Disparities in childhood human capital investments in the United States

Abstract

We analyze data from 10 nationally representative surveys in the United States (2010–2023) to track public, private, and family time investments in children, all theorized to enhance human capital. Average per-child investments from birth to age 18 total $502,152 (2024 USD), with overall disparities by household income and race/ethnicity of 6% to 15%. Early childhood shows the largest gaps—often over 50%—driven mainly by housing and child care. Investment levels converge substantially around age five due to near-universal public schooling, which serves over 90% of children, compared with far lower access to publicly supported early care (roughly 25%) and housing subsidies (less than 5%). Like early childhood, though, later-childhood investments exhibit differences in the types of services and resources children from different backgrounds receive (e.g., compulsory versus compensatory education, office-based versus emergency health services, whole fruit versus fruit juice). Accounting for parent and family time-focused investments (e.g., informal educational activities, meals) narrows disparities by income and race/ethnicity when valued at a constant wage. Disparities increase 50-100% when valued at caregivers’ age- and education-specific wages, likely overstating differences in the quality of family time.

Introduction

Understanding and alleviating inequality in opportunity is one of the most critical challenges currently facing policymakers in the United States (U.S.) and around the world^1,2,3. Research across many academic fields consistently argues that life course inequality in educational attainment, health, labor market participation and earnings, happiness, and other measures of social wellbeing is shaped by disparate opportunities, resources, and human capital investments directed towards children⁴. The childhood resources that are needed to produce long-run outcomes are diverse and include: housing⁵, health care⁶, nutritional food⁷, preschool and other care environments⁸, high-quality education⁹, time parents and families spend reading to and playing with children¹⁰, after-school and enrichment activities¹¹, access to public spaces like libraries¹², and many others.

Within each of these sectors and domains, existing scholarship indicates that children from lower socioeconomic backgrounds and racial/ethnic minorities tend to obtain fewer investments⁴. Differential investment, in turn, contributes to large gaps in educational achievement, economic mobility, health, and overall wellbeing^13,14,15. For example, in the U.S., test-score gaps between children from the top and bottom quartiles (i.e., top versus bottom 25%) of household income and between black or Hispanic versus white children are estimated to be three-quarters of a standard deviation, if not larger^14,16, equivalent to over a year of learning¹⁷. In young adulthood, income-based and race/ethnicity gaps persist in educational attainment and employment¹⁸, as well as in health and happiness¹⁹. Gaps in median lifetime earnings and life expectancy are similarly large across income quartiles and race/ethnicity groups²⁰. At least in the U.S., gaps in educational outcomes and opportunity between females and males are substantially smaller, though gaps by sex are large for other metrics (e.g., earnings), as well as in other countries and contexts^21,22.

Despite longstanding interest in quantifying disparities in childhood human capital investments, the empirical literature remains piecemeal in nature. Much of the existing scholarship tends to focus on investments in one sector (e.g., education or health), setting (e.g., home, school, or community), or childhood stage²³. More comprehensive assessments that aggregate information across sectors, settings, and ages tend to focus on public^24,25 or private expenditures²⁶ or nonmonetary investments such as parent and family time^27,28. Each type of analysis, on its own, narrows the definition of childhood investment and thus leaves out a large share of the total resources directed toward children.

The siloed nature of existing literature is misaligned with longstanding theory and consensus across academic disciplines that the investment mix is as important as any given investment. Studies using credible designs show that a range of goods—medical services, childhood nutrition, housing, early childhood education, teachers, and many others—all improve long-term outcomes^4,24. Literature on the effectiveness of family-based investments that are produced mainly with parent and caregiver time is less developed, focusing more on the unequal distribution of time across children from different backgrounds^27,28. However, theory and some empirical work suggests that structured family time in informal educational activities (e.g., reading, visiting a library) and in recreation activities (e.g., sports, projects, arts and crafts) produce cognitive (e.g., language acquisition) and socio-emotional skills (e.g., secure social attachments, perseverance)^29,30. Unstructured time between children and caregivers that includes a high density of language can have similar effects^{31,32,33,34,35}, which in our data include activities, such as discussion of heritage and meal and transportation time. Notably, while levels of parent and family time differ by household income and race/ethnicity, existing literature suggests that their effects on child outcomes do not^36,37.

Further, economists, developmental psychologists, and sociologists all argue that all of these resources are needed to produce long-term outcomes, that resources interact with each other, and that the timing of investment relative to developmental stage is critical^{13,38,39,40,41}. Thus, the siloed character of existing work that attempts to quantify investment disparities has limited our ability to understand disparities in the total investment mix that theory predicts is important^4,42,43,44.

We produce estimates of the quantity of childhood investments that U.S. children receive between birth and age 18, by household income, race/ethnicity, and sex, and at what specific ages children receive them. Our primary contribution is our comprehensiveness. We compile and analyze data from 10 federally sponsored and nationally representative surveys, spanning 2010–2023 (see Supplementary Methods Tables 1 and 2 for details on the data). We examine investment levels across six sectors and 10 domains (with some sectors including multiple domains): child care, clothing, education (i.e., formal education, informal educational experiences, and college preparation), health (i.e., health care, nutrition, and exercise), housing, and transportation. In total, we include 77 highly granular individual investments from out-patient versus emergency medical visits, whole fruit versus fruit juice, and compulsory education services (e.g., teachers) versus compensatory ones (e.g., special education, tutoring) meant to remediate academic progress (Table 1).

Table 1 Summary of investments included in 10 nationally representative surveys

Table 2 Investments by childhood stage (2024 dollars)

We track activities that families purchase or are provided via public subsidy and those that they produce with caregiver time, all captured in dollar units that are straightforward to aggregate and easily interpreted. Broadly, our “ingredients method” starts by identifying and specifying the ingredients (e.g., personnel, materials, and fees) for a given investment, then prices each ingredient, and sums across ingredients (for details, see Supplementary Data and Supplementary Methods Table 3)^45,46,47. In addition to providing a method to combine monetary and nonmonetary investments, our ingredients approach allows us to compare children who receive similar investments but at different market prices (e.g., food purchased in a grocery store versus restaurant), and to compare children who receive similar investment totals but with different inputs (e.g., caregiver time versus market expenditures). An exception is clothing, where we only can track family expenditures without adjusting for market prices, which we still include given its centrality in related analyses²⁶.

Table 3 Childhood investments by type of investment and costing approach

In this study, we show that the average total investment per child from birth to age 18 in the U.S. is $502,152 (in 2024 USD). Income-based and racial/ethnic disparities in the total investment, which range from 6% to 15%, are smaller than some previous studies indicate²⁶. However, the relative homogeneity in total investments across the childhood lifecycle masks two important patterns: (1) a differential mix of expenditures versus unpaid caregiver time; and (2) substantial disparities from birth to age five that converge at later ages due to the universal nature of public education and differential levels of compensatory spending at older ages (e.g., special education). Finally, we show that the mix of public and privately financed investments vary across groups and that investment disparities are largest in sectors with relatively low levels of public support (e.g., housing, child care). While our results are purely descriptive, the variation in the public-private mix that we observe is consistent with public investments serving an equalizing role.

Results

Total investments in children and youth

We begin by describing the total investment mix, averaging across all children. Table 1 shows that the average total investment from birth to age 18 is $502,152 (in 2024 USD). Informal educational experiences represent the largest domain (23%), followed by nutrition (22%), housing (16%), and formal education (15%). Each of the remaining six domains (health care, exercise, clothing, child care, transportation, and college preparation) individually represent less than 10% of the total. While informal educational experiences, transportation, and exercise are primarily or exclusively comprised of activities where the primary ingredient is parent and family member time, housing, formal education, health care, child care, college preparation, and clothing are primarily or exclusively expenditure based. Nutrition is more mixed, including investments in caloric intake from specific foods and the large amount of time that parents and families spend with children during meals.

Figure 1, panel A, documents these trends visually, while panels B and C show age gradients in how investments are allocated between birth and age 18 (for expenditure- and parent and family time-focused investments, respectively). In the first year of life, time spent on feeding (i.e., breastfeeding, formula feeding, other meals) is the single largest investment, just shy of $20,000 and roughly equivalent to 1.8 h of parent or family time per day. By age one, parent and family time-focused investments in meals levels out and decreases as children grow, while expenditure-focused investments in food increase slightly over the childhood years. Like nutrition, health care investments are higher in the first year of life, which includes costs associated with birth and frequent doctor visits. Clothing expenditures track very closely with investments in food and health care. Housing is a large investment across all childhood years, with a slight increase over time driven by a reduction in shared rooms as children age.

