeted policies and practices of earlier moments in history.

The National Academies noted in a 2017 report on racial health disparities (National Academies, 2017d, p. 104):

Historical Roots of Racial Disparities in Intergenerational Mobility

As noted in the main text, Native Americans and Black Americans stand out as groups subjected to centuries of structural racism rooted in beliefs about White supremacy. The notion of intergenerational poverty has a presentist bias, a narrow time band, and a limited definition of well-being. In other words, we measure only economic status (wages, income, wealth), across at most three generations, in the most recent time period for which we have the best economic data. For example, research on multigenerational poverty finds that 1 in 5 Black families experience poverty across three generations, compared with roughly 1 in 100 White families (Winship et al., 2021; also see Collins & Wanamaker, 2022; Pfeffer & Killewald, 2018). A much longer historical view, however, shows that African and Native American peoples were not “poor” before European contact, but rather sustained themselves and often prospered for millennia in complex societies with functioning governing and economic systems (Carlos et al., 2022; Rodney, 2018).

The plunge into poverty and its persistence across generations in what became the United States is the result of successive waves of theft, destruction, and exploitation of people, land, and property into the present. Among the most glaring forms of historical structural racism that set Black and Native Americans on a course of socioeconomic disadvantage relative to other groups are (a) forced migration and land theft, (b) chattel slavery, labor exploitation, and property theft, (c) scientific racism, and (d) forced assimilation and legalized racial discrimination enforced by racially oppressive institutions. These mechanisms are discussed in detail in the sections that follow.

the United States, had been given away to more than 1.4 million claimants, virtually all of whom were White (approximately 3,500 Black people received land), for a trivial filing fee (Merritt, 2016). Claimants took legal possession of the land after 5 years, conditional on 5 years of continuous residence on the land, building a home, and farming the land. As of 2000, an estimated 46 to 93 million people were descendants of families who took up this “free land” (Shanks, 2005) and the wealth it has generated.

The General Allotment Act (or the Dawes Act) of 1887 aimed to allot federal lands to individual Native American families for private ownership. Its execution, however, resulted in the transfer of roughly 27 million acres of tribal land to non-Native owners (Royster, 1995), a checkerboard ownership pattern frustrating tribal governance and development, and a fractionated pattern of ownership between multiple heirs and the federal government. Comparing a Minnesota reservation that was allotted to one that was not, Akee (2020) found that allotment decreased Native American homeownership and displaced people into wage labor as opposed to

self-employed farming from 1900 to 1910. Leonard et al. (2020) found in a national study of allotted lands that fractionation was associated with decreased per capita income among Native Americans as recently as 2000. Frye and Parker (2021) show that tribal areas with constrained sovereignty over their lands—one result of the Indian Reorganization Act of 1934—have lower per capita incomes. Contemporary forms of discrimination and dispossession are illustrated in the withholding of loans to Native American (and Black) farmers and ranchers, resulting in disproportionate foreclosure and property loss (Carpenter, 2012), and in the higher prices of mortgage loans to Native American home buyers (Cattaneo & Feir, 2021).

The Indian Appropriations Act of 1851 created the U.S. reservation system, which increased the government’s control of Native American people and natural resources and expanded territories for White settlements. In the first of several appropriation acts, funds were allocated to move Native people living in the West onto reservations. Most Native lands today are trust lands, meaning the federal government holds the legal title to the

land, and that Native tribes or individual tribal members lack ownership and control over the land.

More recently, in the 1950s and 1960s, Congress passed 12 “termination” bills, which ended federal responsibilities for tribes in several states and turned over governing power to the states. By 1957, 2.5 million acres had been removed from federal trust protection (Philp, 1983, p. 166). However, federal trust protection is itself no protection against dubious practices of extraction (Fixico, 2011), as illustrated by a $3.4 billion settlement with the U.S. government in Cobell v. Salazar (2009), which found that the Department of Interior and other agencies had breached their trust obligations with hundreds of thousands of Native American plaintiffs in the class. Overall, roughly 56 million acres are held in federal trust as Native reservations, a mere fraction of the 1.9 billion acres that make up the contemporary United States, once occupied by Native peoples.

In all, Native Americans experienced a 98.9% reduction in their access to land from the period of European arrival to the present. The remaining lands are more susceptible to climate risks and less abundant in mineral resources than the territory on which they historically resided (Farrell et al., 2021). These collective experiences of dispossession remain salient. Whitbeck et al. (2004) report that roughly 42% of a sample of Native American respondents in the U.S. Midwest and Canada think about this loss of land at least monthly, and greater proportions think about the loss of language, spiritual ways, and culture.

Chattel Slavery, Labor Exploitation, and Property Theft

The story for Black Americans begins not with land dispossession but with theft of labor and personhood. The first enslaved Africans arrived to what would become the United States in the early 1500s (Guasco, 2014; Johnson, 1923), prior to the 1619 English settlement at Jamestown, Virginia. The Anglo-centered history records slavery as flourishing in the United States for 244 years until the Emancipation Proclamation of 1863, though the timeline is likely much longer given bondage on U.S. shores decades before 1619 and for 2 years after 1863.

Indentured servitude began in early colonial America in response to the need for labor. Most indentured servants were poor Europeans—Irish, Scots, and English—but some were Africans who had been sold into bondage. Typically, servants worked off their indentures and were freed 4 to 7 years after they began. During the earliest years of settlement, no slave laws were in place. However, with increasing demands for labor, rising costs of indentured servants, and increasing demands for land from newly freed servants, colonists established laws in the mid-1600s to hold Africans and their descendants in perpetual servitude. Statutes decreed slavery as a lifelong

and heredity condition and enslaved people as the legal property (i.e., chattel) of their “owners” to be bought, sold, traded, or willed as owners wished (Bridgewater, 2005; Franklin, 1969). This system made labor more profitable and readily renewable across generations. Laws ensured that any children born to an enslaved woman belonged not to the mother but to the White man who owned the mother. Enslavers’ biological children—born of the rape of enslaved women—had no legal right to any of the father’s property, which was ordinarily granted via paternity (Bridgewater, 2005).

When the 13th Amendment was adopted as part of the U.S. Constitution in 1865, officially abolishing chattel slavery, the newly freed people had no land, capital, or equipment for farming. Ida B. Wells, a well-known Black journalist who lived in the late 19th and early 20th century, aptly summed up the situation, noting that the end of slavery left Black people “homeless, penniless, ignorant, nameless, and friendless… We were turned loose to starvation, destitution, and death” (cited in Darity & Mullen, 2020, p. 9). After leading the Union army to victory over the Confederate states, General Sherman ordered the redistribution of 400,000 acres of land along coastal areas of South Carolina and Georgia (40 acres and a mule per family) to help newly freed Black individuals gain economic independence. Shortly thereafter, President Lincoln’s successor Andrew Johnson rescinded the order, and the land remained in White possession (Darity & Mullen, 2020; Saito, 2020). Darity and Mullen (2020) argue that had Sherman’s order been carried out, it “would have dramatically reversed black asset poverty and reduce blacks’ economic vulnerability across generations” (p. 175).

Just as the economic prosperity stolen from Native Americans through land and people theft is incalculable, so are the profits wrought from slavery, which touched every colonial and U.S. institution, from law to the economy to social mores to universities. Economic historians estimate that the present value for lost wages during the period of slavery range from $2.1 to $4.7 trillion, not accounting for land loss or the opportunity costs of educational and discriminatory practices (Darity & Mullen, 2020; Marketti, 1990). Labor exploitation and the stealing of property, land, and assets from Black Americans continued after the formal dissolution of slavery. Between 1865 and the beginning of World War II, Black Americans in the South experienced widespread labor theft and exploitation reminiscent of slavery that impoverished generations of Black families (e.g., through sharecropping, convict leasing, and peonage; Blackmon, 2008; Carper, 1976; Mancini, 1996). Sharecropping was a system of land tenancy in the South that kept Black people working the land they had worked as enslaved people. With little to no capital of their own, they could not even afford the provisions to plant and were thus forced to buy them on credit from White landowners, with agreements to repay the costs upon harvest. Planters kept Black tenants in perpetual debt through unfair contracts and accounting

practices, backed by Jim Crow laws that accorded few to no rights to Black people and by White terrorist violence. Box C-3-2 illustrates the experience of a sharecropper in Florida in the 1920s.

The forced labor of prisoners, overwhelmingly Black men, was also rampant during this period. In southern states in the period after the Civil War—and in compliance with the 13th Amendment that allows for “involuntary servitude” in the case of criminal conviction—southern jurisdictions arrested and convicted hundreds of thousands of Black men and some women (LeFlouria, 2015) and sold their labor to private corporations without pay, or with remuneration going to the local government. To replace the social controls of slavery removed by the Emancipation Proclamation, state legislatures in the south passed an array of interlocking laws referred to as “Black Codes” in 1865–1866 that barred Black citizens from voting, serving on juries, testifying against White people in court, and working in skilled jobs. These laws criminalized many aspects of Black daily life (e.g.,

unemployment, indigency, and disrespect of White people were made illegal) to provide pretexts for jail terms. Black people were often unable to pay even minor fines and, as a result, were sentenced to labor (Carper, 1976; Mancini, 1996).

In his book, Slavery by Another Name, Douglas Blackmon defines convict leasing as “a system in which armies of free men, guilty of no crimes and entitled by law to freedom, were compelled to labor without compensation, were repeatedly bought and sold, and were forced to do the bidding of white masters through the regular application of extraordinary physical coercion” (2008, p. 4). Supposed convicts were sold to work in mines, on railroads, on plantations, and in timber fields. The system was highly profitable for private firms who paid no wages and for the state systems that sold the labor of people convicted of “crimes.” One contract to the Walker Coal and Iron Company in Georgia in 1874 “called for the leasing of 100 prisoners for five years at $11 per convict per year” (Mancini, 1978, p. 341), or about 7 cents per day, when the average daily wage for miners in the United States at the time was $1.97 (Abbott, 1905, Table XII).

Convict leasing was used not only to enrich private and public coffers but also to impoverish newly freed Black people. Sociologist Christopher Muller (2018) shows that in Georgia after emancipation, Black imprisonment in the convict lease system for property crimes increased most in parts of the state where Black people were gaining an economic foothold through valuable land ownership and leaving sharecropping through urban residence, while White incarceration remained steady. White southerners used punishment as a method to interrupt this progress and to instead return Black men and some women to the status of unpaid laborers. Not only did disproportionate incapacitation stymie economically productive activities, but convict labor was essential to the South’s agricultural production and industrialization, and it enriched the jailers and their jurisdictions.

Even Black Americans who were able to escape this exploitative system had to contend with discriminatory laws and lending practices that largely barred Black people from land ownership, apprenticeship programs for skilled training, trade unions, and other routes to upward mobility (Lancaster, 2000). During Reconstruction (1865–1877), the period after the Civil War when efforts were made to reintegrate Confederate states and redress the inequities of slavery, Black Codes were abolished. However, after Reconstruction ended, many of the provisions of these codes were reenacted in Jim Crow laws. Adopted in southern states in the 1870s and 1880s and enforced until 1965, Jim Crow laws legalized and mandated racial segregation in all public facilities (e.g., schools, transportation, hospitals, prisons, morgues), relegating Black people to inferior treatment, jobs, and facilities (Franklin, 2013; Lancaster, 2000).

the Reconstruction era. In 1898, White residents staged what historians now label as a violent coup in order to regain power (Zucchino, 2020). Estimates are that between 60 and 250 Black people were killed. The economic effects were evident in Black employment and labor status. All of Wilmington’s Black city workers were fired and replaced by White employees. The number of Black business owners declined, and Black businesses were disproportionately displaced from the downtown area (Hamilton & Darity, 2006).

The economic toll is even more quantifiable for the Tulsa massacre of 1921, in which as many as 300 Black people were killed. White mobs—deputized and armed by the local police—destroyed roughly 35 acres of the Greenwood section of the city, called Black Wall Street for its concentration of thriving Black-owned businesses (Messer et al., 2018). The Oklahoma Commission to Study the Tulsa Race Riot of 1921 (2001) reported claims for property damage of $1.8 million, or nearly $30 million in today’s dollars. Messer et al. (2018) figure that if the same 1,256 homes were destroyed in Tulsa in 2018, the cost would be roughly $150 million. None of the property claims were ever repaid to Black families.

