What can explain cross-cohort trends in long-term childhood poverty?

While some studies investigate why individuals or households remain in poverty persistently (e.g., Biewen, Reference Biewen2009), there is relatively little discussion of the factors driving long-term poverty trends at the aggregate level. Assuming that the determinants of overall child poverty trends also shape long-term poverty, two broad explanations can be considered: redistribution and predistribution.

In the UK, redistribution for families with children changed significantly during our observation period through two major reform periods. The first was the post-1997 welfare reforms, which substantially expanded the level and coverage of child benefits and tax credits.Footnote ³ For example, child benefit for the first child rose by 25.3 per cent in real terms between April 1997 and April 2003. Reforms to child-related tax credits made these income top-ups more generous by increasing maximum entitlements, reducing taper rates, and introducing more generous childcare rebates; coverage was also widened by removing asset tests and work requirements for child-specific tax credits (Brewer & Hoynes, Reference Brewer and Hoynes2019).Footnote ⁴ Overall, spending on child-related benefits and tax credits rose by about 60 per cent in real terms between 1999 and 2006, an increase of roughly £11 billion (Stewart, Reference Stewart, Hills, Sefton and Stewart2009). These reforms substantially reduced child poverty between 1997 and 2010 (Joyce & Sibieta, Reference Joyce and Sibieta2013), helped counteract rising earnings inequality and contain household income inequality (Cribb et al., Reference Cribb, Joyce and Wernham2023), and likely contributed to the decline in long-term childhood poverty.

The second period began with the 2013 austerity reforms and was followed by harder measures in 2016 (Beatty and Fothergill, 2018; House of Commons Work and Pensions Committee, 2019) . These included the benefit freeze, the benefit cap, the bedroom tax, and the introduction of the two-child limit. The benefit freeze limited annual uprating of most working-age benefits to 1 per cent (2013–15) and then 0 per cent (2016–19) despite rising prices. This represented a substantial real-term cut for low-income households. The benefit cap set a maximum total amount a household could receive in benefits, regardless of assessed need, for example £26,000 per year initially, lowered to £20,000 (£23,000 in London) in 2016. It primarily affected larger families and renters in high-cost areas as the cap removed entitlement above the limit by cutting Housing Benefit or Universal Credit (UC) payments. Officially called the ‘under-occupancy penalty’, the bedroom tax reduced the Housing Benefit for social tenants deemed to have ‘spare’ bedrooms 14 percent cut for one extra room and 25 per cent for two or more. It applied only to working-age households, not pensioners. The policy aimed to encourage downsizing but often reduced incomes because many tenants had no smaller properties available. Introduced in April 2016, the two-child limit restricts the Child Tax Credit and UC so that families can only receive the child benefits for two children, with exceptions in limited cases (e.g., multiple births, non-consensual conception); so, the entitlements for the third and following children born after April 6, 2017 were permanently removed, regardless of income level. It is widely regarded as the main driver of rising child poverty, specifically among large families (Stewart et al., Reference Stewart, Patrick and Reeves2025).

The roll-out of UC during this period also reduced benefit levels. UC replaced six existing means-tested benefits and tax credits for those of working age, with one integrated benefit through the aim of simplification, effectiveness, and increasing incentives to work while potentially making the benefits more generous. It introduced a range of policy changes to taper rates, eligibility, conditionality, and earnings disregards that affected the generosity of the benefit system in complex ways, creating both winners and losers (Brewer et al., Reference Brewer, Joyce and Waters2019; Browne et al., Reference Browne, Hood, Joyce, Emmerson, Johnson and Joyce2016).Footnote ⁵ Since its initial proposal in 2013, UC’s design has changed substantially. In particular, work allowances, the amounts recipients can earn before benefits begin to be withdrawn, were significantly cut (Browne et al., Reference Browne, Hood, Joyce, Emmerson, Johnson and Joyce2016). Moreover, increasing conditionality and sanctions and expanding such conditions to working families likely lead some recipients to lose their benefits (Browne et al., Reference Browne, Hood, Joyce, Emmerson, Johnson and Joyce2016; Wright & Dwyer, Reference Wright and Dwyer2022).

Taken together, these changes are likely to have increased long-term childhood poverty for children spending most of their early years during this period. These measures disproportionately affected low-income households with children (Brewer et al., Reference Brewer, Joyce and Waters2019; De Agostini et al., Reference De Agostini, Hills and Sutherland2018). For example, Bourquin et al. (Reference Bourquin, Norris Keiller and Waters2019a) estimate that the poorest 10 per cent of households with children lost around 20 per cent of their income (about £4000) between 2010 and 2019. Moreover, Brewer et al. (Reference Brewer, Joyce and Waters2019) argue that UC hit the persistently poor harder than those with higher, longer-term incomes.

