Intergenerational persistence of poverty in five high-income countries

Abstract

Childhood poverty increases the likelihood of adult poverty. However, past research offers conflicting accounts of cross-national variation in the strength of—and mechanisms underpinning—the intergenerational persistence of poverty. Here the authors investigate differences in intergenerational poverty in the United States, Australia, Denmark, Germany and the United Kingdom using administrative- and survey-based panel datasets. Intergenerational poverty is decomposed into family background effects, mediation effects, tax and transfer insurance effects and a residual poverty penalty. The intergenerational persistence of poverty is 0.43 in the United States (95% confidence interval (CI) = 0.40–0.46; P < 0.001), compared with 0.16 in the United Kingdom (95% CI = 0.07–0.25; P < 0.001) and 0.08 in Denmark (95% CI = 0.08–0.08; P < 0.001). The US disadvantage is not channelled through family background, mediators, neighbourhood effects or racial or ethnic discrimination. Instead, the United States has comparatively weak tax and transfer insurance effects and a more severe residual poverty penalty. If the United States were to adopt the tax and transfer insurance effects of its peer countries, its intergenerational poverty persistence could decrease by more than one-third.

Main

Poor children are more likely to become poor adults, but less so in some countries compared with others. Previous research offers conflicting accounts of the factors that promote stronger or weaker intergenerational persistence of poverty (IGPov), as well as the mechanisms through which it is channelled. Moreover, IGPov research has generally focused on a single institutional context, often excludes women and the lowest-income-receiving men from analyses, uses incomplete income measures and/or lacks the data needed to observe the social processes through which disadvantage persists. This study addresses these concerns using comparative panel datasets derived from administrative and survey records covering the United States, Australia, Denmark, Germany and the United Kingdom. We investigate cross-national differences in IGPov and explain the economic, social and institutional factors that drive these differences.

Our study of IGPov differs from conventional studies of economic mobility in several ways¹. Rather than studying upward and downward mobility across the entire set of parent and child income distributions, we focus on mobility out of poverty, defined as the share of years during childhood with a post-tax and -transfer household income below the poverty threshold. As is well documented, poverty during childhood is associated with poorer health conditions, weaker learning outcomes and less favourable later-life economic outcomes^2,3. Poverty at any age is associated with a higher likelihood of food insufficiency and other forms of material hardship⁴. Understanding the processes that facilitate mobility out of childhood poverty is thus of considerable importance.

However, research on IGPov is scarce relative to the broader literature on intergenerational income mobility. Whereas widely studied intergenerational income elasticities and rank–rank slopes capture economic mobility across the full parent and child income distributions, a focus on IGPov captures mobility from a state of deprivation. Furthermore, rank–rank slopes target persistence in relative positions in income distribution rather than households’ consumption capabilities, leading to different magnitudes, mechanisms and conceptual meanings compared with IGPov. Local versions of these measures, such as point elasticities and transition matrices, are likewise inadequate for the study of persistence of poverty, as they fail, respectively, to compare individuals potentially far away in the income distribution and, again, to measure material wellbeing⁵ (see section N of the Supplementary Information).

The few studies that do explicitly measure IGPov tend to be single-country studies (often focused on the United States) and are thus of limited value if our aim is to compare across different political and institutional contexts^6,7. Prominent studies that claim to examine IGPov tend to focus on individual (rather than family or household) earnings, often restrict the analysis to fathers and sons and do not include a full set of taxes and transfers in their income definition^8,9,10,11. Moreover, in studies of earnings correlations using logged incomes, zero values (representing individuals without current earnings) are either dropped or manipulated to an artificial income (a value of 1 to ensure inclusion in a log transformation), probably leading to biased estimates of intergenerational disadvantage^12,13. Administrative estimates of intergenerational mobility are also vulnerable to this problem, particularly in the United States: tax-based records of intergenerational mobility exclude many of the poorest households, given that the low-income population is not mandated to file taxes. Further justification for a focus on IGPov is evidence of nonlinearity in intergenerational income elasticities, particularly in studies that account for censoring bias and include very-low-income-receiving adults in their analysis¹². These shortcomings have led scholars to lament the lack of research—and comparative research especially—on IGPov^14,15.

Our study addresses these concerns and documents cross-national variation in the strength of—and mechanisms underpinning—IGPov. We harmonize panel datasets across five high-income countries with different institutional features, allowing for direct comparisons of IGPov using post-tax and -transfer household income measures for each country. We produce estimates of poverty rather than income (limiting censoring bias) over at least five years of childhood (limiting attenuation bias) and over multiple years between 25 and 35 years of age (limiting life cycle bias).

Conceptually, we identify seven competing perspectives on IGPov (see the section ‘Conceptual framework for studying IGPov’) and explicitly test those that focus on how family characteristics, market outcomes and welfare state institutions explain IGPov. Moreover, we access restricted data for the United States to investigate the role of county-level mobility variation in shaping the country’s comparative rate of IGPov^16,17.

Methodologically, we present a non-causal decomposition framework with which to empirically adjudicate the competing perspectives on IGPov. The framework fully decomposes IGPov into family background effects (for example, the roles of parental education and employment), mediation effects (for example, the roles of education, employment and family structure in adulthood), tax and transfer insurance effects (for example, the role of the state in decreasing income penalties stemming from certain education, employment or family structure features associated with childhood poverty) and a residual poverty penalty. We further decompose our mediation effects into two sub-components: benchmark attainment (how childhood poverty is associated with adult benchmarks, such as higher education); and benchmark returns (the pre-tax and -transfer returns to those benchmarks). In doing so, we are able to not only provide high-quality descriptive evidence on cross-national variation in IGPov, but can also distinguish the specific mechanisms through which poverty persists across high-income countries.

As discussed in the Methods, we test our results across multiple types of relative and absolute poverty measure, but follow standard practice in the comparative poverty literature of prioritizing a relative poverty measure with country-year poverty thresholds set at 50% of the national equivalized median household income after accounting for taxes and transfers¹⁸. We present our findings first, followed by our conceptual and empirical framework.

Findings

Table 1 presents descriptive data on post-tax and -transfer poverty and unconditional estimates of IGPov across five countries. We define IGPov as the slope in a regression of the share of adulthood (25–35 years of age) spent in poverty on the share of childhood (birth through to 17 years of age) spent in poverty, as explained in the Methods.

Table 1 Poverty in childhood and adulthood by country

The second column of Table 1 shows that the value of IGPov is 0.43 in the United States, indicating that spending all of one’s childhood in poverty is associated with a 43 percentage point higher mean poverty rate during early adulthood relative to an adult with no child poverty (95% confidence interval (CI) = 0.40–0.46; P < 0.001). In contrast, the degree of poverty persistence is 0.21 in Australia (95% CI = 0.13–0.29; P < 0.001), 0.15 in Germany (95% CI = 0.02–0.27; P = 0.024), 0.08 in Denmark (95% CI = 0.08–0.08; P < 0.001) and 0.16 in the United Kingdom (95% CI = 0.07–0.25; P < 0.001). Importantly, cross-national variation in IGPov is not mechanically related to childhood poverty. As one example, the United Kingdom features notably higher child poverty rates than Australia or Germany, yet comparable poverty persistence.