Fig. 1: Investments in children, by domain, type, and age.

a Documents the mean total investment between birth and age 18, by domain and type of investment (i.e., expenditure- versus parent and family time-focused investments). Dollar units are on the left y-axis, and percent of total investment (summing across all domains) is on the right y-axis. b, c Disaggregate means by investment type and age. Child care, clothing, college preparation, and housing only include expenditure-based investments. Transportation only includes time-based investments. In b, c, 95% confidence intervals (CI) for the means are included. N = 1,202,932 in the full sample.

Informal educational experiences begin at birth—and increase substantially at age one—with early literacy (e.g., reading, telling stories, and singing to children), arts and crafts, and other activities. The peak exactly at age five is somewhat of a data artifact, reflecting how roughly two-thirds of 5-year-olds in our data start kindergarten while others wait a year. As a result, some 5-year-olds receive home-based informal educational experiences from parents and families only captured in the data through age five (e.g., early literacy activities) in addition to other time- and expenditure-focused informal educational activities only captured starting at age five (e.g., projects, lessons, help with homework) (see Supplementary Data). Both exercise (e.g., time at parks/playing sports) and child care also start relatively high at birth and then decline at age five with the start of formal school. Our data include three types of child care, all of which are provided by someone other than a parent: center-based care, home-based care from a nonrelative (e.g., nanny), and home-based relative care, most often provided by grandparents. In our primary analyses, we exclude relative care because it is likely duplicative with informal educational experiences facilitated by family members. For formal education, investments can start at birth with special education services. From kindergarten through 12th grade, the slight decrease over time is driven primarily by larger class sizes in middle and high school, which reduce per-child personnel costs. Time spent transporting children (often to/from school) is a much smaller but steady investment over time, while college preparation is a small, one-time investment that we plot at age 16 (but could occur at any time in high school).

In the discussion, we describe how our calculations likely underestimate the total investment in children and youth, given some investments that we cannot capture reliably in our data (e.g., facilities, supplementary teachers). However, the primary contribution of our analyses in disparities in investments reduces the importance of underestimating the overall level.

Disparities in the total investment

We describe differences in investments by household income, race/ethnicity, and sex. Income is measured at the household level, and we group children by quartiles. Race/ethnicity is self- or guardian-reported using survey questions that ask about race and Hispanic origin in separate questions. We combine these questions to form categories for non-Hispanic Asian American and Pacific Islander, non-Hispanic black, non-Hispanic white (henceforth “AAPI”, “black”, and “white”, respectively), and Hispanic. Sample sizes for American Indian/Alaska Natives, other races, and children identifying with multiple races are too small to generate statistically reliable results, and their exclusion remains a limitation of our study. Sex (male/female) is based on self- or guardian-reported survey questions that ask about sex assigned at birth.

Table 2, panel A describes disparities in total investment amounts. Relative to children in the top quartile of household income, whose total investment is $557,382, children in the bottom quartile receive investments that are roughly $86,000 less (z = 23.81, p < 0.001, d = −2.37, 95% CI = [−93,450, −79,237], all tests are two-tailed). This absolute difference in dollar units represents a relative difference of −15%. While we conduct hypothesis testing of between-group differences in dollar units, discussion of relative differences is useful for describing disparities when the dollar value for a particular group, sector, domain, individual investment, or childhood stage/age changes. Differences in standard deviation units (d) achieve a similar goal. The total investment gap between children from the second versus top quartiles of household income is similar (−$79,475, z = 22.0, p < 0.001, d = −2.20, 95% CI = [−86,556, −72,395]), as is the difference between Hispanic and white children (−$73,135, z = 20.17, p < 0.001, d = −2.06, 95% CI = [−80,241, −66,029])—both representing relative gaps of −14%.

In contrast, investment gaps are smaller between children from the third versus top quartiles of household income (−$44,654, z = 11.17, p < 0.001, d = −1.18, 95% CI = [−52,487, −36,822]), black versus white children (−$55,007, z = 14.96, p < 0.001, d = −1.56, 95% CI = [−62,212, −47,802]), and AAPI versus white children (−$33,005, z = 6.58, p < 0.001, d = −0.90, 95% CI = [−42,842, −23,169]). Relative gaps fall between −6% and −10%. The difference between males and females is small in relative terms (−1%), and the absolute difference is not statistically significantly different from zero (−$3983, z = 1.35, p = 0.176, d = −0.11, 95% CI = [−9748, 1783]). Therefore, we focus our primary analyses on investment differences by household income and race/ethnicity.

Table 2, panels B and C, demonstrate that disparities by income and race/ethnicity are more pronounced for younger versus older children. We split childhood stages into two groups at age five, given visual evidence in Fig. 2 that this is where between-group differences change course. For example, the relative difference for children from the bottom versus top quartiles of household income from birth to age five is −24% (absolute difference: −$44,424, z = 21.99, p < 0.001, d = −1.67, 95% CI = [−48,384, −40,464]), compared to a −12% relative gap for older children through age 18 (absolute difference: −$43,824, z = 15.32, p < 0.001, d = −1.58, 95% CI = [−49,431, −38,216]). The relative gap for AAPI versus white children also is cut roughly in half in later versus early childhood, while the relative gap for black versus white children in early childhood is roughly three times the gap in later childhood. In Supplementary Methods Table 4, we further disaggregate investment disparities between children ages 5–11 and 12–18, finding them to be quite similar. Supplementary Methods Fig. 1 documents age trends by sex.

Fig. 2: Investments in children, by subgroup, type, and age.

a, b, c Show mean investment estimates at a given age by quartiles of household income, while d, e, f show means by race/ethnicity. Race/ethnicity subgroups are ordered alphabetically. AAPI = Asian American and Pacific Islander. a, d Include all investments, while b, e focus on expenditure-focused investments and c, f focus on parent and family time-focused investments. Ninety-five percent confidence intervals (CI) for the means are included. See Supplementary Methods Fig. 1 for time trends by sex. N = 1,202,932 in the full sample.

In Table 3, we explore whether household income and race/ethnicity groups invest in children using similar or different amounts of expenditures versus parent and family member time. Groups may differ based on their preferences. Or, as incomes rise, families may substitute time for market expenditures⁴⁸. For children from the top versus bottom quartiles of household income, relative differences are −16% for expenditure-focused investments (absolute difference: −$49,821, z = 17.18, p < 0.001, d = −1.88, 95% CI = [−55,505, −44,138]) and −15% for parent and family time-focused investments (absolute difference: −$37,414, z = 18.21, p < 0.001, d = −1.44, 95% CI = [−41,440, −33,388]). The similarity between the two perhaps reflects public subsidies (see below).

For children from other income quartiles and for most race/ethnicity groups, gaps in expenditure-focused investments are larger than gaps in parent and family time-focused focused investments. The Hispanic-white gap in expenditure-focused investments of −19% (absolute difference: −$55,495, z = 18.35, p < 0.001, d = −2.26, 95% CI = [−61,423, −49,567]) is over twice as large as the gap in parent and family time-focused investments of −8% (absolute difference: −$17,732, z = 9.34, p < 0.001, d = −0.67, 95% CI = [−21,455, −14,010])—aligning with prior literature that also finds parent and family member time to be inequality reducing for Hispanic children⁴². This pattern is even more pronounced for AAPI children, who receive 16% less expenditure-focused investments (absolute difference: −$46,258, z = 13.15, p < 0.001, d = −2.05, 95% CI = [−53,150, −39,366]) but 6% more parent and family time-focused investments compared to white children (absolute difference: $13,149, z = 3.96, p < 0.001, d = 0.44, 95% CI = [6643, 19,655]).