Predistribution, in the context of poverty, refers to the ability of households to avoid poverty through earnings; that is, the adequacy of earnings relative to the poverty threshold. Shifts in the labour market and family dynamics can strongly influence predistribution. First, the decline in worklessness and single-earner households during the 1990s, particularly among single-parent families (Bourquin et al., Reference Bourquin, Cribb, Waters and Xu2019b), likely reduced long-term poverty. This trend was partly driven by increased conditionality and the expansion of in-work benefits under the 1997 reforms (Avram et al., Reference Avram, Brewer and Salvatori2018; Blundell and Hoynes, Reference Blundell, Hoynes, Card, Blundell and Freeman2004). Second, over the past three decades, growing inequality in household earnings, combined with rising pensioner incomes, has pushed up median household income and thus the poverty threshold; as a result, the effectiveness of employment as a route out of poverty has declined (Cribb et al., Reference Cribb, Joyce and Wernham2023; Joyce & Ziliak, Reference Joyce and Ziliak2019).Footnote ⁶ Finally, the introduction of the National Minimum Wage in 2003, which became more prominent as a primary tool for boosting incomes in the 2010s, may have had only a limited impact on long-term childhood poverty, given that it is not well targeted toward low-income households and overall income gains are limited due to withdrawal of means-tested benefits as earnings increase (Brewer et al., Reference Brewer, May and Phillips2009; Cribb et al., Reference Cribb, Giupponi, Joyce, Lindner, Waters, Wernham and Xu2021).

It is useful to distinguish between the periods before and after 2008. The Great Recession, at least in the early aftermath until 2011, significantly reduced employment and increased earnings inequality in Britain, weakening the effectiveness of employment as a long-term route out of poverty (Belfield et al., Reference Belfield, Blundell, Cribb, Hood and Joyce2017; Brewer et al., Reference Brewer, Browne, Joyce and Sibieta2010; Joyce & Ziliak, Reference Joyce and Ziliak2019). Two factors partially offset these effects. First, benefit levels rose substantially in the aftermath of the crisis, well above inflation, resulting in real income growth in the bottom quartile of the distribution (Pacitti & Smith, Reference Pacitti and Smith2019).Footnote ⁷ Second, from 2011, the minimum wage increased sharply, likely boosting earnings for some of the households at the lower end of the distribution. However, its poverty-reducing impact was likely limited, as gains were largest in the middle of the income distribution, potentially raising the poverty threshold, and were offset by benefit withdrawal as earnings increased (Cribb et al., Reference Cribb, Giupponi, Joyce, Lindner, Waters, Wernham and Xu2021; Joyce & Ziliak, Reference Joyce and Ziliak2019). In this post-crisis context, especially for the later period, the more decisive factor was likely the impact of austerity reforms, which, as described above, likely had a substantial negative effect on long-term childhood poverty.

In terms of family dynamics, living in a single-parent household or in a large family (three or more children) is widely recognised as a major risk factor for child poverty, largely due to higher needs relative to earning potential, often linked to employment constraints such as care responsibilities (Stewart et al., Reference Stewart, Patrick and Reeves2025). While the number of single-parent households has increased in the UK, their effect on poverty has been limited (Brewer et al., Reference Brewer, Browne, Joyce and Sibieta2010), partly due to greater labour market participation among these families. Nevertheless, households with higher needs and lower earning potential are increasingly competing in a dual-earner society, where poverty thresholds largely reflect the income of two-earner households (Nieuwenhuis, Reference Nieuwenhuis2022), which in general might increase their risk of poverty. The proportion of households with three or more children fell during the 1990s and 2000s, likely limiting their aggregate impact. However, the poverty risk for large families fell sharply after the 1997 reforms but has risen substantially since 2013 (Stewart et al., Reference Stewart, Patrick and Reeves2025).

Bringing these together, the interplay between predistribution and redistribution is central to explaining long-term poverty trends. Many have argued that the impact of rising earnings inequality on poverty (and the resulting increase in the poverty threshold) was contained by the expansion of social transfers following the 1997 reforms, particularly in the 2000s (Brewer et al., Reference Brewer, Browne, Joyce and Sibieta2010; Cribb et al., Reference Cribb, Joyce and Wernham2023). In the 2010s, however, the rise in the minimum wage was insufficient to meaningfully strengthen predistribution, and the erosion of redistribution likely led to a rise in long-term childhood poverty for those who spent most of their early years during this period.