The third column of Table 1 displays mean childhood poverty rates for our sample of adults. The average adult in our US sample spent 18.6% of their childhood in poverty, which is slightly higher than the UK rate (15.6%) and notably higher than the German rate (4.7%).

The fourth column of Table 1 describes the mean poverty rate in early adulthood for each country and the fifth column presents the predicted poverty rate for a young adult who experienced no childhood poverty. In the United States, the mean poverty rate in early adulthood is 17.9% (the highest among the five countries examined); however, the mean adult poverty rate for a person with no childhood poverty is 9.9%, which is comparable to the rates in the United Kingdom and Germany. The higher overall poverty rate in early adulthood in the United States is thus attributable to the higher childhood poverty rates and stronger IGPov.

We now apply our decomposition framework to explain these cross-national differences in IGPov. Table 2 presents the results from equations (3)–(5) (Methods), which allow us to measure the role of family background (F) in influencing IGPov.

Table 2 Association of childhood poverty with pre-tax and -transfer adult poverty by country

Model 1 shows the association between post-tax and -transfer childhood poverty and the pre-tax and -transfer poverty rate in adulthood, conditional on our baseline controls (maximum and minimum age observed, sex, share of adulthood living with parents, the last calendar year during which the individual was observed and the share of ages 0–5, 6–10 and 11–17 years during which the individual was observed in our sample). We ran separate regressions for each country. Moreover, we started with a pre-tax and -transfer poverty measure and later evaluate how switching to a post-tax and -transfer measure affects persistence. The United States, Australia and United Kingdom feature comparable associations of childhood poverty with pre-tax and -transfer adult poverty (0.45, 0.47 and 0.48, respectively), whereas the magnitudes are smaller across Germany and Denmark (0.25 and 0.17, respectively). For all countries, the strength of this relationship decreases when incorporating family background characteristics (see model 2), albeit at varying intensities. The final column displays values of F (that is, the difference between the coefficients of interest without and with the inclusion of family background controls). In the United States and United Kingdom, family background effects absorbed, respectively, nine and ten percentage points of the IGPov relationship. This contrasts with the small role of family background in Germany (0.02 from a base of 0.25) and its dominant role in Denmark (0.13 from a base of 0.17).

Figure 1 addresses the second component of our decomposition framework: mediators of IGPov. Analogous to the family background component, mediation is understood here as changes in the slope of poverty persistence associated with the inclusion of a given variable in the model, conditional on family background and all other mediators. Figure 1 presents the total mediation effect by benchmark, whereas section A of the Supplementary Information presents figures that decompose the total mediation effect into its benchmark attainment versus benchmark return components. In Fig. 1a–h, each bar represents the overall mediation effect of the given indicator for the given country (conditional on other mediators, as all mediators were observed concurrently). Figure 1i presents their sums, equivalent to the overall mediation effect of our benchmarks in shaping IGPov (which we refer to as M). In all countries, employment stands out: individual employment (Fig. 1c), full- rather than part-time employment (Fig. 1d) and having others in the household who are employed (Fig. 1e) tend to be the strongest mediators. Single parenthood (Fig. 1f) follows in magnitude for several countries, with marriage or partnership, partner’s education and own education trailing behind.

Fig. 1: Combined mediation effects by country.

a–h, Total effects of various benchmarks (including attainment of a high school degree (a), qualifications beyond a high school degree (b), employment (c), full-time work (d), employment of others in the household (e), single parenthood (f), marriage or partnership (g) and having a partner who has qualifications beyond a high school degree (h)) in mediating the relationship between childhood and adult poverty, according to Gelbach’s decomposition⁶⁹. See equation (7) for the full model specification. All mediators were observed concurrently. i, Sum of a–h, equivalent to the overall mediation effect of the aforementioned benchmarks. The estimates in a–h do not sum up perfectly to i because the combined mediation also includes missing dummies for each variable; the differences are negligible. Exact values and confidence intervals are provided in Supplementary Table 2 in section C of the Supplementary Information. The sample includes respondents who were observed during childhood (0–17 years of age) and adulthood (25–35 years of age). Exact sample sizes were 9,561 (United States (US)), 962 (United Kingdom (UK)), 1,563 (Australia (AU)), 1,708 (Germany (DE)) and 1,801,813 (Denmark (DK)). The data are presented as means ± 95% CIs.

The United Kingdom, Australia and United States feature the largest absolute mediation effects. Recall that these mediators are conditional on family background characteristics (which pre-date the mediators) and that family background has a greater influence in Denmark, contributing to the lower overall mediation effect observed in Denmark.

Figure 2 presents the full results of our decomposition framework. Unlike in Table 1, these models control for the final age of an individual’s observation, so the rate of persistence may vary slightly from the results in Table 1. The results confirm that the magnitude and sources of IGPov differ considerably by country.

Fig. 2: Decomposition of IGPov across countries.

The black bars with numerical labels show IGPov (see equation (1)). The remaining bars show the contributions of four sub-components to IGPov (that is, family background (F), mediators (M), taxes and transfers (T) and the residual poverty penalty (R); see equations (2)–(10)). T captures how taxes and transfers received in adulthood reduce the part of the poverty persistence slope associated with family background and mediators (see the section ‘Empirical strategy’ and section B of the Supplementary Information). R is the residual coefficient in the post-tax and -transfer specification with all controls included. The data are presented as means ± 95% CIs. The confidence intervals for F and M were calculated using the method described by Gelbach⁶⁹, which we adapted to our sequential framework. The confidence intervals for T were estimated by bootstrapping with 1,000 repetitions. Exact values and confidence intervals are provided in Supplementary Table 3 in section C of the Supplementary Information. This sample included respondents who were observed during childhood (0–17 years of age) and adulthood (25–35 years of age). Exact sample sizes were 9,561 (United States), 962 (United Kingdom), 1,563 (Australia), 1,708 (Germany) and 1,801,813 (Denmark).

Germany and Denmark feature the lowest magnitudes of poverty persistence (0.10 and 0.07, respectively); however, the mechanisms through which poverty persists vary across the two countries. In Germany, mediators account for nearly all of IGPov, with a small role for family background effects. Taxes and transfers play a small insurance role (three percentage points), whereas a very small residual poverty penalty persists (0.04 percentage points). In Denmark, family background effects account for most of IGPov, whereas mediators play a minor role. Taxes and transfers received in adulthood reduce poverty persistence in Denmark by nearly ten percentage points, offsetting most of the family background effects. As in Germany, a small residual persists (1.7 percentage points).

In Australia, family background effects and mediators contribute roughly evenly—around 16–17 percentage points each—to the country’s IGPov of 0.20. However, the tax and transfer insurance effect pushes the magnitude of poverty persistence downward by around 12 percentage points, nearly offsetting the contribution of the mediating factors. The residual poverty penalty is again relatively small in Australia, suggesting that the vast majority of poverty persistence is accounted for by our observed indicators.

The most similar country to Australia is the United Kingdom, where a poverty persistence of 0.15 is primarily channelled through mediators, followed by family background effects, but strongly decreased through tax and transfer insurance effects. In fact, without the downward effect of tax and transfer insurance, both Australia and the United Kingdom would have poverty persistence magnitudes above 0.30, nearing the US rate.