In Fig. 3, we show that adding parent and family time-focused investments to expenditure-focused investments reduces the relative difference between AAPI or Hispanic versus white children, compared to differences that focus on expenditure-focused investments only (bars 1 versus 2). However, for black children, the pattern is reversed: the black-white gap in expenditure-focused investments of −8% (absolute difference: −$24,826, z = 8.49, p < 0.001, d = −0.98, 95% CI = [−30,559, −19,093]) is smaller than the gap in parent and family time-focused investments of −13% (absolute difference: −$29,892, z = 13.66, p < 0.001, d = −1.17, 95% CI = [−34,182, −25,602]). We explore this pattern in more detail below by disaggregating disparities to the domain and individual investment levels.

Fig. 3: Disparities in investments in children, by the costing approach.

Bars represent relative differences (in percent units) in investments in children by household income (with top quartile as the reference group) and race/ethnicity (with white as the reference group). Relative differences are calculated by taking the mean dollar difference between each group and their referent, dividing by the mean dollar value of the reference group, and multiplying by 100. The first bar focuses on relative differences in expenditure-focused investments only (see also Table 3, Panel A). The next two bars show relative differences in total investments that sum across expenditure- and time-focused investments from parents and families, with two different costing approaches for parent time: constant parent wage (see also Table 2, Panel A) and differentiated wages by parents’ age and education (see also Supplementary Methods Table 6, Panel A). Differences by sex are excluded as the disparity is close to zero.

The monetized value of parent and family time-focused investments depends on the hourly wage rate used for adult time. In our primary analyses, we use an average hourly wage for U.S. adults, as reported by the Bureau of Labor Statistics ($32.38/h; Supplementary Methods Table 3). With this approach, we simply multiply hours for each parent and family time-focused investment by a constant, adding on supplemental costs for additional ingredients (e.g., arts and crafts materials, entrance fees to museums). Because the supplemental expenses are small, this approach nearly replicates relative differences when measuring parent and family time-focused investments in hours (Supplementary Methods Table 5).

We also consider how an alternative approach to monetizing parent and family time changes our estimates of disparities in children. The alternative approach costs parent and family member time using average wages conditional on parents’ age and education. We show between-group differences just for parent and family time-focused investments in Table 3, panel C and for the total investment mix that also adds on expenditure-focused investments in Fig. 3 (and in Supplementary Methods Table 5 for test statistics). For AAPI children, differentiated parent wages increase their advantage over white children because the average implied wages of AAPI parents are higher than other race/ethnicity groups. However, we observe the opposite for other children. For Hispanic children, parent and family time-focused investments, when valued at a constant wage, reduce overall disparities, but they increase disparities by over 50% when valued at a differentiated wage rate. Income-based investment disparities roughly double when using differentiated wages, relative to disparities using a constant wage (or when using hours).

In other words, differentiated wages suggest that the quality of AAPI, white, and/or high-income parent time is higher than that of lower-income, black, or Hispanic parent time. We are unaware of any empirical evidence that supports such an assertion, and a handful of studies instead suggest that the benefits of parent and family time to child outcomes is similar regardless of parent and family characteristics^36,37. As such, we prefer estimates that use a constant wage rate, which we use in the remaining analyses. See additional discussion in the “Methods” section.

Disparities in individual investments

Next, we increase the granularity of our analyses by plotting the relative difference between groups for each of 10 domains and the 77 specific investments we can measure in the available data, by household income quartile and race/ethnicity. We sort domains by their total dollar amount. The two domains with the largest dollar amount and the largest number of individual investments (i.e., informal education and nutrition) are shown in Fig. 4. The remaining domains are shown in Fig. 5. We sort individual investments within domain the same way. (For all descriptive and test statistics of between-group differences at the domain and individual investment levels, see Supplementary Methods Figs. 2–8.)

Fig. 4: Disparities in individual investments: informal education and nutrition.

For each investment (left y-axis), we present the mean total dollar amount for the reference group (top quartile of household income and white, right y-axes) and plot the relative difference in percent units for the other groups. Relative differences are calculated by taking the mean dollar difference between each group and their referent, dividing by the mean dollar value of the reference group, and multiplying by 100. Ninety-five percent confidence intervals (CI) for the relative differences are calculated by dividing the standard error of the between-group difference in dollar units by the mean dollar value of the reference group, multiplying by 100, and then calculating upper and lower bounds using a z distribution. Domains and investments within each domain are sorted by the total amount in the full sample. See Supplementary Methods Figs. 2–8 for the full set of test statistics and differences by sex. N = 1,202,932 in the full sample.

Fig. 5: Disparities in individual investments: housing, formal school, health care, exercise, clothing, child care, and transportation.

See note to Fig. 4.

For some domains, including housing, child care, exercise, and clothing, individual investments generally move in the same direction and contribute to inequalities. Under the housing domain, where between-group differences range from the black-white gap of −12% (absolute difference: −$10,969, z = 39.81, p < 0.001, d = −1.46, 95% CI = [−11,509, −10,429]) to the Hispanic-white gap of −30% (absolute difference: −$28,187, z = 146.77, p < 0.001, d = −3.95, 95% CI = [−28,563, −27,811]), disparities in bedrooms, communal rooms, and fixed housing costs (utilities, etc.) are generally equivalent. Under child care, we observe large income-based gaps in both center-based child care and home-based care from a nonrelative (e.g., nanny). Domain-level differences in child care that sum across both types are −61% for children from the bottom versus top quartiles of household income (absolute difference: −$17,291, z = 17.71, p < 0.001, d = −1.51, 95% CI = [−19,205, −15,378]), −40% for Hispanic versus white children (absolute difference: −$8,507, z = 8.54, p < 0.001, d = −0.75, 95% CI = [−10,459, −6555]), and −26% for black versus white children (absolute difference: −$5568, z = 4.74, p < 0.001, d = −0.46, 95% CI = [−7869, −3266]). Care provided by a nonparent relative is higher for AAPI, black, and Hispanic versus white children. It is also higher for the three bottom quartiles of household income compared to the top quartile. As noted above, we do not include relative care in the domain or grand total because of its overlap with informal educational experiences. We discuss possible tradeoffs across child care alternatives in the discussion.

For other domains, Figs. 4 and 5 show that domain-level averages mask substantive heterogeneity at the specific investment level. Some of these differences may reflect differences in preferences or cultural norms. For example, under informal education, parent and family time talking about heritage, as well as visits to zoos, are larger investments for AAPI, black, and Hispanic children compared to white children. The reverse is true for school meetings and community groups, which favor white and high-income children. At the domain level, these differences wash out to a large degree, such that between-group differences in informal education as a whole are minimal.

In other instances, domain-level averages mask heterogeneity that may be more consequential. Under nutrition, whole citrus, melons, and berries are more common for children from the top versus the bottom three quartiles of household income and white versus black children. In contrast, fruit juice is more common for children from the bottom two versus top quartiles of household income and for black and Hispanic versus white children, potentially reflecting access to fresh produce. Similarly, children from the bottom two quartiles of household income, as well as AAPI, black, and Hispanic children, receive larger investments in bottled water compared to their reference groups, potentially reflecting access to safe tap water. In the health care domain, children from the bottom three quartiles of household income receive larger investments in hospital stays and emergency room visits, while high-income children receive larger investments in dental care, glasses, and office visits. White children also receive larger investments in dental care, glasses, office visits, outpatient visits, and prescription drugs, while black and Hispanic children receive larger investments in emergency room visits. Summing across investments, domain-level disparities in nutrition range from a 18% gap that favors AAPI over white children (absolute difference: $20,783, z = 9.27, p < 0.001, d = 0.84, 95% CI = [16,388, 25,179]) to a −21% gap between black and white children (absolute difference: −$24,044, z = 16.15, p < 0.001, d = −1.33, 95% CI = [−26,963, −21,126]). Differences in health care are as large as −37% for black versus white children (absolute difference: −$13,071, z = 12.18, p < 0.001, d = −1.02, 95% CI = [−15,173, −10,958]).

Under formal schooling, the largest dollar investment is teachers, where we observe small between-group differences. However, compared to children from the top quartile of household income, children in the lowest income quartile receive 72% larger investments in tutoring (absolute difference: $3933, z = 4.57, p < 0.001, d = 0.54, 95% CI = [2246, 5620]) and 68% larger investments in special education (absolute difference: $4877, z = 5.63, p < 0.001, d = 0.51, 95% CI = [3179, 6575]). Black children receive 134% larger investments in tutoring than white children (absolute difference: $6956, z = 7.68, p < 0.001, d = 0.88, 95% CI = [5180, 8732]), but statistically indistinguishable investments in special education (absolute difference: $1250, z = 1.50, p = 0.177, d = 0.14, 95% CI = [−380, 2879]). Overall, the lowest-income children receive slightly more formal schooling investments than the highest-income children, and black children receive slightly more than white children.