Analytical approach

To explain trends in long-term childhood poverty across cohorts, we apply two main approaches: (i) the prevalence-and-penalty framework using KOB decomposition (Brady et al., Reference Brady, Finnigan and Hübgen2017), and (ii) sequential income accounting (Caminada et al., Reference Caminada, Goudswaard, Wang and Wang2021).

The prevalence and penalty framework allows us to assess the role of work-family risk factors in explaining the variation in long-term poverty across cohorts. This method essentially compares poverty levels between two groups (e.g., a younger vs. an older cohort) and decomposes the difference into: (i) prevalence effects, differences in the share of population with a given risk between the two groups, and (ii) penalty effects, differences in the probability of poverty associated with that risk between the two groups. If cohort differences are mainly due to changes in the composition or levels of risk factors, the prevalence component will dominate; if they are due to changes in the extent to which risk groups are penalised, the penalty component will be larger. Penalty effects can be due to changes in earnings inequality and the poverty threshold (e.g., if it rises disproportionally relative to the earnings of risk groups) or the effects of taxes and transfers.Footnote ⁸ Following Brady et al. (Reference Brady, Finnigan and Hübgen2017), we include all risk factors in the model simultaneously to control for their overlap.

To directly assess the role predistribution vs. redistribution, we first apply the sequential accounting technique (Caminada et al., Reference Caminada, Goudswaard, Wang and Wang2021). We calculate poverty rate against the original poverty threshold (based on equivalised net household income) using different income definitions, starting with gross household income as the primary source, sequentially adding different income components. The differences in poverty rate between these definitions indicate the poverty-reducing effect of: (i) private transfers, (ii) social transfers, and (iii) taxes and social contributions (see Table 1).Footnote ⁹

Table 1.

Sequential income definitions and corresponding poverty-reducing effects

Notes: P₁ to P₄ represent the share of individuals with income below the poverty threshold (60 per cent of median net household income), which is fixed across all income definitions. Social transfers include both means-tested and non-means-tested transfers available in relevant periods. Private transfers include inter-house transfers, child maintenance, and investment income. Investment income is usually not considered as part of private transfers, but for simplicity we include them in this category, considering that their effect is usually limited for poverty reduction. Taxes also include social contributions, for example, those paid for the National Health Service.

While sequential income accounting is a common strategy to assess the role of redistribution in the field, it assumes that earnings are the primary income source and that all other components are added in a fixed order. Different ordering, for example, starting with social transfers and then adding earnings, can lead to different conclusions. To avoid this normative dependence on sequencing, we use a Shapley decomposition, which evaluates all possible orderings of income components and averages their marginal contributions (Shorrocks, Reference Shorrocks2013). We compare two cohorts and decompose the change in childhood poverty rates into the contributions of each income source, allowing us to assess the role of redistribution and predistribution in explaining cross-cohort trends.

Data and measures

We use data from the British Household Panel Survey (BHPS) and Understanding Society (UKHLS) for the period of 1991–2024 (University of Essex, 2024). The BHPS began in 1991 with 5,000 households in Great Britain and expanded to the UK in 2001 with a Northern Ireland sample; it was succeeded by UKHLS in 2009, which scaled up to around 40,000 households. The BHPS sample members have continued to participate as part of UKHLS (starting from the second wave). Combined, these two surveys provide over three decades of continuous, nationally representative longitudinal data. This anonymised data is made available from UK Data Service and secondary analysis of this data does not require ethics approval.Footnote ¹⁰

The analytic sample comprises children aged zero to ten, reflecting a life-course definition of childhood. Of the 33,864 children enumerated at least once across the thirty-two waves, several restrictions are applied to ensure temporal comparability and sufficient longitudinal information. We exclude the Northern Ireland sample because it was not observed in the first ten survey years. UKHLS children born before 2009 are also excluded to maintain a clear separation between the BHPS and UKHLS cohorts. We apply two conditions to select our sample. First, we remove children with five or fewer years of valid income observations, given our focus on long-term poverty experience.Footnote ¹¹ Second, we keep only those who have household income information at age zero or one as this information is crucial in defining birth cohorts and using survey weights. Finally, we exclude cases with missing data on risk factor variables (around 2 per cent). These restrictions yield a final sample of 6,218 children in Great Britain, born between 1991 and 2017.Footnote ¹²