This historical violence and property loss reverberates in present-day socioeconomic well-being. Albright et al. (2021) show that the Tulsa massacre lowered the occupational status of Black Tulsans into the 1940s and lowered their homeownership rates up to 2000, the last year of observations. Moreover, Black homeownership rates were also lower in Black areas across the country that received significant newspaper coverage of the Tulsa massacre. Just as the failure of the Freedman’s Bank made Black people leery of banks (Stein & Yannelis, 2020), the Tulsa massacre “provided a warning of the danger of the accumulation of wealth through home ownership” (Albright et al., 2021, p. 31) that has persisted for decades. In addition to the reverberations from historical forms of expropriation, more recent and contemporary practices of asset extraction are manifest through the racially disparate use of eminent domain (U.S. Commission on Civil Rights, 2014), unfair property tax assessments (Atuahene & Berry, 2018), and Black home loss through foreclosure and institutional purchases (Hwang, 2019), just to name a few.

Scientific Racism

During the mid to late 1600s, race ideology developed as justification for White supremacy, land theft and genocidal wars against Native peoples, and permanent enslavement of Black people. Native people were regarded as fierce, as evil, as savages fated by God for conquest, and Black people as inferior subhumans without capacity for reasoning, imagination, and sentiment (Franklin, 1969; Smedley, 1998; Stannard, 1992). Science played

a supporting role in hardening early folk beliefs among White colonists and settlers about “race” and inequality. From the 18th century and well into the 20th century, anthropologists, biologists, and psychologists developed techniques to measure differences in physical characteristics as a basis for racial classification and subordination (Gould, 1996). For example, during slavery, a physician invented the diagnosis of “drapetomania,” a mental illness hypothesized to cause enslaved Africans, who were thought to be naturally servile, to flee captivity, for which the prescribed treatment was severe whipping (Opara et al., 2022). “Studies” using these techniques purported to confirm the inferiority of Black and Native people, and to support the belief that “White blood” increased the mental capacity of Black and Native children of mixed racial backgrounds (Guthrie, 1976; Smedley & Smedley, 2005).

Integral to the social construction of race ideology that developed during this period was the valuation of phenotypic traits associated with European ancestry and the devaluation of those associated with non-White people, biases that exist to this day among Black Americans (e.g., Adams et al., 2016; Maddox & Gray, 2002) as well as some Native tribal groups (Brown et al., 2018). Color and phenotypic hierarchies established during slavery also cast a long shadow in establishing practices of anti-Blackness (Franklin, 1969; Frazier, 1957). Studies have found that lighter skin tone among Black people is positively correlated with socioeconomic outcomes (Monk, 2021), higher self-esteem among Black youth (Adams et al., 2020), attribution of more positive traits (Maddox & Gray, 2002), lower rates of school suspension, better physical health, greater upward mobility, and increased odds of full-time employment and college attendance (Han, 2020; Hannon et al., 2013; Hargrove, 2019; Keith & Herring, 1991; Ryabov, 2013). Analyzing archival records of capital murder cases, Eberhardt et al. (2006) showed that defendants whose appearance was perceived as more stereotypically Black were significantly more likely to receive a death sentence than defendants whose appearance was perceived as less stereotypically Black, controlling for numerous factors. Persistent skin tone stratification illustrates the lasting effects of the ideologies created to justify racism and colonialism.

Forced Assimilation and Legalized Racial Discrimination

The original U.S. constitution directed that for purposes of representation and taxes, the population would be determined “excluding Indians not taxed, [and including] three fifths of all other Persons.” This separation and erasure of Native Americans and sub-humanization of Black people is built into the fabric of the United States and has clear contemporary manifestations, such as in the “willful blindness toward Native American

victimization” by law enforcement (Perry, 2006, p. 412; also see Fryberg & Stephens, 2010), and the digital association of Black people with apes in facial recognition internet searches (Noble, 2018, p. 6).

Anti-Indigeneity undergirded a variety of U.S. government policies that pressured Native Americans to assimilate. The Dawes Act of 1887, for example, authorized a division of tribal lands into individual plots to encourage Natives to farm and ranch like White homesteaders (https://www.nps.gov/articles/000/dawes-act.htm). Laws such as the Civilization Fund Act of 1819 funded schools and forced Native children to attend boarding schools where displays of Native culture and identity were forbidden. By 1925, some 60,000 Native children—about 80% of Native school-age children—had been forced to attend boarding schools. In keeping with the philosophy of assimilation, “Kill the Indian, Save the Man,” children in these schools were forbidden to speak their native language, wear traditional clothes (which were replaced by uniforms), or perform tribal practices (replaced by Christian practices). Long hair was cut, braids were prohibited, and tribal names were replaced by English-language names (Adams, 1995; Pember, 2019). Contact with family and community members was discouraged and sometimes forbidden altogether (Adams, 1995; American Indian Relief Council, n.d.; Pember, 2019). Evidence of abuses of students in off-reservation boarding schools led to passage of the Indian Child Welfare Act in 1978, which gave Native parents the legal right to refuse their child’s placement in off-reservation schools. Many large Native boarding schools closed in the 1980s and early 1990s. Some located on reservations were taken over by tribes. Still, as of 2021, 15 such boarding schools remain open (Blakemore, 2021; National Native American Boarding School Healing Coalition, 2020).

Forced assimilation is a form of structural racism whose psychological consequences for Native Americans is the focus of a growing number of studies. Compared with all other racial groups, Indigenous youth and adults have higher rates of suicide, substance use disorders, and mental health problems. Research has linked these disparities to both current and historical racial discrimination (Skewes & Blume, 2019) or “historical trauma,” conceptualized as “cumulative emotional and psychological wounding over the lifespan and across generations, emanating from massive group trauma experiences” (Yellow Horse Brave Heart, 2003, p. 7). The wounds include the loss of religion, language, and culture—as experienced by Native youth removed from their families to live in boarding schools—as well as loss of ancestral homelands.

Observational work suggests that Native parents who are alienated from Native cultural and spiritual traditions may be less able to provide their children with a supportive, nurturing family environment, putting their children at increased risk for mental and substance abuse disorders (Zimmerman &

Shannon, 2013). Evans-Campbell et al.’s (2012) study of Native Americans found that former boarding school attendees, compared with non-attendees, reported higher rates of current illicit drug use, alcohol use disorder, suicidal ideation, and attempted suicide. In addition, adults raised by former attendees of boarding schools, when compared with their counterparts, were more likely to have an anxiety disorder, PTSD symptoms, and suicidal thoughts in their lifetime. In a similar vein, there is evidence of deleterious multigenerational effects of relocation experiences on Indigenous families. In Walls and Whitbeck’s (2012) longitudinal study, grandparent-generation participation in government relocation programs, in which reservation-dwelling Native Americans moved to large urban areas for vocational training and job opportunities, not only harmed the well-being of the grandparent generation (e.g., alcohol and drug problems, depressive symptoms) but also rippled downward to harm the mental health of the parent and youth generations. These studies suggest that historical trauma could contribute to higher rates of intergenerational poverty among Native Americans by undermining the psychological functioning and nurturing capacities in the grandparent and parent generations, in turn increasing substance use and mental health problems in the child generation that ultimately lower educational and occupational attainment and reduce upward mobility.

For Black Americans, slavery and Jim Crow laws throughout the United States continued to shape economic and social opportunity with lasting impacts into the present day. Baker (2022) finds that a composite measure of Black people’s state-level exposure to slavery, sharecropping, disenfranchisement, and resistance to desegregation is significantly correlated with contemporary Black poverty and Black-White disparities in poverty. Similarly, Althoff and Reichardt (forthcoming) find that the socioeconomic status of Black families today depends strongly on their historical exposure to racially oppressive institutions. Research (Althoff & Reichardt, 2023) shows that Black people whose ancestors were enslaved up until the Civil War have lower education, income, and wealth today as compared with Black people who were free before the Civil War. While the direct effects of enslavement on these families continued through 1940, the ongoing effects are due to the disproportionate exposure to Jim Crow laws among those families who were enslaved until the Civil War. Thus, state-specific factors perpetuated the socioeconomic disparities that slavery had created among Black families. According to Althoff and Reichardt (forthcoming, pp. 3–4), Black families freed in states with more oppressive regimes experienced significantly lower rates of economic progress starting in the Jim Crow era (1877–1964). For example, consistent with Louisiana’s Jim Crow regime being far stricter than Texas’s, we find that families freed in Louisiana attained 1.2 fewer years of education by 1940 than families freed only a few miles away in Texas. The magnitudes of those border discontinuities are

virtually identical to the general state differences in how families fared after slavery, suggesting that Jim Crow single-handedly shaped the geography of Black economic progress. Althoff and Reichardt (2023, p. 5) conclude: “This result implies that systemic discrimination—the higher exposure to ongoing discrimination because of past discrimination (Bohren et al., 2022)—is at the core of the persisting legacy of racially oppressive institutions in the US.”

Contemporary Drivers of Racial Disparities in Intergenerational Poverty

As with the section above, we offer additional discussion and detail about some but not all of the domains covered in the main text, and add an additional section on Black immigrants. The first two subsections here are titled to correspond with their respective section in the main text.

Crime, Victimization, and Criminal Justice

Slavery and colonialism are intertwined systems of racialized economic oppression to benefit those in power. Power is maintained through violence and the threat of violence, often justified under the guise of punishment for supposed wrongdoing or crime (for historical overviews on punishment against Black and Native people, see Ross [2010], Thompson [2019], Ulmer & Bradley [2019], Hinton & Cook [2021]). Early regimes of punishment against Black and Native Americans were manifested in the physical acts of whippings, scalpings, and murders, and also in economic practices of dispossession and indebting. There was punishment for slow work, punishment for speaking one’s native language, punishment for playing musical instruments, punishment for leaving or staying on designated lands (territories, plantations, reservations, hunting grounds), punishment for reading or refusing to read, and punishment for using resources from the land for survival (Ross, 2010). Punishment is linked to intergenerational poverty because it limits or eliminates the ability of the parental generation to invest in children economically, socially, psychologically, and culturally. When state punishment is enacted against children, it stifles their educational, psychological, and economic growth.

The historical relationship between punishment and economic disfranchisement has been illustrated above through the example of convict leasing. Punishment was similarly used against Native Americans as a pretext for economic and land dispossession. California Statute Chapter 133 of 1850—very wrongly named “An Act for the Government and Protection of Indians”—deputized White settlers as police, judge, and jury, with the purpose of expropriating labor. The text of the law makes clear the connection

between criminal punishment and economic exploitation authorized in the 13th amendment:

Unlike the convict leasing system in the South, individual settlers were enriched by the labor of supposedly criminal Native Americans, who were in turn deprived of the ability to sustain themselves and their families. A frequent grounds for conviction was the “hunger offense of stealing cattle” (p. 26), a crime that resulted from the systematic elimination of Native American food sources (Ross, 2010). In 1861, in Northern California, several hundred Native American men were killed for stealing cattle. In similar cases, the wives and daughters of slain Native American men were taken by White settlers for forced sexual and physical labor.

Law is used strategically to define “crime” in ways that facilitate the maintenance of power and economic hierarchies. Punishment against Black and Native peoples can be exacted to extract labor and land, or for managing periods when their labor was not required (Muller & Schrage, 2021; Western et al., 2006). The “crimes” for which punishment has been levied have been in response to Black and Native people’s attempts at bodily, cultural, economic, and spiritual freedom. These historical practices of shaping law and defining crime in anti-Black and anti-Indigenous ways reverberate today in widespread stereotypes about Black and Native lawlessness, disorder, aggression, and criminality (Muhammad, 2019; Ross, 2010), which justify tough-on-crime policies that target Black and Native communities for policing, punishment, and removal from their families and communities, including through supposedly supportive institutions such as schools and the child welfare system (Jacobs, 2014; Roberts, 2022).

This brief history challenges a common starting point when discussing race, crime, and punishment: the notion that Black and Native people commit more crime and therefore experience more punishment. The definition of crime and the accordant punishments are themselves tools of control—from the vagrancy and trespassing laws of the 19th century to the continued 18-to-1 sentencing disparity between crack and cocaine possession. Even using conventional definitions of crime, there appears to be a

weak and small relationship between an individual’s race and many areas of self-reported offending (Sohoni et al., 2021).