Home-schooling is one investment where income and race/ethnicity gaps do not run in tandem. White children receive larger investments in homeschooling than all other race/ethnicity groups, while children from the top quartile of household income receive smaller investments compared to children from the bottom two quartiles. However, very few students in the U.S. are home-schooled overall (no more than 4%).

The contribution of specific domains to overall disparities

Next, in Fig. 6, we descriptively examine the extent to which each of the 10 domains contributes to total disparities, accounting for the relative between-group difference for each domain and the absolute dollar value across domains. For example, while we observe some of the largest relative differences between income groups in child care, the total dollar value of child care is roughly one-third of the housing investment, where relative differences between groups are roughly half as large. In Fig. 6, the value of the bars sums to 100% for each group, with positive values signaling that the domain contributes to the between-group disparity and negative values indicating that the domain reduces disparities. We disaggregate these same analyses by investment type (expenditure- versus parent and family time-focused investments) in Supplementary Methods Fig. 9.

Fig. 6: Contributors to investment disparities, by domain.

Each bar captures the extent to which a given domain contributes to or detracts from between-group disparities, which is calculated as the between-group mean dollar difference for a given domain divided by the total mean dollar difference across all 10 domains. Bars for a given group sum to 100. a Shows disparities by quartiles of household income, where the top quartile is the reference group; b shows disparities by race/ethnicity, where white is the reference group. Race/ethnicity subgroups are ordered alphabetically, and investment domains are ordered from largest to smallest total dollar amount. See Supplementary Methods Fig. 9 for similar figures disaggregated by investment type (i.e., expenditure-versus parent and family time-focused investments). Differences by sex are excluded as the disparity is close to zero.

Housing is the largest contributor to income-based investment disparities. Housing contributes more to investment disparities between children from the third versus top quartiles of household income (30%) compared to the disparity between children from the bottom versus top quartiles (21%). This is because other domains, including child care also are large contributors to investment disparities for children from the bottom versus top quartiles of household income (20%), but less so for children from the third versus top quartiles (12%). After child care, nutrition is the third largest contributor to income-based investment disparities (18% for children from the top versus bottom quartiles), driven entirely by parent and family time during meals rather than expenditure-based investments in food (Supplementary Methods Fig. 9). Clothing and informal educational experiences contribute to income-based investment disparities to a similar degree for children from the bottom and second versus top quartiles of household income. That said, we are cautious in interpreting the estimate for clothing because our data capture not just the amount of clothing purchased but also differences in prices, and we expect that higher-income households pay more than lower-income households for similar goods (e.g., brand-name versus discount clothing). In contrast, health care contributes no more than 10% to income-based investment disparities, while investments in formal schooling favor children from the bottom two versus the top quartiles of household income.

The primary contributors to total disparities based on race/ethnicity are overlapping with, but not the same as those for household income. Housing also is the primary contributor to investment disparities for Hispanic (39%) and AAPI (55%) versus white children, but contributes less to investment disparities for black children (20%). Across race/ethnicity groups, child care contributes to investment disparities to a moderate degree (10–16%). Formal schooling reduces investment disparities for black children—as it does for low-income children—and contributes minimally to investment disparities for Hispanic children. However, formal schooling contributes substantially to investment disparities for AAPI versus white children. Part of the large positive value for schooling for AAPI children is mechanically driven by the large negative value for nutrition. Because the bars sum to 100%, the inequality-reducing role of parent and family time during meals for AAPI children forces the positive values for other domains to be larger than they would be without accounting for nutrition. In contrast, black children receive far smaller investments in family time during meals (Supplementary Methods Fig. 9), which explains why parent and family time-focused investments are not inequality reducing for black children, but they are for others.

For both AAPI and black children, health care is the second largest contributor to investment disparities, relative to white children, whereas it contributes minimally to income-based investment disparities. For black children, health care makes up 24% of total investment disparities and 52% of expenditure-focused investment disparities (Supplementary Methods Fig. 9).

Public versus private sources of investment

Finally, in Fig. 7, we examine between-group differences in the public versus private source of expenditure-based investments, which provides additional insight as to why some domain- and investment-level gaps are large and others much smaller. All parent and family time-based investments are fully private and thus excluded from this analysis. Here, we focus on nine investment categories in which the public-private split is reasonably captured in our data. For example, the national surveys do not include information on public subsidies for clothing. Some of the categories in Fig. 7 are at the domain level (i.e., health care, nutrition, housing) while others are at the individual investment level (e.g., center care and home-based nonrelative care under child care) due to artifacts of the available data. (For complete information on how we allocate expenditure-focused investments to a public source, see Supplementary Methods Table 6; for tests of between-group differences, see Supplementary Methods Table 7).

Fig. 7: Share of children who receive investment support from public social safety-net programs.

a Plots the share (mean) of children who receive investment support from public social safety-net programs by household income, while b plots means by race/ethnicity. Child subgroups are ordered alphabetically. Ninety-five percent confidence intervals (CI) for the means are logit transformed with bounds [0, 100]. See Supplementary Methods Fig. 10 for patterns by sex. N = 1,202,932 in the full sample.

We find substantial income- and race/ethnicity-based differences in the share of child investments coming from public sources. The most common source of public investment is for formal schooling, which is a common service available to all children in the U.S. Children from the top quartile of household income receive 14.5 percentage points [pp] (z = 6.17, p < 0.001, 95% CI = [9.88, 19.08]) less public investment relative to children from the bottom quartile because they are more likely to be enrolled in private school. Differences in publicly supported formal schooling are also apparent in public support for tutoring and school-related activities, such as team sports.

Beyond schooling, and consistent with means-tested public programs, 80% of children from the bottom quartile and 45% from the second quartile of household income receive public support for health care (e.g., Medicaid or the Children’s Health Insurance Program), compared to 5% of children from the top quartile (bottom versus top quartile difference: 75.06pp, z = 102.98, p < 0.001, 95% CI = [73.63, 76.49]; second versus top quartile difference: 40.49pp, z = 38.3, p < 0.001, 95% CI = [38.42, 42.57]). Seventy-four percent of children from families from the bottom quartile of household income receive public support for food expenditures (e.g., Supplemental Nutrition Assistance Program [SNAP] or Supplemental Nutrition Program for Women, Infants, and Children [WIC]) (difference compared to top quartile: 71.44pp, z = 66.34, p < 0.001, 95% CI = [69.33, 73.55]), and 62% receive publicly financed center-based child care (e.g., Head Start, public pre-K, Temporary Assistance to Needy Families [TANF]) (difference compared to top quartile: 54.04pp, z = 33.18, p < 0.001, 95% CI = [50.85, 57.23]). However, fewer low-income children receive housing assistance, including 13% for the bottom quartile (difference compared to top quartile: 12.93pp, z = 53.89, p < 0.001, 95% CI = [12.46, 13.40]) and 2% for the second quartile (difference compared to top quartile: 1.81pp, z = 18.69, p < 0.001, 95% CI = [1.62, 2.00]). Given the correlation between household income and race/ethnicity, we observe similar patterns in black-white and Hispanic-white gaps for these same programs. Supplementary Methods Fig. 10 shows differences by sex in the source of investment, which are minimal and consistent with findings in Tables 2 and 3 that show minimal investment gaps by sex.

Discussion

Despite widespread consensus that children require a diverse bundle of investments that act in concert to produce long-term outcomes^4,13,39,49, existing work quantifying how much human capital investments U.S. children obtain—and disparities in those investments—has largely occurred in silos focused on narrowly defined investment categories. This study produced a comprehensive picture of child monetary and nonmonetary investment, from public and private sources, for several broad investment sectors and domains that include 77 individual investments, at each age from birth to 18.

Our work is generally consistent with previous literature. For example, studies using different data on investments in caregiver and family time estimate that parents spend between 7 and 13 h per week engaging with children as a primary activity⁵⁰. We estimate a total amount of parent and family time-focused investments in children under 18 years of 6810 h (Supplementary Methods Table 7), or an average of 7.3 h per week.