The United States features the highest magnitude of poverty persistence at 0.42. In contrast with Australia and the United Kingdom, however, the residual poverty penalty accounts for the largest share of poverty persistence in the United States (more than 0.22; over half of the overall persistence). Mediators and family background effects account for around 13 and nine percentage points, respectively. The other notable difference in the United States is its lack of tax and transfer insurance, which diminishes poverty persistence by two percentage points—less than one-fifth of the absolute rate observed in Australia, the United Kingdom or Denmark (and even smaller in relative terms when compared with the baseline levels of IGPov in each country). Put simply, the United States has a higher IGPov relative to the other countries examined, due to its comparatively weak tax and transfer insurance effects and strong residual poverty penalty.

Further analyses

We conducted several further analyses to add insight on the United States’ outlier status with respect to IGPov. First, we assessed the extent to which racial discrimination and place (that is, county of residence during childhood) affect the United States’ comparatively high poverty persistence. Given that Black individuals are exposed to much higher levels of child poverty relative to White individuals in the United States¹⁹, it was hypothesized that a stronger IGPov among Black individuals could be driving the US findings. In the left panel of Fig. 3, we examine whether IGPov is similar for White and Black individuals in our sample. The results show substantively similar magnitudes and mechanisms of IGPov for Black and White individuals. Although racial discrimination contributes to poverty and economic opportunity¹⁹, we do not find evidence that it explains the US performance relative to other high-income countries.

Fig. 3: Decomposition of IGPov by race and mobility tercile of childhood county of residence in the United States.

IGPov (black bars with numerical labels) and the contributions of four sub-components to IGPov (that is, F, M, T and R; see equations (2)–(10)) for Black versus White US residents (left) and high-, medium- and low-mobility counties (right; based on economic mobility estimates from Opportunity Insights). The confidence intervals for F and M were calculated using the method described by Gelbach⁶⁹, which we adapted to our sequential framework. The confidence intervals for T were estimated by bootstrapping with 1,000 repetitions. Exact values and confidence intervals are provided in Supplementary Table 4 in section C of the Supplementary Information. This sample included respondents who were observed during childhood (0–17 years of age) and adulthood (25–35 years of age). Exact sample sizes were 4,173 (Black), 4,795 (White) and 3,187 for the county-level analyses. The data are presented as means ± 95% CIs.

The literature on neighbourhood effects has consistently shown that where a child grows up in the United States has a strong influence on their later-life economic outcomes^20,21. To test whether spatial differences in mobility can help to explain the outlier status of the United States, we use restricted-access data from the Panel Study of Income Dynamics (PSID) to evaluate whether counties with higher levels of economic mobility, according to Opportunity Insights¹⁶, have notably different levels of poverty persistence. Specifically, we incorporated the Opportunity Insights estimates of mean adult income rank per county for children born to parents with incomes at the 25th percentile of income distribution. To preserve power, we classified all counties as having low, medium or high mobility based on their respective tercile within the distribution of all observed counties. Figure 3 shows that IGPov is lower in high-mobility counties but the point estimate of 0.33 is still higher than for all other countries examined (95% CI = 0.30–0.36; P < 0.001). We thus conclude that other factors—including tax and transfer insurance effects—account for more of the US performance.

Second, we evaluated whether country-specific mediators (potential observable mediators that exist in our US data but are not comparably available across all countries) help to explain the high residual poverty penalty for the United States. In section G of the Supplementary Information, we provide additional information on university degree, union membership, household wealth decile, self-reported health, asthma and/or high blood pressure, past incarceration and home ownership. Inclusion of these mediators only lowers the US residual from 0.22 to 0.20 percentage points.

Third, we corroborated our US findings with an alternative measure of adult deprivation: food insecurity. Specifically, we measured the share of young adulthood with low or very low food security, which is defined by the US Department of Agriculture as “reports of reduced quality, variety, or desirability of diet” and/or “reports of multiple indications of disrupted eating patterns and reduced food intake”. If we had found effects of childhood poverty on adult poverty but not adult food insecurity, our results would have reflected broader measurement error in our US income data. Instead, childhood poverty was strongly linked to adult food hardship, even when controlling for the extended mediators described above (see section G of the Supplementary Information).

In the Methods, we present many sensitivity analyses to corroborate our primary findings and highlight the usefulness of studying IGPov beyond intergenerational income mobility.

Discussion

Given the large personal and societal costs of poverty, countries have invested considerable state capacity in reducing it, whereas researchers have long debated poverty’s primary causes^22,23,24. Nevertheless, there are several shortcomings of previous research on the topic in revealing the mechanisms through which upward mobility out of poverty is achieved.

Moving beyond previous studies of IGPov, this study (1) harmonizes panel datasets across five high-income countries with different institutional features, allowing for direct comparisons of IGPov measured with post-tax and -transfer household income; (2) conceptually distinguishes competing perspectives on the sources of IGPov; (3) introduces a decomposition framework with which to empirically dissect IGPov into the sum of family background effects, adult mediators, tax and transfer insurance effects and a residual poverty penalty; and (4) addresses the lack of cross-country evidence on nonlinearities of income-based measures of mobility. These advancements help to answer our two key questions: how do countries vary with respect to IGPov and what can explain the cross-national variation that we find?

In answering the first, we found that the United States has a much stronger IGPov than the four other high-income countries examined. Spending all of one’s childhood in poverty in the United States is associated with a 42 percentage point increase in the mean poverty rate during early adulthood. This is more than four times stronger than in Denmark and more than twice as strong as in Australia or the United Kingdom. These findings hold when equalizing the calendar years in adulthood during which respondents were observed, averaging outcomes over 30–35 years of age instead of 25–35 years of age and accounting for potential variation in attrition bias. Moreover, cross-national variation in life cycle bias should not affect our conclusions (see section D of the Supplementary Information). Our evidence shows that cross-national variation in IGPov is not systematically related to levels of child poverty; thus, the study of why poverty persists from childhood into adulthood is not analogous to the study of why certain levels of poverty exist. Moreover, we argue conceptually (section N of the Supplementary Information) and demonstrate empirically (section I of the Supplementary Information) that the study of IGPov is distinct from broader analyses of intergenerational income mobility.

In answering our second question, we found that countries vary meaningfully in the mechanisms through which IGPov is channelled, with important lessons that could inform policies to decrease poverty persistence. Seven perspectives on IGPov are outlined in the Methods section ‘Conceptual framework for studying IGPov’ and discussed in depth in light of previous literature. These relate to: (1) family resources; (2) family background characteristics; (3) place; (4) mediation through benchmark attainment; (5) mediation through benchmark returns; (6) tax and transfer insurance effects; and (7) the residual poverty penalty (see Fig. 4).

Fig. 4: Interrelatedness between the seven perspectives on IGPov and the factors that influence it.

Schematic showing how seven perspectives on IGPov relate to the influences of several key factors: poverty in childhood; poverty in adulthood; education, employment and family structure; parental education, employment and presence; and neighbourhood. These perspectives, discussed in detail in the Methods, relate to: (1) family resources and investment; (2) family background; (3) place effects; (4) mediation through benchmark attainment; (5) mediation through benchmark returns; (6) tax and transfer insurance effects; and (7) a residual poverty penalty.