Disparities in incarceration rates between White and Black people are longstanding, dating back to the earliest record keeping in the 1800s. Figure C-3-1 (from Muller, 2012) compares Nonwhite and White incarceration rates from 1880 to 1950 (separate data for Blacks are not available). At the beginning of the time period, the Nonwhite incarceration rate is roughly two times the White rate, while by the end of the time period the gap had widened, with the Nonwhite rate roughly five times the White rate. The ratio (or disparity) of the Black:White incarceration rates peaked at about 7:1 in the early 2000s (National Research Council, 2014), and has declined to roughly 5:1 in the most recent period (Nellis, 2021). In other words,

contemporary Black/White disparities in incarceration are roughly equal to what they were in 1950. Figure C-3-2 shows that contemporary racial disparities in incarceration rates are also evident for juveniles, with Native and Black juveniles experience much higher rates than White, Asian, and Latino youth.

Racial disparities in offending and cumulative disadvantages across multiple domains reverberate through the stages of criminal processing (e.g., policing, arrest, prosecution, sentencing, incarceration) to produce racial inequality in the criminal justice system (Alexander, 2010; National Academies, 2022b). Researchers have discussed three policy changes that have been important for how structural racism has contributed to these racial disparities in the criminal justice system. The first policy change, collectively known as the War on Drugs, intensively criminalized drug use and drug sales disproportionately in Black communities (Provine, 2011; Tonry, 1996; Tonry & Melewski, 2008). The punitive effect of the War on Drugs can be seen in the increasing probability of imprisonment given a drug arrest and the growing share of people in prison convicted of drug crimes (Beck & Blumstein, 2018; Blumstein & Beck, 1999; Tonry & Melewski, 2008). Given the large racial disparity in drug arrests and prison

admissions, the growth in drug incarceration also tended to increase the racial disparity in incarceration through the 1980s and 1990s. The second major policy change contributing to racially disparate criminalization was the War on Crime, which encompassed a variety of changes in sentencing policy at the state and federal levels that increased the duration of prison sentences, particularly for violent offenses. Third, policing strategies changed in the final decades of the 20th century to focus more on crime prevention and to allocate resources more intensively to areas and people who were viewed as high risk (National Academies, 2022b).

Involvement in the criminal legal system has long-term economic effects for young people and adults. People with a conviction experience a cumulative loss of roughly $100,000 in earnings, and people who have experienced incarceration experience nearly $500,000 in lost earnings over their lifetime (Craigie et al., 2020). Incarceration and conviction have similar negative effects on wealth (Maroto, 2015; Schneider & Turney, 2015; Sykes & Maroto, 2016). Given the racial disparities in criminal justice processing, the negative effects on employment and earnings exacerbate racial gaps in socioeconomic outcomes (Gordon et al., 2021; Lyons & Pager, 2007; Pettit, 2011; Western & Sirois, 2019). There is little research on the socioeconomic outcomes after incarceration for Native Americans.

Housing and Neighborhood Environments

Racial disparities in housing are driven by the acts of both private citizens and state actors in coordinating efforts to exclude Black people from property ownership and White neighborhoods. For example, the Servicemen’s Readjustment Act of 1944, more commonly known as the GI Bill, offered preferred mortgage financing, tuition for college and vocational training, and enhanced unemployment benefits to returning veterans. While it was federal law, it was administered locally. Black veterans returning to the Jim Crow South in the 1940s, 1950s, and 1960s were denied access to mortgages—as well as to the college tuition benefits (Turner & Bound, 2003)—and thus to the wealth-enhancing possibilities of homeownership and education. Black veterans in the North faced less absolute exclusion, but the federally promoted practice of redlining (discussed in Chapter 3) similarly made the GI Bill widely inaccessible for broadly improving Black people’s housing situation (Agbai, forthcoming; Cohen, 2003; Delmont, 2022). Using conservative assumptions, Meschede et al. (2022) estimate that “Black veterans received at most 70% of the value that white veterans received,” and the gap was greatest for government spending on housing benefits.

decline in values due to the colored element now controlling the district” (p. 200).

While the HOLC may have been relatively fair in some cities in apportioning home loans to Black households (Fishback et al., 2022; Hillier, 2003), the Federal Housing Authority (FHA) was certainly not. In Chicago in 1938, the Chicago Housing Authority created a map of mortgage-lending risk based on the FHA’s evaluations. The entirety of the city’s Black community was colored in red. The Chicago Housing Authority summarized the message of the map bluntly: “All Negro census tracts fall within the area where loans have not been made by the major loaning agencies, and loans will not be made” (quoted in Pattillo, 2007, p. 331).

Capitalizing on the exclusion of Black people from the conventional housing finance market, private investors created a shadow market selling “on contract” (Satter, 2009). White “sellers” retained the deed and imposed stringent requirements for maintenance and repayment. With the slightest infraction, owners repossessed properties and kept all of the payments made by the Black “buyer.” Contract selling extracted an estimated $3 billion (in 2019 dollars) in wealth from Black families in Chicago in the 1960s (Dubois Cook Center, 2019), and has re-emerged in the post-2008 housing collapse period in predominately Black cities like Atlanta (Immergluck, 2018).

The most recent episode of housing discrimination is the subprime and foreclosure crisis (Hwang et al., 2015; Rugh & Massey, 2010). In December of 2011, the U.S. Department of Justice issued a press release with the headline, “Justice Department Reaches $335 Million Settlement to Resolve Allegations of Lending Discrimination by Countrywide Financial Corporation; More than 200,000 African-American and Hispanic Borrowers who Qualified for Loans were Charged Higher Fees or Placed into Subprime Loans.” This was just one of many lawsuits and settlements to come.

Subprime loans are characterized by higher interest rates, payment plans that assume upward value trajectories, deferred or “balloon” payments, interest-only payments, prepayment penalties, and other complicated and disadvantageous arrangements that are often not fully disclosed or explained to the homebuyer. Contrary to the exclusionary practices wrought by redlining, predatory inclusion marks the targeting of Black households for financial instruments that prove detrimental to their socioeconomic well-being.

The rash of subprime lending eventually led to the foreclosure crisis. Black and Latino households were roughly twice as likely to be affected by foreclosure through home loss or serious arrears (Bocian et al., 2011), and Black and Latino neighborhoods were also disproportionately affected (Hall et al., 2015). Roughly 8% of Black households (240,000 households) lost their homes between 2005 and 2008 (Bocian et al., 2010). Disparities in

foreclosures and steep declines in housing values had a cascading effect on household wealth, since homes represent the major portion of most people’s wealth profile. The median net worth of Black Americans in the United States declined by 53% from 2005 to 2009 (and by an even larger 66% for Latino households), but declined by only 16% for White households. This represented a doubling of the racial wealth gap (Pew Research Center, 2011). The foreclosure crisis and recession were particularly pronounced among Black families with children, widening an already large Black-White wealth gap among such households. Percheski and Gibson-Davis (2020, p. 10) report that by 2016, “black child households had 1 cent of median wealth for every dollar of wealth held by non-Hispanic white child households.” The targeting of subprime mortgages to Black buyers negatively impacted Black wealth and Black neighborhoods, continuing a cycle of lowering household resources that might promote intergenerational mobility.

Comparisons with Black Immigrants

Although not included in the main text, a discussion of the socioeconomic status and mobility outcomes of Black immigrants offers additional evidence on racism and discrimination as drivers of intergenerational poverty. Black immigrants have been heralded as a “model minority” (Ukpokodu, 2018), and their success is often advanced as a counter-explanation for the poor social position of Black Americans. The logic goes like this: because both Black Americans and Black immigrants share similar phenotypic characteristics, both groups must suffer from similar levels of discrimination. Therefore, if Black immigrants can succeed despite the barriers of racism and discrimination, then cultural differences, rather than racism, must account for observed differences (Patterson, 2006, 2015; Sowell, 1979, 1981). Such arguments, however, omit important factors discussed here that help explain disparities between immigrant and native-born Black people (Hamilton, 2019; Model, 2008).

The Hart-Cellar Immigration Act of 1965 paved the way for large waves of Black immigrants to migrate to the United States (Portes & Rumbaut, 2014). In 1960, Black immigrants accounted for less than 1% of Black people residing in the United States. By 2010, almost 10% of Black people in the United States were foreign-born, a 10-fold increase in 50 years (Hamilton, 2019). When these new waves of Black immigrants arrived, scholars noted that some Black immigrants had better outcomes than Black Americans. For example, data from the 1980 U.S. Census showed that prior to adjusting for relevant human capital characteristics, Black immigrants from the English-speaking Caribbean were more likely to be in the labor force, had greater employment rates, and earned more than Black

Americans, findings that led some to conclude that culture, rather than racism, explained the poor outcomes of Black Americans (Sowell, 1979, 1981).

These early accounts of labor market disparities between Black immigrants and Black Americans severely overstated the advantages of Black immigrants. After controlling for a standard set of human characteristics, most subgroups of Black immigrants have similar or lower employment rates and earnings than Black Americans (Hamilton, 2019; Model, 2008). Black immigrants’ labor force participation rates are higher than those of Black and White Americans, which suggests that disparities in labor force participation likely result from different reservation wages between immigrants and natives, in general, rather than differences in cultural practices (Hamilton, 2019).

Early studies of labor market disparities between Black immigrants and Black Americans also ignored selection bias issues (Patterson, 2006; Sowell, 1981). Like other contemporary immigrants, Black immigrants are a self-selected group of movers (Feliciano, 2005; Feliciano & Lanuza, 2017; Hamilton, 2019; Model, 2008). Most Black immigrants are selected on a range of factors that are positively correlated with better labor market outcomes, including age, health, education, and social class position (Hamilton, 2019). For example, Black Jamaican U.S. immigrants have 13.13 mean years of education, compared with 9.64 years among individuals residing in Jamaica. Similarly, Nigerian U.S. immigrants have 15 mean years of education compared with 6 years among adults residing in Nigeria (Hamilton, 2019). Given the dramatic difference in the home country education distribution between the United States and most Black immigrant sending countries, most Black immigrants also occupy a more favorable social class position in their home country than in the United States (Feliciano, 2020; Hamilton, 2019). Studies have consistently found that the outcomes of Black immigrants are more similar to those of Black Americans who have also made a move across states since birth than to those of Black American nonmovers, which suggests that the favorable outcomes of Black immigrants result from unobserved factors associated with migration (both domestic and international) than from cultural differences between Black immigrants and Black Americans (Butcher, 1994; Hamilton, 2019; Model, 2008).

Any advantages experienced by Black immigrants do not apply to all Black immigrants, even those from the same region or country. For example, among Black women from the English-speaking Caribbean, early arrivals tended to have better labor market outcomes than more recent arrivals. Model (2008) argues that when the Hart-Cellar Act passed in 1965, there were relatively few immigrants from the English-speaking Caribbean residing in the United States to sponsor the visas of family members. As a

result, immigrants with skills needed in the United States were among the early drivers of migration after 1965. This fact generated a flow of women from the English-speaking Caribbean who were positive-selected on skills and education (many were employed in the health care sector) as well as risk taking and motivation, given that these women arrived in the country with a small or no established co-ethnic community. Over time, however, family reunification drove an increase in migration, which resulted in a less highly selected group of migrants.

Indeed, patterns of intergenerational mobility among immigrants highlight the role of contemporary discrimination in affecting the outcomes of Black immigrants. Using a century of U.S. censuses and contemporary tax records, Abramitzky and Boustan (2022) find that sons of immigrants raised in the bottom 25th percentile of the income distribution were in a more favorable position in the income distribution than the sons of U.S.born fathers. For the post-1965 immigration era, the only countries of origin whose immigrants’ sons were in a lower position than sons of U.S.born fathers were Haiti, Trinidad and Tobago, and Jamaica. Hence, Black immigrants differ from non-Black immigrants and seem to face similar barriers to upward mobility as Black natives.

and Evaluation contracted to conduct a regular systemic review of home visiting research, the Home Visiting Evidence of Effectiveness (HomVee) review, and publish findings on its website (Avellar & Paulsell, 2011).

The 2021 HomVee review (Office of Planning, Research, and Evaluation, 2021) included 22 programs that reported statistically positive outcomes in randomized control trials (RCTs) or rigorous quasi-experimental studies. As of September 2021, the review included evaluations of 53 home visiting models, focusing on the 22 models that met the stringent Office of Planning, Research, and Evaluation criteria for evidence-based models. The impacts of those 22 programs on child outcomes were found to be ineffective much more often than effective: Approximately 130 impacts were “positive,” in 600 cases there was “no difference,” and in 7 cases the impact was “negative.” A broader look at the results shows limited evidence that programs changed parenting or child outcomes. Overall, they suggest that home visiting programs can, but typically do not, increase cognitive stimulation in the home and decrease the use of punitive, harsh punishment. Even when such programs are successful in changing parenting behavior, the size of their impacts is modest—about 0.10 standard deviations (Michalopoulos et al., 2019). In most cases, their positive impacts were limited to shorter-run improvements in birth outcomes, reductions in hospitalization, declines in behavior problems, or increased access to services like Temporary Assistance for Needy Families or Special Supplemental Nutrition Program for Women, Infants, and Children.