Our estimates of expenditure-focused investments also are consistent with other studies that use different data sources and aggregation strategies. The clearest example is an analysis of 2015 data aggregated by the U.S. Department of Agriculture (USDA), which focused primarily on private expenditures from parents and families. While our analyses purposefully combine expenditure-based investments from private and public sources—and we do not have a way to disaggregate—domains that are overlapping in our data and theirs and come primarily or exclusively from private sources (e.g., housing, clothing) produce similar estimates. Formal schooling includes expenditure-focused investments that are publicly subsidized for most children, where our estimates also are consistent with other literature. We estimate a per-child investment in formal education of roughly $6000 per year (Fig. 1), which is slightly more than half of federally reported average per-pupil expenditures on instructional resources⁵¹. Our estimates are driven by students’ primary teacher(s)—for whom we have direct teacher-child links—who, in our data, account for roughly half of total teachers in schools (also including gym, music, art, reading specialists, etc.).

We contribute to these literatures by aggregating a more comprehensive set of investments. In doing so, our results identify several important facts about child investments in the U.S. First, total investments over the course of childhood are roughly $86,000 more for children from the top quartile of household income versus the bottom quartile, and roughly $55,000–$75,000 more for the race/ethnicity group with the highest investment (white) compared to the two groups with the lowest investment (black and Hispanic). These differences are meaningful (upwards of 15% gaps) but far smaller than other studies of investment disparities. For example, the USDA study described above identifies income-based gaps of roughly 50% or larger in family expenditures. We observe smaller disparities because we count parent and family time-focused investments that are inequality-reducing for AAPI and Hispanic children⁴². Another explanation is that, for expenditure-focused investments, our ingredients-based costing approach uses nationally normed prices (except for clothing) that likely reduce disparities, as opposed to capturing differences in prices that families are willing or able to spend on similar resources.

Second, relative equality in total investment quantities masks important heterogeneity in the timing and type of investments children receive. White children and those from higher-income households receive substantially more investment from birth to age five, compared to their counterparts. These differences are largely explained by white and higher-income families investing more in child care from someone other than a relative, as well as housing, where large gaps start at birth and continue throughout childhood. These patterns are concerning, as early childhood is a critical period in which investments have large returns^13,24. This pattern reflects a U.S. policy choice to constrain public spending on 0–5 year olds relative to its Organization for Economic Co-operation and Development (OECD) peers—a decision the U.S. has not made for older children⁵².

Third, investment levels converge abruptly at age five, a process largely driven by formal schooling. That said, equalizing levels of investment during middle and later childhood also appear to reflect increased compensatory investments in groups that are under-invested in during early childhood. Low-income, black, and Hispanic children are more likely to participate in activities, such as tutoring and special education, often meant to remediate academic progress or offset the disabilities that result in a special education classification. Differences in compensatory investment in school-age youth is consistent with large investment gaps in the pre-school years and is similar to patterns of investment observed in health care. Low-income, black, and Hispanic children are less likely to obtain care in settings associated with primary and secondary preventive services (office visits, dental, and pharmaceutical services), but more likely to obtain care in hospitals and emergency departments. In other words, equity in total investments does not reflect an equitable investment mix.

Fourth, for some sectors and domains, a large share of low-income (particularly bottom quartile), black, and Hispanic children receive investments assisted by public sources and programs. The central example is formal schooling, which is a publicly provided service available to all children in the U.S.—and taken up by over 90% of kids. Health care and food also are highly subsidized for children from the bottom quartile of household income (around or above 75%) and, in turn, for black and Hispanic children. While we do not measure what counterfactual investment levels would be in the absence of public support, our estimates are consistent with the idea that the system of universal or means-tested public services is an equalizing force. We hypothesize that similar investment levels for children in the bottom and second quartiles of household income across the childhood years (see Table 1 and Fig. 2), likely is due to means-tested public assistance specifically targeting the lowest quartile. This evidence also echoes and extends prior work showing how growing inequality in the U.S. is driven by the top end of the income distribution rather than losses at the bottom end⁴. A natural corollary is that in sectors and domains where public assistance is smaller, investment disparities are larger. This is true for two of the largest contributors to investment disparities: child care, where public assistance reaches roughly 60% of the lowest-income children (as opposed to 75% or higher for some other domains), and housing, where public assistance reaches no more than 15% of any of the income or race/ethnicity subgroups. Health care is a large contributor to investment disparities for black children, even though they are more likely to receive public support than white children.

Our approach uses a comprehensive framework that allows us to capture both time and material investments across a broad set of sectors and domains. However, it is not without limitations. We are unable to track investments in facility costs and expenditure-focused investments on transportation, nor are we able to apply our ingredients method to clothing investments. As described above, we include teacher salaries for students’ primary teacher(s) but not for supplemental teachers. These exclusions lead us to underestimate the total investment in children between birth and age 18 and may cause us to miss some sources of investment disparity. We also could not track insults and contaminants, such as environmental exposures or violence, which impact child development and differ across groups⁴¹. For investments we do track, estimates come from survey sources where reporting errors may differ across groups. More broadly, the conversion of raw units into a common dollar-denominated metric is not without controversies and uncertainties, including the price of parent and family time^46,53. Finally, our measurement of public support does not include tax expenditure programs like the earned income tax credit (EITC) and does not account for the fact that private spending in nearly all domains is supplemented with public subsidies. One example is the tax exclusion of employer sponsored health insurance, which is regressive in nature and far larger than the EITC⁵⁴. This leads us to underestimate the share of investment from public sources.

We made the deliberate choice to be agnostic about the quality of investments children receive. As a result, our estimates reflect the raw quantity of time and material that is directed towards children of different groups. We measure hours spent in tutoring, for example, but not the relative productivity of a given hour of tutoring. We measure the number of physicians visits a child had at a given age, but not whether those visits produced more health. This approach has several advantages. It allows us to convert all metrics to a dollar scale that is readily understood and easily interpreted by diverse audiences. Further, it requires relatively few assumptions and allows us to apply common decision rules to inputs captured in different ways across varied surveys. In contrast, there is often a lack of consensus on what constitutes high- versus low-quality education or health care. To our knowledge, no available data source has information on the quality of childhood investments across a broad range of investment categories for a nationally representative sample. Price may be a proxy for quality in some but not all instances⁵⁵. Under health care, for example, we find that unit prices for outpatient and emergency room visits are quite similar (Supplementary Methods Table 3), despite considerable debate about the quality of each, as measured vis-a-vis patient feedback and subsequent health outcomes⁵⁶. Under nutrition meals purchased at full-service restaurants (and fast-food restaurants) often are more expensive than similar meals prepared at home with ingredients purchased at the grocery store, even if the nutritional value is the same⁵⁷. It is even more challenging to quantify the quality of time-focused investments from parents and caregivers (e.g., reading to children) from large-scale datasets¹¹.

At the same time, advocates and researchers focused on the “opportunity gap” clearly have in mind differences in both quantity and quality. Quality of investment impacts its return, and there is substantial evidence that education, health, economic, socio-emotional, and other metrics of wellbeing and human capital accumulation are quite uneven across groups based on household income and race/ethnicity⁴. There are some signals of quality differences in our data. Whole fruit, which is a larger investment for high-income and white children, has somewhat more nutritional value (e.g., fiber) compared to 100% fruit juice⁵⁸, which is a larger investment for lower-income, black, and Hispanic children. Under child care, low-income, black, and Hispanic children receive the largest investment in relative care, most often from grandparents, compared to other types of care. However, these same groups receive smaller investments than high-income and white children in early literacy activities (e.g., reading, stories) from family members. While only indirect evidence of quality, these patterns align with other literature showing how quality of care varies across informal and formal settings and that, regardless of race/ethnicity, grandparents tend to offer fewer structured activities to children than other home-based, nonrelative caregivers (e.g., nannies) and center-based caregivers^59,60. In formal school, our main analyses document comparable investments in teacher salaries across income and racial/ethnic groups, which is promising given the importance of teachers to short- and long-run outcomes⁹. However, differences arise in teacher experience, education level, and class size (Supplementary Methods Table 8), which are the “ingredients” that determine the per-child investment in teacher salary and directly impact student outcomes⁹.