We found evidence consistent with past research that family resources during childhood can have consequences for later-life opportunities². As noted, this is particularly true in the United States, where the consequences of poverty during childhood are particularly severe for the likelihood of poverty in adulthood. In all countries, family background characteristics—such as parental education, employment and family structure—carry some weight in explaining poverty persistence, but with large variation across contexts. In Denmark, family background characteristics account for most of the positive relationship between childhood and adult poverty. This is consistent with our theoretical expectation: in a context in which welfare state and labour market institutions more forcefully equalize economic opportunity—through the provision of affordable childcare, free access to higher education, compressed earnings distributions, universal healthcare and more—variations in family background characteristics are likely to carry more weight, particularly relative to adult mediators, in explaining variation in later-life outcomes. In contrast, family background effects account for a smaller share of IGPov in countries such as the United States and United Kingdom.

Although we could not incorporate place effects consistently across all countries in this analysis, we used restricted-access PSID data to investigate whether spatial differences in economic mobility across the United States help to explain the country’s outlier status with respect to IGPov. Although we found evidence of variations in IGPov across US counties, consistent with previous work on US neighbourhood effects^17,25, we did not find evidence that the perspective related to place carries more weight than the other six perspectives in explaining the comparatively high rate of poverty persistence in the United States. Even in the top one-third of the most economically mobile counties in the United States, IGPov is still 0.33, which is higher than in our other high-income countries. This finding does not negate the importance of spatial segregation within the United States; however, it does emphasize that should the United States want to match peer nations with respect to IGPov, it must seek solutions beyond the equalization of opportunity across neighbourhoods.

The two perspectives relating to mediation—benchmark attainment and benchmark returns (see section A of the Supplementary Information)—also vary in importance by country, but with some commonalities: the conditional mediating effect of educational attainment is relatively small. Although education plays a large role in intergenerational income mobility more broadly¹, its effects on IGPov are smaller, whereas employment tends to carry more relative weight.

More so than mediators, however, variation in the tax and transfer insurance effect separates the United States from its peer countries. As elaborated on previously, taxes and transfers are often ignored in studies of economic mobility; when they are studied, they are often evaluated with respect to their role in boosting family income during childhood. This study, in contrast, explicitly measured the role of taxes and transfers in insuring against risks in adulthood. Even if adults from disadvantaged backgrounds do not meet certain benchmarks associated with economic success, the state can still intervene to limit their poverty risks in adulthood. In the United Kingdom, the tax and transfer insurance effect decreases IGPov by around 16 percentage points, nearly counteracting the positive effect of the country’s mediating factors. In Denmark and Australia, tax and transfer insurance effects also contribute to a decreased IGPov. The comparatively weak welfare state of the United States, however, does relatively little to reduce poverty persistence. If the United States instead had the tax and transfer insurance effect of the United Kingdom, its overall IGPov might fall by more than 33% of its observed value (from 0.42 to 0.28, keeping everything else constant, although this should not be taken as a causal estimate).

There is, of course, an argument that generous tax and transfer effects may generate moral hazards, in which state aid reduces the incentive to acquire more education or employment²⁶. Relatedly, we acknowledge that stronger tax and transfer systems may affect pre-tax and -transfer outcomes, so that our estimates of T are clearly non-causal. More broadly, governments may prefer that adults achieve economic self-sufficiency rather than receiving state income transfers. Our evidence does not contradict these claims and cannot generally speak them. What matters with respect to poverty, however, is the total level of resources that households command in order to consume basic necessities and participate fully in society; the inclusion of taxes and transfers and the recognition of their ability to lessen risks of poverty in adulthood are thus essential components of the study of IGPov. Future research should investigate how changes to tax and transfer systems thus affect IGPov. The American welfare state, for example, has grown in size in recent decades, albeit with a shift in policy logic from means-tested cash support for low-income families to work-conditional income supplements^27,28. Moreover, two of the highest-cost programmes in the American welfare state—Social Security benefits and Medicare spending—are not intended to mitigate IGPov given that they target an older-age population.

Finally, our residual poverty penalty measures the remaining variation in the child to adult poverty relationship that is not captured in our observable characteristics or in the tax and transfer insurance effect. In Australia, Denmark, Germany and the United Kingdom, our residual poverty penalty is relatively small. In the United States, in contrast, the residual penalty contributes more strongly than other components to explaining IGPov. We demonstrate in section G of the Supplementary Information that the inclusion of wealth, home ownership, union membership, health and past incarceration in the United States only marginally reduces the residual; moreover, it persists when using a direct measure of adult disadvantage (food insecurity). The large residual poverty penalty suggests that childhood poverty is particularly severe in the United States and—more so than in other countries—operates through a set of mechanisms that we do not observe in the present study. As one example, less access to quality healthcare among low-income-receiving US residents may strengthen the penalty associated with growing up in poverty, particularly if poverty is operating through unobserved health outcomes to influence poverty in adulthood. Future research should further investigate the sources of the US residual, perhaps in a single-country study that uses other data sources (the National Longitudinal Survey of Youth for example) with more information on health outcomes, parenting practices, cultural and cognitive traits and financial literacy and/or access to financial institutions^29,30,31.

Our findings offer several broader contributions to the poverty, stratification and mobility literature. In directly testing competing theories of the roles of family, market and state in shaping IGPov, we are able to provide a more comprehensive account of the sources of IGPov. Recall that many studies in this field have instead examined pre-tax and -transfer individual earnings elasticities, often only among fathers and sons, effectively eliminating considerations about the family and state^9,10,11. This study, in contrast, has developed a decomposition framework to account for the weight of each of these factors and emphasizes the necessity of bringing the state into the study of IGPov.

Within the broader economic mobility literature, our study demonstrates, both conceptually and empirically, the need for further research on IGPov. Arguably, the study of upward mobility from poverty carries greater welfare consequences than general mobility across the broader income distribution. Indeed, if one is interested in the intergenerational transmission of welfare, under decreasing marginal returns to the utility of money one should place a higher weight on transmission at the bottom of the distribution. Regardless, our findings emphasize that the mechanisms facilitating mobility from poverty are not necessarily the same as those reducing parent–offspring rank correlations (see section I of the Supplementary Information).

We acknowledge several limitations and opportunities for future research. Due to our cross-national focus, we depend on indicators that can be used in a comparable manner across our countries. As such, some relevant indicators, such as school performance and cognitive- or non-cognitive skills, are not included in our models. Should these traits be differentially inherited based on parental income, and should they affect later-life economic outcomes, they may contribute to the decomposition’s residual component.

A related limitation is the lack of consistent information across our five countries to account for first- or second-generation immigrant status. First-generation immigrants who did not spend at least 5 years of childhood in their host country were excluded from the analysis due to our restriction of measuring childhood poverty over at least 5 years of data. Second-generation immigrants tend to experience levels of intergenerational mobility that match or exceed those of children with native-born parents³²; if the same were true for IGPov, and if our samples under-represent the share of second-generation immigrants in a country, the level of IGPov for the country may be overstated. Given the small share of second-generation immigrants in our five countries relative to the population at large, it is unlikely that any cross-national variation in sampling frame bias related to immigration status will affect our observed cross-national differences in IGPov. Nonetheless, we acknowledge that the study of IGPov by immigration status deserves more attention in future research.