Table C-4-1 provides a broader overview of the HomVee results from the most promising home visiting programs that were evaluated between 2019 and 2020. Because the evaluations include so many parenting and child outcomes, we set a statistical significance threshold of p < 0.10 and simply count the number of impact coefficients that fall below that threshold. Given the nature of significance testing, we would expect to see 10% of the coefficients below p < 0.10 even if there were no true impacts.

Promising evidence of longer-term impacts on problem behaviors was reported for both the Nurse Family Partnership (NFP) and Attachment and Biobehavioral Catch-up–Infant (ABC). The NFP provides one-on-one home visits by nurses who focus on improving maternal and infant health and promoting family economic self-sufficiency during pregnancy and the child’s first 2 years. Follow-up studies of the NFP study participants in the early implementations of the programs showed reduced behavior problems at 15 years, reduced substance use at 12 years, and increased vocabulary and academic skills at 1 to 2 years post-intervention (Kitzman et al., 2010; Olds et al., 2002).

But these findings were not replicated in other NFP evaluation studies. As seen in Table C-4-1, a review of child impacts from a broader implementation of NFP found significant results less than 10% of the time—which

could easily have occurred by chance. ABC–Infant provides trained parent coaches to mothers of infants, many of whom are either foster parents or parents deemed to be abusive. ABC focuses on promoting mutually responsive interactions by having parents and coaches watch videos of interactions together (Office of Planning, Research, and Evaluation, 2021). Follow-up studies of early implementations of ABC–Infant reported higher rates of attachment security at 9 years, higher levels of emotional regulation at 8 years, and higher vocabulary skills at 3 and 5 years (Raby et al., 2018; Zajac et al., 2019). Again, these promising findings were not replicated in other studies. Across all ABC-Infant studies, Table C-4-1 shows significant child impacts less than half of the time (Office of Planning, Research, and Evaluation, 2021).

1.9 million preschoolers, most of whom are from low-income families (Friedman-Krauss & Barnett, 2020).

Perry and Abecedarian

The ability of intensive model programs to improve the life chances of disadvantaged children is illustrated by the well-known Perry Preschool and Abecedarian interventions. Perry was implemented in the 1960s in Ypsilanti, Michigan, while Abecedarian ran during the 1970s in Chapel Hill, North Carolina. Perry provided center care and home visiting to 65 3- and 4-year-olds from low-income Black families. The first Perry cohort received one year of program services, and the remaining four cohorts received 2 years. Costs across the five cohorts averaged $23,000 per child (Heckman et al., 2010). Heckman et al. (2010) estimate that the dollar value of benefits generated by the program amounted to between six and nine times the cost of the program, with benefits driven in roughly equal measure by increases in earnings and reductions in crime. When considering that high benefit/cost ratio, however, it is important to recognize that the home environments of the comparison group of children were of much lower quality relative to the home environments of today’s children. Only 21% of Perry mothers and 11% of Perry fathers had graduated from high school, and family sizes averaged 6.7—much larger than today (Schweinhart, 1993).¹

Abecedarian provided considerably more services than Perry and over a much longer period. These included center care and pediatric care for about 100 low-income, predominantly Black children from 3 months of age to kindergarten entry. The per-child cost of this 5-year program has been estimated at $105,000 (García et al., 2021. Even given the program’s expense, its long-run benefits have been estimated to total more than six times its costs: More than two-thirds of the benefits were driven by crime reductions, and the remainder reflect differences in adult health and the

___________________

¹ In the 1960s and 1970s, children from low-income families, especially Black children, faced conditions best described as deplorable. In the mid-1960s, the Food Stamp program and tax credit programs such as the Earned Income Tax Credit had not yet been introduced. Racial discrimination in parts of the country denied Blacks access to quality schools and hospital care, including childbirth in hospitals with a physician present. Parental schooling levels were much lower than they are today, and family sizes were much larger (Duncan & Magnuson, 2013). Today families, including low-income families, are about twice as likely to use child care, especially center-based child care (Duncan & Magnuson, 2013). These factors produced conditions that made it much easier for a Head Start or model program like Perry to demonstrate effectiveness for enrolled children when compared with children experiencing business-as-usual conditions. This explanation probably accounts, at least in part, for the fact that end-of-treatment impacts are substantially smaller in more recent evaluations (Duncan & Magnuson, 2013).

labor income of participants and their parents (the 5 years of full-time child care enabled parents to establish and maintain more continuous and higher-paying careers; García et al., 2020). As in the case of Perry, the low-quality nature of the home environments of children in the Abecedarian evaluation is reflected in the low average levels of mothers’ completed schooling (10.2 years), IQ (85.5), and age at their child’s enrollment (19.9 years; Ramey et al., 2000).

Both Perry and Abecedarian focused on encouraging adult-child interactions, hands-on learning activities, and frequent conversations between children and teachers (Schweinhart & Weikart, 1980; Ramey et al., 2012). The teachers, their supervisors, and researchers collaborated in developing the interventions and classroom activities that were later incorporated into many early childhood curricula. However, the evaluations were not designed to enable researchers to disentangle the separate contributions of these components to improving children’s development, and these programs differ in many ways from today’s early care and education programs (as discussed in more detail below).

Results from these two early RCT interventions provide convincing evidence that ECE programs can improve educational attainment, income, and health in adulthood (Heckman, 2011). The key policy and program issue for the committee is whether much larger-scale and less expensive versions of ECE programs—run or supervised by governments rather than researchers, provided to children from families with higher incomes and schooling levels, and living in communities with a much richer set of center-based child care options in programs that do provide the same types of educational experiences as Perry and Abecedarian—can generate impacts comparable to those of Perry and Abecedarian.

Head Start

Head Start began in 1965 as part of the War on Poverty and provided part-time center-based ECE to low-income children (Office of Headstart, 2023). It was offered to thousands of children in its early years, with a focus on promoting health and social skills. Strong quasi-experimental studies have demonstrated long-term benefits for children who attended Head Start during the period from 1965 to 1980 in terms of higher rates of college enrollment (Bailey et al., 2021; see also Johnson, 2011; Ludwig & Miller, 2007). Evidence from cohorts that entered Head Start in the 1980s and 1990s tends to show higher levels of education and reductions in special education and grade retention, even if impacts on academic and social skills fade out during elementary school (Deming, 2009; Garces et al., 2002).

lower scores (effect sizes of 0.24 to 0.28 standard deviations in math, executive functioning, and social skills) than non-pre-K children. However, these differences were not statistically significant after adjusting for the large number of comparisons in the study (Peisner-Feinberg et al., 2020).

Less rigorous studies provide inconsistent evidence regarding differences between pre-K attenders and nonattenders.² The most promising results are from a follow-up study of children who were eligible in 2005–2006 for the Tulsa Universal Pre-K. Gormley et al. (2023) reports that pre-K attendees were more likely than nonattending children with similar demographic characteristics subsequently to enroll in college (increase of 12 percentage points). In other studies that attempted to match attenders and nonattenders using propensity score methods on school-age demographic characteristics, pre-K attenders in Boston, North Carolina, New Jersey, Maryland (Minervino, 2014) and Georgia (Early et al., 2019) showed slightly higher reading and math scores on state-mandated reading and math tests in third and fifth grade.

ECE programs in other countries provide some evidence of positive impacts on adult outcomes of their at-scale ECE programs. One of the ECE programs showing long-term impacts is the Norwegian program, which requires a college-educated teacher, good adult-child ratios (typically 16 children and 3 adults), and a play-oriented experiential orientation to instruction (Havnes & Mogstad, 2011). Uneven expansion of the Norwegian program during the 1970s allowed for a difference-in-difference analysis that demonstrated that attending preschool between the ages of 3 and 6 led to a substantial increase in completed schooling, labor market attachment, and earnings (Havnes & Mogstad, 2011).

Possible Explanations for Discrepant Findings on Long-Term ECE Impacts

Whereas the two large publicly funded ECE programs, Head Start and public pre-K, have shown positive impacts on school readiness, the evidence for medium- and longer-term impacts is mixed. The earliest ECE programs, like Perry and Abecedarian, were clearly effective at improving earnings, educational attainment, and health, as well as at decreasing crime and incarceration—all of which are factors that tend to lower the rate of intergenerational poverty (Campbell et al., 2012; Heckman et al., 2010). Earlier Head Start and pre-K programs also showed promising long-term impacts on educational attainment and crime (Bailey et al., 2021; Deming,

___________________

² These studies use propensity score matching without having access to pre-treatment assessment of skills. The committee judged that this method was not as rigorous as random assignment or regression discontinuity methods.

2009; Gary-Lobe, 2021). In contrast, the most rigorous evaluations of today’s Head Start and pre-K programs suggest that they have null to negative medium-term impacts. It is possible that longer-run follow-up will show the kinds of positive impacts seen in Perry and Abecedarian. However, until the evaluation evidence shows more consistently positive impacts for these programs, the committee cannot, based on the evidence, confidently recommend expansions as an approach that is likely to reduce intergenerational poverty.

Transforming Model Programs into At-Scale Public Programs

Programs like Perry and Abecedarian were small, conducted by developmental researchers and led by trained ECE teachers. Classrooms were carefully monitored to ensure that the program was being implemented successfully and teachers were actively engaged in developing the curriculum. It is naïve to believe that findings from those programs will generalize to large public programs that serve millions of children each year (Tseng, 2017). Even replications like the Infant Health and Development Program, which used the Abecedarian curriculum and included almost 1,000 low-birthweight children, predominantly from low-income families, were led by researchers who continued to monitor the implementation carefully (Ramey et al., 2012). Today’s programs rely on performance standards that are less focused, yet more comprehensive. They include substantially less monitoring and supportive collaborative coaching than the early RCT studies (Friedman-Strauss & Barnett, 2020). Again, these factors are probably one reason why today’s programs have smaller impacts on the outcomes collected in early RCT studies, although they are unlikely to explain fade-out.

Targeting Recipients of ECE Services

Evidence shows that Head Start impacts were larger and more likely to be sustained when the children who attended Head Start would otherwise have stayed home with their parents or been cared for in a home-based ECE setting (Feller et al., 2016; Kline & Walters, 2016), and this suggests that increasing funding for Head Start might have longer-term impacts and reduce intergenerational poverty if the program could specifically target those children. However, a Head Start expansion could not be designed to exclude children who would otherwise be in other center-based care.

Growing evidence indicates that children who do not speak English at home may gain more and maintain those gains longer when they attend ECE programs (Phillips et al., 2017). In the RCT HSIS, children whose home language was not English showed larger gains in vocabulary skills if they attended Head Start, and those gains remained statistically significant

through first grade (Feller et al., 2016). The RCT evaluation of North Carolina pre-K reports statistically significant impacts on kindergarten language skills for children who did not speak English at home, even when those impacts faded in the sample as a whole (Peisner-Feinberg et al., 2020). These impacts were small in both studies (< 0.20 standard deviations), however, raising questions about whether they were sufficiently large to reduce intergenerational poverty even if maintained past first grade.

To What Extent Subsequent Experiences Support Initial Gains

The extent to which communities and schools support and build on the skills children have acquired in ECE programs is clearly important for maintaining the programs’ impacts and may play a role in fade-out. Children from low-income families who attend public ECE programs, especially if they are members of racially minoritized groups, tend to transition to lower-quality schools, encounter lower teacher expectations, and experience harsh punitive discipline in schools and by the local police (for details, see the sections on K-12 schools and juvenile justice). A lack of “sustaining environments” after low-income children leave their ECE programs makes it difficult for children to maintain the gains they have made in those programs.

The lack of sustaining environments to maintain ECE short-term impacts has been widely cited as an explanation for the fade-out of those impacts over time (Abenavoli, 2019). Despite some studies reporting that pre-K attenders continued to show higher levels of skills than nonattenders when they transition from pre-K programs into more effective schools or have more effective teachers (Abenavoli, 2019), this conclusion was not supported in a comprehensive meta-analysis of all studies testing the sustaining environment hypothesis published through 2018 (B. Bailey et al., 2020). Others argue that redundancy in instruction between pre-K and kindergarten, especially in literacy and math, accounts for the marked convergence in those skills, but pre-K fade-out did not diminish when pre-K and kindergarten instruction included less redundancy, despite wide-scale evidence of considerable overlap in early reading and math instruction (Burchinal et al., 2022). Thus, findings do not suggest that current “sustaining” elementary schools support learning more effectively for pre-K attenders than nonattenders. Nevertheless, it is logical to assume that there would be less fade-out of pre-K impacts if kindergarten teachers were encouraged to differentiate instruction based on entry skills (Cohen-Vogel et al., 2021).