Ultimately, the value of our analyses comes from comparing well-established evidence of achievement and other outcome gaps^16,20,61 to the comparably small gap in total investments that our comprehensive approach reveals. Our findings speak directly to the U.S. context, which often is viewed as an outlier amongst wealthy nations for its high child poverty rates and comparably low public expenditures directed toward children³. The U.S. spends more than the average OECD country on 6–17 year olds, but substantially less on 0–5 year olds—the ages where we observe the largest investment gaps⁵². Ultimately, our analyses suggest that policy should focus on the timing and character of investment. Furthermore, it suggests that public spending, when it occurs, is an inequality-reducing force.

Methods

Our study was reviewed by the University of Maryland Institutional Review Board (IRB) and was deemed exempt from full committee review because we rely on de-identified publicly available survey records. Informed consent and subject compensation was managed by the individual surveys we describe below. Detailed information about survey operations can be found online via the weblinks listed in the Data availability section. Across all surveys, our sample includes 1,202,932 observations of children age zero to 18 (Supplementary Methods Table 1).

Survey inclusion criteria and coverage of investments

Our analyses of cross-sector investments in children and youth in the U.S. are based on data from 10 publicly available surveys that cover six broad sectors of investment in children and youth described in prior literature^25,26 (ordered alphabetically by sector):

• 1. Child care (Early Childhood Program Participation survey [ECPP] and Parent and Family Involvement in education [PFI] that are both part of the National Household Education Survey [NHES])

• 2. Clothing (Consumer Expenditure Survey [CES])

• 3. Education, including four domains:

  • a. Formal education (NHES: PFI, Early Childhood Longitudinal Study Kindergarten Class of 2010–11 [ECLSK], and High School Longitudinal Study [HSLS])

  • b. Informal educational experiences (ECLSK, HSLS, NHES: ECPP, NHES: PFI, PSID)

  • c. College preparation (HSLS)

• 4. Health, including three domains:

  • a. Health care (Medical Expenditure Panel Survey [MEPS] and ATUS)

  • b. Nutrition (National Health and Nutrition Examination Survey [NHANES] and ATUS)

  • c. Exercise (Panel Survey of Income Dynamics [PSID] and NHES: PFI)

• 5. Housing (American Community Survey [ACS] and American Housing Survey [AHS]

• 6. Transportation (American Time-use Survey [ATUS])

To identify these surveys for analysis, we use four inclusion criteria. First, we require that surveys are nationally representative in order to provide generalizable conclusions across the U.S. population of children. Second, we focus on surveys (or survey waves) that cover the time period between 2010 and 2023, in order to speak to recent patterns. For health care data from MEPS, we focus more narrowly on 2014 onward that reflect health care arrangements in the U.S. after the implementation of the Affordable Care Act. We compile as many years and waves of available data as possible from each survey and within this time span to maximize sample size. However, we exclude 2020 because most surveys did not collect data from this covid year, and those that did either cannot ensure national representativeness (i.e., ACS, ATUS) or provide insight into investments during covid for only a small subset of investments (i.e., clothing from CES, health care from MEPS).

Third, surveys must be able to link investments directly to individual children, as opposed to parent or family reports of household-wide investments. One exception to this rule is housing, from AHS, which is by its nature a household-level expenditure. However, as described below, we assign each child a share of housing expenses. Clothing data from CES also is provided at the household level, but with sufficient information on child characteristics that allows us to attach purchases to individual children (see Supplementary data for details).

Finally, surveys must include information on children’s household income, race/ethnicity, and sex, which are subgroups we focus on when examining investment disparities. Reporting criteria and coding of race/ethnicity is described above in the Results section. Throughout the analyses, we refer to “sex” because most surveys use this terminology and ask about a child’s sex assigned at birth. However, three surveys (i.e., ECLSK, NHANES, and PSID) refer to self-identified “gender” in at least some survey waves. For household income, as a general rule, data collected by the U.S. Census Bureau measure “money income” that is pre-tax and does not include transfers. Due to the self-reported nature of income in some surveys, though, there could differences across respondents in what people include⁶². For example, in NHES, respondents are asked to select an income category based on the following prompt: “Include your own income. Include money from jobs or other earnings, pensions, interest, rent, Social Security payments, and so on”. That said, in the notes to Table 3, we show that estimates of average parent wages, derived from the same self-reported data, increases monotonically across quartiles of household income, suggesting that we likely are categorizing families appropriately into the categories we use in our analyses.

We do not set a requirement that surveys include information on the source of investment and whether it is covered by public social safety net programs, which is one part of our analyses. That said, all surveys we use have this information for the majority of investments examined.

In a couple of instances, we identify multiple surveys that meet these criteria and cover the same investment areas. We select the survey with the most detailed information to convert raw survey units into dollars (a process we describe below) and, in some instances, include multiple surveys if they capture investments in similar ways and cover different age levels and years. A slight exception to this rule is the ATUS, which provides very detailed time information in minutes per day on a range of activities and is used in several studies tracking time-based investments in children^27,28,42. However, ATUS is primarily concerned with the time that adults spend rather than the time that children receive. While the survey contains unique child identifiers in instances when they accompany their parent or primary caregiver for a given activity, it is not possible to track children’s time when spent with anyone else. Likely as a result, we estimate much smaller child investments in activities, such as reading, playing games, help with homework, etc. from the ATUS compared to other surveys that include similar data (i.e., ECLSK, NHES). Ultimately, we only use ATUS as a primary data source for three specific investments that are not captured in other surveys (i.e., health care support time, meal time, transportation time).

Supplementary Methods Table 1 provides details on the surveys—including design features, age coverage, year coverage, and sample sizes—while Supplementary data lists the full set of variables we cull from them. Together, the 10 surveys cover a large swath of human capital investments described in the academic literature. However, there are some gaps. We exclude a couple of one-off investments, such as summer activities (e.g., summer school, camp), because we only have this information for children aged five to seven. Given prior literature documenting investment gaps during summertime⁴³, we consider the sorts of investments that may occur over the summer (e.g., child care, tutoring, lessons) that we do observe in our data, and we multiply these out across 12 months (rather than 10 months that cover the formal schooling period).

Similarly, in an effort to be as conservative as possible in what we consider to be an investment, we focus on what children use rather than what they have access to. For example, we include the time children spend in public libraries but exclude this and other community resources if we do not know that children used them directly. Applying this conservative perspective, we also avoid double counting investments that may show up in different surveys or in different sectors (e.g., informal educational experiences versus relative care).

Samples and imputation

Some surveys include children and youth at all ages from birth to age 18 (e.g., those covering housing, health care, nutrition; Supplementary Methods Table 1). Other surveys focus on more narrow age bands. For example, the two sub-surveys within the NHES each focus on a different age group: birth through age five for early childhood education and care from the ECPP survey, and ages five to 18 for school-aged investments from the PFI survey.

In order to ensure that we capture all observed investments across all ages during childhood and adolescence, there are some instances where we impute data. While the ECLSK and HSLS surveys cover a primary education investment (teachers and other schooling personnel) in the elementary and high school levels, respectively, we are not aware of a similar nationally representative survey covering the middle school years. The Middle Grade Longitudinal Survey was designed to cover similar information to ECLSK and HSLS in the middle grades. However, the survey design and data collection was interrupted by covid. To address this gap, we take a similar imputation strategy as other studies²⁶ to carry forward values from the highest age band/grade level from ECLSK (age 10–11/5th grade) and carry backward values from the lowest age band/grade level from HSLS (age 13–14/9th grade). Using the same strategy, we duplicate observations to impute forward two of four high school grade levels in the HSLS data where students were not surveyed (10th and 12th grades). For PSID, there is no information for children ages zero to one, given that data from the child development supplement sub-survey that we use is drawn from a sample of households and individuals that engaged in the main PSID survey one year earlier. We do not impute backwards here, as investments for children ages one to two can be quite different from those ages zero to one (see Fig. 1).