We measure poverty using information on equivalised net household income. Net household income is calculated as the sum of the net monthly incomes of all household members, including net labor income, miscellaneous income, private benefits, investment income, pension income, and social benefits.Footnote ¹³ From this total, council tax liabilities are subtracted, and adjustments are made for housing benefits.Footnote ¹⁴ To account for differences in household size and composition, income is equivalised using the Organisation for Economic Co-operation and Development (OECD)-modified scale, which assigns a weight of 1.0 to the first adult, 0.5 to each additional adult, and 0.3 to each child. The household’s total net income is then divided by the sum of equivalence weights to produce the equivalised net household income. For the main analysis, we use a threshold of 60per cent of the median equivalised disposable household income, but also present results for alternative thresholds in the Appendix.Footnote ¹⁵

Long-term childhood poverty: Trends for 1991–2017 cohorts

Figure 1 shows the rate of long-term childhood poverty across birth cohorts from 1991–93 to 2016–17. On average, 17 per cent of children experienced long-term childhood poverty. This significantly varies across cohorts, revealing three distinct periods. First, there was a marked decline for cohorts born in 1990s, from 25 per cent for those born in 1991–93 to only 13–14 per cent for the 1998–99 cohort. A similar trend was also noted in previous research by Jenkins (Reference Jenkins2011). These children were among the first to benefit from 1997 reforms, which may have raised employment and social transfers for some in these groups. Indeed, as we show in Table A2 in the Appendix, poverty rates for all cohorts start to decline after 1999 when the Working Family Tax Credit replaced the Family Credit with a lower taper rate and more generous earnings disregards and rebates against childcare costs.

Figure 1.

Trends in long-term childhood poverty for 1991–2017 birth cohorts in Britain.

Notes: The figure shows the rate of long-term childhood poverty, the proportion of children living at least half of their childhood in poverty. It is based on a poverty definition using 60 per cent of median equivalised net household income as the threshold. The equivalisation is held using the modified OECD scale. The estimates are based on imputed data and using longitudinal weights. The shaded areas represent 95 per cent confidence intervals.

Second, long-term childhood poverty remained relatively stable among cohorts born in the 2000s. Although there were some fluctuations, overall levels were similar across these cohorts, despite rising unemployment and widening earnings inequality following the 2008 financial crisis. This stability in long-term rate masks important yearly variation: poverty rates for these groups increased in the years leading up to 2008 and then declined sharply afterward (see Table A2 in the Appendix), likely due to policy responses to the crisis, such as increases in benefit generosity through changes to uprating.

Finally, long-term childhood poverty has surged among cohorts born after 2012–13, reaching 23 per cent for those born in 2016–17. These children spent their entire childhoods under post-2013 austerity measures, and the increase likely reflects the erosion of the real value of benefits during this period, considering substantial rises in the minimum wage. Apart from a brief improvement during the COVID-19 response, which likely reflects the temporary £20 uplift, yearly poverty rates generally rise after 2013 and, in particular, after 2016 for the latest cohorts (see Table A2 in the Appendix).

We test the sensitivity of these findings to several methodological choices. The results are reported in the Appendix. First, we vary the threshold for defining long-term poverty, using one-third and three-quarters of childhood spent in poverty as alternative thresholds. The trends remain consistent, showing a marked decline for children born in the 1990s and a sharp increase for those born after 2013 (see Figure A2). We also confirm these trends using a broader measure of duration (i.e., the proportion of childhood spent in poverty) (see Figure A3).

Second, we test the impact of using alternative poverty thresholds. Instead of the 60 per cent of median equivalised household disposable income used in the main analysis, we estimate long-term childhood poverty based on 50 per cent and 70 per cent thresholds. As expected, the absolute levels vary, but the overall trends remain robust: declines for 1990s cohorts and substantial increases after 2013 (see Figure A4).

Third, we assess the impact of varying the number of years used to calculate long-term childhood poverty. Our main analysis uses data from the first eleven years of life, but for recent cohorts who have not yet reached age eleven, we use shorter observation windows, with a minimum of seven years for those born in 2017. These later cohorts also show the largest increases in long-term poverty. To test whether the shorter observation period influences our results, we re-estimate rates for all cohorts using only the first seven years of childhood. The overall levels and trends remain consistent with our main findings, except for a few fluctuations for certain cohorts (see Figure A5). This shows that using at least seven years of data already gives enough indication about the childhood poverty trends.