___________________

³ For more information, see www.buildinitiative.org

K-12 education of around $60 billion. It seems implausible that the federal role could expand that much.

We explore one potential federal policy change short of closing the entire adequacy gap. We base it on Lafortune et al.’s, (2018) analysis of post-1990 state finance reforms. They estimate that a typical reform raised spending in the lowest-income fifth of districts in the state by $1,377 per pupil (in 2013 dollars), or about 12% of the average of $11,595. Total public-school enrollment is about 50 million, so approximately 10 million students attend schools in the bottom fifth of districts. Thus, increasing spending in these districts by $1,000 per pupil would increase total spending by $10 billion per year.

As noted above, not all federal funding would “stick.” If we assume that changes in state and local effort reduce the effectiveness or intended concentration of federal spending by one-third, federal expenditures would need to rise by $15 billion to achieve the above increase in expenditures in low-income districts.

What would be the impact of this? Jackson et al.’s (2016) results indicate that a 12% increase in spending would reduce the adult poverty rates of low-income children growing up in these districts by 0.12 × 0.61 = 7 percentage points and increase their adult family incomes by 21%. Using more recent reforms, Rothstein and Schanzenbach (2022) find that exposure to a typical reform for 12 years raises average earnings of students in a state by 4% but note that this combines effects on low- and non-low-income students and on high- and low-income districts.

How many children would be affected? Approximately 17% of public-school students (8.5 million) are in poverty (Digest of Education Statistics, Table 102.70). Lafortune et al., (2018) finds that about one-third of free and reduced-price lunch students (a proxy for student poverty) attend school in the lowest-income fifth of districts. Thus, the above reform would reach approximately 2.8 million children from families below the poverty line. (Though note that above we assumed that states would divert one-third of spending to other districts than the lowest income. These districts have children in poverty as well, so the number of children in poverty who would be affected would be larger.) If this reform reduced their adult poverty rate by 7 percentage points, that would reduce intergenerational poverty by 16,000 students out of each birth cohort.

Promising Approaches Within K-12

As discussed in the report, there is limited evidence about long-term effects of specific programs or practices within the K-12 system, but there are a number of promising approaches that have been shown to have strong shorter-term impacts. If these early impacts are found to persist,

these approaches would be candidates for recommended interventions; in the meantime, we see them as promising and worth consideration, but not meeting our evidentiary standards for committee recommendations. We discuss several here, acknowledging that this list is not exhaustive.

High-dosage tutoring for struggling students

Evaluations have demonstrated that programs providing frequent tutoring sessions, individually or in small groups, to students with skills below grade level improves academic skills (Fryer, 2016). Although improved academic skills do not translate directly into lower poverty in adulthood, the evidence on shorter-run improvements from carefully crafted tutoring programs is promising enough to include in our list of profitable ways districts might promote school success among disadvantaged students.

The tutoring programs that appear to be successful involve sessions provided by trained volunteers or educators to struggling students, individually or in groups of six or fewer, for at least 50 hours over the school year (Fryer, 2016). They range from programs like Reading Recovery for early elementary students (D’Agostino et al., 2017) to tutoring sessions for high school students (Guryan et al., 2023). Rigorous evaluations summarized in a meta-analysis indicate that these programs have moderate short-term impacts and, in the studies with longer-term follow-ups, those impacts were maintained for at least a year and translated into higher rates of high school graduation. Tutoring programs that did not meet these two criteria showed neither short- nor long-term impacts (Fryer, 2016).

Based on this evidence, Kraft and Falken (2021) propose creating a national tutoring network that could be adopted by school districts and encouraged by federal funding. In the proposal, tutors could include high school students who tutor in elementary schools as an elective class, college students who tutor in middle schools via the federal work-study program, and full-time 2- and 4-year college graduates who tutor in high schools via AmeriCorps. Their estimates suggests that targeted approaches to scaling schoolwide tutoring nationally, such as focusing on K–8 Title I schools, would cost between $5 and $15 billion annually. They do not provide estimates of the impacts on either the tutors or the students being tutored.

Several large evaluations estimated both long-term impacts and program costs. A meta-analysis examined Reading Recovery, a widely implemented early-grade intervention in which struggling elementary school students receive 20 weeks of individually designed diagnostic teaching by trained professionals (D’Agostino et al., 2017). Synthesizing results from RCTs or high-quality quasi-experimental studies yielded an effect on reading achievement of 0.59 standard deviations at a cost of $2,500 to $9,000 per student. Guryan et al. (2023) report on a multi-study evaluation of

the Saga Education tutoring program, which provides similar high-dosage tutoring to 9th and 10th graders at a relatively low cost ($3,500 to $4,300 per participant per year) that yielded higher math test scores and increased grades in math and non-math courses, with estimated impacts of 0.37 standard deviation that persisted several years later.

In sum, high-dosage tutoring for struggling K-12 students that focuses on matching instruction to skills level appears effective at improving skills in the short-term and improving important young adult outcomes such as high school graduation. These programs may be cost-effective when educated young adults serve as the tutors and follow carefully crafted instructional plans.

Improving teacher quality

It has long been known that schools serving low-income students have a more difficult time attracting and retaining high-quality teachers and that students of color do better in school when matched with teachers of color (Gershenson et al., forthcoming, 2016; Lindsay & Hart, 2017). So, the problem is both one of allocating teachers across schools and school districts as well as increasing the supply of high-quality teachers, and in particular high-quality teachers of color.

One promising model for increasing teacher quality is university programs that encourage and facilitate science, technology, engineering, and mathematics (STEM) undergraduate majors to be certified as public-school teachers. The UTeach program at the University of Texas-Austin is an exemplar; as of 2018 programs similar to it are also available at 44 universities in 21 states. As explained in Backes et al. (2018), UTeach recruits math and science majors to pursue careers in teaching and offers free field-based courses that enable interested students to try out teaching before committing their early careers to it. It was funded by grants from, among other places, the nonprofit National Math and Science Initiative, but could presumably be supported in some way with federal money.

The Backes et al. (2018) non-experimental evaluation found that Texas students taught by UTeach graduates perform significantly better on end-of-grade tests in math in middle school and end-of-course tests in math and science in high school by 8% to 15% of a standard deviation on the test, depending on grade and subject. As to teacher diversity, UTeach increased the fraction of Hispanic (but not Black) science and math teachers in the schools they studied.

Other approaches to diversifying the teacher labor forces include urban teacher residency models, such as Alder Graduate School of Education and the Boston Teacher Residency programs. An evaluation of the latter found that program graduates were more diverse than Boston Public School teachers, were more likely to remain teachers and, after several years,

outperformed veteran teachers in the district (Papay et al., 2012). Teacher training and certification programs drawing students from community colleges and minority-serving institutions—such as the Department of Education’s Center of Educational Excellence for Black Teachers Program at Historically Black Colleges and Universities—are also likely to draw and train minority group teachers than traditional programs, but the evaluation literature on this is thin.

Reducing punitive school discipline

Experimental and correlational evidence reviewed in the main text shows that exposure to harsh discipline in schools leads to worse adult educational and criminal legal outcomes. Given the disproportionate experience of school discipline by low-income and Black, Latino, and Native American children—which is not accounted for by behavioral differences (Skiba et al., 2011)—this experience likely also contributes to these children’s lower rates of intergenerational mobility. Teacher-student race matching has been shown in both correlational and experimental studies to reduce punitive discipline for students of color (Lindsay & Hart, 2017; Shirrell et al., 2021). Another promising approach—and one of the few that includes outcomes for Native American students—is restorative justice practices (Anyon et al., 2016; Gregory et al., 2016).

A meta-analysis of RCT interventions to reduce harsh discipline found that students’ academic skills, counseling, mentoring programs, and teacher training all showed promising results for reducing in-school or out-of-school suspensions (Valdebenito et al., 2019). For example, Okonofua et al. (2016) tested a brief randomly assigned on-line intervention with math teachers in five middle schools in three school districts that “encouraged teachers to understand and value students’ experiences and negative feelings that can cause misbehavior and to sustain positive relationships when students misbehave.” This “empathetic mindset” reduced suspension of students of the treated teachers by half during the academic year as compared with the control group. The effect was consistent across racial, gender, and prior-year suspension groups. While this study had a 1-year observation period, Valdebenito et al. (2019) find that the effects of most interventions faded after 6 months, suggesting the importance of ongoing and repeated training and awareness.

A practice known as Positive Behavioral Interventions and Supports (PBIS) has been shown in RCTs to reduce exclusionary discipline (Bradshaw et al., 2010). PBIS is a schoolwide practice in which “schools establish a set of positively stated, schoolwide expectations for student behavior, which are taught to all students and staff” (Bradshaw et al., 2012, p. e1137; also see www.pbis.org). Descriptive studies in Canada and Oregon showed that

PBIS schools showed no Native American-White disparities in punitive discipline (Greflund et al., 2014; Vincent et al., 2015), and an RCT of PBIS with a racial equity focus showed greater reductions in punitive discipline for Black students (McIntosh et al., 2021). PBIS with added school mental health supports shows additional effectiveness. Weist et al. (2022) report RCT results that this “Interconnected Systems Framework” compared with either PBIS or just mental health services showed “reduced office discipline referrals (ODRs) and in-school suspensions, as well as reduced ODRs and out-of-school suspensions for African American students.”

There are no studies of the direct effects of discipline practices and interventions on children’s later adult outcomes. However, the evidence reviewed here and in the main section establishes a clear indirect relationship between school discipline and intergenerational mobility. There is causal evidence on policies and practices that effectively reduce punitive discipline and causal evidence that children who experience more punitive school discipline are less likely to graduate from high school and attend college and more likely to have criminal legal contact.

Smaller class sizes in the early grades

The well-known STAR evaluation used randomization to study the effect of small classes in K–3 on test scores of low-income students in Tennessee. Assignment to a smaller class had positive effects on test scores through these grades (Krueger 1999, 2003). Although there has been some criticism of the experiment on the grounds that it did not incorporate modern practices like careful collection of baseline data, subsequent reanalysis has shown both that the treatment arms were balanced (supporting the interpretation of the treatment effects as causal) and that the early-grade effects persisted to later grades and even to college enrollment (Chetty et al., 2011; Krueger & Whitmore 2001). There is also suggestive evidence of effects on adult earnings, though the sample was not large enough to measure these precisely (Chetty et al., 2011). In the case of class size reduction, there is little evidence about whether this is more or less effective than alternative ways of spending the same resources.

When California attempted to quickly reduce class sizes in the 1990s, districts had to hire many new, inexperienced teachers to fill the new classrooms, which may have offset the benefits of the small classes (Jepsen & Rivkin, 2009). This appears to be an implementation issue; in the longer run, there is no evidence that a gradual increase in teacher hiring requires bringing in lower-quality teachers.

Expand high-quality (“no excuses”) charter schools

Several recent evaluations of a group of charter schools in Boston have used admissions lotteries to identify the causal effect of enrollment in these

schools. The schools have in common a “no excuses” model, characterized by high expectations, rigid curricula, and strict discipline, and serve a high-poverty population. The lottery studies find very large positive effects on student test scores (Abdulkadiroğlu et al., 2011). Studies of other schools using similar models elsewhere also indicate large effects (e.g., Dobbie & Fryer, 2011). This stands in stark contrast to estimates of the effect of the average charter school, which is generally near zero (CREDO, 2013). Longer-term follow-ups indicate positive effects of the Boston no-excuses schools on students’ eventual 4-year college enrollment and persistence, an outcome that these schools specifically target (Angrist et al., 2016). There is not yet evidence of effects on college completion or adult poverty.

This evidence is strongly suggestive that it would be beneficial to expand the no-excuses charter school sector, though the absence of longer-term evidence prevents us from adopting this as a recommendation. Moreover, the rigorous studies to date examine high-poverty urban populations in a few large cities with struggling public school systems, and it remains unclear how specific the no-excuses effects are to these settings (Angrist et al., 2013).