A separate potential limitation is our measurement of taxes and transfers when calculating childhood poverty. We measure post-tax and -transfer poverty in childhood to capture a more complete set of parental resources and consumption capabilities (relative to pre-tax and -transfer income). This is different from studying how policy-driven increases in transfer income during childhood affect IGPov (which would generally require a single-country study applying an identification strategy around a particular policy change). To the extent that there are nonlinearities in how childhood transfers affect adult poverty (for example, an equivalent increase in transfers may have stronger IGPov effects depending on levels of childhood disadvantage, the age of the child when the transfer is made or how parents tend to use such resources)³³ and that these nonlinearities vary across countries, including taxes and transfers in our childhood poverty measure may bias cross-national comparisons of the association of post-tax and -transfer childhood poverty and adult poverty.

Finally, although we emphasize cross-national comparisons, future work could apply our framework to study (as one example) within-country changes in intergenerational mobility over time or across regions within a given country. Moreover, future work should continue to expand the set of countries for which comparable estimates of IGPov can be estimated. To facilitate this, the Supplementary Information includes a variable construction codebook and full replication code with which to reproduce and extend our findings. As such, this study’s data, conceptual, methodological and empirical contributions offer a foundation for continued research on the social processes that generate upward mobility from childhood poverty.

Methods

This research complies with all of the relevant ethical regulations. The study protocol was approved by the Bocconi University Research Ethics Committee (FA0006190). Below, we elaborate on our data, conceptual framework and empirical strategy.

Data sources

Our primary data sources are panel data files for the United States, United Kingdom, Australia, Germany and Denmark (Table 3). For the United States, United Kingdom, Australia and Germany, we used harmonized data files from the Cross-National Equivalent File (CNEF) database, which we then further harmonized to allow for comparable estimates of IGPov. We supplemented our US dataset with a restricted-access version of the PSID providing geographic identifiers on where our residents had lived. We also supplemented the UK files with variables extracted from the British Household Panel Survey and the UK Household Longitudinal Survey.

Table 3 Overview of the sample and data sources

For Denmark, we used administrative (registry) data harmonized to match the input variables of the CNEF files. Comparing estimates from Danish register data with survey-based estimates from other countries has precedent in previous research (most notably, see Landersø and Heckman²⁶); we address potential challenges of comparing across these data sources, such as differential rates of selective attrition, in the Supplementary Information.

For all countries, we generated an added list of variables that were not present in the CNEF-provided files. We detail these in the codebook in the Supplementary Information. Some indicators required proxying to achieve harmonization across all countries. For example, we did not consistently observe who was the mother of an individual in our base samples, but we were able to observe all of the relationships within the household with respect to the household head. Thus, we considered the mother a female household head or partner of the head. Nonetheless, most variables of interest were readily comparable.

The harmonized files allow us to measure concepts—such as post-tax and -transfer income, employment rates and demographic indicators—similarly across countries throughout the lives of each of our respondents. Our measure of post-tax and -transfer income includes near-cash transfers and refundable tax credits, such as Supplemental Nutrition Assistance Program and Earned Income Tax Credit benefits provided in the United States. In each of the countries, we can observe a subset of respondents’ income throughout their childhood (between birth and 17 years of age) and during their early adulthood (between 25 and 35 years of age).

We measured each adult’s poverty rate as the mean poverty rate between 25 and 35 years of age. For adults not observed through to 35 years of age, we took the mean poverty rate between 25 years of age and the final age of observation below 35 years of age. Observing poverty over several possible years, rather than a single year (for example, only at 25 years of age), decreases the risk of single-year measurement error (transitory fluctuation bias), income volatility³⁴ or life cycle bias influencing our results^1,35. Given that poverty risks are likely to decrease between 25 and 35 years of age, we controlled for the final age of observation in all models. In our final sample, the mean age of the final observation for each adult ranged from 28 years in Australia to 33 years in Denmark.

We measured childhood poverty as the mean poverty rate between birth and 17 years of age and restricted our sample to adults who were observed for at least 5 years during their childhood to ensure that we could reliably estimate their poverty status during childhood.

For all poverty measurements, we followed the standard practice of applying a relative poverty measure in which poverty thresholds are set at 50% of the national post-tax and -transfer equivalized median household income for the given country and year. We applied a square-root equivalent scale. We calculated the thresholds using nationally representative country-year data before constructing our sample. In the case of Germany, the poverty threshold was calculated separately for East and West Germany until 2014, after which CNEF data do not provide information on the individual’s region of residence, so we used a whole-Germany threshold. We explore the different patterns of poverty persistence across regions in Germany in Supplementary Appendix F. Given that the poverty threshold is year and country specific, our analytical focus is on mobility out of relative poverty rather than absolute poverty. This means, for instance, that a case in which incomes uniformly increase in percentage terms would yield minor effects on exits out of poverty. In section E of the Supplementary Information, however, we present alternative results using an absolute poverty threshold, adjusted for purchasing power parities and inflation, across all country-years. We also present results using a variation of the relative poverty threshold set to 60% of the median for each country-year.

Our final sample therefore comprises the set of individuals for whom we observed income for at least 5 years in childhood and at least once after 25 years of age. We collapsed our panel data at the individual level, obtaining adult outcomes (means and maximums over 25–35 years of age) and childhood characteristics (means from birth to 17 years of age). For the other covariates, we set missing values to zero and included missing dummies corresponding to each to preserve the power. We provide further details on variable construction in the Supplementary Information and summary statistics in section C of the Supplementary Information. We used Stata version 17 for all of the analyses.

Conceptual framework for studying IGPov

How should we understand the association between childhood poverty and adulthood poverty and how the strength of this association varies by place? Below, we elaborate on the seven dominant and often competing perspectives on IGPov (see Fig. 4). We then describe our decomposition framework, which empirically adjudicates (in a descriptive, non-causal manner) the relative strength of each perspective in shaping cross-national variation in IGPov.

Family resources and investment

The first perspective on IGPov is that family resources have direct consequences on child development and, in turn, the likelihood of poverty in adulthood.

In the economics literature, the Becker–Tomes model (1979) argues that parental resources affect parents’ investments in their children’s human capital, or their education and health in particular. Families with higher incomes can better balance demands for current consumption (for example, paying rent and feeding the family) with investments that generate longer-run rewards for children (for example, tutoring and schooling). Solon³⁵ suggests that parental investments are particularly consequential where public investments into children are lower, and when returns to education are higher, both of which tend to be true in the United States compared with other high-income countries³⁶.

The Family Investment Model of child development similarly acknowledges that more resources also allow for more time with children, including time spent reading books and other enrichment activities associated with more favourable development outcomes^35,37,38. A lack of resources, meanwhile, can generate psychological distress and harsher parenting, further inhibiting child development (for example, the Family Stress Model)³⁹ and can also result in bio-social developmental hindrances⁴⁰. In line with these models, previous research has established that public income transfers can contribute to improvements in children’s health, wellbeing, test scores, high school completion rates and college attendance rates^41,42,43,44, whereas policy-driven reductions in income lead to more unstable living conditions for children in low-income households⁴⁵.

In Fig. 4, the resources perspective is represented by an encircled number 1 alongside our measure of childhood poverty. We used a post-tax and -transfer measure of childhood poverty that captures a comprehensive set of available household resources that can be used for investments into the wellbeing of children and families.