Finally, we explore the impact of housing costs. We compare long-term childhood poverty before and after accounting for housing costs, which include self-reported rent and mortgage payments. Until 2008, the poverty rates are similar across both measures, but post-2008, poverty rates after housing costs rise significantly and diverge from the before-housing-cost figures. The levels, however, converge again for the latest cohort. While this is not a definitive test as median incomes may also reflect rising housing costs, it suggests that accounting for changes in cost-of-living may alter conclusions (see Figure A6). We return to this point in the conclusion.

Explaining trends: Prevalence and penalty of risk factors

To understand the drivers of change in long-term childhood poverty, particularly for the earlier and later cohorts, we begin with a KOB decomposition. We first identify relevant risk factors by examining the profiles of children in long-term poverty. As shown in Appendix Figure A1, long-term poverty is significantly more common in single-parent households and households with three or more children, with disability, as well as with no employment, low pay, or low working hours. In contrast, being a single-earner household, having an informal carer in the household, living in private rental housing, or residing in London are not associated with a higher risk of poverty.Footnote ¹⁷ Therefore, we only use the significant risk factors to decompose the cross-cohort changes.

Figure 2 presents our KOB decomposition results for the decline in long-term childhood poverty among 1990s cohorts and the increase among the latest 2010s cohorts, focusing on whether these changes are driven by shifts in the prevalence of risk factors or in their penalties.

Figure 2.

KOB decomposition of cross-cohort changes in long-term childhood poverty.

Notes: The figure shows the results of a KOB decomposition applied to examine the sources of cross-cohort changes in long-term childhood poverty. The estimates are based on imputed data and using longitudinal weights. The error bars represent 95 per cent confidence intervals.

The left panel shows the decomposition of changes in long-term childhood poverty between cohorts born in 1991–93 and 1998–2001. The gap between these two groups shows an 11-percentage-point decline, with one-third explained by changes in the prevalence of risk factors and two-thirds by the changes in their penalties. The main contributors were the reduction in the prevalence of workless and low-pay households, likely reflecting rising employment rates during this period, and decreases in the penalties associated with living in a low-pay household. As discussed earlier, shifts in penalties may reflect changes in benefit generosity or rising earnings among at-risk groups relative to the poverty threshold (i.e., median income). The declining penalty for low pay suggests that these households may have experienced income gains, either through higher earnings or social benefits, that lifted later cohorts above the poverty line in at least some years.

The right panel presents a similar decomposition for the later period, focusing on the differences between cohorts born in 2010–11 and 2016–17. The sharp rise in long-term childhood poverty across these cohorts appears to be driven only by changes in the penalties associated with certain risk factors, particularly having 3+ kids, single parenthood, low pay, and low working hours. Given that this period saw increases in the minimum wage, the growing impact of these risk factors is likely explained by declining benefit levels due to austerity reforms, including the two-child limit. Rising penalties for single parents might also be related to stricter job-search conditions applied with 2008–12 reforms (Hoynes et al., 2024).

Explaining trends: sequential income accounting

While the KOB decomposition analysis helps explain whether changes in the prevalence and penalty of risk factors drive long-term childhood poverty trends, it does not directly assess the role of redistribution vis-à-vis predistribution. To address this, we first apply sequential income accounting, calculating poverty rates under different income definitions and attributing changes to specific income components specifically related to redistribution.

Figure 3 shows the poverty rate for different income definitions on the left and the extent to which long-term childhood poverty is reduced by social transfers, taxes and social transfers, and private transfers when they are added to gross household earnings on the right.

Figure 3.

The role of redistribution and predistribution in explaining trends in long-term childhood poverty for 1991–2017 birth cohorts in Britain: sequential income accounting.

Notes: The figure shows the results of sequential income accounting analysis to examine the role of redistribution and predistribution. All income definitions are equivalised for household size and composition. The estimates are based on imputed data and using longitudinal weights. The error bars represent 95 per cent confidence intervals.