Increase racial, ethnic, and socioeconomic diversity within schools

Another driver of disparities in educational quality is school segregation. In particular, majority non-White schools tend to have less funding, fewer resources, and less skilled teachers (Bischof & Owens, 2019). In 1954, the Supreme Court passed its landmark Brown v. Board of Education, and other subsequent court orders further reinforced it. This resulted in court-ordered desegregation in school districts around the country, which led to slow but substantial racial integration throughout the United States (Orfield et al., 2016; Reardon & Owens, 2014; Reardon et al., 2012). In studies using national data that exploited quasi-random variation in the timing of the initial court orders, the resulting desegregation in the 1960s and 1970s was found to improve educational and occupational attainment among Black adults (Johnson, 2011). Specifically, each additional year of exposure to court-ordered desegregation led to a 1.8 percentage-point increase in the likelihood of high school graduation, and the average effects of a 5-year exposure to court-ordered school desegregation led to about a 15% increase in wages. This study also found effects on potential mechanisms, including increased per-pupil spending and reduced class sizes, and it showed improvements in other downstream outcomes like overall self-reported health (Johnson & Nazaryan, 2019).

In 1991, the Supreme Court ruled in Board of Education v. Dowell that earlier court-ordered desegregation plans were not intended to be permanent. Roughly 600 of the nearly 1,000 school districts that were under court-ordered desegregation were subsequently released from oversight.

These more recent court decisions resulted in “resegregation” of the released school districts (Reardon & Yun, 2002; Reardon et al., 2012), and highly segregated schools with less than 10% White enrollment have more than tripled in recent decades to nearly 20% (Orfield et al., 2016). While no studies to our knowledge have examined the effects of these recent changes on adult earnings or poverty, a handful of studies have leveraged the longitudinal and geographic variation in the timing of these local court decisions to demonstrate worsened health among affected children while they are still in school (Wang et al., 2022) as well as negative effects on health as they age into adulthood (Kim et al., 2022; Shen, 2018). Another study examined the end of race-based busing in North Carolina during this period, exploiting changes in maps for school boundaries, and found that attending a more segregated school district resulted in decreased high school exam scores among both White and racial/ethnic minority students, deceased high school graduation and college attendance among White students, and increased crime among minority boys (Billings et al., 2013).

Postsecondary Education

Community college completion provides low-socioeconomic status (SES) students with meaningful economic returns relative to noncompleting students (Mountjoy, 2022), and 4-year degree completion provides low-SES students substantially larger returns (Card, 1999; Zimmerman, 2014). Even for low-SES students, higher education provides relatively greater returns when students enroll at higher-value universities (Black et al., 2023; Bleemer, 2021a, 2022) and in higher-paying fields of study (Bleemer & Mehta, 2021). Federal higher education policy would thus maximally promote economic mobility by increasing aggregate college-going, improving educational quality at institutions and in programs where low-SES students enroll, and shifting low-SES students toward higher-value institutions and programs.

Effective Financial Aid for Low-Income Students

The evidence on the effectiveness of the Pell grant program has been quite mixed (Eng & Matsudaira, 2021), though there is some support for increasing its maximum value while limiting offsets by states and localities by reducing their financial support (Denning et al., 2019). Stronger evidence on the effectiveness of scholarship assistance to low-income students include the HAIL program by the University of Michigan (Dynarski et al., 2021, 2022a) and the Buffett Scholarship in Nebraska (Angrist et al., 2022).

proportion of students from the lowest income quintiles and exhibit twice the mobility rates—defined as moving students from the lowest to the highest income quintiles by age 30—compared with predominately White institutions, despite having fewer resources (Espinosa et al., 2018). MSIs enroll and graduate a disproportionate number of Black, Hispanic, and Native American students. Chetty et al. (2017) show that five of the top 10 colleges and universities that promote upward intergenerational mobility are Hispanic-serving institutions. Black students who attend historically Black colleges and universities with high proportions of Pell-eligible students are more likely to graduate than Black students at similar but predominately White institutions (Education Trust, 2019). And Native Americans who attended tribal colleges and universities had lower debt loads compared with peers at non-tribal colleges and universities (Gallup, 2019). A National Academies of Sciences report (2019, p. 125) suggested seven steps to improve student success at MSIs, with a particular focus on improving STEM education and increasing STEM majors.

Beyond the upward mobility of their own students, MSIs can serve as important pipelines into professions that can increase the upward mobility of the next generation of low-income youth. Rigorous causal studies show that Black students—especially low-income Black boys—have better educational outcomes when they are taught by Black teachers in elementary school (Gershenson et al., forthcoming, 2016; Lindsay & Hart, 2017), just as racial concordance between patients and doctors has been found to improve health outcomes for adults and children (Alsan et al., 2019; Greenwood et al., 2020; Saha et al., 1999; Shen et al., 2018; Takeshita et al., 2020; Thornton et al., 2011; Traylor et al., 2010).

Adjustments to Federal Financial Aid Formulas and Integrated Postsecondary Education Data (IPEDS) Data Collection

Free Application for Student Aid completion is a substantial barrier to low-SES students’ college application and enrollment (Bettinger et al., 2012). Low-SES students become more likely to apply to and enroll in college if universities reveal their expected costs of attendance, net of financial aid, prior to application, at least in circumstances where those costs are low, as is often true for low-SES students (Dynarski et al., 2021, 2022b).

College major attainment is a first-order determinant of the return to higher education, with the relative economic return to certain majors (like economics and computer science) rivaling the baseline return to a college education (Altonji et al., 2016). Many universities restrict access to lucrative college majors on the basis of prior academic preparation and performance, and these restrictions disproportionately exclude low-SES students from lucrative fields of study (Bleemer & Mehta, 2021). Providing information

on major restriction policies through Integrated Postsecondary Education Data System (IPEDS) could better inform students about their access to lucrative fields of study and incentivize universities to widen access for low-SES students.

Target Federal Higher Education Funding Toward High-Value-Added College Programs

Information on which universities and university programs provide high-value education to low-SES students is improving over time (e.g. Chetty et al., 2020). It is important that federal support for higher education institutions be increasingly targeted toward high-value programs as such information becomes available. For example, community college nursing programs tend to provide outsized economic returns to relatively low-SES students (Grosz, 2020, 2021).

Career Education

Career and Technical Education Pathways in High School

There is rigorous evidence that high-quality career and technical education (CTE) improves higher education and labor market outcomes for disadvantaged students from a number of studies, all of which are based either on RCTs or lotteries among students who had applied to high school programs that were oversubscribed. For example:

1. Career academies—Career Academies are programs within comprehensive high schools that orient CTE and work experience toward specific sectors of the economy—like health care, information technology (IT), or finance. Kemple and Willner (2008) shows that, 8 years after random assignment (and 5 years after the completion of high school), Career Academies raised the earnings of males by 18% and of students overall by 11%. They also improved marriage rates and stable household formation. But they had no lasting impact on educational attainment, either through dropout rates or higher education. But Hemelt and Lenard (2019) show that an IT academy in North Carolina increased high school graduation (by 7–8 percentage points) and postsecondary enrollments 3 to 6 years after entry in the 9th grade.
2. Technical high schools—Studies by Dougherty (2018), Ecton and Dougherty (2023), and Brunner et al. (2021) of technical high schools in Massachusetts and Connecticut—the enrollments of which were oversubscribed—show evidence of lower high school

dropout rates and higher earnings for students who did not enroll in college. Seven years after enrollment, low-income students from technical high schools in Massachusetts had annual earnings approximately $3,500 higher than similar students from other high schools, among those who did not attend college (Ecton & Doughtery, 2023); on-time graduation rates were also 7–10 percentage points higher (Doughtery, 2018). Among male technical school attendees in Connecticut, dropout rates were 10 percentage points lower and earnings were $1,600 higher (Brunner et al., 2021).

3. P-Tech and other pathways—P-Tech is a program which began as a collaboration in New York between high schools, the City University of New York, and IBM. It is a program covering grades 9–14, where students engage in IT training and work experience along with their academic work. The model is now spreading to other locations and industries. According to Rosen et al. (2020), high school students in P-Tech had passed their New York regents exams and enrolled in postsecondary programs more frequently than students in other high schools. Bonilla (2020), in a study of a competitive grant to build pathways to postsecondary education and work in high-demand industries in California high schools, also found significantly lower rates of dropping out.

If the impacts on dropping out and high school completion persist, then these programs generate large positive impacts. The total social and economic cost of high school dropout, analyzed by Belfield and Levin (2007) and updated to current dollars, suggests social costs of approximately $1 million in present discounted value over the life cycle for each dropout, in terms of lost earnings, higher costs to taxpayers, and higher nonpecuniary costs (with the last two primarily associated with crime and poor health).

Moreover, the costs of these CTE approaches per student are not terribly high. For instance, the marginal costs for each Career Academy student are roughly $1,100–2,400 per year (Hemelt & Lenard, 2018), while the overall costs of technical high schools are $6,000; assuming that the cost of high school without either program is nearly $5,000, the two figures suggest the costs of the two approaches are quite comparable. So, the expected value of a program that raises high school graduation rates by 10 percentage points would be at least $100,000 per person while costing vastly less. Also, while some models of CTE have traditionally produced evidence of lower enrollments in 4-year colleges, the studies above mostly show little evidence of this.

One concern about CTE programs has been that they might “track” students away from 4-year colleges and universities. As indicated, they appear to raise enrollments in 2-year colleges, especially when “career

pathways” are created that explicitly link students to programs there. But the studies above show little evidence that high-quality CTE lowers enrollment in 7-year programs or attainment of bachelor of arts degrees.

Postsecondary Sectoral Training Programs

A number of sector-based training program models have been developed in the past 20 years and have been subjected to rigorous random-assignment evaluations. Among the most successful, Per Scholas (one of the WorkAdvance⁴ programs), Project Quest,⁵ and Year Up⁶ generated average earnings gains for participants of approximately $5,000–$8,000 that last for at least 5 to 11 years in RCT evaluations, while direct costs of training and supports (but not employment under Year Up) are about $6,000 to $12,500 (see Table C-4-3).

It is important to note that not all sectoral training programs are as effective as these. For example, Per Scholas is one of four sectoral training programs evaluated by MDRC. Across all four, earnings impacts averaged about $2,500 per year. We do not understand enough about what made

___________________

⁴ For more information see https://www.mdrc.org/publication/employment-and-earnings-effects-workadvance-demonstration-after-seven-years

⁵ For more information see https://economicmobilitycorp.org/eleven-year-gains-project-quests-investment-continues-to-pay-dividends/

⁶ For more information see https://www.yearup.org/about/newsroom/press/year-announces-significant-sustained-earnings-gains-young-adults-five-year

Per Scholas so effective, so it is important that the process of scaling up the program models we discuss be done in a way that carefully monitors implementation quality and program effectiveness.

Also noteworthy is that these programs have not been proven for general populations of disadvantaged youth. All employed careful screening, with some reporting acceptance rates ranging between 10% and 30% (Kazis & Molina, 2016; Maguire, 2016).

And finally, in addition to the direct costs listed in the table, all of these kinds of training programs carry an opportunity cost of lost earnings during the training period for participants, which can last six months to 2 years.

Details of Our Calculations

In arriving at our aggregate cost estimate, we assume an average direct cost of about $10,000 per trainee.⁷ Year Up—which involves about 6 months of classroom training and 6 months of paid internship at a private company for low-income high school graduates—costs a lot more if you include the costs of the internship. Employer payments tend to cover about 60% of its costs and we assume that would also be true for our scaled-up programs. (Year Up is also working on variations of its model, to see whether it can generate lower-cost versions.)

So, comparing benefits to costs, the very best programs can pay for themselves in roughly 3–4 years (if earnings are foregone for 1–2 years), while a somewhat wider range of programs can take more than 5–6 years to pay for themselves. Assuming no further fade-out of impacts, the present discounted values of future earnings streams for program participants generate substantial benefits for them and the public (especially if the earnings gains can reduce crime, poor health, or the need for participants to rely on public benefit programs in the future). Assuming some moderate fade-out can still generate public benefits that clearly exceed costs

At what scale would one need to provide these programs, and at what total cost?