All surveys ensure national representativeness through complex sampling designs. We use weights provided by each survey to produce population estimates. We divide the weights by the number of waves such that the weights sum to the population count in an average year (versus person-years). This approach follows guidelines from survey developers. The approach changes the weighted population counts, but not proportions, means, and other descriptive statistics that we derive in our analyses. For the longitudinal surveys that follow the same children over time, ECLSK and HSLS, the weights provided already are adjusted for the number of waves, such that the weighted estimates are representative of the population cohort that is the target of the survey.

In all surveys, weights are proportional to the likelihood of being selected to participate in the survey, with adjustments made to account for nonresponse and coverage error. All surveys also use multi-stage sampling designs, but differ in terms of the primary sampling unit (PSU). Surveys administered through the U.S. Census Bureau (ACS, AHS, ATUS, MEPS, NHES) use the same broad sampling approach: first to stratify or sample geographies and then randomly select households within geographies, with either all or a random subset of children in each household included in the data. The two longitudinal education surveys administered by the U.S. Department of Education (ECLSK, HSLS) first randomly select schools within strata based on school type (public versus private), region, and urbanicity; then, classes are randomly selected within schools. Across all our analyses, standard errors account for these complex sample designs using Taylor series linearization.

Survey sample sizes also matter for our analyses. Subgroup cells are a function of age and household income, race/ethnicity, or sex. Most surveys have average cell sizes over 300. However, in most surveys, AAPI children are the smallest subgroup, and PSID includes some cell sizes for this group that are below 30. Because PSID is the only nationally representative survey that includes information on tutoring and some other investments—and no additional waves of data are available at the time of publication—we still include these data and interpret results more cautiously. The total sample size in the ACS is by far the largest across surveys. To decrease computation times, we identified a 5% random sample from each ACS wave and multiplied the weights by 20 to ensure that the final sample still represents the population of U.S. children.

The characteristics of children

Supplementary Methods Table 2 describes characteristics of the (unweighted) samples gathered from nine surveys, excluding the ACS (panel A), which is much larger than all other surveys combined, and then just the ACS (weighted) on its own (panel B), which is our best estimate for the characteristics of the U.S. population of children. In the 2010–2023 period (excluding 2020), population estimates indicate that (in alphabetical order) 4.8% of children were AAPI, non-Hispanic, 13.4% were black, non-Hispanic, just under a quarter were Hispanic (24.8%), just over half were white, non-Hispanic (50.7%), and the remaining 6.4% were another race. In the population, AAPI and white children were overrepresented in the top 25% of the household income distribution, while black and Hispanic children were underrepresented. The unweighted characteristics of the samples generally agree with these patterns.

Investment accounting

To measure cumulative, cross-sector investments, we use a multi-stage process that first puts all investments on a common dollar scale. We use 2024 USD since the last year of available survey data tracks investments through the end of 2023. The primary framework we use to convert investment units from their raw scales to dollars is the “ingredients method”⁴⁶. This method starts by identifying and specifying the ingredients (e.g., personnel, materials, fees) for a given investment, then prices each ingredient, and sums across ingredients. Each investment has different sets of ingredients and requires different steps and assumptions to convert to dollars per year. We provide detailed decision rules for each variable/investment in Supplementary data, and we describe broad categories of decision rules below.

For clothing data from the CES, we are not able to apply our ingredients method because the information provided only includes the total family cost for a broad category of clothing (e.g., pants, shirts, and footwear) without the number of units purchased. Therefore, we cannot adjust clothing expenditures to national prices and so treat analyses of clothing investments as exploratory. Similarly, in the health care data, glasses and “other” health expenses remain in expenditure units because we do not have units of use. In these two instances, the total investment is quite small.

For many investment domains (child care, exercise, formal education, informal educational experiences, transportation), the primary ingredient is adult or parent time. This is true for all parent and family time-focused investments, as well as for many expenditure-focused investments that are produced through time from someone other than a parent or family member (e.g., tutor, coach). We never count the time of the child. For some investments, adult or parent time has a direct cost (e.g., salaries for teachers, tutors, child care workers), while other time-based investments from parents and family are not directly priced in a market.

The first step in converting time to dollars is to transform the raw data to hours per year. A couple of investments (e.g., child care, home-schooling, reading, and transportation) already are measured in fine-grained time units (e.g., hours, minutes). For the remainder of the time-based investments, raw scales generally refer to event counts: times per week or times per month. Here, we use the ATUS to infer the time in minutes for a given activity during a single event. These estimates are included in the Supplementary data. As noted above, ATUS censors time-based investments in children from individuals other than parents and primary caregivers. However, it provides very detailed time information on a range of activities that children and parents engage in together. We assume that these time-per-event estimates with parents are a reasonable stand-in for the time-per-event that children spend with other adults. After collecting these data from the ATUS, we multiply event counts per year by hours per event.

Next, we convert time to dollars using hourly wages derived from additional nationally representative sources, including the Bureau of Labor Statistics (BLS). Wages for different positions and all other “standard values” of prices are reported in Supplementary Methods Table 3 along with the source. In several instances, the occupation from which we derive the hourly wage is obvious, such as with child care workers for center-based child care and home-based child care from a nanny ($14.94/h) and recreation workers for sports teams ($16.56/h). For several investments that are led by an experienced professional (e.g., lessons, religious activities, community group), we use the broad occupation category of educator. For full-time teachers, we use yearly wages that are reported in national data from the School and Staffing Survey and vary along three dimensions: degree, experience, and school type (public, private). Because U.S. principals generally are required to have a master’s degree, salary schedules only include two dimensions: experience and school type. For school counselors, our data do not provide information on education and experience, and so we use the average yearly wage as reported in BLS.

Identifying the hourly price of parent or family member time for relevant investments (e.g., reading, transportation) is more complicated and subjective, given the different activities that parents engage in and the fact that these are non-paid activities. We follow the U.S. Bureau of Economic Analysis and use a single generalist wage⁶³, as opposed to specialist wages that considers the market substitute for specific activities (e.g., education-related activities versus recreational ones) and are used in other analyses⁴². While the U.S. Bureau of Economic Analysis uses the wages of paid household workers, such as nannies or other child care workers, we prefer the average hourly wage of U.S. adults.

In a supplementary analysis, we consider an approach that differentiates parent wages, which also is a common approach in the literature⁴². Here, we use the current population survey (CPS) to estimate average wages within age-education cells. Based on the CPS, we bucket age in roughly 5-year increments, with some adjustments for young parents and those right around retirement: younger than 21, 22–24, 25–29, 30–34, 35–39, 40–44, 45–49, 50 –54, 55–59, 60–61, 62–64, 65–69, 70–74, and 75 or older. Education categories include: less than high school diploma, high school diploma (or equivalent), associate’s degree/vocational/some college, bachelor’s degree, and professional degree. All five surveys that include parent and family time-focused investments (i.e., ATUS, ECLSK, NHANES, NHES, PSID) include this information for at least one parent, and most have it for up to two (the exception is NHANES). After calculating the wage for each parent, we average the two and use that to cost out child-level investments.

This supplementary analysis using the implied parent wages of each child probes some of our costing assumptions, which we describe in the Results section. However, as we describe above, we prefer using the average hourly wage across all U.S. adults. Our preferred strategy assumes that the value and productivity of parent and family member time is the same for different groups, no matter their household income and race/ethnicity. For example, parents and family members with different wages may be equally “good” at reading to their children. Conversely, an oil executive who earns a high wage may not be as “good” at taking care of a child than a professional child care worker who earns a lower wage. Further, because household income varies substantially across race/ethnicity groups (Supplementary Methods Table 2), as do hourly wages (see notes to Table 3), differentiating wages suggests that time for AAPI and white families is more valuable than time for black and Hispanic families. By using a common wage rate in our primary analyses, we avoid these concerns. This approach is consistent with a small body of literature showing that the effect of parent and family time on child outcomes is similar no matter the characteristics of parents, families, and households^36,37.

Except for parent and family member time, we multiply wages or salaries by 1.315 to add on benefits. This is not controversial for professional services. However, there is debate in the field about whether or not to add benefits for part-time workers and volunteer positions, which includes many of the parent- or family-led activities that we examine in our analyses⁵³. The literature suggests that benefits should be added if one thinks about parent and family time as substitutable with child care worker time (e.g., hiring someone in their place), which would require benefits. However, as described above, in our preferred analyses we cost out parent time with an average adult/parent wage rate and so do not add on benefits.