We separately interpret three periods. First, the decline among those born in the 1990s seems largely due to improvements in predistribution, defined here as the ability to avoid poverty through household earnings alone. The dotted line for gross household earnings shows how this measure has evolved over time. The decline in long-term childhood poverty among 1990s cohorts coincides with a period of rising predistribution: poverty based on household earnings fell from 33 per cent for children born in 1991–93 to 25 per cent for those born in 1997–98. A KOB decomposition shows that more than half of this decline was due to reduced prevalence of workless, low-pay, and low-work households, while another half is explained by declining penalties for low-pay households, likely driven by rising employment and wage growth among these households relative to the median (see Figure A7 in the Appendix). Moreover, for the same groups, the reduction of poverty by social transfers seems to grow, but only slightly. Taxes seem to increase poverty for all cohorts, echoing recent findings on fiscal impoverishment (Schechtl & O’Brien, Reference Schechtl and O’Brien2024)Footnote ¹⁸ , yet this regressive effect also declines slightly for those born in the late 1990s compared to the earlier cohorts, possibly as a result of rising tax credits. The role of private transfers (difference between market income and gross household earnings) appears relatively negligible for these cohorts in this analysis.

Second, stable trends for those born in the 2000s are a result of counteracting effects of predistribution and redistribution. Predistribution trends reversed for children born in the 2000s. Long-term poverty based on earnings rose from 25 per cent for the 2000–01 cohort to 40 per cent for the 2008–09 cohort. Despite this deterioration in predistribution, overall long-term childhood poverty remained stable during this period, largely due to growing poverty reduction via social transfers. This suggests that growing reliance on benefits may have offset worsening predistribution conditions, such as rising earnings inequality, helping to maintain stable poverty levels. In the Appendix, we examine the drivers of declining predistribution (see Figure A7). The results point to rising prevalence of workless, low-pay, and low-work households, alongside rising penalties for single-earner, low-pay, and low-work hours households. These trends are likely linked to declining earnings for low-income households (mainly due to rising part-time work and declining work hours among men) against a rising poverty threshold as well as the rising earnings inequality in the aftermath of the 2008 financial crisis, which led to higher unemployment and increased earnings inequality (Cribb et al., Reference Cribb, Joyce and Wernham2023; Joyce & Ziliak, Reference Joyce and Ziliak2019).

Third, the surge in long-term childhood poverty for those born in the 2010s seems largely due to worsening redistribution, despite slightly improving predistribution. For children born in 2008–09, adding social transfers to gross household earnings leads to a 31 percentage points reduction in poverty, but this figure steadily declines to 15 percentage points (pp) for the 2016–17 cohort. The slight improvement in predistribution is likely due to significant rises in minimum wage, which became the main anti-poverty tool in this period, yet our findings show that its effectiveness was limited and did not translate into lower long-term childhood poverty, and the resurgence in long-term childhood poverty was most likely driven by cuts to social transfers. During this period, the poverty-reducing effect of private transfers and the poverty-increasing effect of taxes seem to increase slightly when they are evaluated after adding them to gross household earnings.

Explaining trends: Shapley decomposition

We now turn to the Shapley decomposition to explain the 1990s decline and the 2010s surge in long-term childhood poverty by assessing the contribution of different income sources across all possible order combinations. This method evaluates the average marginal contribution of each income component to observed changes in poverty trends. Figure 4 presents the results: the left panel compares the 1991–93 and 1998–99 cohorts (the 1990s decline), and the right panel compares the 2004–05 and 2016–17 cohorts (the 2010s surge).

Figure 4.

The role of redistribution and predistribution in explaining cross-cohort trends in long-term childhood poverty: Shapley decomposition.

Notes: The figure shows the results of Shapley decomposition of poverty rates between two cohorts described in subtitles. The analysis is held on pooled data, so reflect a decomposition of changes in average poverty rates during cohorts’ observation windows, rather than changes in long-term childhood poverty rate. The estimates are based on imputed data and using longitudinal weights.

The 1990s decline is mainly associated with rising predistribution. Consistent with earlier findings on increasing employment and lower penalties for low-pay households. Changes in household earnings alone seem to reduce long-term childhood poverty by around 11 percentage points. However, these gains were partly offset by higher taxes associated with increased earnings, which contributed around 6 percentage points in the opposite direction. Declining private transfers also had a small negative effect, pushing poverty upwards, while social transfers contributed modestly to the overall decline, reducing poverty by around 3 percentage points.

In contrast, the 2010s surge in long-term childhood poverty was primarily associated with falling redistribution. Consistent with our previous results, austerity measures substantially reduced the generosity of social benefits, and these changes in social transfers seem to explain roughly two-thirds of the increase in poverty among the later cohorts. Changes in household earnings also contributed to the rise, likely reflecting growing earnings inequality and the worsening position of low earners relative to the median, despite significant increases in the minimum wage. Taxes further added to the rise in poverty, likely due to the freeze in nominal tax thresholds, which gradually expanded the tax base. Finally, declines in private transfers seem to also modestly contribute to the increase.