• Youth—About 3.2 million students graduated from high school in 2019, and two-thirds of them enrolled right away in college. So just

___________________

⁷ For more information see https://www.mdrc.org/publication/employment-and-earnings-effects-workadvance-demonstration-after-seven-years. We include it in these options for the following reasons: (a) earnings impacts have been estimated over longer time periods than for the other programs; (b) students earn college credentials which might draw returns for longer periods and with greater portability; and (c) as a community college program, it might be easier to scale. And although Year Up is considerably more expensive than Per Scholas, we include the latter because of its exclusive focus on youth, its broad industry range, and its larger impacts to date on youth earnings.

over 1 million had no immediate chance of getting a postsecondary credential. If we add in high school dropouts who might get a GED, plus enrollees who will not complete any credential or earn enough credits to raise their earnings, plus those who obtain very weak credentials, we end up with about 1.5–2.0 million students every year who could benefit. On the other hand, Year Up and the other sectoral programs screen out many students (with poor cognitive skills and other barriers to work), so many of these youth would not be eligible to enter these programs, or might not even apply for other reasons. If we offer Year Up or another program to 1 million of these students per year, it would cost $10 billion annually (not including foregone earnings of the students)—or less if fewer are enrolled.

• Adults—In this case, we use a stock estimate of the population of potential low-income participants in these programs, rather than a flow. We can consider offering something like Project Quest or Per Scholas to adults at ages 25–44 (or perhaps a bit lower) who have moderately young children (perhaps below age 13) and have incomes below 200% of poverty. Roughly 30% or so of such adults meet the income criteria, and we assume that around 60% have children in this age range (though the true number might differ). There are 88 million Americans in that age range, so about 15–16 million fit the additional eligibility criteria. If we offered one million of them entry into Per Scholas or Project Quest each year, that would also cost $10 billion a year.

Scaling Challenges

In all cases, scaling these programs up while maintaining quality would be challenging—so the scaling would need to take place slowly and deliberately. The original sites that have been evaluated to date are fairly small. Each of the programs is now trying to replicate itself, but these efforts will not generate nearly enough scale to achieve our goals. The programs are also exploring ways to achieve more scale while maintaining quality, such as the use of online training to reduce cost, and partnerships with other providers (like community colleges) to expand their reach. The efforts of the city of San Antonio to scale up Project Quest have been described in presentations.⁸

___________________

⁸ For more information see https://www.bexar.org/DocumentCenter/View/30320/Project-QUEST---Bexar-County-SBED-Webinar

2022). Differences in suicide rates by gender, race, and ethnicity underscore the disproportionate burden borne by Native American populations, for whom rates can be more than six times higher than the group with the lowest rate, Latino youth. The group with the next highest rate is White adolescents. While suicides among Black and Latino youth are less frequent, the rate for Black youth has risen more quickly over time.

There is some research showing that mental health treatment can make a difference. Analyses of multiple cognitive behavioral therapy (CBT) interventions, including CBT and interpersonal psychotherapy for adolescents, have found them to be effective for the treatment of adolescent depression (Klein et al., 2007). Research on the effectiveness of pharmacological treatments is more varied, with some of the stronger evidence for selective serotonin reuptake inhibitors (Cipriani et al., 2016; Strawn et al., 2015). Most research on the effectiveness of treatment focuses on mental health outcomes. Given the strong evidence that mental health affects educational attainment and earnings, research linking mental health treatment with future economic outcomes would be very informative.

The Centers for Disease Control and Prevention (CDC) and the Substance Abuse and Mental Health Services Administration (SAMHSA) have each developed evidence-based guidelines and strategies to address gaps in youth mental health care. The CDC has outlined the importance of early identification, providing resources to support families and identifying gaps in workforce development funding. This includes training professionals from fields that are connected to mental health and also integrating mental health care with routine health care. SAMHSA has focused on reducing emergency department utilization for children with mental health issues, reducing criminal justice interactions, and encouraging community-based care coordination.

With research pointing to the importance of location-based policy levers (e.g., So et al., 2019), SAMHSA has prioritized Certified Community Behavioral Health Clinics, which are a central part of the agency’s strategy, with 400 grantees nationally, but is also supporting school-based services. Another area where there is growing evidence of effectiveness in addressing both physical and mental health care needs, especially among low-income communities, is in school-based health centers (SBHCs). A review of the evidence on the effectiveness of SBHCs on educational and health outcomes led the Community Preventive Services Task Force to recommend continued implementation and maintenance of SBHCs (Community Preventive Services Task Force, 2016). SBHCs receive funding from multiple sources (50% from the federal government and the rest from local and state sources). From 1996 to 2017, total funds dedicated to SBHCs have increased from $42 to $91 million, and the number of SBHCs has grown from 900 to 2,584. However, this growth masks significant geographic

disparities: the number of states with SBHC programs declined over this period from 34 to 17.

Pollution

Federal clean-up of Superfund sites has been linked, causally, with important medium-term outcomes. Using data from 1994 to 2002 on the timing of Superfund site clean-up and comparing the outcomes of siblings born to the same family before and after a clean-up, researchers have linked Superfund site clean-ups with improved cognitive test scores and halving in the rate of children with a cognitive disability (Persico et al., 2020). If one only considers the reduction in special education expenditures associated with the clean-up, the researchers calculate that Superfund clean-ups would pay for themselves within 40 years. Researchers have also linked Superfund Site clean-ups with reduced child blood lead levels and improved infant health (Currie et al., 2011a; Klemick et al., 2020).

Children living in households below the poverty line and Black children are more likely to live near one of the 300,000 facilities that emit toxic chemicals, known as toxic release inventory (TRI) sites (Perlin et al., 1999). There is research linking childhood exposure to toxic chemicals at these sites to short-term, and more recently long-term, outcomes including educational attainment and wages. Researchers address the nonrandom placement of TRI sites in various ways. In the short term, evidence suggests that proximity to TRI sites increases infant mortality and reduces birthweight. In a recent working paper, Persico (2022) exploits variation in exposure across siblings derived from the opening of a plant and/or a family moving. Children exposed to a TRI site in utero complete 1.2 fewer years of school and have nearly 60% lower income in adulthood than their unexposed siblings. It is important to note that the TRI program, managed by the Environmental Protection Agency, tracks the management of 770 toxic chemicals and publishes this information for the public.¹ The TRI program does not set standards for chemical emission levels.

Nutrition and Food Insecurity

For pregnant individuals, accessing healthy nutrition is critical to ensure healthy birth outcomes, such as lowering the risk of low birthweight (da Silva Lopes et al., 2017), as well as developmental outcomes into

___________________

¹ Facilities that manufacture, process, or otherwise use these chemicals in amounts above established levels must annually report the amount released (i.e., emitted into the air or water) and/or managed through recycling, energy recovery, and treatment. According to Persico (2022), 221 million Americans lived in a zipcode with a TRI as of 2016.

childhood, such as cognition (Borge at al., 2017; Ramakrishnan et al., 2012; Veena et al., 2016). Nutritional status during pregnancy is also associated with overweight and obesity in children. For example, there is a correlation between obesity during pregnancy and poor neonatal outcomes, such as higher risks of having preterm birth, large for gestational age, and perinatal death (Aviram et al., 2011; Marchi et al., 2015). Obesity during pregnancy is also correlated with poor infant and childhood outcomes, including child obesity, and with poor health outcomes into adulthood (Langley-Evans, 2015; Poston et al., 2011).

The federal nutrition programs have been shown to be effective at reducing food insecurity in childhood and improving child health and future economic outcomes. Supplemental Nutrition Assistance Program (SNAP) in particular has been linked to improvements in newborn health, improved child outcomes, reductions in poverty, and improved economic and health outcomes in adulthood (Almond et al., 2011; M. Bailey et al., 2020; Hoynes & Schanzenbach, 2015; Hoynes et al., 2011, 2015, 2016). However, SNAP benefits are often consumed before the end of the month, resulting in increased food insecurity, decreased food consumption and other spending, and impaired dietary quality (Calloway et al., 2015; Franckle et al., 2019; Gregory & Smith, 2019; Hamrick & Andrews, 2016; Todd, 2015; Weinstein et al., 2009; Whiteman et al., 2018), which in turn have negative impacts on academic achievement (Bond et al., 2022; Cotti et al., 2018; Gassman-Pines & Bellows, 2018; Gennetian et al., 2016). Another complication is that SNAP benefits are to a large degree fungible with cash incomes, making it difficult to know to what extend intergenerational impacts are the result of increased food consumption or increased economic resources.

The Special Supplementation Nutrition Program for Women, Infants, and Children (WIC) is associated with positive impacts on nutrition, dietary intake, food security, and health, including birth outcomes, the latter of which is causally associated with educational attainment and later economic outcomes (Black et al., 2007; Royer, 2009). Improvements in 2009 to the nutritional content of WIC food packages—including the addition of a fruit and vegetable voucher and requirements that bread be whole grain and milk be low fat—have been found to improve dietary quality, perinatal outcomes, and early child development (Guan et al., 2021; Hamad et al., 2019a,b; Tester et al., 2016).

Despite these numerous benefits, participation rates among eligible children decline significantly after infancy (> 98%) across 1–4 years of age, (65%, 49%, 44%, 25%, respectively), and only 52% of eligible pregnant individuals participate (USDA, 2022). These numbers increased during the pandemic due to the multiple waivers that existed (remote benefits issuance, physical presence, allowance of uploading electronic certification

documents, etc.) that streamlined certification, but quantitative data on national impact does not yet exist. COVID-19 waivers increasing the WIC fruit/vegetable benefit also improved dietary intake, but evidence is still needed to prove its positive impact on health.

Factors contributing to the high numbers of children who are WIC eligible but not enrolled include concerns regarding immigration status, lack of transportation to attend in-person appointments, and administrative barriers that made it more difficult for families to enroll or stay enrolled in WIC (National Academies, 2017e; Vargas & Pirog, 2016). For SNAP, take-up rates are generally high (82% nationally, though they can be as low as 55% in some states). The main barrier to SNAP utilization among children is the ineligibility of many immigrant groups. Among immigrants, only those with permanent residency and those who are refugees can access SNAP benefits. The undocumented and those awaiting permanent residency are not eligible. Having parents or siblings in the latter categories affects the benefit levels of eligible children who are U.S. citizens. As a result, children in immigrant families are at higher risk for food insecurity (Capps et al., 2009; East, 2020; Kaushal et al., 2013; Van Hook & Balistari, 2006).

There is causal evidence that expanding coverage to immigrant families is effective in improving child outcomes. East (2020) examines the experience of U.S.-born children of immigrants whose parents were subject to changes in eligibility for SNAP over time, though the children maintained their eligibility throughout. Eliminating (restoring) parental eligibility for SNAP while maintaining child eligibility effectively reduced (increased) total monthly SNAP benefit payments. East (2020) also explores the impact of having SNAP from infancy through age 5 on health outcomes at ages 6–16, including parent reports of child health. She documents effects of SNAP benefit levels on parent-reported child health that are similar to the effects on self-reported health generated from the roll-out of SNAP as documented by Hoynes et al. (2018). An additional year of eligibility for SNAP reduces the probability of fair/poor health by 5% at ages 6–16 in the more recent study, compared with a 3% reduction in the same measure of self-reported health by adults associated with SNAP roll-out.

Paid Family and Medical Leave

There are some interventions that may be promising avenues for increasing intergenerational mobility through improvement in child and maternal health. These include paid family and medical leave. Only 23% of U.S. civilian workers report having access to paid leave for medical, caregiving, or parental obligations (Beach & Walsh, 2021). Lack of access to paid

leave disproportionately impacts low-income and racial/ethnic minority workers (Bartel et al., 2019; Boyens et al., 2022). Paid family and medical leave policies improve neonatal health outcomes, including reductions in rates of low birthweight and the risk of prematurity especially for unmarried and Black mothers (Rossin, 2011; Stearns, 2015), which in turn affect health in childhood (Lichtman-Sadot & Bell, 2017) and educational attainment and earnings later in life.

to their parents’ level of wealth (Feiveson & Sabelhaus, 2018). As a result, the life chances of children from families at the lower end of the wealth distribution, on average, follow significantly different trajectories from those whose families were at the middle and top end (Pfeffer & Killewald, 2019). Lack of wealth constrains opportunities, inhibits choice, and increases economic vulnerability.

Trends in Wealth and Wealth Inequality

Wealth in the United States is highly skewed. Average household wealth was approximately $748,800 in 2019, while the median value was $121,700.¹ These numbers are explained by two related phenomena. First, most households are concentrated at the lower end of the wealth distribution with a wealthy minority at the top. And second, the share of wealth held by the top 1% of U.S. families was 33.1%, while the top 10% held 71% of U.S. wealth, compared with the 27% and 2% held by the bottom 50–90th percentile and bottom 50% of U.S. households. The wealth distribution has become more unequal in recent decades, with the top 1% owning an increasing share of overall U.S. wealth. From 1989 to 2019 the share owned by the top 1% increased from 25 to 34% (Bricker et al., 2020). The Gini coefficient, a measure ranging from zero for perfect equality to one for maximum inequality captures how evenly distributed wealth is in an economy, provides another way of showing this trend. For the United States, the coefficient peaked in 2016 at 0.86 (last year calculated) yet it remained relatively unchanged between 1950 (0.83) and 2007 (0.82), rising most recently post-Great Recession (Kuhn et al., 2020).