Many of the investments described above are shared amongst several children, in which case we divide the total dollar amount by the number of children that are likely to share in it. For example, for the time that adults, parents, or guardians spend over meals, we divide by the number of children who were present during that meal. For home-based child care by a relative or nonrelative, we divide by the number of children in the household who receive care from that same person. And, for center-based child care, we divide by the maximum number of children who can receive care from a single worker, as mandated by law (e.g., four children per adult for ages 0–1). Using a similar approach, we divide the yearly wages for U.S. teachers by class size (in ECLSK) or school-level student-teacher ratio (in HSLS, where class size is not captured). We divide principal salaries by school size and counselor salaries by student-counselor ratios. Finally, we add on additional ingredients (e.g., school supplies), where average national costs also are pulled from nationally representative surveys, such as the CES.

For nutrition, health care, and housing investments, we also apply an ingredients-based framework, but to data where the raw unit generally is not time. Nutrition data from the NHANES survey is reported in their food ingredients (e.g., apples, broccoli, rice) that are collected from meal diaries. Survey developers then converted this information to volume units (e.g., grams, ounces) for the five primary food groups (dairy, fruit, vegetable, grain, protein), in addition to other areas (e.g., formula, beverages, excess sugar). Each of these groups has subcategories (e.g., dark green versus starchy vegetables), which allows us to convert these units to dollars using national price estimates collected through the Quarterly Food-at-Home Price Database (QFHPD). All prices assume food ingredients are purchased in a grocery store and do not reflect mark-up rates when meals are instead purchased in a restaurant or cafeteria. Time during meals (and breast- or formula-feeding infants) also is a large parent and family time-focused investment under nutrition. The ATUS includes detailed time-diary information on meal time when parents or guardians are engaging with children as a primary activity. Meal-diary data from NHANES identifies whether a child was breast- or formula-fed and how many times a day each occurred. We cost out infant feeding by multiplying the number of feedings per day by an estimate of the per-feeding time (15 min), and then across days in a year.

For health care, the primary raw units are event counts for office visits, emergency room visits, and other health care services. While the MEPS dataset also includes exact expenditures paid for by families or insurers, we prefer the count data for two reasons. First, starting with event counts aligns with the ingredients approach taken elsewhere. We capture the number of office visits that children attend, for example, in a similar way as we capture the number of times a child goes to a museum, bookstore, or library. Second, the exact dollar amount of medical expenditures, as reported in the survey, depends not only on the number of events experienced but also differences in unit prices negotiated between providers and insurers. Therefore, to generate standard price values per event, we use the same MEPS dataset to calculate the average Medicaid rate for a given event type. While Medicaid rates also vary between states, our estimated rates reflect the national average. The Medicaid rate is the lowest rate, which aligns with our attempt to be as conservative as possible. After obtaining the average Medicaid rates, we then multiply event counts by the nationally normed Medicaid price for that event.

Similarly, for special education, we first identify children who receive these services and then apply a common national price for each disability category identified under the Individuals with Disabilities Education Act (IDEA) (e.g., learning disability, traumatic brain injury). In Supplementary Methods Table 3, we refer to these as “pooled costs” because the price already sums across individual ingredients, including personnel, materials, and fees.

One limitation of our strategy is that we do not have a consistent way to capture costs associated with facilities, which can be an additional ingredient in the investment function. Formal schooling, for example, requires use of a school building, but the available data sets do not include information to calculate the per-child costs (e.g., the number of rooms in the school). The same is true for visits to libraries, zoos, museums, etc. Exclusion of facility fees in these instances will understate the total investment we make in children and youth and perhaps misstate differences across groups.

For home-based investments (e.g., home-schooling, help with homework, child care in the home, home-based learning activities), facility investments are covered by housing, which also is a sector in and of itself. Housing data from ACS includes the total yearly housing costs from rent or mortgage payments and utilities (and other associated owner costs, such as condominium fees or property taxes). Rather than taking direct expenditures, we focus on the housing resource available to a child (bedroom, share of other rooms in the house) and estimate a price for that resource using guidelines in the literature²⁶. We first estimate per-room costs. We then allocate the number of rooms dedicated to children as the number of bedrooms per child, which can vary between 0 and 1, plus an equal share of all remaining rooms in the house (e.g., 5 people in a 4 common room house would each get 0.8 rooms).

Public versus private sources

In addition to capturing the dollar amount of investments, we collect information from the same surveys on the source of investment: public versus private. We define public investments as those that come from the public school system or from local, state, or federal government safety net programs. These programs include: public housing, multi-family housing, and housing choice vouchers from the Department of Housing and Urban Development; Medicaid, Medicare, Veteran Administration, or other government insurers for health care; Supplemental Nutrition Assistance Program (SNAP) or Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) for food; Head Start or Temporary Assistance for Needy Families (TANF) for child care; and public schooling.

In Supplementary Methods Table 6, we provide details on the source of payment data available in each survey and how we create the measures that we use in our analysis. In most instances, available data allow us to create binary measures of whether a child or family received assistance from one of these sources, but not the specific amount. Therefore, analyses of these data should be interpreted as the percent of children from different backgrounds who receive public support. For health care, the data allows us to identify whether or not children received support from public insurers, as well as the total amount of expenditures covered by public versus private sources. However, we exclude this information so that estimates are consistent across all surveys and sectors. These estimates are available upon request.

For all investments that rely primarily on parent or family time, we infer that the source is fully private. For investments in college preparation, community groups, religious activities, and clothing, we do not have any information on the source of investment, which could be split between public and private sources. Therefore, we exclude these investments from our analyses.

Accumulation of investments across types and ages

After combining and converting the survey data into a common dollar scale, our empirical strategy is straightforward. We provide simple descriptive statistics on investments and how they differ between children from different backgrounds based on household income, race/ethnicity, and sex. To do so, we estimate means for each investment by each age (e.g., 0–1, 1–2, and 3–4) and subgroup, applying the survey weights and accounting for the multi-stage sampling design to ensure national representativeness. We also capture sample sizes, standard errors, and variances of the means that we use to sum up investments and conduct hypothesis testing. Next, we take sums of the estimated means across ages for a single investment, across multiple investments within a given domain or sector, and then across all domains and sectors. We use the delta method to estimate variances and standard errors for the sums, executed in Stata 18 using the lincom and nlcom commands. We calculate standard deviations for the sums by multiplying the estimated variances by the sample size, summing the variances across age groups and investments, and taking the square root. Our final estimates are average per-child investments, from birth to age 18.

Finally, we use the standard errors, variances, and sample sizes to conduct simple two-tailed hypothesis testing of between-group differences in means for different sets of investments (e.g., individual investments, domains, the total). We calculate p-values based on a normal distribution rather than a t-distribution, given that the sample size to calculate a degrees of freedom correction is not obvious. Summed estimates come from many different surveys where sample sizes vary. We believe that this is a reasonable assumption given that sample sizes across surveys are all large (Supplementary Methods Table 1), at which point the t-distribution approximates a normal distribution. If we instead apply a very conservative approach that uses the smallest sample size across surveys for a degrees of freedom correction, test statistics and p-values are almost all identical to those reported here (to three decimal places). Within each table or figure, we apply a Benjamini–Hochberg adjustment to account for multiple hypothesis testing⁶⁴.

While we conduct hypothesis testing in absolute dollar units, we characterize between-group differences primarily in relative terms or percent units. To calculate relative differences, we divide the mean difference in dollar units by the total mean dollar value for the reference group and multiply by 100. In some analyses, we also estimate 95% confidence intervals around the relative differences by dividing the standard error of the difference, in dollar units, by the dollar value for the reference group, multiplying by 100, and estimating upper and lower bounds using a test statistic of 1.96 (assuming a normal distribution). The logic here is that, if we assume the mean for the reference group is given (no sampling error), then we only need to scale the standard error of the absolute difference by this constant because the constant contributes nothing to sampling variance—much like if we scaled the standard error of a proportion to a percentage, we would simply multiply by 100. These estimates underestimate the width of the true interval. However, underestimation has little practical significance because the referent mean is precisely estimated in our data in most instances. We also present estimates of between-group differences in standard deviation units, which we interpret as Cohen’s d, by dividing the mean dollar difference between groups by the standard deviation of the group of interest.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.