Family background

A second perspective argues that it is not merely family income that matters in shaping opportunity, but also other characteristics of the family that may be associated with their income, such as the presence of both parents or the education and employment of the parent(s)⁴⁶. This correlated disadvantages argument³⁰ suggests that even if two families have equivalent incomes, children in the family with comparatively lower-educated parents, as one example, may nonetheless experience poorer later-life economic outcomes⁴⁷.

The family background perspective is central to What Money Can’t Buy: Family Income and Children’s Life Chances, in which Susan Mayer⁴⁸ argues that parental preferences and traits, more so than income, affect children’s later-life opportunities. Mayer argues that investments into children’s human capital, such as the purchase of books, are relatively affordable, and influxes of cash payments to parents with children often lead to consumption on leisure goods and services rather than investments into children. Family traits thus matter as much as, if not more than, family income, according to the argument.

There are several other potential pathways through which non-monetary characteristics of the parents and family can affect their children’s later-life economic opportunity. Evidence on role model effects suggests that children frequently take cues from their parents’ behaviours that carry over into adulthood, including whether to pursue higher education or employment^30,46. Analyses of intergenerational welfare receipt, for example, often point to culture as a mediating device through which a child may view receipt of welfare as a socially legitimate income source in adulthood³⁰. Moreover, parenting styles, more so than material wellbeing, may have a stronger influence on children’s cognitive skills^9,49, and family instability has been linked directly to downward mobility⁵⁰. Regardless of the specific pathways through which family background matters (which are beyond the scope of the present study), past research has convincingly demonstrated that it is likely to matter for children’s later-life outcomes.

In Fig. 4, the family background perspective is represented by an encircled number 2 alongside our measures of parental education and employment and the presence of both parents (as a non-exhaustive list of family background characteristics). Empirically, the strength of family background in shaping IGPov can be descriptively estimated based on the weakened association of childhood poverty and adult poverty when accounting for family background characteristics.

Place effects

A set of related arguments moves beyond income and family background to instead focus on the characteristics of the geographic area where the child is raised. Many studies have demonstrated that neighbourhood effects carry implications for economic and social wellbeing that cannot be reduced to family resources or family background²⁰. In the United States, in particular, several studies have documented the persistent negative consequences of living in areas where disadvantage is more concentrated^25,51.

Perhaps most prominently, Wilson⁵² has demonstrated how labour market changes that induce decreasing demand for industrial workers can lead to a cascading set of challenges—from decreasing employment opportunities to decreasing revenues for maintaining public services—that generate a persistent, geographic concentration of poverty. More recently, Sharkey¹⁷ documented that younger Black adults are ten times more likely to live in poor neighbourhoods than younger White adults and that growing up in one of these high-poverty neighbourhoods affects children’s cognitive development. How much a neighbourhood affects children’s economic opportunity is probably conditional on the neighbourhood’s characteristics and how persistently exposed a child is to the neighbourhood²⁰.

Outside of the United States, evidence of negative place effects is less consistent. A study from Toronto, for example, found no strong effects of low-income children’s neighbourhoods on future earnings⁵³. A review by Musterd⁵⁴ of neighbourhood effects in Europe, where welfare state and labour market institutions tend to be stronger, similarly concluded that “[N]eighborhood effects in Europe seem to be milder than in the USA”. Although US-specific evidence of place effects is strong, it nonetheless remains unclear how spatial variation in IGPov across the United States compares with the overall US performance relative to other high-income countries. Such an analysis has implications for evaluating how neighbourhood effects rank among these competing perspectives in shaping IGPov in the United States.

In Fig. 4, the place effects are represented by an encircled number 3 alongside our measure of the neighbourhood in which a child is raised. The only country for which we accessed place effects is the United States (using restricted-access PSID data). We assessed whether variation in IGPov across low- and high-mobility neighbourhoods across the United States affects the United States’ overall performance relative to other high-income countries.

Mediation through benchmark attainment

The next two perspectives focus broadly on mediation effects, but from two different perspectives: differential attainment of benchmarks associated with economic success and differential returns to given benchmarks. Here we focus on the fourth perspective: benchmark attainment, or the association of childhood poverty with achieving adult benchmarks.

Specifically, benchmarks refer to measurable milestones in adulthood that are often associated with a decreased likelihood of poverty, such as completing high school, finding full-time employment and entering into a stable family arrangement⁷. More poverty during childhood often leads to a lower likelihood of meeting these benchmarks. As such, benchmark attainment is probably lower for adults with more disadvantaged childhoods, although the strength of this relationship probably varies by place.

Although the accessibility of a university degree is perhaps the most commonly studied mediation effect in the intergenerational mobility literature^55,56,57, the completion of secondary school (a high school degree or equivalent) may be more directly relevant for avoiding poverty. Brady et al.⁵⁸ identified the lack of a high school degree as a key risk factor of poverty. Across most of the 29 rich countries they studied, the authors found an increased likelihood of poverty associated with not having a high school degree, even after accounting for employment, family structure and other potential risk factors. Notably, Brady et al.⁵⁸ found that the penalty for lacking this level of education was stronger in the United States than in the other 28 countries that they evaluated. Relatedly, policy recommendations for improving upward economic mobility often focus on access to education^1,59. Beyond education, studies have linked childhood poverty to a lower likelihood of employment in adulthood and a higher likelihood of having a child before marriage, poorer health and more³. In Fig. 4, these benchmark attainment effects are represented by an encircled number 4, connecting childhood poverty to adult benchmarks.

Mediation through benchmark returns

It is not merely differential attainment of benchmarks that matter, but also the returns to these benchmarks (for example, the earnings advantages associated with completing a high school degree). Consider that if attainment of a high school degree were unequal but completing high school had no relative economic benefit compared with that of not completing high school the differential attainment of education would not have any weight in explaining IGPov. Put differently, benchmark attainment and returns interact to shape overall mediation effects. Although the previous perspective discussed attainment, this perspective focuses on differential returns (based on pre-tax and -transfer earnings) associated with a given benchmark.

Education is central to this perspective. Where returns to education are higher, theory suggests that market earnings differentials and intergenerational earnings elasticities should be larger³⁵. To exemplify, the United States has particularly high university wage premiums, as well as comparatively strong intergenerational earnings elasticities^1,8,60. From an IGPov perspective, the negative association of greater educational attainment with poverty in adulthood is likely to be conditional on prevailing labour market institutions. In countries where organized labour is stronger and where minimum wages are higher, a high school or university degree may be of less necessity for avoiding poverty⁶¹. Still, even in such contexts, education-based differences in economic opportunity are likely to persist⁶².

Employment and family structure in adulthood also entail their own set of returns: lack of employment and working few hours are linked to lower household earnings, even conditional on the earnings received by other members of the household. Household and family structure should also help to explain differential outcomes: two-adult households are more likely than single-adult households, and single-parent households especially, to have larger pre-tax and -transfer incomes⁶³. Given rising assortative mating, returns may also be increasingly concentrated in households with two higher-educated adults^64,65, further reducing the likelihood of poverty among higher-educated individuals.

In Fig. 4, benchmark returns are represented by an encircled number 5, connecting adult benchmarks to adult poverty status.

Tax and transfer insurance effects

Our sixth perspective captures the insurance effect of taxes and transfers against the attainment of our observed mediators and family background characteristics (we provide an empirical specification of the insurance effect later).