Wealth holdings vary by many of the same group traits that cause income to vary including age, education, household composition, geography, and race/ethnicity. U.S. households with children are less wealthy than households without children. In 2019 they had a median wealth level of $64,050 compared with $114,850 of households with nonresidential children. They are also more unequal than nonchild households, with a Gini coefficient of 0.90 compared with 0.86 for nonchild households (Gibson-Davis & Hill, 2021).

Trends in Wealth and Wealth Inequality By Race/Ethnicity

A focus on wealth reveals extreme economic inequality by race/ethnicity. In 2019, median wealth for White households was $188,200 compared

___________________

¹ These values do not include defined benefit pension assets. The Federal Reserve estimates that median wealth increases to $172,000 when inclusive of defined benefit reserves (Bricker et al., 2020).

with $24,100 and $36,100 for Black and Latino households, respectively (Bhutta et al., 2020). Few datasets are available that specifically track the wealth of Native Americans, but research from the National Longitudinal Survey of Youth estimated that in 2000, the median wealth for Native Americans in the survey was $5,700, compared with the median wealth of $65,000 for the sample overall (Zagorsky, 2006).

Percheski and Gibson-Davis (2020) found that between 2004 and 2016, Black-White wealth inequality grew faster in households with children than among all households. By 2016, they found, non-Hispanic White child households had a median wealth of $47,250, compared with a median wealth of $294 among Black child households. Pfeffer and Killewald (2019) found that Black children born in the middle 22% of the wealth distribution are 2.5 times more likely to fall into the bottom 20% of the wealth distribution than White children born into the same tier of wealth.

Causes of Wealth Inequality Trends

Kuhn et al. (2020) explored trends in household wealth inequality between 1950 and 2016, and the growing concentration of wealth at the higher end of the distribution. They found that private business revenue explains much of the increase for the wealthy minority. Growth in wealth among households in the middle of the wealth distribution has been driven by increases in homeownership and, more notably, home equity.

Several recent papers have examined the importance of intergenerational wealth on present wealth inequality (Adermon et al., 2018; Pfeffer and Killewald, 2018; Toney, 2022). Pfeffer and Killewald (2018) found that grandparent wealth is an independent predictor of adult grandchildren’s wealth using U.S. panel data, and Adermon et al. (2018) found the same with Swedish data. Pfeffer and Killewald (2018) found that for both White and Black families, education explains more of the intergenerational wealth gaps and social-class persistence than homeownership, bequests or inheritances, business ownership, or marriage. Toney (2022) showed that the composition of assets within wealth portfolios is static across generations, although the relationship is weaker among Black households.

Causes of Wealth Inequality By Race and Ethnicity

Racial/ethnic wealth gaps reflect historical and contemporary processes and policies that have predominantly supported wealth accumulation for White Americans and impeded or exploited wealth opportunities for Black Americans from slavery and sharecropping and debt traps to more recent evidence of biased home appraisal practices and targeted subprime

mortgage lending (Bayer et al., 2016; Berger, 2018; Faber, 2013; O’Connell, 2012; Perry et al., 2018; Reece, 2020). Black and Hispanic homeowners had much higher rates of delinquency and default in the downturn. These estimated differences are especially pronounced for loans originated near the peak of the housing boom. These findings suggest that black and Hispanic homeowners drawn into the market near the peak were especially vulnerable to adverse economic shocks and raise concerns about homeownership as a mechanism for reducing racial disparities in wealth. Racial differences in wealth accumulation and growing wealth gaps over time have been also attributed to differences in asset holdings and returns on those investments. In recent decades, greater amounts of financial asset ownership among White households and appreciation in stock equity have rewarded them over time. Returns to housing, where Black households hold much of their wealth, have appreciated less due to longstanding patterns of residential racial segregation (Boulware & Kuttner, 2020; Derenoncourt et al., 2022). Other contributing factors include differences in the incidence and size of inheritances (Bhutta et al., 2020). Not only are White households more likely to receive financial inheritances, but they are also more likely to expect them (Addo & Darity, 2021).

Obstacles to wealth accumulation have also operated partly through disparities in income throughout the life course, as income from employment and capital are the primary ways households are able to set aside money to save and build wealth via investments, educational attainment, and asset purchases (Elmi & Lopez, 2021). Derenoncourt et al. (2022) showed that Black-White differences in savings rates are explained by initial wealth levels, age, and education, and Gittleman and Wolff (2004) found no differences in savings after accounting for income.

Background Section for Interventions

EITC Proposal

Graphs depicting the three examples of modifications to the EITC are shown in text Figure 6-7. In the panels for Options 1-3, the 2022 EITC payment for a single parent with two children is shown as a solid line. The phase-in and phase-out rates (the slopes of the two lines), the maximum credit, and the exact locations of the earnings thresholds all vary with family size, but the general pattern is similar. The final graph shows only the incremental credit payments associated with each one.

Taking the example of Option 1, the solid line in the figure shows the current (2022) value of the credit as a function of family earned income. Families with zero earnings are not eligible for any EITC benefits. As family

earnings increase up to $15,410 (the “phase-in” portion of the schedule), EITC payments increase as well—up to a maximum payment in 2022 of $6,164. Between $15,410 and $20,130 (the “flat” portion), this maximum payment is constant. Once earnings exceed $20,130, the credit payment is gradually reduced, so that when earned income reaches about $50,000, EITC payments fall to zero. The dashed line shows how much higher tax credit payments would be if all payment amounts from the 2022 schedule were increased by 40%.

Combining the EITC with Other Programs to Increase Income Stability

The committee considered expansions along the lines of the 2021 Child Tax Credit, which made credits available to families without earnings without increasing the incentive to work. On the plus side, it would provide resources to more low-income families with children than the EITC alone, but it would probably lead to a reduction in maternal employment relative to the status quo. Option 3 in Figure 6-7 combines the addition of credits for families without earnings with an increase in the credit for those with earnings, in order to preserve (and indeed increase) the incentive to work relative to current policy.

The EITC’s annual payments are typically distributed as tax refunds 1 or 2 months after the end of the year in which the income is earned, which may limit the credit’s usefulness in financing recurrent monthly expenses, like child care. Prior work has shown that the lump-sum annual nature of the EITC leads to fluctuations in spending and food security across the year (Batra & Hamad, 2021; McGranahan & Schanzenbach, 2013). It also could limit the work-promoting effects insofar as this annual structure restricts the ability of new mothers to use the credit to finance child care during their first year as new parents. At the same time, the annual structure of the EITC makes it a form of involuntary savings, which people use to help purchase durable goods that are likely to increase workforce participation, such as vehicles (Goodman-Bacon & McGranahan, 2008). The temporary, 6-month-long expansion of the Child Tax Credit (CTC) in 2021 was implemented as a monthly advance payment, and early quasi-experimental evidence indicates that the 2021 expanded CTC improved household food sufficiency and parental mental health (Batra et al., 2023; Shafer et al., 2022). This indirect evidence suggests that more regular payments could help to provide more income stability, though there is no rigorous evidence of the effect of a monthly payment structure on employment, earnings, or multiyear poverty.

on employment. Even in these models, though, sufficiently large minimum wage increases will reduce employment. These models are well reviewed by Dube (2019a) and Belman and Wolfson (2014).

It is thus an empirical question whether increases in the minimum wage will reduce employment, and by how much. There is no consensus in the economics literature about the exact amount, with high-quality studies reaching opposite conclusions. Neumark and Shirley (2021) observe that the minimum wage is an unusual research topic in that the state of the literature itself is under fierce debate.

Neumark and Shirley assemble the entire set of published minimum wage studies from the past three decades. They identify the “core estimates” from each study, in most cases identified by the authors of the studies. They find that 79.2% of estimated employment elasticities are negative, and that the evidence of negative employment effects is even stronger for teens and young adults, lesser-educated workers, and workers directly affected by the minimum wage. This is surprisingly consistent with other literature reviews. For example, among the 36 papers he considers, Dube (2019a) obtains a median estimate of the own-wage elasticity (the percentage employment reduction induced by a 1% increase in average wages induced by a higher minimum wage) of −0.17. Of those 36 studies, 26 are based on “narrow subgroups,” including teens, restaurant or retail workers, and lower-educated immigrants. Among those 26 studies, Dube reports a median own-wage elasticity of −0.19.

This negative average effect reflects a number of estimates indicating positive or zero effects. One prominent recent study (Cengiz et al., 2019) finds that each 1% increase in average wages induced by increases in the minimum wage actually raises employment by 0.41%, though the confidence interval includes some negative effects, as large as −0.45%. This implies that employment reductions offset no more than half of the effect of wage increases. Other studies come to similar conclusions as Dube’s (2019a) review. Clemens and Strain (2021) estimate own-wage elasticities of −0.26 for less-educated workers and −0.23 for younger workers.

In the committee’s view, the evidence supports an own-wage elasticity in a range that includes near-zero, that is centered around −0.2, and that includes larger magnitudes for certain subgroups of workers.

Some studies find that larger minimum wage increases are associated with relatively larger employment reductions—that is, the employment elasticity is nonlinear. The average minimum wage increase in the 138 state-level changes studied by Cengiz et al. (2019) was around eight log points. Clemens and Strain (2018, 2021) studied the last decade of minimum wage increases and found little evidence that “small” increases of under $1 are associated with disemployment effects. For “small” increases, Clemens and Strain’s results are qualitatively similar to the results in Cengiz et al. Dube

(2019a, p. 3) concludes that “the best evidence suggests that the employment effects are small up to around 59% of the median wage.”

However, Clemens and Strain find larger disemployment effects of “large” increases of over $1. Among individuals ages 16–25 with less education than a high-school degree, they find that a one percent increase in wages induced by a higher minimum wage reduces employment by around one percent. For workers ages 16–21, they find a reduction in employment of −0.4%.

Jardim et al. (2017, 2018, 2022) studied Seattle’s path to a $15-per-hour minimum wage, which reached as high as $11 in 2015 and $13 in 2016. Using administrative data from Washington State, Jardim (2022) found larger employment effects than are typically found. However, their point estimates have wide confidence intervals, and their results are also consistent with a selection effect deriving from the booming Seattle economy, or even spillovers from the minimum wage itself (Rothstein & Schanzenbach, 2017).

A contrasting result comes from Godoy and Reich (2021), who studied low-wage counties where minimum wages reached as high as 82% of the median wage. They did not find adverse effects of minimum wage increases on employment, hours, or annual weeks of work.

Another outstanding question is whether the long-run effects of minimum wage increases differ from the short-run effects that have been more studied. Clemens and Strain (2021) found that the medium-run disemployment effects of relatively large minimum wage increases are larger than the short-run effects. Meer and West (2016) argue that minimum wage increases affect employment by reducing its growth over time rather than by quickly reducing its level. Adjustment costs—driven in part by firm entries and exits within an industry—might help explain larger longer-term disemployment effects (Aaronson et al., 2015; Sorkin, 2015). Brummund and Strain (2020) found larger disemployment effects from the practice of indexing minimum wage increases to inflation, and speculate that modest, nominal increases are easier for firms to absorb without adjusting headcount than longer-lasting increases.

A minimum wage increase could reduce poverty even if it also reduces employment, if the benefits to the still-employed workers are larger than the costs to those who lose employment. (Note that for any own-wage elasticity above −1, the aggregate earnings of near-minimum-wage workers rise; the literature is clear that employment responses are much smaller than this threshold.) The picture is clearer here; studies generally find at least some poverty reduction (as measured by the Official Poverty Measure) from minimum wage increases, though there is disagreement about the magnitude. Dube (2019b) found that over three or more years, each 10% increase in the minimum wage reduces poverty by between 2.2% and 4.6%, a large effect. Godoy and Reich (2021) found substantial declines in both household and child poverty.

___________________

² https://eda.gov/arpa/good-jobs-challenge

³ https://www.dol.gov/newsroom/releases/osec/osec20220121#:~:text=The%20%E2%80%9CGood%20Jobs%E2%80%9D%20initiative%2C,all%20workers%20and%20job%20seekers

⁴ https://www.congress.gov/bill/116th-congress/senate-bill/223/text?q=%7B%22search%22%3A%5B%22S.+524%22%5D%7D