Although earnings returns to benchmarks such as education or employment have received considerable attention in the intergenerational mobility literature, market earnings are only one component of income. As the cross-sectional literature on poverty and inequality has regularly demonstrated, taxes and transfers also matter considerably in shaping poverty outcomes. When taxes and transfers decrease the penalty associated with not meeting a given benchmark (that is, decreasing the likelihood of poverty for an unemployed individual), they effectively insure against social risks and reduce the relevance of differential earnings returns. This leads to our sixth perspective: the tax and transfer insurance effect.

Direct income transfers from the state are among the most powerful interventions in addressing poverty and inequality. Cross-nationally, country patterns of the strength of tax and transfer systems are well documented: the Nordic countries tend to reduce poverty more through income transfers, whereas the United States has generally lagged behind, even compared with the United Kingdom^36,66.

The intergenerational mobility literature has had relatively little to say about the role of taxes and transfers in shaping mobility. As noted before, a large share of the intergenerational mobility literature focuses solely on earnings and excludes government income support from analyses altogether (but see Landersø and Heckman²⁶). In studies that do incorporate this role of the state, analyses tend to focus on how income transfers boost family income during childhood, or on facilitating benchmark attainment, rather than insuring against risk in adulthood^6,42. In describing the potential role of the state in shaping mobility, for example, Torche¹ writes that “public policy could foster mobility in two ways: investing in the human capital development of disadvantaged children…and financing higher education to ameliorate the effect of credit constraints”. However, this perspective, echoed in other mobility research⁹, does not acknowledge the insurance effect of taxes and transfers in adulthood: even when children grow up in disadvantaged homes and lack access to higher education, the state can still intervene to reduce poverty in adulthood.

Our focus on income transfers as potentially decreasing IGPov also contrasts strongly with the tradition of US-centric literature identifying intergenerational social assistance receipt as a proxy of intergenerational disadvantage^30,67. Poverty concerns the level of resources that a household has to consume basic necessities and participate in society; in contexts where public transfers boost household income, poverty research should acknowledge this rather than using receipt of transfers as a proxy for disadvantage. This is particularly true for the non-US countries in our sample, where receipt of income support is near-universal among some subpopulations (for example, universal child allowance).

As described later, we can produce pre- and post-tax and -transfer poverty estimates for each of our comparative datasets, allowing us to measure the insurance effect of taxes and transfers. In Fig. 4, this effect is represented by an encircled number 6, connecting adult benchmarks to adult poverty status alongside benchmark returns.

Residual poverty penalty

Finally, childhood poverty may operate beyond the pathways identified above by influencing poverty in adulthood. In the intergenerational mobility literature, the “direct effects of social origins” refer to the residual association of childhood and adulthood socioeconomic status⁶⁸. Commonly, a residual association exists even after accounting for—as one example—access to education. This may also be the case in our study of IGPov. We label this potential residual effect as a residual poverty penalty.

A residual penalty reflects the influence of unobserved characteristics that mediate the relationship of IGPov. For example, wealth (an observable characteristic, but unobserved in our data framework due to a lack of information across countries) could play a role in shaping IGPov and will probably not be fully explained by education, employment and family structure mediators. Alternatively, non-cognitive traits, such as communication ability or personality, may be handed down from parents to children and are not fully aligned with other mediators.

Should these unobserved features universally affect IGPov, we may see similarly sized residual poverty penalties across the five countries of our study. In contrast, a large variation in the residuals would suggest that other institutional features may matter. Specifically, a stronger residual poverty penalty in one country versus another could reflect real differences in the severity of child poverty. In contexts where childhood poverty is coupled with low-quality and low-access public services, for example, the penalty of that time in poverty may persist into adulthood independent of our other observed characteristics. As one example, lack of access to affordable and quality healthcare, which is more common in the United States, might generate conditions under which poverty is more strongly associated with adverse health outcomes; to the extent that these health outcomes are difficult to measure and yet affect later-life economic opportunity, they may appear in our residual term, alongside other unobserved correlates of poverty. Publicly provided childcare services represent another service that affects our residuals, alongside variations in school quality, differential access to financial institutions and differential financial literacy, parenting norms, cultural traits and more.

In Fig. 4, the encircled number 7 represents residual poverty, connecting childhood poverty to adult poverty independent of the other mechanisms examined.

We adjudicated these seven competing perspectives to make sense of cross-national variation in IGPov. To do so, we converted the perspectives into a four-part decomposition framework that dissects IGPov into family background effects, mediation effects, the tax and transfer insurance effect and the residual poverty penalty.

Empirical strategy

We introduce an accounting framework to descriptively decompose the sources and magnitudes of IGPov into four primary components: family background (F), mediating benchmarks (M), taxes and transfers (T) and a residual term (R) that captures the persistent effect of childhood poverty on adult poverty that is not channelled through F, M or T. We further decompose our mediating benchmarks (that is, M) into two components: benchmark attainment and benchmark returns. Related to the seven perspectives presented above, our decomposition framework captures how the association between family resources and adult poverty status (perspective 1) is channelled through F (perspective 2), M (perspectives 4 and 5), T (perspective 6) and R (perspective 7). We check for place effects (perspective 3) in the United States by comparing the results of our decomposition for counties with different levels of overall economic mobility.

Equation (1) provides our baseline equation for measuring IGPov and equation (2) documents how we decompose IGPov into its four sub-components. We estimate all models separately for each country’s respective sample.

$${\rm{Pov}}_{{\rm{Post}}}={\beta }_{1}{\rm{ChPov}}+\varepsilon$$

(1)

$${\beta }_{1}={\rm{IGPov}}=F+M+T+R$$

(2)

In equation (1), β₁ captures the association of (post-tax and -transfer) child poverty (ChPov) with post-tax and -transfer poverty in adulthood (Pov_Post) for individuals in a given country. β₁ thus captures the mean association of childhood poverty with adult poverty (that is, IGPov). Recall that childhood poverty is averaged over birth through to 17 years of age, whereas adult poverty is averaged over 25–35 years of age. Equation (2) formalizes this understanding and details the four components into which we will subsequently decompose IGPov.

To isolate the effect of family background (F), we estimate equations (3)–(5) as follows:

$${\rm{Pov}}_{{\rm{Pre}}}={\rho }_{1}{\rm{ChPov}}+\varepsilon$$

(3)

$${\rm{Pov}}_{{\rm{Pre}}}={\delta }_{1}{\rm{ChPov}}+{\delta }_{2}{\rm{Fam}}+\varepsilon$$

(4)

$$F={\rho }_{1}-{\delta }_{1}$$

(5)

Equation (3) is similar to equation (1), but instead applies a pre-tax and -transfer version of adult poverty status (we will estimate the effects of taxes and transfers in a later step). The parameter ρ₁ gives us the association of childhood poverty with pre-tax and -transfer young adult poverty; in equation (4), the parameter δ₁ provides the conditional association after incorporating family background controls (Fam) into the model. We present all of the indicators included in this vector in Table 4. The difference between these two parameters gives us F (that is, the influence of family background characteristics on IGPov). As before, we run models separately for each of our five countries.

Table 4 Summary of decomposition framework and indicators of interest

To isolate the overall effect of mediating benchmarks (M), we estimate:

$${\rm{Pov}}_{{\rm{Pre}}}={\gamma }_{1}{\rm{ChPov}}+{\gamma }_{2}{\rm{Fam}}+{\gamma }_{3}{\rm{Med}}+\varepsilon$$

(6)

$$M={\delta }_{1}-{\gamma }_{1}$$

(7)

Equation (6) builds on equation (4), but adds our set of mediating benchmarks. We discuss how we operationalize the mediators in Table 4. As equation (7) describes, the influence of the mediators, M, in shaping poverty persistence is captured by the decline in the association of childhood poverty with pre-tax and -transfer adult poverty across these two equations. Recall that M itself is decomposable into two components: benchmark attainment and benchmark returns. We provide a formal definition of these two terms in section A of the Supplementary Information.

To identify the effect of the tax and transfer insurance effect—and also the residual term—we conclude our framework with:

$${\rm{Pov}}_{{\rm{Post}}}={\theta }_{1}{\rm{ChPov}}+{\theta }_{2}{\rm{Fam}}+{\theta }_{3}{\rm{Med}}+\varepsilon$$

(8)

$$T=({\beta }_{1}-{\rho }_{1})-({\theta }_{1}-{\gamma }_{1})$$

(9)

$$R={\theta }_{1}$$

(10)

In equation (8), the only difference from equation (6) is the switch to a post-tax and -transfer outcome indicator (capturing poverty in adulthood with a full income definition). Combined with information from the previous models, we can compute how taxes and transfers insure against our measures of family background disadvantage and not meeting our mediating benchmarks. Specifically, the first part of equation (9), (β₁ − ρ₁), captures the overall effect of taxes and transfers in reducing the association of childhood poverty with adult poverty. We then subtract from that the insurance effect against unobserved background and benchmarks, computed as θ₁ − γ₁. The outcome provides a direct estimate of how taxes and transfers provided to adults reduce the penalty associated with observable characteristics—such as low educational attainment or joblessness—that may also be linked to child poverty. We elaborate on the calculation of this tax and transfer insurance effect in section A of the Supplementary Information.

As equation (10) specifies, the residual poverty penalty in our decomposition framework is simply θ₁ (that is, the remaining association of childhood poverty with post-tax and -transfer adult poverty independent of F, M and T). The residual can include omitted variable bias, differential returns to omitted mediating benchmarks and/or real variation in the severity of child poverty not channelled through our observable characteristics. Given that we capture the residual after including taxes and transfers, differences in the residual could also be due to differences in the role of taxes and transfers in insuring against unobserved characteristics. The pre-tax and -transfer (uninsured) residual is represented by the value of γ₁.

Table 4 summarizes the decomposition framework, parameters of interest and specific indicators used to estimate each parameter. One can validate that the sum of equations (5), (7), (9) and (10) is equal to β₁, or IGPov. We weight all of our analyses using the mean of each adult’s weight during childhood, following Bastian and Michelmore⁴².

We acknowledge that our framework is potentially sensitive to the order in which the given components are added; however, the framework is sequenced in its logical order: family background naturally occurs before mediating factors in adulthood, and taxes and transfers only apply to a family after its mediating benchmarks (for example, whether one is currently employed or not and other factors shaping current income) are determined. A Gelbach⁶⁹ decomposition, which ignores sequentiality, confirms that the value of M generally grows relative to F if we disregard the sequence of inputs. We argue, however, that the more conceptually sound approach is to acknowledge that F occurs before M, validating the sequence of our primary decomposition analysis.

We also acknowledge that—in line with most of the intergenerational mobility literature—our methods are not designed to infer causality; instead, our framework offers useful non-causal evidence on the social processes through which childhood and adulthood poverty are associated. Finally, we acknowledge that F may moderate the returns to M; in section K of the Supplementary Information, we present an alternative specification in which we interact the main family background variables with each M but do not find that it meaningfully alters our conclusions.

Sensitivity tests, data validation and potential objections

We offer several sensitivity tests and additional analyses to corroborate our findings and clarify the usefulness of studying IGPov.

In section D of the Supplementary Information, we: (1) present evidence that cross-national variation in selective attrition is unlikely to bias our findings; (2) demonstrate that cross-national variation in life cycle bias is unlikely to meaningfully affect our findings; (3) demonstrate that variation in the years during which young adults are observed does not meaningfully affect our findings; and (4) document that measuring family resources during childhood is an appropriate measure of parental circumstances and can be reliably applied in an analysis of IGPov.

In section H of the Supplementary Information, we assess the sensitivity of our results for the United Kingdom depending on the two income concepts available in the British Household Panel Survey: current income in the previous month versus annual income in the previous 12 months. We show in section H of the Supplementary Information that the results are broadly consistent when using monthly or annual income. Moreover, our UK results do not vary notably if we limit the UK sample to the British Household Panel Survey (effectively ending the sample in 2008 and excluding the UK Household Longitudinal Study transition).

In section F of the Supplementary Information, we assess the sensitivity of our German results conditional on our inclusion of the East German sample that appeared later in the SOEP. Due to changes in the SOEP sampling procedures in the early 1990s, we must exclude children who grew up entirely in East Germany from our primary estimates for Germany. This is consistent with past use of the SOEP data in mobility estimates^70,71,72, but is unfortunate given that poverty is more concentrated in East Germany.

In section J of the Supplementary Information, we provide evidence that levels of child poverty do not mechanically affect our estimates of IGPov. In section K of the Supplementary Information, we show that omitted interactions between family background characteristics and our mediator and background measures do not bias our findings.

We also take several steps to demonstrate that our estimates of IGPov are conceptually and empirically distinct from estimates of intergenerational income mobility (using rank–rank slopes obtained by regressing offspring ranks of income distribution on their parents’)^{73,74,75,76,77,78}. First, in section I of the Supplementary Information, we show that switching to rank–rank estimates strongly increases intergenerational income associations in the United Kingdom, Germany and Australia, consistent with cross-national mobility evidence; meanwhile, the residual term increases for all countries, whereas the role of tax and transfer insurance effects decreases. These results are unsurprising: taxes and transfers rarely affect income ranks, as we elaborate on in section N of the Supplementary Information.

Second, we address two points related to the (non)linearity of the relationship between adult and childhood poverty. In section L of the Supplementary Information, we identify nonlinearities in the likelihood of adult poverty for individuals exposed to at least some childhood poverty. For this group, average poverty in adulthood is substantially higher than for others and the slope translating childhood to adulthood resources is also notably stronger. These discontinuities in levels and slopes reinforce our conceptual argument of poverty as a distinct economic state worth studying. Nonetheless, we also present evidence that, within groups (that is, among those exposed or not to childhood poverty), a linear model is an appropriate estimator of intergenerational persistence, since it captures well the association between childhood and adult disadvantage.

Third, we assess cross-national differences in IGPov with two alternative poverty thresholds: an absolute poverty threshold and an alternative relative threshold set at 60% (rather than 50%) of the national median equivalized income (see section E of the Supplementary Information). Our results are consistent across these alternative poverty definitions. The absolute poverty findings help to ensure that our results using a relative poverty measure are not mechanically driven by differences in country-level income inequality, as we elaborate on in section M of the Supplementary Information.

In short, although there are many potential challenges in comparing IGPov across countries, the evidence we present in the Supplementary Information is able to directly address several of these concerns while ruling out that others are likely to meaningfully affect our findings.

Reporting summary

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